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/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62 static int ext4_mballoc_ready;
  63
  64 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  65                              unsigned long journal_devnum);
  66 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  67 static int ext4_commit_super(struct super_block *sb, int sync);
  68 static void ext4_mark_recovery_complete(struct super_block *sb,
  69                                         struct ext4_super_block *es);
  70 static void ext4_clear_journal_err(struct super_block *sb,
  71                                    struct ext4_super_block *es);
  72 static int ext4_sync_fs(struct super_block *sb, int wait);
  73 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  75 static int ext4_unfreeze(struct super_block *sb);
  76 static int ext4_freeze(struct super_block *sb);
  77 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  78                        const char *dev_name, void *data);
  79 static inline int ext2_feature_set_ok(struct super_block *sb);
  80 static inline int ext3_feature_set_ok(struct super_block *sb);
  81 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 MODULE_ALIAS_FS("ext2");
  96 MODULE_ALIAS("ext2");
  97 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  98 #else
  99 #define IS_EXT2_SB(sb) (0)
 100 #endif
 101
 102
 103 static struct file_system_type ext3_fs_type = {
 104         .owner          = THIS_MODULE,
 105         .name           = "ext3",
 106         .mount          = ext4_mount,
 107         .kill_sb        = kill_block_super,
 108         .fs_flags       = FS_REQUIRES_DEV,
 109 };
 110 MODULE_ALIAS_FS("ext3");
 111 MODULE_ALIAS("ext3");
 112 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 113
 114 static int ext4_verify_csum_type(struct super_block *sb,
 115                                  struct ext4_super_block *es)
 116 {
 117         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 118                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 119                 return 1;
 120
 121         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 122 }
 123
 124 static __le32 ext4_superblock_csum(struct super_block *sb,
 125                                    struct ext4_super_block *es)
 126 {
 127         struct ext4_sb_info *sbi = EXT4_SB(sb);
 128         int offset = offsetof(struct ext4_super_block, s_checksum);
 129         __u32 csum;
 130
 131         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 132
 133         return cpu_to_le32(csum);
 134 }
 135
 136 static int ext4_superblock_csum_verify(struct super_block *sb,
 137                                        struct ext4_super_block *es)
 138 {
 139         if (!ext4_has_metadata_csum(sb))
 140                 return 1;
 141
 142         return es->s_checksum == ext4_superblock_csum(sb, es);
 143 }
 144
 145 void ext4_superblock_csum_set(struct super_block *sb)
 146 {
 147         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 148
 149         if (!ext4_has_metadata_csum(sb))
 150                 return;
 151
 152         es->s_checksum = ext4_superblock_csum(sb, es);
 153 }
 154
 155 void *ext4_kvmalloc(size_t size, gfp_t flags)
 156 {
 157         void *ret;
 158
 159         ret = kmalloc(size, flags | __GFP_NOWARN);
 160         if (!ret)
 161                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 162         return ret;
 163 }
 164
 165 void *ext4_kvzalloc(size_t size, gfp_t flags)
 166 {
 167         void *ret;
 168
 169         ret = kzalloc(size, flags | __GFP_NOWARN);
 170         if (!ret)
 171                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 172         return ret;
 173 }
 174
 175 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 176                                struct ext4_group_desc *bg)
 177 {
 178         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 179                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 180                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 181 }
 182
 183 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 184                                struct ext4_group_desc *bg)
 185 {
 186         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 187                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 188                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 189 }
 190
 191 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 192                               struct ext4_group_desc *bg)
 193 {
 194         return le32_to_cpu(bg->bg_inode_table_lo) |
 195                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 196                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 197 }
 198
 199 __u32 ext4_free_group_clusters(struct super_block *sb,
 200                                struct ext4_group_desc *bg)
 201 {
 202         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 203                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 204                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 205 }
 206
 207 __u32 ext4_free_inodes_count(struct super_block *sb,
 208                               struct ext4_group_desc *bg)
 209 {
 210         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 211                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 212                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 213 }
 214
 215 __u32 ext4_used_dirs_count(struct super_block *sb,
 216                               struct ext4_group_desc *bg)
 217 {
 218         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 219                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 220                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 221 }
 222
 223 __u32 ext4_itable_unused_count(struct super_block *sb,
 224                               struct ext4_group_desc *bg)
 225 {
 226         return le16_to_cpu(bg->bg_itable_unused_lo) |
 227                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 228                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 229 }
 230
 231 void ext4_block_bitmap_set(struct super_block *sb,
 232                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 233 {
 234         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 235         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 236                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 237 }
 238
 239 void ext4_inode_bitmap_set(struct super_block *sb,
 240                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 241 {
 242         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 243         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 244                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 245 }
 246
 247 void ext4_inode_table_set(struct super_block *sb,
 248                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 249 {
 250         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 251         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 252                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 253 }
 254
 255 void ext4_free_group_clusters_set(struct super_block *sb,
 256                                   struct ext4_group_desc *bg, __u32 count)
 257 {
 258         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 259         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 260                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 261 }
 262
 263 void ext4_free_inodes_set(struct super_block *sb,
 264                           struct ext4_group_desc *bg, __u32 count)
 265 {
 266         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 267         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 268                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 269 }
 270
 271 void ext4_used_dirs_set(struct super_block *sb,
 272                           struct ext4_group_desc *bg, __u32 count)
 273 {
 274         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 275         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 276                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 277 }
 278
 279 void ext4_itable_unused_set(struct super_block *sb,
 280                           struct ext4_group_desc *bg, __u32 count)
 281 {
 282         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 283         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 284                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 285 }
 286
 287
 288 static void __save_error_info(struct super_block *sb, const char *func,
 289                             unsigned int line)
 290 {
 291         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 292
 293         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 294         if (bdev_read_only(sb->s_bdev))
 295                 return;
 296         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 297         es->s_last_error_time = cpu_to_le32(get_seconds());
 298         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 299         es->s_last_error_line = cpu_to_le32(line);
 300         if (!es->s_first_error_time) {
 301                 es->s_first_error_time = es->s_last_error_time;
 302                 strncpy(es->s_first_error_func, func,
 303                         sizeof(es->s_first_error_func));
 304                 es->s_first_error_line = cpu_to_le32(line);
 305                 es->s_first_error_ino = es->s_last_error_ino;
 306                 es->s_first_error_block = es->s_last_error_block;
 307         }
 308         /*
 309          * Start the daily error reporting function if it hasn't been
 310          * started already
 311          */
 312         if (!es->s_error_count)
 313                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 314         le32_add_cpu(&es->s_error_count, 1);
 315 }
 316
 317 static void save_error_info(struct super_block *sb, const char *func,
 318                             unsigned int line)
 319 {
 320         __save_error_info(sb, func, line);
 321         ext4_commit_super(sb, 1);
 322 }
 323
 324 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 325 {
 326         struct super_block              *sb = journal->j_private;
 327         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 328         int                             error = is_journal_aborted(journal);
 329         struct ext4_journal_cb_entry    *jce;
 330
 331         BUG_ON(txn->t_state == T_FINISHED);
 332         spin_lock(&sbi->s_md_lock);
 333         while (!list_empty(&txn->t_private_list)) {
 334                 jce = list_entry(txn->t_private_list.next,
 335                                  struct ext4_journal_cb_entry, jce_list);
 336                 list_del_init(&jce->jce_list);
 337                 spin_unlock(&sbi->s_md_lock);
 338                 jce->jce_func(sb, jce, error);
 339                 spin_lock(&sbi->s_md_lock);
 340         }
 341         spin_unlock(&sbi->s_md_lock);
 342 }
 343
 344 /* Deal with the reporting of failure conditions on a filesystem such as
 345  * inconsistencies detected or read IO failures.
 346  *
 347  * On ext2, we can store the error state of the filesystem in the
 348  * superblock.  That is not possible on ext4, because we may have other
 349  * write ordering constraints on the superblock which prevent us from
 350  * writing it out straight away; and given that the journal is about to
 351  * be aborted, we can't rely on the current, or future, transactions to
 352  * write out the superblock safely.
 353  *
 354  * We'll just use the jbd2_journal_abort() error code to record an error in
 355  * the journal instead.  On recovery, the journal will complain about
 356  * that error until we've noted it down and cleared it.
 357  */
 358
 359 static void ext4_handle_error(struct super_block *sb)
 360 {
 361         if (sb->s_flags & MS_RDONLY)
 362                 return;
 363
 364         if (!test_opt(sb, ERRORS_CONT)) {
 365                 journal_t *journal = EXT4_SB(sb)->s_journal;
 366
 367                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 368                 if (journal)
 369                         jbd2_journal_abort(journal, -EIO);
 370         }
 371         if (test_opt(sb, ERRORS_RO)) {
 372                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 373                 /*
 374                  * Make sure updated value of ->s_mount_flags will be visible
 375                  * before ->s_flags update
 376                  */
 377                 smp_wmb();
 378                 sb->s_flags |= MS_RDONLY;
 379         }
 380         if (test_opt(sb, ERRORS_PANIC))
 381                 panic("EXT4-fs (device %s): panic forced after error\n",
 382                         sb->s_id);
 383 }
 384
 385 #define ext4_error_ratelimit(sb)                                        \
 386                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 387                              "EXT4-fs error")
 388
 389 void __ext4_error(struct super_block *sb, const char *function,
 390                   unsigned int line, const char *fmt, ...)
 391 {
 392         struct va_format vaf;
 393         va_list args;
 394
 395         if (ext4_error_ratelimit(sb)) {
 396                 va_start(args, fmt);
 397                 vaf.fmt = fmt;
 398                 vaf.va = &args;
 399                 printk(KERN_CRIT
 400                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 401                        sb->s_id, function, line, current->comm, &vaf);
 402                 va_end(args);
 403         }
 404         save_error_info(sb, function, line);
 405         ext4_handle_error(sb);
 406 }
 407
 408 void __ext4_error_inode(struct inode *inode, const char *function,
 409                         unsigned int line, ext4_fsblk_t block,
 410                         const char *fmt, ...)
 411 {
 412         va_list args;
 413         struct va_format vaf;
 414         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 415
 416         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 417         es->s_last_error_block = cpu_to_le64(block);
 418         if (ext4_error_ratelimit(inode->i_sb)) {
 419                 va_start(args, fmt);
 420                 vaf.fmt = fmt;
 421                 vaf.va = &args;
 422                 if (block)
 423                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 424                                "inode #%lu: block %llu: comm %s: %pV\n",
 425                                inode->i_sb->s_id, function, line, inode->i_ino,
 426                                block, current->comm, &vaf);
 427                 else
 428                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 429                                "inode #%lu: comm %s: %pV\n",
 430                                inode->i_sb->s_id, function, line, inode->i_ino,
 431                                current->comm, &vaf);
 432                 va_end(args);
 433         }
 434         save_error_info(inode->i_sb, function, line);
 435         ext4_handle_error(inode->i_sb);
 436 }
 437
 438 void __ext4_error_file(struct file *file, const char *function,
 439                        unsigned int line, ext4_fsblk_t block,
 440                        const char *fmt, ...)
 441 {
 442         va_list args;
 443         struct va_format vaf;
 444         struct ext4_super_block *es;
 445         struct inode *inode = file_inode(file);
 446         char pathname[80], *path;
 447
 448         es = EXT4_SB(inode->i_sb)->s_es;
 449         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 450         if (ext4_error_ratelimit(inode->i_sb)) {
 451                 path = file_path(file, pathname, sizeof(pathname));
 452                 if (IS_ERR(path))
 453                         path = "(unknown)";
 454                 va_start(args, fmt);
 455                 vaf.fmt = fmt;
 456                 vaf.va = &args;
 457                 if (block)
 458                         printk(KERN_CRIT
 459                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 460                                "block %llu: comm %s: path %s: %pV\n",
 461                                inode->i_sb->s_id, function, line, inode->i_ino,
 462                                block, current->comm, path, &vaf);
 463                 else
 464                         printk(KERN_CRIT
 465                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 466                                "comm %s: path %s: %pV\n",
 467                                inode->i_sb->s_id, function, line, inode->i_ino,
 468                                current->comm, path, &vaf);
 469                 va_end(args);
 470         }
 471         save_error_info(inode->i_sb, function, line);
 472         ext4_handle_error(inode->i_sb);
 473 }
 474
 475 const char *ext4_decode_error(struct super_block *sb, int errno,
 476                               char nbuf[16])
 477 {
 478         char *errstr = NULL;
 479
 480         switch (errno) {
 481         case -EIO:
 482                 errstr = "IO failure";
 483                 break;
 484         case -ENOMEM:
 485                 errstr = "Out of memory";
 486                 break;
 487         case -EROFS:
 488                 if (!sb || (EXT4_SB(sb)->s_journal &&
 489                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 490                         errstr = "Journal has aborted";
 491                 else
 492                         errstr = "Readonly filesystem";
 493                 break;
 494         default:
 495                 /* If the caller passed in an extra buffer for unknown
 496                  * errors, textualise them now.  Else we just return
 497                  * NULL. */
 498                 if (nbuf) {
 499                         /* Check for truncated error codes... */
 500                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 501                                 errstr = nbuf;
 502                 }
 503                 break;
 504         }
 505
 506         return errstr;
 507 }
 508
 509 /* __ext4_std_error decodes expected errors from journaling functions
 510  * automatically and invokes the appropriate error response.  */
 511
 512 void __ext4_std_error(struct super_block *sb, const char *function,
 513                       unsigned int line, int errno)
 514 {
 515         char nbuf[16];
 516         const char *errstr;
 517
 518         /* Special case: if the error is EROFS, and we're not already
 519          * inside a transaction, then there's really no point in logging
 520          * an error. */
 521         if (errno == -EROFS && journal_current_handle() == NULL &&
 522             (sb->s_flags & MS_RDONLY))
 523                 return;
 524
 525         if (ext4_error_ratelimit(sb)) {
 526                 errstr = ext4_decode_error(sb, errno, nbuf);
 527                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 528                        sb->s_id, function, line, errstr);
 529         }
 530
 531         save_error_info(sb, function, line);
 532         ext4_handle_error(sb);
 533 }
 534
 535 /*
 536  * ext4_abort is a much stronger failure handler than ext4_error.  The
 537  * abort function may be used to deal with unrecoverable failures such
 538  * as journal IO errors or ENOMEM at a critical moment in log management.
 539  *
 540  * We unconditionally force the filesystem into an ABORT|READONLY state,
 541  * unless the error response on the fs has been set to panic in which
 542  * case we take the easy way out and panic immediately.
 543  */
 544
 545 void __ext4_abort(struct super_block *sb, const char *function,
 546                 unsigned int line, const char *fmt, ...)
 547 {
 548         va_list args;
 549
 550         save_error_info(sb, function, line);
 551         va_start(args, fmt);
 552         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 553                function, line);
 554         vprintk(fmt, args);
 555         printk("\n");
 556         va_end(args);
 557
 558         if ((sb->s_flags & MS_RDONLY) == 0) {
 559                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 560                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 561                 /*
 562                  * Make sure updated value of ->s_mount_flags will be visible
 563                  * before ->s_flags update
 564                  */
 565                 smp_wmb();
 566                 sb->s_flags |= MS_RDONLY;
 567                 if (EXT4_SB(sb)->s_journal)
 568                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 569                 save_error_info(sb, function, line);
 570         }
 571         if (test_opt(sb, ERRORS_PANIC))
 572                 panic("EXT4-fs panic from previous error\n");
 573 }
 574
 575 void __ext4_msg(struct super_block *sb,
 576                 const char *prefix, const char *fmt, ...)
 577 {
 578         struct va_format vaf;
 579         va_list args;
 580
 581         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 582                 return;
 583
 584         va_start(args, fmt);
 585         vaf.fmt = fmt;
 586         vaf.va = &args;
 587         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 588         va_end(args);
 589 }
 590
 591 #define ext4_warning_ratelimit(sb)                                      \
 592                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 593                              "EXT4-fs warning")
 594
 595 void __ext4_warning(struct super_block *sb, const char *function,
 596                     unsigned int line, const char *fmt, ...)
 597 {
 598         struct va_format vaf;
 599         va_list args;
 600
 601         if (!ext4_warning_ratelimit(sb))
 602                 return;
 603
 604         va_start(args, fmt);
 605         vaf.fmt = fmt;
 606         vaf.va = &args;
 607         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 608                sb->s_id, function, line, &vaf);
 609         va_end(args);
 610 }
 611
 612 void __ext4_warning_inode(const struct inode *inode, const char *function,
 613                           unsigned int line, const char *fmt, ...)
 614 {
 615         struct va_format vaf;
 616         va_list args;
 617
 618         if (!ext4_warning_ratelimit(inode->i_sb))
 619                 return;
 620
 621         va_start(args, fmt);
 622         vaf.fmt = fmt;
 623         vaf.va = &args;
 624         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 625                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 626                function, line, inode->i_ino, current->comm, &vaf);
 627         va_end(args);
 628 }
 629
 630 void __ext4_grp_locked_error(const char *function, unsigned int line,
 631                              struct super_block *sb, ext4_group_t grp,
 632                              unsigned long ino, ext4_fsblk_t block,
 633                              const char *fmt, ...)
 634 __releases(bitlock)
 635 __acquires(bitlock)
 636 {
 637         struct va_format vaf;
 638         va_list args;
 639         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 640
 641         es->s_last_error_ino = cpu_to_le32(ino);
 642         es->s_last_error_block = cpu_to_le64(block);
 643         __save_error_info(sb, function, line);
 644
 645         if (ext4_error_ratelimit(sb)) {
 646                 va_start(args, fmt);
 647                 vaf.fmt = fmt;
 648                 vaf.va = &args;
 649                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 650                        sb->s_id, function, line, grp);
 651                 if (ino)
 652                         printk(KERN_CONT "inode %lu: ", ino);
 653                 if (block)
 654                         printk(KERN_CONT "block %llu:",
 655                                (unsigned long long) block);
 656                 printk(KERN_CONT "%pV\n", &vaf);
 657                 va_end(args);
 658         }
 659
 660         if (test_opt(sb, ERRORS_CONT)) {
 661                 ext4_commit_super(sb, 0);
 662                 return;
 663         }
 664
 665         ext4_unlock_group(sb, grp);
 666         ext4_handle_error(sb);
 667         /*
 668          * We only get here in the ERRORS_RO case; relocking the group
 669          * may be dangerous, but nothing bad will happen since the
 670          * filesystem will have already been marked read/only and the
 671          * journal has been aborted.  We return 1 as a hint to callers
 672          * who might what to use the return value from
 673          * ext4_grp_locked_error() to distinguish between the
 674          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 675          * aggressively from the ext4 function in question, with a
 676          * more appropriate error code.
 677          */
 678         ext4_lock_group(sb, grp);
 679         return;
 680 }
 681
 682 void ext4_update_dynamic_rev(struct super_block *sb)
 683 {
 684         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 685
 686         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 687                 return;
 688
 689         ext4_warning(sb,
 690                      "updating to rev %d because of new feature flag, "
 691                      "running e2fsck is recommended",
 692                      EXT4_DYNAMIC_REV);
 693
 694         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 695         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 696         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 697         /* leave es->s_feature_*compat flags alone */
 698         /* es->s_uuid will be set by e2fsck if empty */
 699
 700         /*
 701          * The rest of the superblock fields should be zero, and if not it
 702          * means they are likely already in use, so leave them alone.  We
 703          * can leave it up to e2fsck to clean up any inconsistencies there.
 704          */
 705 }
 706
 707 /*
 708  * Open the external journal device
 709  */
 710 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 711 {
 712         struct block_device *bdev;
 713         char b[BDEVNAME_SIZE];
 714
 715         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 716         if (IS_ERR(bdev))
 717                 goto fail;
 718         return bdev;
 719
 720 fail:
 721         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 722                         __bdevname(dev, b), PTR_ERR(bdev));
 723         return NULL;
 724 }
 725
 726 /*
 727  * Release the journal device
 728  */
 729 static void ext4_blkdev_put(struct block_device *bdev)
 730 {
 731         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 732 }
 733
 734 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 735 {
 736         struct block_device *bdev;
 737         bdev = sbi->journal_bdev;
 738         if (bdev) {
 739                 ext4_blkdev_put(bdev);
 740                 sbi->journal_bdev = NULL;
 741         }
 742 }
 743
 744 static inline struct inode *orphan_list_entry(struct list_head *l)
 745 {
 746         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 747 }
 748
 749 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 750 {
 751         struct list_head *l;
 752
 753         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 754                  le32_to_cpu(sbi->s_es->s_last_orphan));
 755
 756         printk(KERN_ERR "sb_info orphan list:\n");
 757         list_for_each(l, &sbi->s_orphan) {
 758                 struct inode *inode = orphan_list_entry(l);
 759                 printk(KERN_ERR "  "
 760                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 761                        inode->i_sb->s_id, inode->i_ino, inode,
 762                        inode->i_mode, inode->i_nlink,
 763                        NEXT_ORPHAN(inode));
 764         }
 765 }
 766
 767 static void ext4_put_super(struct super_block *sb)
 768 {
 769         struct ext4_sb_info *sbi = EXT4_SB(sb);
 770         struct ext4_super_block *es = sbi->s_es;
 771         int i, err;
 772
 773         ext4_unregister_li_request(sb);
 774         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 775
 776         flush_workqueue(sbi->rsv_conversion_wq);
 777         destroy_workqueue(sbi->rsv_conversion_wq);
 778
 779         if (sbi->s_journal) {
 780                 err = jbd2_journal_destroy(sbi->s_journal);
 781                 sbi->s_journal = NULL;
 782                 if (err < 0)
 783                         ext4_abort(sb, "Couldn't clean up the journal");
 784         }
 785
 786         ext4_es_unregister_shrinker(sbi);
 787         del_timer_sync(&sbi->s_err_report);
 788         ext4_release_system_zone(sb);
 789         ext4_mb_release(sb);
 790         ext4_ext_release(sb);
 791         ext4_xattr_put_super(sb);
 792
 793         if (!(sb->s_flags & MS_RDONLY)) {
 794                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 795                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 796         }
 797         if (!(sb->s_flags & MS_RDONLY))
 798                 ext4_commit_super(sb, 1);
 799
 800         if (sbi->s_proc) {
 801                 remove_proc_entry("options", sbi->s_proc);
 802                 remove_proc_entry(sb->s_id, ext4_proc_root);
 803         }
 804         kobject_del(&sbi->s_kobj);
 805
 806         for (i = 0; i < sbi->s_gdb_count; i++)
 807                 brelse(sbi->s_group_desc[i]);
 808         kvfree(sbi->s_group_desc);
 809         kvfree(sbi->s_flex_groups);
 810         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 811         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 812         percpu_counter_destroy(&sbi->s_dirs_counter);
 813         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 814         brelse(sbi->s_sbh);
 815 #ifdef CONFIG_QUOTA
 816         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 817                 kfree(sbi->s_qf_names[i]);
 818 #endif
 819
 820         /* Debugging code just in case the in-memory inode orphan list
 821          * isn't empty.  The on-disk one can be non-empty if we've
 822          * detected an error and taken the fs readonly, but the
 823          * in-memory list had better be clean by this point. */
 824         if (!list_empty(&sbi->s_orphan))
 825                 dump_orphan_list(sb, sbi);
 826         J_ASSERT(list_empty(&sbi->s_orphan));
 827
 828         sync_blockdev(sb->s_bdev);
 829         invalidate_bdev(sb->s_bdev);
 830         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 831                 /*
 832                  * Invalidate the journal device's buffers.  We don't want them
 833                  * floating about in memory - the physical journal device may
 834                  * hotswapped, and it breaks the `ro-after' testing code.
 835                  */
 836                 sync_blockdev(sbi->journal_bdev);
 837                 invalidate_bdev(sbi->journal_bdev);
 838                 ext4_blkdev_remove(sbi);
 839         }
 840         if (sbi->s_mb_cache) {
 841                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 842                 sbi->s_mb_cache = NULL;
 843         }
 844         if (sbi->s_mmp_tsk)
 845                 kthread_stop(sbi->s_mmp_tsk);
 846         sb->s_fs_info = NULL;
 847         /*
 848          * Now that we are completely done shutting down the
 849          * superblock, we need to actually destroy the kobject.
 850          */
 851         kobject_put(&sbi->s_kobj);
 852         wait_for_completion(&sbi->s_kobj_unregister);
 853         if (sbi->s_chksum_driver)
 854                 crypto_free_shash(sbi->s_chksum_driver);
 855         kfree(sbi->s_blockgroup_lock);
 856         kfree(sbi);
 857 }
 858
 859 static struct kmem_cache *ext4_inode_cachep;
 860
 861 /*
 862  * Called inside transaction, so use GFP_NOFS
 863  */
 864 static struct inode *ext4_alloc_inode(struct super_block *sb)
 865 {
 866         struct ext4_inode_info *ei;
 867
 868         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 869         if (!ei)
 870                 return NULL;
 871
 872         ei->vfs_inode.i_version = 1;
 873         spin_lock_init(&ei->i_raw_lock);
 874         INIT_LIST_HEAD(&ei->i_prealloc_list);
 875         spin_lock_init(&ei->i_prealloc_lock);
 876         ext4_es_init_tree(&ei->i_es_tree);
 877         rwlock_init(&ei->i_es_lock);
 878         INIT_LIST_HEAD(&ei->i_es_list);
 879         ei->i_es_all_nr = 0;
 880         ei->i_es_shk_nr = 0;
 881         ei->i_es_shrink_lblk = 0;
 882         ei->i_reserved_data_blocks = 0;
 883         ei->i_reserved_meta_blocks = 0;
 884         ei->i_allocated_meta_blocks = 0;
 885         ei->i_da_metadata_calc_len = 0;
 886         ei->i_da_metadata_calc_last_lblock = 0;
 887         spin_lock_init(&(ei->i_block_reservation_lock));
 888 #ifdef CONFIG_QUOTA
 889         ei->i_reserved_quota = 0;
 890         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 891 #endif
 892         ei->jinode = NULL;
 893         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 894         spin_lock_init(&ei->i_completed_io_lock);
 895         ei->i_sync_tid = 0;
 896         ei->i_datasync_tid = 0;
 897         atomic_set(&ei->i_ioend_count, 0);
 898         atomic_set(&ei->i_unwritten, 0);
 899         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 900 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 901         ei->i_crypt_info = NULL;
 902 #endif
 903         return &ei->vfs_inode;
 904 }
 905
 906 static int ext4_drop_inode(struct inode *inode)
 907 {
 908         int drop = generic_drop_inode(inode);
 909
 910         trace_ext4_drop_inode(inode, drop);
 911         return drop;
 912 }
 913
 914 static void ext4_i_callback(struct rcu_head *head)
 915 {
 916         struct inode *inode = container_of(head, struct inode, i_rcu);
 917         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 918 }
 919
 920 static void ext4_destroy_inode(struct inode *inode)
 921 {
 922         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 923                 ext4_msg(inode->i_sb, KERN_ERR,
 924                          "Inode %lu (%p): orphan list check failed!",
 925                          inode->i_ino, EXT4_I(inode));
 926                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 927                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 928                                 true);
 929                 dump_stack();
 930         }
 931         call_rcu(&inode->i_rcu, ext4_i_callback);
 932 }
 933
 934 static void init_once(void *foo)
 935 {
 936         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 937
 938         INIT_LIST_HEAD(&ei->i_orphan);
 939         init_rwsem(&ei->xattr_sem);
 940         init_rwsem(&ei->i_data_sem);
 941         inode_init_once(&ei->vfs_inode);
 942 }
 943
 944 static int __init init_inodecache(void)
 945 {
 946         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 947                                              sizeof(struct ext4_inode_info),
 948                                              0, (SLAB_RECLAIM_ACCOUNT|
 949                                                 SLAB_MEM_SPREAD),
 950                                              init_once);
 951         if (ext4_inode_cachep == NULL)
 952                 return -ENOMEM;
 953         return 0;
 954 }
 955
 956 static void destroy_inodecache(void)
 957 {
 958         /*
 959          * Make sure all delayed rcu free inodes are flushed before we
 960          * destroy cache.
 961          */
 962         rcu_barrier();
 963         kmem_cache_destroy(ext4_inode_cachep);
 964 }
 965
 966 void ext4_clear_inode(struct inode *inode)
 967 {
 968         invalidate_inode_buffers(inode);
 969         clear_inode(inode);
 970         dquot_drop(inode);
 971         ext4_discard_preallocations(inode);
 972         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 973         if (EXT4_I(inode)->jinode) {
 974                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 975                                                EXT4_I(inode)->jinode);
 976                 jbd2_free_inode(EXT4_I(inode)->jinode);
 977                 EXT4_I(inode)->jinode = NULL;
 978         }
 979 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 980         if (EXT4_I(inode)->i_crypt_info)
 981                 ext4_free_encryption_info(inode, EXT4_I(inode)->i_crypt_info);
 982 #endif
 983 }
 984
 985 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 986                                         u64 ino, u32 generation)
 987 {
 988         struct inode *inode;
 989
 990         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 991                 return ERR_PTR(-ESTALE);
 992         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 993                 return ERR_PTR(-ESTALE);
 994
 995         /* iget isn't really right if the inode is currently unallocated!!
 996          *
 997          * ext4_read_inode will return a bad_inode if the inode had been
 998          * deleted, so we should be safe.
 999          *
1000          * Currently we don't know the generation for parent directory, so
1001          * a generation of 0 means "accept any"
1002          */
1003         inode = ext4_iget_normal(sb, ino);
1004         if (IS_ERR(inode))
1005                 return ERR_CAST(inode);
1006         if (generation && inode->i_generation != generation) {
1007                 iput(inode);
1008                 return ERR_PTR(-ESTALE);
1009         }
1010
1011         return inode;
1012 }
1013
1014 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1015                                         int fh_len, int fh_type)
1016 {
1017         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1018                                     ext4_nfs_get_inode);
1019 }
1020
1021 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1022                                         int fh_len, int fh_type)
1023 {
1024         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1025                                     ext4_nfs_get_inode);
1026 }
1027
1028 /*
1029  * Try to release metadata pages (indirect blocks, directories) which are
1030  * mapped via the block device.  Since these pages could have journal heads
1031  * which would prevent try_to_free_buffers() from freeing them, we must use
1032  * jbd2 layer's try_to_free_buffers() function to release them.
1033  */
1034 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1035                                  gfp_t wait)
1036 {
1037         journal_t *journal = EXT4_SB(sb)->s_journal;
1038
1039         WARN_ON(PageChecked(page));
1040         if (!page_has_buffers(page))
1041                 return 0;
1042         if (journal)
1043                 return jbd2_journal_try_to_free_buffers(journal, page,
1044                                                         wait & ~__GFP_WAIT);
1045         return try_to_free_buffers(page);
1046 }
1047
1048 #ifdef CONFIG_QUOTA
1049 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1050 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1051
1052 static int ext4_write_dquot(struct dquot *dquot);
1053 static int ext4_acquire_dquot(struct dquot *dquot);
1054 static int ext4_release_dquot(struct dquot *dquot);
1055 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1056 static int ext4_write_info(struct super_block *sb, int type);
1057 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1058                          struct path *path);
1059 static int ext4_quota_off(struct super_block *sb, int type);
1060 static int ext4_quota_on_mount(struct super_block *sb, int type);
1061 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1062                                size_t len, loff_t off);
1063 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1064                                 const char *data, size_t len, loff_t off);
1065 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1066                              unsigned int flags);
1067 static int ext4_enable_quotas(struct super_block *sb);
1068
1069 static struct dquot **ext4_get_dquots(struct inode *inode)
1070 {
1071         return EXT4_I(inode)->i_dquot;
1072 }
1073
1074 static const struct dquot_operations ext4_quota_operations = {
1075         .get_reserved_space = ext4_get_reserved_space,
1076         .write_dquot    = ext4_write_dquot,
1077         .acquire_dquot  = ext4_acquire_dquot,
1078         .release_dquot  = ext4_release_dquot,
1079         .mark_dirty     = ext4_mark_dquot_dirty,
1080         .write_info     = ext4_write_info,
1081         .alloc_dquot    = dquot_alloc,
1082         .destroy_dquot  = dquot_destroy,
1083 };
1084
1085 static const struct quotactl_ops ext4_qctl_operations = {
1086         .quota_on       = ext4_quota_on,
1087         .quota_off      = ext4_quota_off,
1088         .quota_sync     = dquot_quota_sync,
1089         .get_state      = dquot_get_state,
1090         .set_info       = dquot_set_dqinfo,
1091         .get_dqblk      = dquot_get_dqblk,
1092         .set_dqblk      = dquot_set_dqblk
1093 };
1094 #endif
1095
1096 static const struct super_operations ext4_sops = {
1097         .alloc_inode    = ext4_alloc_inode,
1098         .destroy_inode  = ext4_destroy_inode,
1099         .write_inode    = ext4_write_inode,
1100         .dirty_inode    = ext4_dirty_inode,
1101         .drop_inode     = ext4_drop_inode,
1102         .evict_inode    = ext4_evict_inode,
1103         .put_super      = ext4_put_super,
1104         .sync_fs        = ext4_sync_fs,
1105         .freeze_fs      = ext4_freeze,
1106         .unfreeze_fs    = ext4_unfreeze,
1107         .statfs         = ext4_statfs,
1108         .remount_fs     = ext4_remount,
1109         .show_options   = ext4_show_options,
1110 #ifdef CONFIG_QUOTA
1111         .quota_read     = ext4_quota_read,
1112         .quota_write    = ext4_quota_write,
1113         .get_dquots     = ext4_get_dquots,
1114 #endif
1115         .bdev_try_to_free_page = bdev_try_to_free_page,
1116 };
1117
1118 static const struct export_operations ext4_export_ops = {
1119         .fh_to_dentry = ext4_fh_to_dentry,
1120         .fh_to_parent = ext4_fh_to_parent,
1121         .get_parent = ext4_get_parent,
1122 };
1123
1124 enum {
1125         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1126         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1127         Opt_nouid32, Opt_debug, Opt_removed,
1128         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1129         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1130         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1131         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1132         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1133         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1134         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1135         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1136         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1137         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1138         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1139         Opt_lazytime, Opt_nolazytime,
1140         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1141         Opt_inode_readahead_blks, Opt_journal_ioprio,
1142         Opt_dioread_nolock, Opt_dioread_lock,
1143         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1144         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1145 };
1146
1147 static const match_table_t tokens = {
1148         {Opt_bsd_df, "bsddf"},
1149         {Opt_minix_df, "minixdf"},
1150         {Opt_grpid, "grpid"},
1151         {Opt_grpid, "bsdgroups"},
1152         {Opt_nogrpid, "nogrpid"},
1153         {Opt_nogrpid, "sysvgroups"},
1154         {Opt_resgid, "resgid=%u"},
1155         {Opt_resuid, "resuid=%u"},
1156         {Opt_sb, "sb=%u"},
1157         {Opt_err_cont, "errors=continue"},
1158         {Opt_err_panic, "errors=panic"},
1159         {Opt_err_ro, "errors=remount-ro"},
1160         {Opt_nouid32, "nouid32"},
1161         {Opt_debug, "debug"},
1162         {Opt_removed, "oldalloc"},
1163         {Opt_removed, "orlov"},
1164         {Opt_user_xattr, "user_xattr"},
1165         {Opt_nouser_xattr, "nouser_xattr"},
1166         {Opt_acl, "acl"},
1167         {Opt_noacl, "noacl"},
1168         {Opt_noload, "norecovery"},
1169         {Opt_noload, "noload"},
1170         {Opt_removed, "nobh"},
1171         {Opt_removed, "bh"},
1172         {Opt_commit, "commit=%u"},
1173         {Opt_min_batch_time, "min_batch_time=%u"},
1174         {Opt_max_batch_time, "max_batch_time=%u"},
1175         {Opt_journal_dev, "journal_dev=%u"},
1176         {Opt_journal_path, "journal_path=%s"},
1177         {Opt_journal_checksum, "journal_checksum"},
1178         {Opt_nojournal_checksum, "nojournal_checksum"},
1179         {Opt_journal_async_commit, "journal_async_commit"},
1180         {Opt_abort, "abort"},
1181         {Opt_data_journal, "data=journal"},
1182         {Opt_data_ordered, "data=ordered"},
1183         {Opt_data_writeback, "data=writeback"},
1184         {Opt_data_err_abort, "data_err=abort"},
1185         {Opt_data_err_ignore, "data_err=ignore"},
1186         {Opt_offusrjquota, "usrjquota="},
1187         {Opt_usrjquota, "usrjquota=%s"},
1188         {Opt_offgrpjquota, "grpjquota="},
1189         {Opt_grpjquota, "grpjquota=%s"},
1190         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1191         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1192         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1193         {Opt_grpquota, "grpquota"},
1194         {Opt_noquota, "noquota"},
1195         {Opt_quota, "quota"},
1196         {Opt_usrquota, "usrquota"},
1197         {Opt_barrier, "barrier=%u"},
1198         {Opt_barrier, "barrier"},
1199         {Opt_nobarrier, "nobarrier"},
1200         {Opt_i_version, "i_version"},
1201         {Opt_dax, "dax"},
1202         {Opt_stripe, "stripe=%u"},
1203         {Opt_delalloc, "delalloc"},
1204         {Opt_lazytime, "lazytime"},
1205         {Opt_nolazytime, "nolazytime"},
1206         {Opt_nodelalloc, "nodelalloc"},
1207         {Opt_removed, "mblk_io_submit"},
1208         {Opt_removed, "nomblk_io_submit"},
1209         {Opt_block_validity, "block_validity"},
1210         {Opt_noblock_validity, "noblock_validity"},
1211         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1212         {Opt_journal_ioprio, "journal_ioprio=%u"},
1213         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1214         {Opt_auto_da_alloc, "auto_da_alloc"},
1215         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1216         {Opt_dioread_nolock, "dioread_nolock"},
1217         {Opt_dioread_lock, "dioread_lock"},
1218         {Opt_discard, "discard"},
1219         {Opt_nodiscard, "nodiscard"},
1220         {Opt_init_itable, "init_itable=%u"},
1221         {Opt_init_itable, "init_itable"},
1222         {Opt_noinit_itable, "noinit_itable"},
1223         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1224         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1225         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1226         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1227         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1228         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1229         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1230         {Opt_err, NULL},
1231 };
1232
1233 static ext4_fsblk_t get_sb_block(void **data)
1234 {
1235         ext4_fsblk_t    sb_block;
1236         char            *options = (char *) *data;
1237
1238         if (!options || strncmp(options, "sb=", 3) != 0)
1239                 return 1;       /* Default location */
1240
1241         options += 3;
1242         /* TODO: use simple_strtoll with >32bit ext4 */
1243         sb_block = simple_strtoul(options, &options, 0);
1244         if (*options && *options != ',') {
1245                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1246                        (char *) *data);
1247                 return 1;
1248         }
1249         if (*options == ',')
1250                 options++;
1251         *data = (void *) options;
1252
1253         return sb_block;
1254 }
1255
1256 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1257 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1258         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1259
1260 #ifdef CONFIG_QUOTA
1261 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1262 {
1263         struct ext4_sb_info *sbi = EXT4_SB(sb);
1264         char *qname;
1265         int ret = -1;
1266
1267         if (sb_any_quota_loaded(sb) &&
1268                 !sbi->s_qf_names[qtype]) {
1269                 ext4_msg(sb, KERN_ERR,
1270                         "Cannot change journaled "
1271                         "quota options when quota turned on");
1272                 return -1;
1273         }
1274         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1275                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1276                          "when QUOTA feature is enabled");
1277                 return -1;
1278         }
1279         qname = match_strdup(args);
1280         if (!qname) {
1281                 ext4_msg(sb, KERN_ERR,
1282                         "Not enough memory for storing quotafile name");
1283                 return -1;
1284         }
1285         if (sbi->s_qf_names[qtype]) {
1286                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1287                         ret = 1;
1288                 else
1289                         ext4_msg(sb, KERN_ERR,
1290                                  "%s quota file already specified",
1291                                  QTYPE2NAME(qtype));
1292                 goto errout;
1293         }
1294         if (strchr(qname, '/')) {
1295                 ext4_msg(sb, KERN_ERR,
1296                         "quotafile must be on filesystem root");
1297                 goto errout;
1298         }
1299         sbi->s_qf_names[qtype] = qname;
1300         set_opt(sb, QUOTA);
1301         return 1;
1302 errout:
1303         kfree(qname);
1304         return ret;
1305 }
1306
1307 static int clear_qf_name(struct super_block *sb, int qtype)
1308 {
1309
1310         struct ext4_sb_info *sbi = EXT4_SB(sb);
1311
1312         if (sb_any_quota_loaded(sb) &&
1313                 sbi->s_qf_names[qtype]) {
1314                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1315                         " when quota turned on");
1316                 return -1;
1317         }
1318         kfree(sbi->s_qf_names[qtype]);
1319         sbi->s_qf_names[qtype] = NULL;
1320         return 1;
1321 }
1322 #endif
1323
1324 #define MOPT_SET        0x0001
1325 #define MOPT_CLEAR      0x0002
1326 #define MOPT_NOSUPPORT  0x0004
1327 #define MOPT_EXPLICIT   0x0008
1328 #define MOPT_CLEAR_ERR  0x0010
1329 #define MOPT_GTE0       0x0020
1330 #ifdef CONFIG_QUOTA
1331 #define MOPT_Q          0
1332 #define MOPT_QFMT       0x0040
1333 #else
1334 #define MOPT_Q          MOPT_NOSUPPORT
1335 #define MOPT_QFMT       MOPT_NOSUPPORT
1336 #endif
1337 #define MOPT_DATAJ      0x0080
1338 #define MOPT_NO_EXT2    0x0100
1339 #define MOPT_NO_EXT3    0x0200
1340 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1341 #define MOPT_STRING     0x0400
1342
1343 static const struct mount_opts {
1344         int     token;
1345         int     mount_opt;
1346         int     flags;
1347 } ext4_mount_opts[] = {
1348         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1349         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1350         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1351         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1352         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1353         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1354         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1355          MOPT_EXT4_ONLY | MOPT_SET},
1356         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1357          MOPT_EXT4_ONLY | MOPT_CLEAR},
1358         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1359         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1360         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1361          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1362         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1363          MOPT_EXT4_ONLY | MOPT_CLEAR},
1364         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1365          MOPT_EXT4_ONLY | MOPT_CLEAR},
1366         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1367          MOPT_EXT4_ONLY | MOPT_SET},
1368         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1369                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1370          MOPT_EXT4_ONLY | MOPT_SET},
1371         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1372         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1373         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1374         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1375         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1376          MOPT_NO_EXT2 | MOPT_SET},
1377         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1378          MOPT_NO_EXT2 | MOPT_CLEAR},
1379         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1380         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1381         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1382         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1383         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1384         {Opt_commit, 0, MOPT_GTE0},
1385         {Opt_max_batch_time, 0, MOPT_GTE0},
1386         {Opt_min_batch_time, 0, MOPT_GTE0},
1387         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1388         {Opt_init_itable, 0, MOPT_GTE0},
1389         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1390         {Opt_stripe, 0, MOPT_GTE0},
1391         {Opt_resuid, 0, MOPT_GTE0},
1392         {Opt_resgid, 0, MOPT_GTE0},
1393         {Opt_journal_dev, 0, MOPT_GTE0},
1394         {Opt_journal_path, 0, MOPT_STRING},
1395         {Opt_journal_ioprio, 0, MOPT_GTE0},
1396         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1397         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1398         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1399          MOPT_NO_EXT2 | MOPT_DATAJ},
1400         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1401         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1402 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1403         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1404         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1405 #else
1406         {Opt_acl, 0, MOPT_NOSUPPORT},
1407         {Opt_noacl, 0, MOPT_NOSUPPORT},
1408 #endif
1409         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1410         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1411         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1412         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1413                                                         MOPT_SET | MOPT_Q},
1414         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1415                                                         MOPT_SET | MOPT_Q},
1416         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1417                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1418         {Opt_usrjquota, 0, MOPT_Q},
1419         {Opt_grpjquota, 0, MOPT_Q},
1420         {Opt_offusrjquota, 0, MOPT_Q},
1421         {Opt_offgrpjquota, 0, MOPT_Q},
1422         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1423         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1424         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1425         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1426         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1427         {Opt_err, 0, 0}
1428 };
1429
1430 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1431                             substring_t *args, unsigned long *journal_devnum,
1432                             unsigned int *journal_ioprio, int is_remount)
1433 {
1434         struct ext4_sb_info *sbi = EXT4_SB(sb);
1435         const struct mount_opts *m;
1436         kuid_t uid;
1437         kgid_t gid;
1438         int arg = 0;
1439
1440 #ifdef CONFIG_QUOTA
1441         if (token == Opt_usrjquota)
1442                 return set_qf_name(sb, USRQUOTA, &args[0]);
1443         else if (token == Opt_grpjquota)
1444                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1445         else if (token == Opt_offusrjquota)
1446                 return clear_qf_name(sb, USRQUOTA);
1447         else if (token == Opt_offgrpjquota)
1448                 return clear_qf_name(sb, GRPQUOTA);
1449 #endif
1450         switch (token) {
1451         case Opt_noacl:
1452         case Opt_nouser_xattr:
1453                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1454                 break;
1455         case Opt_sb:
1456                 return 1;       /* handled by get_sb_block() */
1457         case Opt_removed:
1458                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1459                 return 1;
1460         case Opt_abort:
1461                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1462                 return 1;
1463         case Opt_i_version:
1464                 sb->s_flags |= MS_I_VERSION;
1465                 return 1;
1466         case Opt_lazytime:
1467                 sb->s_flags |= MS_LAZYTIME;
1468                 return 1;
1469         case Opt_nolazytime:
1470                 sb->s_flags &= ~MS_LAZYTIME;
1471                 return 1;
1472         }
1473
1474         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1475                 if (token == m->token)
1476                         break;
1477
1478         if (m->token == Opt_err) {
1479                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1480                          "or missing value", opt);
1481                 return -1;
1482         }
1483
1484         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1485                 ext4_msg(sb, KERN_ERR,
1486                          "Mount option \"%s\" incompatible with ext2", opt);
1487                 return -1;
1488         }
1489         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1490                 ext4_msg(sb, KERN_ERR,
1491                          "Mount option \"%s\" incompatible with ext3", opt);
1492                 return -1;
1493         }
1494
1495         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1496                 return -1;
1497         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1498                 return -1;
1499         if (m->flags & MOPT_EXPLICIT)
1500                 set_opt2(sb, EXPLICIT_DELALLOC);
1501         if (m->flags & MOPT_CLEAR_ERR)
1502                 clear_opt(sb, ERRORS_MASK);
1503         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1504                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1505                          "options when quota turned on");
1506                 return -1;
1507         }
1508
1509         if (m->flags & MOPT_NOSUPPORT) {
1510                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1511         } else if (token == Opt_commit) {
1512                 if (arg == 0)
1513                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1514                 sbi->s_commit_interval = HZ * arg;
1515         } else if (token == Opt_max_batch_time) {
1516                 sbi->s_max_batch_time = arg;
1517         } else if (token == Opt_min_batch_time) {
1518                 sbi->s_min_batch_time = arg;
1519         } else if (token == Opt_inode_readahead_blks) {
1520                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1521                         ext4_msg(sb, KERN_ERR,
1522                                  "EXT4-fs: inode_readahead_blks must be "
1523                                  "0 or a power of 2 smaller than 2^31");
1524                         return -1;
1525                 }
1526                 sbi->s_inode_readahead_blks = arg;
1527         } else if (token == Opt_init_itable) {
1528                 set_opt(sb, INIT_INODE_TABLE);
1529                 if (!args->from)
1530                         arg = EXT4_DEF_LI_WAIT_MULT;
1531                 sbi->s_li_wait_mult = arg;
1532         } else if (token == Opt_max_dir_size_kb) {
1533                 sbi->s_max_dir_size_kb = arg;
1534         } else if (token == Opt_stripe) {
1535                 sbi->s_stripe = arg;
1536         } else if (token == Opt_resuid) {
1537                 uid = make_kuid(current_user_ns(), arg);
1538                 if (!uid_valid(uid)) {
1539                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1540                         return -1;
1541                 }
1542                 sbi->s_resuid = uid;
1543         } else if (token == Opt_resgid) {
1544                 gid = make_kgid(current_user_ns(), arg);
1545                 if (!gid_valid(gid)) {
1546                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1547                         return -1;
1548                 }
1549                 sbi->s_resgid = gid;
1550         } else if (token == Opt_journal_dev) {
1551                 if (is_remount) {
1552                         ext4_msg(sb, KERN_ERR,
1553                                  "Cannot specify journal on remount");
1554                         return -1;
1555                 }
1556                 *journal_devnum = arg;
1557         } else if (token == Opt_journal_path) {
1558                 char *journal_path;
1559                 struct inode *journal_inode;
1560                 struct path path;
1561                 int error;
1562
1563                 if (is_remount) {
1564                         ext4_msg(sb, KERN_ERR,
1565                                  "Cannot specify journal on remount");
1566                         return -1;
1567                 }
1568                 journal_path = match_strdup(&args[0]);
1569                 if (!journal_path) {
1570                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1571                                 "journal device string");
1572                         return -1;
1573                 }
1574
1575                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1576                 if (error) {
1577                         ext4_msg(sb, KERN_ERR, "error: could not find "
1578                                 "journal device path: error %d", error);
1579                         kfree(journal_path);
1580                         return -1;
1581                 }
1582
1583                 journal_inode = d_inode(path.dentry);
1584                 if (!S_ISBLK(journal_inode->i_mode)) {
1585                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1586                                 "is not a block device", journal_path);
1587                         path_put(&path);
1588                         kfree(journal_path);
1589                         return -1;
1590                 }
1591
1592                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1593                 path_put(&path);
1594                 kfree(journal_path);
1595         } else if (token == Opt_journal_ioprio) {
1596                 if (arg > 7) {
1597                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1598                                  " (must be 0-7)");
1599                         return -1;
1600                 }
1601                 *journal_ioprio =
1602                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1603         } else if (token == Opt_test_dummy_encryption) {
1604 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1605                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1606                 ext4_msg(sb, KERN_WARNING,
1607                          "Test dummy encryption mode enabled");
1608 #else
1609                 ext4_msg(sb, KERN_WARNING,
1610                          "Test dummy encryption mount option ignored");
1611 #endif
1612         } else if (m->flags & MOPT_DATAJ) {
1613                 if (is_remount) {
1614                         if (!sbi->s_journal)
1615                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1616                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1617                                 ext4_msg(sb, KERN_ERR,
1618                                          "Cannot change data mode on remount");
1619                                 return -1;
1620                         }
1621                 } else {
1622                         clear_opt(sb, DATA_FLAGS);
1623                         sbi->s_mount_opt |= m->mount_opt;
1624                 }
1625 #ifdef CONFIG_QUOTA
1626         } else if (m->flags & MOPT_QFMT) {
1627                 if (sb_any_quota_loaded(sb) &&
1628                     sbi->s_jquota_fmt != m->mount_opt) {
1629                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1630                                  "quota options when quota turned on");
1631                         return -1;
1632                 }
1633                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1634                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1635                         ext4_msg(sb, KERN_ERR,
1636                                  "Cannot set journaled quota options "
1637                                  "when QUOTA feature is enabled");
1638                         return -1;
1639                 }
1640                 sbi->s_jquota_fmt = m->mount_opt;
1641 #endif
1642 #ifndef CONFIG_FS_DAX
1643         } else if (token == Opt_dax) {
1644                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1645                 return -1;
1646 #endif
1647         } else {
1648                 if (!args->from)
1649                         arg = 1;
1650                 if (m->flags & MOPT_CLEAR)
1651                         arg = !arg;
1652                 else if (unlikely(!(m->flags & MOPT_SET))) {
1653                         ext4_msg(sb, KERN_WARNING,
1654                                  "buggy handling of option %s", opt);
1655                         WARN_ON(1);
1656                         return -1;
1657                 }
1658                 if (arg != 0)
1659                         sbi->s_mount_opt |= m->mount_opt;
1660                 else
1661                         sbi->s_mount_opt &= ~m->mount_opt;
1662         }
1663         return 1;
1664 }
1665
1666 static int parse_options(char *options, struct super_block *sb,
1667                          unsigned long *journal_devnum,
1668                          unsigned int *journal_ioprio,
1669                          int is_remount)
1670 {
1671         struct ext4_sb_info *sbi = EXT4_SB(sb);
1672         char *p;
1673         substring_t args[MAX_OPT_ARGS];
1674         int token;
1675
1676         if (!options)
1677                 return 1;
1678
1679         while ((p = strsep(&options, ",")) != NULL) {
1680                 if (!*p)
1681                         continue;
1682                 /*
1683                  * Initialize args struct so we know whether arg was
1684                  * found; some options take optional arguments.
1685                  */
1686                 args[0].to = args[0].from = NULL;
1687                 token = match_token(p, tokens, args);
1688                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1689                                      journal_ioprio, is_remount) < 0)
1690                         return 0;
1691         }
1692 #ifdef CONFIG_QUOTA
1693         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1694             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1695                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1696                          "feature is enabled");
1697                 return 0;
1698         }
1699         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1700                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1701                         clear_opt(sb, USRQUOTA);
1702
1703                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1704                         clear_opt(sb, GRPQUOTA);
1705
1706                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1707                         ext4_msg(sb, KERN_ERR, "old and new quota "
1708                                         "format mixing");
1709                         return 0;
1710                 }
1711
1712                 if (!sbi->s_jquota_fmt) {
1713                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1714                                         "not specified");
1715                         return 0;
1716                 }
1717         }
1718 #endif
1719         if (test_opt(sb, DIOREAD_NOLOCK)) {
1720                 int blocksize =
1721                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1722
1723                 if (blocksize < PAGE_CACHE_SIZE) {
1724                         ext4_msg(sb, KERN_ERR, "can't mount with "
1725                                  "dioread_nolock if block size != PAGE_SIZE");
1726                         return 0;
1727                 }
1728         }
1729         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1730             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1731                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1732                          "in data=ordered mode");
1733                 return 0;
1734         }
1735         return 1;
1736 }
1737
1738 static inline void ext4_show_quota_options(struct seq_file *seq,
1739                                            struct super_block *sb)
1740 {
1741 #if defined(CONFIG_QUOTA)
1742         struct ext4_sb_info *sbi = EXT4_SB(sb);
1743
1744         if (sbi->s_jquota_fmt) {
1745                 char *fmtname = "";
1746
1747                 switch (sbi->s_jquota_fmt) {
1748                 case QFMT_VFS_OLD:
1749                         fmtname = "vfsold";
1750                         break;
1751                 case QFMT_VFS_V0:
1752                         fmtname = "vfsv0";
1753                         break;
1754                 case QFMT_VFS_V1:
1755                         fmtname = "vfsv1";
1756                         break;
1757                 }
1758                 seq_printf(seq, ",jqfmt=%s", fmtname);
1759         }
1760
1761         if (sbi->s_qf_names[USRQUOTA])
1762                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1763
1764         if (sbi->s_qf_names[GRPQUOTA])
1765                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1766 #endif
1767 }
1768
1769 static const char *token2str(int token)
1770 {
1771         const struct match_token *t;
1772
1773         for (t = tokens; t->token != Opt_err; t++)
1774                 if (t->token == token && !strchr(t->pattern, '='))
1775                         break;
1776         return t->pattern;
1777 }
1778
1779 /*
1780  * Show an option if
1781  *  - it's set to a non-default value OR
1782  *  - if the per-sb default is different from the global default
1783  */
1784 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1785                               int nodefs)
1786 {
1787         struct ext4_sb_info *sbi = EXT4_SB(sb);
1788         struct ext4_super_block *es = sbi->s_es;
1789         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1790         const struct mount_opts *m;
1791         char sep = nodefs ? '\n' : ',';
1792
1793 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1794 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1795
1796         if (sbi->s_sb_block != 1)
1797                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1798
1799         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1800                 int want_set = m->flags & MOPT_SET;
1801                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1802                     (m->flags & MOPT_CLEAR_ERR))
1803                         continue;
1804                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1805                         continue; /* skip if same as the default */
1806                 if ((want_set &&
1807                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1808                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1809                         continue; /* select Opt_noFoo vs Opt_Foo */
1810                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1811         }
1812
1813         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1814             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1815                 SEQ_OPTS_PRINT("resuid=%u",
1816                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1817         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1818             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1819                 SEQ_OPTS_PRINT("resgid=%u",
1820                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1821         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1822         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1823                 SEQ_OPTS_PUTS("errors=remount-ro");
1824         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1825                 SEQ_OPTS_PUTS("errors=continue");
1826         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1827                 SEQ_OPTS_PUTS("errors=panic");
1828         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1829                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1830         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1831                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1832         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1833                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1834         if (sb->s_flags & MS_I_VERSION)
1835                 SEQ_OPTS_PUTS("i_version");
1836         if (nodefs || sbi->s_stripe)
1837                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1838         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1839                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1840                         SEQ_OPTS_PUTS("data=journal");
1841                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1842                         SEQ_OPTS_PUTS("data=ordered");
1843                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1844                         SEQ_OPTS_PUTS("data=writeback");
1845         }
1846         if (nodefs ||
1847             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1848                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1849                                sbi->s_inode_readahead_blks);
1850
1851         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1852                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1853                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1854         if (nodefs || sbi->s_max_dir_size_kb)
1855                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1856
1857         ext4_show_quota_options(seq, sb);
1858         return 0;
1859 }
1860
1861 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1862 {
1863         return _ext4_show_options(seq, root->d_sb, 0);
1864 }
1865
1866 static int options_seq_show(struct seq_file *seq, void *offset)
1867 {
1868         struct super_block *sb = seq->private;
1869         int rc;
1870
1871         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1872         rc = _ext4_show_options(seq, sb, 1);
1873         seq_puts(seq, "\n");
1874         return rc;
1875 }
1876
1877 static int options_open_fs(struct inode *inode, struct file *file)
1878 {
1879         return single_open(file, options_seq_show, PDE_DATA(inode));
1880 }
1881
1882 static const struct file_operations ext4_seq_options_fops = {
1883         .owner = THIS_MODULE,
1884         .open = options_open_fs,
1885         .read = seq_read,
1886         .llseek = seq_lseek,
1887         .release = single_release,
1888 };
1889
1890 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1891                             int read_only)
1892 {
1893         struct ext4_sb_info *sbi = EXT4_SB(sb);
1894         int res = 0;
1895
1896         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1897                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1898                          "forcing read-only mode");
1899                 res = MS_RDONLY;
1900         }
1901         if (read_only)
1902                 goto done;
1903         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1904                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1905                          "running e2fsck is recommended");
1906         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1907                 ext4_msg(sb, KERN_WARNING,
1908                          "warning: mounting fs with errors, "
1909                          "running e2fsck is recommended");
1910         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1911                  le16_to_cpu(es->s_mnt_count) >=
1912                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1913                 ext4_msg(sb, KERN_WARNING,
1914                          "warning: maximal mount count reached, "
1915                          "running e2fsck is recommended");
1916         else if (le32_to_cpu(es->s_checkinterval) &&
1917                 (le32_to_cpu(es->s_lastcheck) +
1918                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1919                 ext4_msg(sb, KERN_WARNING,
1920                          "warning: checktime reached, "
1921                          "running e2fsck is recommended");
1922         if (!sbi->s_journal)
1923                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1924         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1925                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1926         le16_add_cpu(&es->s_mnt_count, 1);
1927         es->s_mtime = cpu_to_le32(get_seconds());
1928         ext4_update_dynamic_rev(sb);
1929         if (sbi->s_journal)
1930                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1931
1932         ext4_commit_super(sb, 1);
1933 done:
1934         if (test_opt(sb, DEBUG))
1935                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1936                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1937                         sb->s_blocksize,
1938                         sbi->s_groups_count,
1939                         EXT4_BLOCKS_PER_GROUP(sb),
1940                         EXT4_INODES_PER_GROUP(sb),
1941                         sbi->s_mount_opt, sbi->s_mount_opt2);
1942
1943         cleancache_init_fs(sb);
1944         return res;
1945 }
1946
1947 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1948 {
1949         struct ext4_sb_info *sbi = EXT4_SB(sb);
1950         struct flex_groups *new_groups;
1951         int size;
1952
1953         if (!sbi->s_log_groups_per_flex)
1954                 return 0;
1955
1956         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1957         if (size <= sbi->s_flex_groups_allocated)
1958                 return 0;
1959
1960         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1961         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1962         if (!new_groups) {
1963                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1964                          size / (int) sizeof(struct flex_groups));
1965                 return -ENOMEM;
1966         }
1967
1968         if (sbi->s_flex_groups) {
1969                 memcpy(new_groups, sbi->s_flex_groups,
1970                        (sbi->s_flex_groups_allocated *
1971                         sizeof(struct flex_groups)));
1972                 kvfree(sbi->s_flex_groups);
1973         }
1974         sbi->s_flex_groups = new_groups;
1975         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1976         return 0;
1977 }
1978
1979 static int ext4_fill_flex_info(struct super_block *sb)
1980 {
1981         struct ext4_sb_info *sbi = EXT4_SB(sb);
1982         struct ext4_group_desc *gdp = NULL;
1983         ext4_group_t flex_group;
1984         int i, err;
1985
1986         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1987         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1988                 sbi->s_log_groups_per_flex = 0;
1989                 return 1;
1990         }
1991
1992         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1993         if (err)
1994                 goto failed;
1995
1996         for (i = 0; i < sbi->s_groups_count; i++) {
1997                 gdp = ext4_get_group_desc(sb, i, NULL);
1998
1999                 flex_group = ext4_flex_group(sbi, i);
2000                 atomic_add(ext4_free_inodes_count(sb, gdp),
2001                            &sbi->s_flex_groups[flex_group].free_inodes);
2002                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2003                              &sbi->s_flex_groups[flex_group].free_clusters);
2004                 atomic_add(ext4_used_dirs_count(sb, gdp),
2005                            &sbi->s_flex_groups[flex_group].used_dirs);
2006         }
2007
2008         return 1;
2009 failed:
2010         return 0;
2011 }
2012
2013 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2014                                    struct ext4_group_desc *gdp)
2015 {
2016         int offset;
2017         __u16 crc = 0;
2018         __le32 le_group = cpu_to_le32(block_group);
2019
2020         if (ext4_has_metadata_csum(sbi->s_sb)) {
2021                 /* Use new metadata_csum algorithm */
2022                 __le16 save_csum;
2023                 __u32 csum32;
2024
2025                 save_csum = gdp->bg_checksum;
2026                 gdp->bg_checksum = 0;
2027                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2028                                      sizeof(le_group));
2029                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2030                                      sbi->s_desc_size);
2031                 gdp->bg_checksum = save_csum;
2032
2033                 crc = csum32 & 0xFFFF;
2034                 goto out;
2035         }
2036
2037         /* old crc16 code */
2038         if (!(sbi->s_es->s_feature_ro_compat &
2039               cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)))
2040                 return 0;
2041
2042         offset = offsetof(struct ext4_group_desc, bg_checksum);
2043
2044         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2045         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2046         crc = crc16(crc, (__u8 *)gdp, offset);
2047         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2048         /* for checksum of struct ext4_group_desc do the rest...*/
2049         if ((sbi->s_es->s_feature_incompat &
2050              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2051             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2052                 crc = crc16(crc, (__u8 *)gdp + offset,
2053                             le16_to_cpu(sbi->s_es->s_desc_size) -
2054                                 offset);
2055
2056 out:
2057         return cpu_to_le16(crc);
2058 }
2059
2060 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2061                                 struct ext4_group_desc *gdp)
2062 {
2063         if (ext4_has_group_desc_csum(sb) &&
2064             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2065                                                       block_group, gdp)))
2066                 return 0;
2067
2068         return 1;
2069 }
2070
2071 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2072                               struct ext4_group_desc *gdp)
2073 {
2074         if (!ext4_has_group_desc_csum(sb))
2075                 return;
2076         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2077 }
2078
2079 /* Called at mount-time, super-block is locked */
2080 static int ext4_check_descriptors(struct super_block *sb,
2081                                   ext4_group_t *first_not_zeroed)
2082 {
2083         struct ext4_sb_info *sbi = EXT4_SB(sb);
2084         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2085         ext4_fsblk_t last_block;
2086         ext4_fsblk_t block_bitmap;
2087         ext4_fsblk_t inode_bitmap;
2088         ext4_fsblk_t inode_table;
2089         int flexbg_flag = 0;
2090         ext4_group_t i, grp = sbi->s_groups_count;
2091
2092         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2093                 flexbg_flag = 1;
2094
2095         ext4_debug("Checking group descriptors");
2096
2097         for (i = 0; i < sbi->s_groups_count; i++) {
2098                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2099
2100                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2101                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2102                 else
2103                         last_block = first_block +
2104                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2105
2106                 if ((grp == sbi->s_groups_count) &&
2107                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2108                         grp = i;
2109
2110                 block_bitmap = ext4_block_bitmap(sb, gdp);
2111                 if (block_bitmap < first_block || block_bitmap > last_block) {
2112                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2113                                "Block bitmap for group %u not in group "
2114                                "(block %llu)!", i, block_bitmap);
2115                         return 0;
2116                 }
2117                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2118                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2119                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2120                                "Inode bitmap for group %u not in group "
2121                                "(block %llu)!", i, inode_bitmap);
2122                         return 0;
2123                 }
2124                 inode_table = ext4_inode_table(sb, gdp);
2125                 if (inode_table < first_block ||
2126                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2127                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2128                                "Inode table for group %u not in group "
2129                                "(block %llu)!", i, inode_table);
2130                         return 0;
2131                 }
2132                 ext4_lock_group(sb, i);
2133                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2134                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2135                                  "Checksum for group %u failed (%u!=%u)",
2136                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2137                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2138                         if (!(sb->s_flags & MS_RDONLY)) {
2139                                 ext4_unlock_group(sb, i);
2140                                 return 0;
2141                         }
2142                 }
2143                 ext4_unlock_group(sb, i);
2144                 if (!flexbg_flag)
2145                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2146         }
2147         if (NULL != first_not_zeroed)
2148                 *first_not_zeroed = grp;
2149         return 1;
2150 }
2151
2152 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2153  * the superblock) which were deleted from all directories, but held open by
2154  * a process at the time of a crash.  We walk the list and try to delete these
2155  * inodes at recovery time (only with a read-write filesystem).
2156  *
2157  * In order to keep the orphan inode chain consistent during traversal (in
2158  * case of crash during recovery), we link each inode into the superblock
2159  * orphan list_head and handle it the same way as an inode deletion during
2160  * normal operation (which journals the operations for us).
2161  *
2162  * We only do an iget() and an iput() on each inode, which is very safe if we
2163  * accidentally point at an in-use or already deleted inode.  The worst that
2164  * can happen in this case is that we get a "bit already cleared" message from
2165  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2166  * e2fsck was run on this filesystem, and it must have already done the orphan
2167  * inode cleanup for us, so we can safely abort without any further action.
2168  */
2169 static void ext4_orphan_cleanup(struct super_block *sb,
2170                                 struct ext4_super_block *es)
2171 {
2172         unsigned int s_flags = sb->s_flags;
2173         int nr_orphans = 0, nr_truncates = 0;
2174 #ifdef CONFIG_QUOTA
2175         int i;
2176 #endif
2177         if (!es->s_last_orphan) {
2178                 jbd_debug(4, "no orphan inodes to clean up\n");
2179                 return;
2180         }
2181
2182         if (bdev_read_only(sb->s_bdev)) {
2183                 ext4_msg(sb, KERN_ERR, "write access "
2184                         "unavailable, skipping orphan cleanup");
2185                 return;
2186         }
2187
2188         /* Check if feature set would not allow a r/w mount */
2189         if (!ext4_feature_set_ok(sb, 0)) {
2190                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2191                          "unknown ROCOMPAT features");
2192                 return;
2193         }
2194
2195         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2196                 /* don't clear list on RO mount w/ errors */
2197                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2198                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2199                                   "clearing orphan list.\n");
2200                         es->s_last_orphan = 0;
2201                 }
2202                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2203                 return;
2204         }
2205
2206         if (s_flags & MS_RDONLY) {
2207                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2208                 sb->s_flags &= ~MS_RDONLY;
2209         }
2210 #ifdef CONFIG_QUOTA
2211         /* Needed for iput() to work correctly and not trash data */
2212         sb->s_flags |= MS_ACTIVE;
2213         /* Turn on quotas so that they are updated correctly */
2214         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2215                 if (EXT4_SB(sb)->s_qf_names[i]) {
2216                         int ret = ext4_quota_on_mount(sb, i);
2217                         if (ret < 0)
2218                                 ext4_msg(sb, KERN_ERR,
2219                                         "Cannot turn on journaled "
2220                                         "quota: error %d", ret);
2221                 }
2222         }
2223 #endif
2224
2225         while (es->s_last_orphan) {
2226                 struct inode *inode;
2227
2228                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2229                 if (IS_ERR(inode)) {
2230                         es->s_last_orphan = 0;
2231                         break;
2232                 }
2233
2234                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2235                 dquot_initialize(inode);
2236                 if (inode->i_nlink) {
2237                         if (test_opt(sb, DEBUG))
2238                                 ext4_msg(sb, KERN_DEBUG,
2239                                         "%s: truncating inode %lu to %lld bytes",
2240                                         __func__, inode->i_ino, inode->i_size);
2241                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2242                                   inode->i_ino, inode->i_size);
2243                         mutex_lock(&inode->i_mutex);
2244                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2245                         ext4_truncate(inode);
2246                         mutex_unlock(&inode->i_mutex);
2247                         nr_truncates++;
2248                 } else {
2249                         if (test_opt(sb, DEBUG))
2250                                 ext4_msg(sb, KERN_DEBUG,
2251                                         "%s: deleting unreferenced inode %lu",
2252                                         __func__, inode->i_ino);
2253                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2254                                   inode->i_ino);
2255                         nr_orphans++;
2256                 }
2257                 iput(inode);  /* The delete magic happens here! */
2258         }
2259
2260 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2261
2262         if (nr_orphans)
2263                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2264                        PLURAL(nr_orphans));
2265         if (nr_truncates)
2266                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2267                        PLURAL(nr_truncates));
2268 #ifdef CONFIG_QUOTA
2269         /* Turn quotas off */
2270         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2271                 if (sb_dqopt(sb)->files[i])
2272                         dquot_quota_off(sb, i);
2273         }
2274 #endif
2275         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2276 }
2277
2278 /*
2279  * Maximal extent format file size.
2280  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2281  * extent format containers, within a sector_t, and within i_blocks
2282  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2283  * so that won't be a limiting factor.
2284  *
2285  * However there is other limiting factor. We do store extents in the form
2286  * of starting block and length, hence the resulting length of the extent
2287  * covering maximum file size must fit into on-disk format containers as
2288  * well. Given that length is always by 1 unit bigger than max unit (because
2289  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2290  *
2291  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2292  */
2293 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2294 {
2295         loff_t res;
2296         loff_t upper_limit = MAX_LFS_FILESIZE;
2297
2298         /* small i_blocks in vfs inode? */
2299         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2300                 /*
2301                  * CONFIG_LBDAF is not enabled implies the inode
2302                  * i_block represent total blocks in 512 bytes
2303                  * 32 == size of vfs inode i_blocks * 8
2304                  */
2305                 upper_limit = (1LL << 32) - 1;
2306
2307                 /* total blocks in file system block size */
2308                 upper_limit >>= (blkbits - 9);
2309                 upper_limit <<= blkbits;
2310         }
2311
2312         /*
2313          * 32-bit extent-start container, ee_block. We lower the maxbytes
2314          * by one fs block, so ee_len can cover the extent of maximum file
2315          * size
2316          */
2317         res = (1LL << 32) - 1;
2318         res <<= blkbits;
2319
2320         /* Sanity check against vm- & vfs- imposed limits */
2321         if (res > upper_limit)
2322                 res = upper_limit;
2323
2324         return res;
2325 }
2326
2327 /*
2328  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2329  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2330  * We need to be 1 filesystem block less than the 2^48 sector limit.
2331  */
2332 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2333 {
2334         loff_t res = EXT4_NDIR_BLOCKS;
2335         int meta_blocks;
2336         loff_t upper_limit;
2337         /* This is calculated to be the largest file size for a dense, block
2338          * mapped file such that the file's total number of 512-byte sectors,
2339          * including data and all indirect blocks, does not exceed (2^48 - 1).
2340          *
2341          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2342          * number of 512-byte sectors of the file.
2343          */
2344
2345         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2346                 /*
2347                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2348                  * the inode i_block field represents total file blocks in
2349                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2350                  */
2351                 upper_limit = (1LL << 32) - 1;
2352
2353                 /* total blocks in file system block size */
2354                 upper_limit >>= (bits - 9);
2355
2356         } else {
2357                 /*
2358                  * We use 48 bit ext4_inode i_blocks
2359                  * With EXT4_HUGE_FILE_FL set the i_blocks
2360                  * represent total number of blocks in
2361                  * file system block size
2362                  */
2363                 upper_limit = (1LL << 48) - 1;
2364
2365         }
2366
2367         /* indirect blocks */
2368         meta_blocks = 1;
2369         /* double indirect blocks */
2370         meta_blocks += 1 + (1LL << (bits-2));
2371         /* tripple indirect blocks */
2372         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2373
2374         upper_limit -= meta_blocks;
2375         upper_limit <<= bits;
2376
2377         res += 1LL << (bits-2);
2378         res += 1LL << (2*(bits-2));
2379         res += 1LL << (3*(bits-2));
2380         res <<= bits;
2381         if (res > upper_limit)
2382                 res = upper_limit;
2383
2384         if (res > MAX_LFS_FILESIZE)
2385                 res = MAX_LFS_FILESIZE;
2386
2387         return res;
2388 }
2389
2390 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2391                                    ext4_fsblk_t logical_sb_block, int nr)
2392 {
2393         struct ext4_sb_info *sbi = EXT4_SB(sb);
2394         ext4_group_t bg, first_meta_bg;
2395         int has_super = 0;
2396
2397         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2398
2399         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2400             nr < first_meta_bg)
2401                 return logical_sb_block + nr + 1;
2402         bg = sbi->s_desc_per_block * nr;
2403         if (ext4_bg_has_super(sb, bg))
2404                 has_super = 1;
2405
2406         /*
2407          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2408          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2409          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2410          * compensate.
2411          */
2412         if (sb->s_blocksize == 1024 && nr == 0 &&
2413             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2414                 has_super++;
2415
2416         return (has_super + ext4_group_first_block_no(sb, bg));
2417 }
2418
2419 /**
2420  * ext4_get_stripe_size: Get the stripe size.
2421  * @sbi: In memory super block info
2422  *
2423  * If we have specified it via mount option, then
2424  * use the mount option value. If the value specified at mount time is
2425  * greater than the blocks per group use the super block value.
2426  * If the super block value is greater than blocks per group return 0.
2427  * Allocator needs it be less than blocks per group.
2428  *
2429  */
2430 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2431 {
2432         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2433         unsigned long stripe_width =
2434                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2435         int ret;
2436
2437         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2438                 ret = sbi->s_stripe;
2439         else if (stripe_width <= sbi->s_blocks_per_group)
2440                 ret = stripe_width;
2441         else if (stride <= sbi->s_blocks_per_group)
2442                 ret = stride;
2443         else
2444                 ret = 0;
2445
2446         /*
2447          * If the stripe width is 1, this makes no sense and
2448          * we set it to 0 to turn off stripe handling code.
2449          */
2450         if (ret <= 1)
2451                 ret = 0;
2452
2453         return ret;
2454 }
2455
2456 /* sysfs supprt */
2457
2458 struct ext4_attr {
2459         struct attribute attr;
2460         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2461         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2462                          const char *, size_t);
2463         union {
2464                 int offset;
2465                 int deprecated_val;
2466         } u;
2467 };
2468
2469 static int parse_strtoull(const char *buf,
2470                 unsigned long long max, unsigned long long *value)
2471 {
2472         int ret;
2473
2474         ret = kstrtoull(skip_spaces(buf), 0, value);
2475         if (!ret && *value > max)
2476                 ret = -EINVAL;
2477         return ret;
2478 }
2479
2480 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2481                                               struct ext4_sb_info *sbi,
2482                                               char *buf)
2483 {
2484         return snprintf(buf, PAGE_SIZE, "%llu\n",
2485                 (s64) EXT4_C2B(sbi,
2486                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2487 }
2488
2489 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2490                                          struct ext4_sb_info *sbi, char *buf)
2491 {
2492         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2493
2494         if (!sb->s_bdev->bd_part)
2495                 return snprintf(buf, PAGE_SIZE, "0\n");
2496         return snprintf(buf, PAGE_SIZE, "%lu\n",
2497                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2498                          sbi->s_sectors_written_start) >> 1);
2499 }
2500
2501 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2502                                           struct ext4_sb_info *sbi, char *buf)
2503 {
2504         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2505
2506         if (!sb->s_bdev->bd_part)
2507                 return snprintf(buf, PAGE_SIZE, "0\n");
2508         return snprintf(buf, PAGE_SIZE, "%llu\n",
2509                         (unsigned long long)(sbi->s_kbytes_written +
2510                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2511                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2512 }
2513
2514 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2515                                           struct ext4_sb_info *sbi,
2516                                           const char *buf, size_t count)
2517 {
2518         unsigned long t;
2519         int ret;
2520
2521         ret = kstrtoul(skip_spaces(buf), 0, &t);
2522         if (ret)
2523                 return ret;
2524
2525         if (t && (!is_power_of_2(t) || t > 0x40000000))
2526                 return -EINVAL;
2527
2528         sbi->s_inode_readahead_blks = t;
2529         return count;
2530 }
2531
2532 static ssize_t sbi_ui_show(struct ext4_attr *a,
2533                            struct ext4_sb_info *sbi, char *buf)
2534 {
2535         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2536
2537         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2538 }
2539
2540 static ssize_t sbi_ui_store(struct ext4_attr *a,
2541                             struct ext4_sb_info *sbi,
2542                             const char *buf, size_t count)
2543 {
2544         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2545         unsigned long t;
2546         int ret;
2547
2548         ret = kstrtoul(skip_spaces(buf), 0, &t);
2549         if (ret)
2550                 return ret;
2551         *ui = t;
2552         return count;
2553 }
2554
2555 static ssize_t es_ui_show(struct ext4_attr *a,
2556                            struct ext4_sb_info *sbi, char *buf)
2557 {
2558
2559         unsigned int *ui = (unsigned int *) (((char *) sbi->s_es) +
2560                            a->u.offset);
2561
2562         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2563 }
2564
2565 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2566                                   struct ext4_sb_info *sbi, char *buf)
2567 {
2568         return snprintf(buf, PAGE_SIZE, "%llu\n",
2569                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2570 }
2571
2572 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2573                                    struct ext4_sb_info *sbi,
2574                                    const char *buf, size_t count)
2575 {
2576         unsigned long long val;
2577         int ret;
2578
2579         if (parse_strtoull(buf, -1ULL, &val))
2580                 return -EINVAL;
2581         ret = ext4_reserve_clusters(sbi, val);
2582
2583         return ret ? ret : count;
2584 }
2585
2586 static ssize_t trigger_test_error(struct ext4_attr *a,
2587                                   struct ext4_sb_info *sbi,
2588                                   const char *buf, size_t count)
2589 {
2590         int len = count;
2591
2592         if (!capable(CAP_SYS_ADMIN))
2593                 return -EPERM;
2594
2595         if (len && buf[len-1] == '\n')
2596                 len--;
2597
2598         if (len)
2599                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2600         return count;
2601 }
2602
2603 static ssize_t sbi_deprecated_show(struct ext4_attr *a,
2604                                    struct ext4_sb_info *sbi, char *buf)
2605 {
2606         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
2607 }
2608
2609 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2610 static struct ext4_attr ext4_attr_##_name = {                   \
2611         .attr = {.name = __stringify(_name), .mode = _mode },   \
2612         .show   = _show,                                        \
2613         .store  = _store,                                       \
2614         .u = {                                                  \
2615                 .offset = offsetof(struct ext4_sb_info, _elname),\
2616         },                                                      \
2617 }
2618
2619 #define EXT4_ATTR_OFFSET_ES(_name,_mode,_show,_store,_elname)           \
2620 static struct ext4_attr ext4_attr_##_name = {                           \
2621         .attr = {.name = __stringify(_name), .mode = _mode },           \
2622         .show   = _show,                                                \
2623         .store  = _store,                                               \
2624         .u = {                                                          \
2625                 .offset = offsetof(struct ext4_super_block, _elname),   \
2626         },                                                              \
2627 }
2628
2629 #define EXT4_ATTR(name, mode, show, store) \
2630 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2631
2632 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2633 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2634 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2635
2636 #define EXT4_RO_ATTR_ES_UI(name, elname)        \
2637         EXT4_ATTR_OFFSET_ES(name, 0444, es_ui_show, NULL, elname)
2638 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2639         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2640
2641 #define ATTR_LIST(name) &ext4_attr_##name.attr
2642 #define EXT4_DEPRECATED_ATTR(_name, _val)       \
2643 static struct ext4_attr ext4_attr_##_name = {                   \
2644         .attr = {.name = __stringify(_name), .mode = 0444 },    \
2645         .show   = sbi_deprecated_show,                          \
2646         .u = {                                                  \
2647                 .deprecated_val = _val,                         \
2648         },                                                      \
2649 }
2650
2651 EXT4_RO_ATTR(delayed_allocation_blocks);
2652 EXT4_RO_ATTR(session_write_kbytes);
2653 EXT4_RO_ATTR(lifetime_write_kbytes);
2654 EXT4_RW_ATTR(reserved_clusters);
2655 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2656                  inode_readahead_blks_store, s_inode_readahead_blks);
2657 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2658 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2659 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2660 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2661 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2662 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2663 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2664 EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
2665 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2666 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2667 EXT4_RW_ATTR_SBI_UI(err_ratelimit_interval_ms, s_err_ratelimit_state.interval);
2668 EXT4_RW_ATTR_SBI_UI(err_ratelimit_burst, s_err_ratelimit_state.burst);
2669 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_interval_ms, s_warning_ratelimit_state.interval);
2670 EXT4_RW_ATTR_SBI_UI(warning_ratelimit_burst, s_warning_ratelimit_state.burst);
2671 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_interval_ms, s_msg_ratelimit_state.interval);
2672 EXT4_RW_ATTR_SBI_UI(msg_ratelimit_burst, s_msg_ratelimit_state.burst);
2673 EXT4_RO_ATTR_ES_UI(errors_count, s_error_count);
2674 EXT4_RO_ATTR_ES_UI(first_error_time, s_first_error_time);
2675 EXT4_RO_ATTR_ES_UI(last_error_time, s_last_error_time);
2676
2677 static struct attribute *ext4_attrs[] = {
2678         ATTR_LIST(delayed_allocation_blocks),
2679         ATTR_LIST(session_write_kbytes),
2680         ATTR_LIST(lifetime_write_kbytes),
2681         ATTR_LIST(reserved_clusters),
2682         ATTR_LIST(inode_readahead_blks),
2683         ATTR_LIST(inode_goal),
2684         ATTR_LIST(mb_stats),
2685         ATTR_LIST(mb_max_to_scan),
2686         ATTR_LIST(mb_min_to_scan),
2687         ATTR_LIST(mb_order2_req),
2688         ATTR_LIST(mb_stream_req),
2689         ATTR_LIST(mb_group_prealloc),
2690         ATTR_LIST(max_writeback_mb_bump),
2691         ATTR_LIST(extent_max_zeroout_kb),
2692         ATTR_LIST(trigger_fs_error),
2693         ATTR_LIST(err_ratelimit_interval_ms),
2694         ATTR_LIST(err_ratelimit_burst),
2695         ATTR_LIST(warning_ratelimit_interval_ms),
2696         ATTR_LIST(warning_ratelimit_burst),
2697         ATTR_LIST(msg_ratelimit_interval_ms),
2698         ATTR_LIST(msg_ratelimit_burst),
2699         ATTR_LIST(errors_count),
2700         ATTR_LIST(first_error_time),
2701         ATTR_LIST(last_error_time),
2702         NULL,
2703 };
2704
2705 /* Features this copy of ext4 supports */
2706 EXT4_INFO_ATTR(lazy_itable_init);
2707 EXT4_INFO_ATTR(batched_discard);
2708 EXT4_INFO_ATTR(meta_bg_resize);
2709 EXT4_INFO_ATTR(encryption);
2710
2711 static struct attribute *ext4_feat_attrs[] = {
2712         ATTR_LIST(lazy_itable_init),
2713         ATTR_LIST(batched_discard),
2714         ATTR_LIST(meta_bg_resize),
2715         ATTR_LIST(encryption),
2716         NULL,
2717 };
2718
2719 static ssize_t ext4_attr_show(struct kobject *kobj,
2720                               struct attribute *attr, char *buf)
2721 {
2722         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2723                                                 s_kobj);
2724         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2725
2726         return a->show ? a->show(a, sbi, buf) : 0;
2727 }
2728
2729 static ssize_t ext4_attr_store(struct kobject *kobj,
2730                                struct attribute *attr,
2731                                const char *buf, size_t len)
2732 {
2733         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2734                                                 s_kobj);
2735         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2736
2737         return a->store ? a->store(a, sbi, buf, len) : 0;
2738 }
2739
2740 static void ext4_sb_release(struct kobject *kobj)
2741 {
2742         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2743                                                 s_kobj);
2744         complete(&sbi->s_kobj_unregister);
2745 }
2746
2747 static const struct sysfs_ops ext4_attr_ops = {
2748         .show   = ext4_attr_show,
2749         .store  = ext4_attr_store,
2750 };
2751
2752 static struct kobj_type ext4_ktype = {
2753         .default_attrs  = ext4_attrs,
2754         .sysfs_ops      = &ext4_attr_ops,
2755         .release        = ext4_sb_release,
2756 };
2757
2758 static void ext4_feat_release(struct kobject *kobj)
2759 {
2760         complete(&ext4_feat->f_kobj_unregister);
2761 }
2762
2763 static ssize_t ext4_feat_show(struct kobject *kobj,
2764                               struct attribute *attr, char *buf)
2765 {
2766         return snprintf(buf, PAGE_SIZE, "supported\n");
2767 }
2768
2769 /*
2770  * We can not use ext4_attr_show/store because it relies on the kobject
2771  * being embedded in the ext4_sb_info structure which is definitely not
2772  * true in this case.
2773  */
2774 static const struct sysfs_ops ext4_feat_ops = {
2775         .show   = ext4_feat_show,
2776         .store  = NULL,
2777 };
2778
2779 static struct kobj_type ext4_feat_ktype = {
2780         .default_attrs  = ext4_feat_attrs,
2781         .sysfs_ops      = &ext4_feat_ops,
2782         .release        = ext4_feat_release,
2783 };
2784
2785 /*
2786  * Check whether this filesystem can be mounted based on
2787  * the features present and the RDONLY/RDWR mount requested.
2788  * Returns 1 if this filesystem can be mounted as requested,
2789  * 0 if it cannot be.
2790  */
2791 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2792 {
2793         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2794                 ext4_msg(sb, KERN_ERR,
2795                         "Couldn't mount because of "
2796                         "unsupported optional features (%x)",
2797                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2798                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2799                 return 0;
2800         }
2801
2802         if (readonly)
2803                 return 1;
2804
2805         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_READONLY)) {
2806                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2807                 sb->s_flags |= MS_RDONLY;
2808                 return 1;
2809         }
2810
2811         /* Check that feature set is OK for a read-write mount */
2812         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2813                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2814                          "unsupported optional features (%x)",
2815                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2816                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2817                 return 0;
2818         }
2819         /*
2820          * Large file size enabled file system can only be mounted
2821          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2822          */
2823         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2824                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2825                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2826                                  "cannot be mounted RDWR without "
2827                                  "CONFIG_LBDAF");
2828                         return 0;
2829                 }
2830         }
2831         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2832             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2833                 ext4_msg(sb, KERN_ERR,
2834                          "Can't support bigalloc feature without "
2835                          "extents feature\n");
2836                 return 0;
2837         }
2838
2839 #ifndef CONFIG_QUOTA
2840         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2841             !readonly) {
2842                 ext4_msg(sb, KERN_ERR,
2843                          "Filesystem with quota feature cannot be mounted RDWR "
2844                          "without CONFIG_QUOTA");
2845                 return 0;
2846         }
2847 #endif  /* CONFIG_QUOTA */
2848         return 1;
2849 }
2850
2851 /*
2852  * This function is called once a day if we have errors logged
2853  * on the file system
2854  */
2855 static void print_daily_error_info(unsigned long arg)
2856 {
2857         struct super_block *sb = (struct super_block *) arg;
2858         struct ext4_sb_info *sbi;
2859         struct ext4_super_block *es;
2860
2861         sbi = EXT4_SB(sb);
2862         es = sbi->s_es;
2863
2864         if (es->s_error_count)
2865                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2866                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2867                          le32_to_cpu(es->s_error_count));
2868         if (es->s_first_error_time) {
2869                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2870                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2871                        (int) sizeof(es->s_first_error_func),
2872                        es->s_first_error_func,
2873                        le32_to_cpu(es->s_first_error_line));
2874                 if (es->s_first_error_ino)
2875                         printk(": inode %u",
2876                                le32_to_cpu(es->s_first_error_ino));
2877                 if (es->s_first_error_block)
2878                         printk(": block %llu", (unsigned long long)
2879                                le64_to_cpu(es->s_first_error_block));
2880                 printk("\n");
2881         }
2882         if (es->s_last_error_time) {
2883                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2884                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2885                        (int) sizeof(es->s_last_error_func),
2886                        es->s_last_error_func,
2887                        le32_to_cpu(es->s_last_error_line));
2888                 if (es->s_last_error_ino)
2889                         printk(": inode %u",
2890                                le32_to_cpu(es->s_last_error_ino));
2891                 if (es->s_last_error_block)
2892                         printk(": block %llu", (unsigned long long)
2893                                le64_to_cpu(es->s_last_error_block));
2894                 printk("\n");
2895         }
2896         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2897 }
2898
2899 /* Find next suitable group and run ext4_init_inode_table */
2900 static int ext4_run_li_request(struct ext4_li_request *elr)
2901 {
2902         struct ext4_group_desc *gdp = NULL;
2903         ext4_group_t group, ngroups;
2904         struct super_block *sb;
2905         unsigned long timeout = 0;
2906         int ret = 0;
2907
2908         sb = elr->lr_super;
2909         ngroups = EXT4_SB(sb)->s_groups_count;
2910
2911         sb_start_write(sb);
2912         for (group = elr->lr_next_group; group < ngroups; group++) {
2913                 gdp = ext4_get_group_desc(sb, group, NULL);
2914                 if (!gdp) {
2915                         ret = 1;
2916                         break;
2917                 }
2918
2919                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2920                         break;
2921         }
2922
2923         if (group >= ngroups)
2924                 ret = 1;
2925
2926         if (!ret) {
2927                 timeout = jiffies;
2928                 ret = ext4_init_inode_table(sb, group,
2929                                             elr->lr_timeout ? 0 : 1);
2930                 if (elr->lr_timeout == 0) {
2931                         timeout = (jiffies - timeout) *
2932                                   elr->lr_sbi->s_li_wait_mult;
2933                         elr->lr_timeout = timeout;
2934                 }
2935                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2936                 elr->lr_next_group = group + 1;
2937         }
2938         sb_end_write(sb);
2939
2940         return ret;
2941 }
2942
2943 /*
2944  * Remove lr_request from the list_request and free the
2945  * request structure. Should be called with li_list_mtx held
2946  */
2947 static void ext4_remove_li_request(struct ext4_li_request *elr)
2948 {
2949         struct ext4_sb_info *sbi;
2950
2951         if (!elr)
2952                 return;
2953
2954         sbi = elr->lr_sbi;
2955
2956         list_del(&elr->lr_request);
2957         sbi->s_li_request = NULL;
2958         kfree(elr);
2959 }
2960
2961 static void ext4_unregister_li_request(struct super_block *sb)
2962 {
2963         mutex_lock(&ext4_li_mtx);
2964         if (!ext4_li_info) {
2965                 mutex_unlock(&ext4_li_mtx);
2966                 return;
2967         }
2968
2969         mutex_lock(&ext4_li_info->li_list_mtx);
2970         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2971         mutex_unlock(&ext4_li_info->li_list_mtx);
2972         mutex_unlock(&ext4_li_mtx);
2973 }
2974
2975 static struct task_struct *ext4_lazyinit_task;
2976
2977 /*
2978  * This is the function where ext4lazyinit thread lives. It walks
2979  * through the request list searching for next scheduled filesystem.
2980  * When such a fs is found, run the lazy initialization request
2981  * (ext4_rn_li_request) and keep track of the time spend in this
2982  * function. Based on that time we compute next schedule time of
2983  * the request. When walking through the list is complete, compute
2984  * next waking time and put itself into sleep.
2985  */
2986 static int ext4_lazyinit_thread(void *arg)
2987 {
2988         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2989         struct list_head *pos, *n;
2990         struct ext4_li_request *elr;
2991         unsigned long next_wakeup, cur;
2992
2993         BUG_ON(NULL == eli);
2994
2995 cont_thread:
2996         while (true) {
2997                 next_wakeup = MAX_JIFFY_OFFSET;
2998
2999                 mutex_lock(&eli->li_list_mtx);
3000                 if (list_empty(&eli->li_request_list)) {
3001                         mutex_unlock(&eli->li_list_mtx);
3002                         goto exit_thread;
3003                 }
3004
3005                 list_for_each_safe(pos, n, &eli->li_request_list) {
3006                         elr = list_entry(pos, struct ext4_li_request,
3007                                          lr_request);
3008
3009                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
3010                                 if (ext4_run_li_request(elr) != 0) {
3011                                         /* error, remove the lazy_init job */
3012                                         ext4_remove_li_request(elr);
3013                                         continue;
3014                                 }
3015                         }
3016
3017                         if (time_before(elr->lr_next_sched, next_wakeup))
3018                                 next_wakeup = elr->lr_next_sched;
3019                 }
3020                 mutex_unlock(&eli->li_list_mtx);
3021
3022                 try_to_freeze();
3023
3024                 cur = jiffies;
3025                 if ((time_after_eq(cur, next_wakeup)) ||
3026                     (MAX_JIFFY_OFFSET == next_wakeup)) {
3027                         cond_resched();
3028                         continue;
3029                 }
3030
3031                 schedule_timeout_interruptible(next_wakeup - cur);
3032
3033                 if (kthread_should_stop()) {
3034                         ext4_clear_request_list();
3035                         goto exit_thread;
3036                 }
3037         }
3038
3039 exit_thread:
3040         /*
3041          * It looks like the request list is empty, but we need
3042          * to check it under the li_list_mtx lock, to prevent any
3043          * additions into it, and of course we should lock ext4_li_mtx
3044          * to atomically free the list and ext4_li_info, because at
3045          * this point another ext4 filesystem could be registering
3046          * new one.
3047          */
3048         mutex_lock(&ext4_li_mtx);
3049         mutex_lock(&eli->li_list_mtx);
3050         if (!list_empty(&eli->li_request_list)) {
3051                 mutex_unlock(&eli->li_list_mtx);
3052                 mutex_unlock(&ext4_li_mtx);
3053                 goto cont_thread;
3054         }
3055         mutex_unlock(&eli->li_list_mtx);
3056         kfree(ext4_li_info);
3057         ext4_li_info = NULL;
3058         mutex_unlock(&ext4_li_mtx);
3059
3060         return 0;
3061 }
3062
3063 static void ext4_clear_request_list(void)
3064 {
3065         struct list_head *pos, *n;
3066         struct ext4_li_request *elr;
3067
3068         mutex_lock(&ext4_li_info->li_list_mtx);
3069         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
3070                 elr = list_entry(pos, struct ext4_li_request,
3071                                  lr_request);
3072                 ext4_remove_li_request(elr);
3073         }
3074         mutex_unlock(&ext4_li_info->li_list_mtx);
3075 }
3076
3077 static int ext4_run_lazyinit_thread(void)
3078 {
3079         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
3080                                          ext4_li_info, "ext4lazyinit");
3081         if (IS_ERR(ext4_lazyinit_task)) {
3082                 int err = PTR_ERR(ext4_lazyinit_task);
3083                 ext4_clear_request_list();
3084                 kfree(ext4_li_info);
3085                 ext4_li_info = NULL;
3086                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
3087                                  "initialization thread\n",
3088                                  err);
3089                 return err;
3090         }
3091         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
3092         return 0;
3093 }
3094
3095 /*
3096  * Check whether it make sense to run itable init. thread or not.
3097  * If there is at least one uninitialized inode table, return
3098  * corresponding group number, else the loop goes through all
3099  * groups and return total number of groups.
3100  */
3101 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
3102 {
3103         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
3104         struct ext4_group_desc *gdp = NULL;
3105
3106         for (group = 0; group < ngroups; group++) {
3107                 gdp = ext4_get_group_desc(sb, group, NULL);
3108                 if (!gdp)
3109                         continue;
3110
3111                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
3112                         break;
3113         }
3114
3115         return group;
3116 }
3117
3118 static int ext4_li_info_new(void)
3119 {
3120         struct ext4_lazy_init *eli = NULL;
3121
3122         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
3123         if (!eli)
3124                 return -ENOMEM;
3125
3126         INIT_LIST_HEAD(&eli->li_request_list);
3127         mutex_init(&eli->li_list_mtx);
3128
3129         eli->li_state |= EXT4_LAZYINIT_QUIT;
3130
3131         ext4_li_info = eli;
3132
3133         return 0;
3134 }
3135
3136 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3137                                             ext4_group_t start)
3138 {
3139         struct ext4_sb_info *sbi = EXT4_SB(sb);
3140         struct ext4_li_request *elr;
3141
3142         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3143         if (!elr)
3144                 return NULL;
3145
3146         elr->lr_super = sb;
3147         elr->lr_sbi = sbi;
3148         elr->lr_next_group = start;
3149
3150         /*
3151          * Randomize first schedule time of the request to
3152          * spread the inode table initialization requests
3153          * better.
3154          */
3155         elr->lr_next_sched = jiffies + (prandom_u32() %
3156                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
3157         return elr;
3158 }
3159
3160 int ext4_register_li_request(struct super_block *sb,
3161                              ext4_group_t first_not_zeroed)
3162 {
3163         struct ext4_sb_info *sbi = EXT4_SB(sb);
3164         struct ext4_li_request *elr = NULL;
3165         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3166         int ret = 0;
3167
3168         mutex_lock(&ext4_li_mtx);
3169         if (sbi->s_li_request != NULL) {
3170                 /*
3171                  * Reset timeout so it can be computed again, because
3172                  * s_li_wait_mult might have changed.
3173                  */
3174                 sbi->s_li_request->lr_timeout = 0;
3175                 goto out;
3176         }
3177
3178         if (first_not_zeroed == ngroups ||
3179             (sb->s_flags & MS_RDONLY) ||
3180             !test_opt(sb, INIT_INODE_TABLE))
3181                 goto out;
3182
3183         elr = ext4_li_request_new(sb, first_not_zeroed);
3184         if (!elr) {
3185                 ret = -ENOMEM;
3186                 goto out;
3187         }
3188
3189         if (NULL == ext4_li_info) {
3190                 ret = ext4_li_info_new();
3191                 if (ret)
3192                         goto out;
3193         }
3194
3195         mutex_lock(&ext4_li_info->li_list_mtx);
3196         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3197         mutex_unlock(&ext4_li_info->li_list_mtx);
3198
3199         sbi->s_li_request = elr;
3200         /*
3201          * set elr to NULL here since it has been inserted to
3202          * the request_list and the removal and free of it is
3203          * handled by ext4_clear_request_list from now on.
3204          */
3205         elr = NULL;
3206
3207         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3208                 ret = ext4_run_lazyinit_thread();
3209                 if (ret)
3210                         goto out;
3211         }
3212 out:
3213         mutex_unlock(&ext4_li_mtx);
3214         if (ret)
3215                 kfree(elr);
3216         return ret;
3217 }
3218
3219 /*
3220  * We do not need to lock anything since this is called on
3221  * module unload.
3222  */
3223 static void ext4_destroy_lazyinit_thread(void)
3224 {
3225         /*
3226          * If thread exited earlier
3227          * there's nothing to be done.
3228          */
3229         if (!ext4_li_info || !ext4_lazyinit_task)
3230                 return;
3231
3232         kthread_stop(ext4_lazyinit_task);
3233 }
3234
3235 static int set_journal_csum_feature_set(struct super_block *sb)
3236 {
3237         int ret = 1;
3238         int compat, incompat;
3239         struct ext4_sb_info *sbi = EXT4_SB(sb);
3240
3241         if (ext4_has_metadata_csum(sb)) {
3242                 /* journal checksum v3 */
3243                 compat = 0;
3244                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
3245         } else {
3246                 /* journal checksum v1 */
3247                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3248                 incompat = 0;
3249         }
3250
3251         jbd2_journal_clear_features(sbi->s_journal,
3252                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3253                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
3254                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
3255         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3256                 ret = jbd2_journal_set_features(sbi->s_journal,
3257                                 compat, 0,
3258                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3259                                 incompat);
3260         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3261                 ret = jbd2_journal_set_features(sbi->s_journal,
3262                                 compat, 0,
3263                                 incompat);
3264                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3265                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3266         } else {
3267                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3268                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3269         }
3270
3271         return ret;
3272 }
3273
3274 /*
3275  * Note: calculating the overhead so we can be compatible with
3276  * historical BSD practice is quite difficult in the face of
3277  * clusters/bigalloc.  This is because multiple metadata blocks from
3278  * different block group can end up in the same allocation cluster.
3279  * Calculating the exact overhead in the face of clustered allocation
3280  * requires either O(all block bitmaps) in memory or O(number of block
3281  * groups**2) in time.  We will still calculate the superblock for
3282  * older file systems --- and if we come across with a bigalloc file
3283  * system with zero in s_overhead_clusters the estimate will be close to
3284  * correct especially for very large cluster sizes --- but for newer
3285  * file systems, it's better to calculate this figure once at mkfs
3286  * time, and store it in the superblock.  If the superblock value is
3287  * present (even for non-bigalloc file systems), we will use it.
3288  */
3289 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3290                           char *buf)
3291 {
3292         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3293         struct ext4_group_desc  *gdp;
3294         ext4_fsblk_t            first_block, last_block, b;
3295         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3296         int                     s, j, count = 0;
3297
3298         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3299                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3300                         sbi->s_itb_per_group + 2);
3301
3302         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3303                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3304         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3305         for (i = 0; i < ngroups; i++) {
3306                 gdp = ext4_get_group_desc(sb, i, NULL);
3307                 b = ext4_block_bitmap(sb, gdp);
3308                 if (b >= first_block && b <= last_block) {
3309                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3310                         count++;
3311                 }
3312                 b = ext4_inode_bitmap(sb, gdp);
3313                 if (b >= first_block && b <= last_block) {
3314                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3315                         count++;
3316                 }
3317                 b = ext4_inode_table(sb, gdp);
3318                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3319                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3320                                 int c = EXT4_B2C(sbi, b - first_block);
3321                                 ext4_set_bit(c, buf);
3322                                 count++;
3323                         }
3324                 if (i != grp)
3325                         continue;
3326                 s = 0;
3327                 if (ext4_bg_has_super(sb, grp)) {
3328                         ext4_set_bit(s++, buf);
3329                         count++;
3330                 }
3331                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3332                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3333                         count++;
3334                 }
3335         }
3336         if (!count)
3337                 return 0;
3338         return EXT4_CLUSTERS_PER_GROUP(sb) -
3339                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3340 }
3341
3342 /*
3343  * Compute the overhead and stash it in sbi->s_overhead
3344  */
3345 int ext4_calculate_overhead(struct super_block *sb)
3346 {
3347         struct ext4_sb_info *sbi = EXT4_SB(sb);
3348         struct ext4_super_block *es = sbi->s_es;
3349         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3350         ext4_fsblk_t overhead = 0;
3351         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3352
3353         if (!buf)
3354                 return -ENOMEM;
3355
3356         /*
3357          * Compute the overhead (FS structures).  This is constant
3358          * for a given filesystem unless the number of block groups
3359          * changes so we cache the previous value until it does.
3360          */
3361
3362         /*
3363          * All of the blocks before first_data_block are overhead
3364          */
3365         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3366
3367         /*
3368          * Add the overhead found in each block group
3369          */
3370         for (i = 0; i < ngroups; i++) {
3371                 int blks;
3372
3373                 blks = count_overhead(sb, i, buf);
3374                 overhead += blks;
3375                 if (blks)
3376                         memset(buf, 0, PAGE_SIZE);
3377                 cond_resched();
3378         }
3379         /* Add the internal journal blocks as well */
3380         if (sbi->s_journal && !sbi->journal_bdev)
3381                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3382
3383         sbi->s_overhead = overhead;
3384         smp_wmb();
3385         free_page((unsigned long) buf);
3386         return 0;
3387 }
3388
3389
3390 static ext4_fsblk_t ext4_calculate_resv_clusters(struct super_block *sb)
3391 {
3392         ext4_fsblk_t resv_clusters;
3393
3394         /*
3395          * There's no need to reserve anything when we aren't using extents.
3396          * The space estimates are exact, there are no unwritten extents,
3397          * hole punching doesn't need new metadata... This is needed especially
3398          * to keep ext2/3 backward compatibility.
3399          */
3400         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3401                 return 0;
3402         /*
3403          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3404          * This should cover the situations where we can not afford to run
3405          * out of space like for example punch hole, or converting
3406          * unwritten extents in delalloc path. In most cases such
3407          * allocation would require 1, or 2 blocks, higher numbers are
3408          * very rare.
3409          */
3410         resv_clusters = ext4_blocks_count(EXT4_SB(sb)->s_es) >>
3411                         EXT4_SB(sb)->s_cluster_bits;
3412
3413         do_div(resv_clusters, 50);
3414         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3415
3416         return resv_clusters;
3417 }
3418
3419
3420 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3421 {
3422         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3423                                 sbi->s_cluster_bits;
3424
3425         if (count >= clusters)
3426                 return -EINVAL;
3427
3428         atomic64_set(&sbi->s_resv_clusters, count);
3429         return 0;
3430 }
3431
3432 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3433 {
3434         char *orig_data = kstrdup(data, GFP_KERNEL);
3435         struct buffer_head *bh;
3436         struct ext4_super_block *es = NULL;
3437         struct ext4_sb_info *sbi;
3438         ext4_fsblk_t block;
3439         ext4_fsblk_t sb_block = get_sb_block(&data);
3440         ext4_fsblk_t logical_sb_block;
3441         unsigned long offset = 0;
3442         unsigned long journal_devnum = 0;
3443         unsigned long def_mount_opts;
3444         struct inode *root;
3445         const char *descr;
3446         int ret = -ENOMEM;
3447         int blocksize, clustersize;
3448         unsigned int db_count;
3449         unsigned int i;
3450         int needs_recovery, has_huge_files, has_bigalloc;
3451         __u64 blocks_count;
3452         int err = 0;
3453         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3454         ext4_group_t first_not_zeroed;
3455
3456         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3457         if (!sbi)
3458                 goto out_free_orig;
3459
3460         sbi->s_blockgroup_lock =
3461                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3462         if (!sbi->s_blockgroup_lock) {
3463                 kfree(sbi);
3464                 goto out_free_orig;
3465         }
3466         sb->s_fs_info = sbi;
3467         sbi->s_sb = sb;
3468         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3469         sbi->s_sb_block = sb_block;
3470         if (sb->s_bdev->bd_part)
3471                 sbi->s_sectors_written_start =
3472                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3473
3474         /* Cleanup superblock name */
3475         strreplace(sb->s_id, '/', '!');
3476
3477         /* -EINVAL is default */
3478         ret = -EINVAL;
3479         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3480         if (!blocksize) {
3481                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3482                 goto out_fail;
3483         }
3484
3485         /*
3486          * The ext4 superblock will not be buffer aligned for other than 1kB
3487          * block sizes.  We need to calculate the offset from buffer start.
3488          */
3489         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3490                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3491                 offset = do_div(logical_sb_block, blocksize);
3492         } else {
3493                 logical_sb_block = sb_block;
3494         }
3495
3496         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3497                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3498                 goto out_fail;
3499         }
3500         /*
3501          * Note: s_es must be initialized as soon as possible because
3502          *       some ext4 macro-instructions depend on its value
3503          */
3504         es = (struct ext4_super_block *) (bh->b_data + offset);
3505         sbi->s_es = es;
3506         sb->s_magic = le16_to_cpu(es->s_magic);
3507         if (sb->s_magic != EXT4_SUPER_MAGIC)
3508                 goto cantfind_ext4;
3509         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3510
3511         /* Warn if metadata_csum and gdt_csum are both set. */
3512         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3513                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3514             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3515                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3516                              "redundant flags; please run fsck.");
3517
3518         /* Check for a known checksum algorithm */
3519         if (!ext4_verify_csum_type(sb, es)) {
3520                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3521                          "unknown checksum algorithm.");
3522                 silent = 1;
3523                 goto cantfind_ext4;
3524         }
3525
3526         /* Load the checksum driver */
3527         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3528                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3529                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3530                 if (IS_ERR(sbi->s_chksum_driver)) {
3531                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3532                         ret = PTR_ERR(sbi->s_chksum_driver);
3533                         sbi->s_chksum_driver = NULL;
3534                         goto failed_mount;
3535                 }
3536         }
3537
3538         /* Check superblock checksum */
3539         if (!ext4_superblock_csum_verify(sb, es)) {
3540                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3541                          "invalid superblock checksum.  Run e2fsck?");
3542                 silent = 1;
3543                 goto cantfind_ext4;
3544         }
3545
3546         /* Precompute checksum seed for all metadata */
3547         if (ext4_has_metadata_csum(sb))
3548                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3549                                                sizeof(es->s_uuid));
3550
3551         /* Set defaults before we parse the mount options */
3552         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3553         set_opt(sb, INIT_INODE_TABLE);
3554         if (def_mount_opts & EXT4_DEFM_DEBUG)
3555                 set_opt(sb, DEBUG);
3556         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3557                 set_opt(sb, GRPID);
3558         if (def_mount_opts & EXT4_DEFM_UID16)
3559                 set_opt(sb, NO_UID32);
3560         /* xattr user namespace & acls are now defaulted on */
3561         set_opt(sb, XATTR_USER);
3562 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3563         set_opt(sb, POSIX_ACL);
3564 #endif
3565         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3566         if (ext4_has_metadata_csum(sb))
3567                 set_opt(sb, JOURNAL_CHECKSUM);
3568
3569         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3570                 set_opt(sb, JOURNAL_DATA);
3571         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3572                 set_opt(sb, ORDERED_DATA);
3573         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3574                 set_opt(sb, WRITEBACK_DATA);
3575
3576         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3577                 set_opt(sb, ERRORS_PANIC);
3578         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3579                 set_opt(sb, ERRORS_CONT);
3580         else
3581                 set_opt(sb, ERRORS_RO);
3582         /* block_validity enabled by default; disable with noblock_validity */
3583         set_opt(sb, BLOCK_VALIDITY);
3584         if (def_mount_opts & EXT4_DEFM_DISCARD)
3585                 set_opt(sb, DISCARD);
3586
3587         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3588         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3589         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3590         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3591         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3592
3593         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3594                 set_opt(sb, BARRIER);
3595
3596         /*
3597          * enable delayed allocation by default
3598          * Use -o nodelalloc to turn it off
3599          */
3600         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3601             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3602                 set_opt(sb, DELALLOC);
3603
3604         /*
3605          * set default s_li_wait_mult for lazyinit, for the case there is
3606          * no mount option specified.
3607          */
3608         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3609
3610         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3611                            &journal_devnum, &journal_ioprio, 0)) {
3612                 ext4_msg(sb, KERN_WARNING,
3613                          "failed to parse options in superblock: %s",
3614                          sbi->s_es->s_mount_opts);
3615         }
3616         sbi->s_def_mount_opt = sbi->s_mount_opt;
3617         if (!parse_options((char *) data, sb, &journal_devnum,
3618                            &journal_ioprio, 0))
3619                 goto failed_mount;
3620
3621         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3622                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3623                             "with data=journal disables delayed "
3624                             "allocation and O_DIRECT support!\n");
3625                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3626                         ext4_msg(sb, KERN_ERR, "can't mount with "
3627                                  "both data=journal and delalloc");
3628                         goto failed_mount;
3629                 }
3630                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3631                         ext4_msg(sb, KERN_ERR, "can't mount with "
3632                                  "both data=journal and dioread_nolock");
3633                         goto failed_mount;
3634                 }
3635                 if (test_opt(sb, DAX)) {
3636                         ext4_msg(sb, KERN_ERR, "can't mount with "
3637                                  "both data=journal and dax");
3638                         goto failed_mount;
3639                 }
3640                 if (test_opt(sb, DELALLOC))
3641                         clear_opt(sb, DELALLOC);
3642         }
3643
3644         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3645                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3646
3647         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3648             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3649              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3650              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3651                 ext4_msg(sb, KERN_WARNING,
3652                        "feature flags set on rev 0 fs, "
3653                        "running e2fsck is recommended");
3654
3655         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3656                 set_opt2(sb, HURD_COMPAT);
3657                 if (EXT4_HAS_INCOMPAT_FEATURE(sb,
3658                                               EXT4_FEATURE_INCOMPAT_64BIT)) {
3659                         ext4_msg(sb, KERN_ERR,
3660                                  "The Hurd can't support 64-bit file systems");
3661                         goto failed_mount;
3662                 }
3663         }
3664
3665         if (IS_EXT2_SB(sb)) {
3666                 if (ext2_feature_set_ok(sb))
3667                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3668                                  "using the ext4 subsystem");
3669                 else {
3670                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3671                                  "to feature incompatibilities");
3672                         goto failed_mount;
3673                 }
3674         }
3675
3676         if (IS_EXT3_SB(sb)) {
3677                 if (ext3_feature_set_ok(sb))
3678                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3679                                  "using the ext4 subsystem");
3680                 else {
3681                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3682                                  "to feature incompatibilities");
3683                         goto failed_mount;
3684                 }
3685         }
3686
3687         /*
3688          * Check feature flags regardless of the revision level, since we
3689          * previously didn't change the revision level when setting the flags,
3690          * so there is a chance incompat flags are set on a rev 0 filesystem.
3691          */
3692         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3693                 goto failed_mount;
3694
3695         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3696         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3697             blocksize > EXT4_MAX_BLOCK_SIZE) {
3698                 ext4_msg(sb, KERN_ERR,
3699                        "Unsupported filesystem blocksize %d", blocksize);
3700                 goto failed_mount;
3701         }
3702
3703         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3704                 if (blocksize != PAGE_SIZE) {
3705                         ext4_msg(sb, KERN_ERR,
3706                                         "error: unsupported blocksize for dax");
3707                         goto failed_mount;
3708                 }
3709                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3710                         ext4_msg(sb, KERN_ERR,
3711                                         "error: device does not support dax");
3712                         goto failed_mount;
3713                 }
3714         }
3715
3716         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_ENCRYPT) &&
3717             es->s_encryption_level) {
3718                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3719                          es->s_encryption_level);
3720                 goto failed_mount;
3721         }
3722
3723         if (sb->s_blocksize != blocksize) {
3724                 /* Validate the filesystem blocksize */
3725                 if (!sb_set_blocksize(sb, blocksize)) {
3726                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3727                                         blocksize);
3728                         goto failed_mount;
3729                 }
3730
3731                 brelse(bh);
3732                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3733                 offset = do_div(logical_sb_block, blocksize);
3734                 bh = sb_bread_unmovable(sb, logical_sb_block);
3735                 if (!bh) {
3736                         ext4_msg(sb, KERN_ERR,
3737                                "Can't read superblock on 2nd try");
3738                         goto failed_mount;
3739                 }
3740                 es = (struct ext4_super_block *)(bh->b_data + offset);
3741                 sbi->s_es = es;
3742                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3743                         ext4_msg(sb, KERN_ERR,
3744                                "Magic mismatch, very weird!");
3745                         goto failed_mount;
3746                 }
3747         }
3748
3749         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3750                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3751         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3752                                                       has_huge_files);
3753         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3754
3755         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3756                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3757                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3758         } else {
3759                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3760                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3761                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3762                     (!is_power_of_2(sbi->s_inode_size)) ||
3763                     (sbi->s_inode_size > blocksize)) {
3764                         ext4_msg(sb, KERN_ERR,
3765                                "unsupported inode size: %d",
3766                                sbi->s_inode_size);
3767                         goto failed_mount;
3768                 }
3769                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3770                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3771         }
3772
3773         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3774         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3775                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3776                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3777                     !is_power_of_2(sbi->s_desc_size)) {
3778                         ext4_msg(sb, KERN_ERR,
3779                                "unsupported descriptor size %lu",
3780                                sbi->s_desc_size);
3781                         goto failed_mount;
3782                 }
3783         } else
3784                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3785
3786         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3787         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3788         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3789                 goto cantfind_ext4;
3790
3791         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3792         if (sbi->s_inodes_per_block == 0)
3793                 goto cantfind_ext4;
3794         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3795                                         sbi->s_inodes_per_block;
3796         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3797         sbi->s_sbh = bh;
3798         sbi->s_mount_state = le16_to_cpu(es->s_state);
3799         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3800         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3801
3802         for (i = 0; i < 4; i++)
3803                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3804         sbi->s_def_hash_version = es->s_def_hash_version;
3805         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
3806                 i = le32_to_cpu(es->s_flags);
3807                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3808                         sbi->s_hash_unsigned = 3;
3809                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3810 #ifdef __CHAR_UNSIGNED__
3811                         if (!(sb->s_flags & MS_RDONLY))
3812                                 es->s_flags |=
3813                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3814                         sbi->s_hash_unsigned = 3;
3815 #else
3816                         if (!(sb->s_flags & MS_RDONLY))
3817                                 es->s_flags |=
3818                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3819 #endif
3820                 }
3821         }
3822
3823         /* Handle clustersize */
3824         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3825         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3826                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3827         if (has_bigalloc) {
3828                 if (clustersize < blocksize) {
3829                         ext4_msg(sb, KERN_ERR,
3830                                  "cluster size (%d) smaller than "
3831                                  "block size (%d)", clustersize, blocksize);
3832                         goto failed_mount;
3833                 }
3834                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3835                         le32_to_cpu(es->s_log_block_size);
3836                 sbi->s_clusters_per_group =
3837                         le32_to_cpu(es->s_clusters_per_group);
3838                 if (sbi->s_clusters_per_group > blocksize * 8) {
3839                         ext4_msg(sb, KERN_ERR,
3840                                  "#clusters per group too big: %lu",
3841                                  sbi->s_clusters_per_group);
3842                         goto failed_mount;
3843                 }
3844                 if (sbi->s_blocks_per_group !=
3845                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3846                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3847                                  "clusters per group (%lu) inconsistent",
3848                                  sbi->s_blocks_per_group,
3849                                  sbi->s_clusters_per_group);
3850                         goto failed_mount;
3851                 }
3852         } else {
3853                 if (clustersize != blocksize) {
3854                         ext4_warning(sb, "fragment/cluster size (%d) != "
3855                                      "block size (%d)", clustersize,
3856                                      blocksize);
3857                         clustersize = blocksize;
3858                 }
3859                 if (sbi->s_blocks_per_group > blocksize * 8) {
3860                         ext4_msg(sb, KERN_ERR,
3861                                  "#blocks per group too big: %lu",
3862                                  sbi->s_blocks_per_group);
3863                         goto failed_mount;
3864                 }
3865                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3866                 sbi->s_cluster_bits = 0;
3867         }
3868         sbi->s_cluster_ratio = clustersize / blocksize;
3869
3870         if (sbi->s_inodes_per_group > blocksize * 8) {
3871                 ext4_msg(sb, KERN_ERR,
3872                        "#inodes per group too big: %lu",
3873                        sbi->s_inodes_per_group);
3874                 goto failed_mount;
3875         }
3876
3877         /* Do we have standard group size of clustersize * 8 blocks ? */
3878         if (sbi->s_blocks_per_group == clustersize << 3)
3879                 set_opt2(sb, STD_GROUP_SIZE);
3880
3881         /*
3882          * Test whether we have more sectors than will fit in sector_t,
3883          * and whether the max offset is addressable by the page cache.
3884          */
3885         err = generic_check_addressable(sb->s_blocksize_bits,
3886                                         ext4_blocks_count(es));
3887         if (err) {
3888                 ext4_msg(sb, KERN_ERR, "filesystem"
3889                          " too large to mount safely on this system");
3890                 if (sizeof(sector_t) < 8)
3891                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3892                 goto failed_mount;
3893         }
3894
3895         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3896                 goto cantfind_ext4;
3897
3898         /* check blocks count against device size */
3899         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3900         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3901                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3902                        "exceeds size of device (%llu blocks)",
3903                        ext4_blocks_count(es), blocks_count);
3904                 goto failed_mount;
3905         }
3906
3907         /*
3908          * It makes no sense for the first data block to be beyond the end
3909          * of the filesystem.
3910          */
3911         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3912                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3913                          "block %u is beyond end of filesystem (%llu)",
3914                          le32_to_cpu(es->s_first_data_block),
3915                          ext4_blocks_count(es));
3916                 goto failed_mount;
3917         }
3918         blocks_count = (ext4_blocks_count(es) -
3919                         le32_to_cpu(es->s_first_data_block) +
3920                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3921         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3922         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3923                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3924                        "(block count %llu, first data block %u, "
3925                        "blocks per group %lu)", sbi->s_groups_count,
3926                        ext4_blocks_count(es),
3927                        le32_to_cpu(es->s_first_data_block),
3928                        EXT4_BLOCKS_PER_GROUP(sb));
3929                 goto failed_mount;
3930         }
3931         sbi->s_groups_count = blocks_count;
3932         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3933                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3934         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3935                    EXT4_DESC_PER_BLOCK(sb);
3936         sbi->s_group_desc = ext4_kvmalloc(db_count *
3937                                           sizeof(struct buffer_head *),
3938                                           GFP_KERNEL);
3939         if (sbi->s_group_desc == NULL) {
3940                 ext4_msg(sb, KERN_ERR, "not enough memory");
3941                 ret = -ENOMEM;
3942                 goto failed_mount;
3943         }
3944
3945         if (ext4_proc_root)
3946                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3947
3948         if (sbi->s_proc)
3949                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3950                                  &ext4_seq_options_fops, sb);
3951
3952         bgl_lock_init(sbi->s_blockgroup_lock);
3953
3954         for (i = 0; i < db_count; i++) {
3955                 block = descriptor_loc(sb, logical_sb_block, i);
3956                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3957                 if (!sbi->s_group_desc[i]) {
3958                         ext4_msg(sb, KERN_ERR,
3959                                "can't read group descriptor %d", i);
3960                         db_count = i;
3961                         goto failed_mount2;
3962                 }
3963         }
3964         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3965                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3966                 goto failed_mount2;
3967         }
3968
3969         sbi->s_gdb_count = db_count;
3970         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3971         spin_lock_init(&sbi->s_next_gen_lock);
3972
3973         setup_timer(&sbi->s_err_report, print_daily_error_info,
3974                 (unsigned long) sb);
3975
3976         /* Register extent status tree shrinker */
3977         if (ext4_es_register_shrinker(sbi))
3978                 goto failed_mount3;
3979
3980         sbi->s_stripe = ext4_get_stripe_size(sbi);
3981         sbi->s_extent_max_zeroout_kb = 32;
3982
3983         /*
3984          * set up enough so that it can read an inode
3985          */
3986         sb->s_op = &ext4_sops;
3987         sb->s_export_op = &ext4_export_ops;
3988         sb->s_xattr = ext4_xattr_handlers;
3989 #ifdef CONFIG_QUOTA
3990         sb->dq_op = &ext4_quota_operations;
3991         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3992                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3993         else
3994                 sb->s_qcop = &ext4_qctl_operations;
3995         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
3996 #endif
3997         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3998
3999         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
4000         mutex_init(&sbi->s_orphan_lock);
4001
4002         sb->s_root = NULL;
4003
4004         needs_recovery = (es->s_last_orphan != 0 ||
4005                           EXT4_HAS_INCOMPAT_FEATURE(sb,
4006                                     EXT4_FEATURE_INCOMPAT_RECOVER));
4007
4008         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
4009             !(sb->s_flags & MS_RDONLY))
4010                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
4011                         goto failed_mount3a;
4012
4013         /*
4014          * The first inode we look at is the journal inode.  Don't try
4015          * root first: it may be modified in the journal!
4016          */
4017         if (!test_opt(sb, NOLOAD) &&
4018             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4019                 if (ext4_load_journal(sb, es, journal_devnum))
4020                         goto failed_mount3a;
4021         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
4022               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4023                 ext4_msg(sb, KERN_ERR, "required journal recovery "
4024                        "suppressed and not mounted read-only");
4025                 goto failed_mount_wq;
4026         } else {
4027                 clear_opt(sb, DATA_FLAGS);
4028                 sbi->s_journal = NULL;
4029                 needs_recovery = 0;
4030                 goto no_journal;
4031         }
4032
4033         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
4034             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
4035                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
4036                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
4037                 goto failed_mount_wq;
4038         }
4039
4040         if (!set_journal_csum_feature_set(sb)) {
4041                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
4042                          "feature set");
4043                 goto failed_mount_wq;
4044         }
4045
4046         /* We have now updated the journal if required, so we can
4047          * validate the data journaling mode. */
4048         switch (test_opt(sb, DATA_FLAGS)) {
4049         case 0:
4050                 /* No mode set, assume a default based on the journal
4051                  * capabilities: ORDERED_DATA if the journal can
4052                  * cope, else JOURNAL_DATA
4053                  */
4054                 if (jbd2_journal_check_available_features
4055                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
4056                         set_opt(sb, ORDERED_DATA);
4057                 else
4058                         set_opt(sb, JOURNAL_DATA);
4059                 break;
4060
4061         case EXT4_MOUNT_ORDERED_DATA:
4062         case EXT4_MOUNT_WRITEBACK_DATA:
4063                 if (!jbd2_journal_check_available_features
4064                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
4065                         ext4_msg(sb, KERN_ERR, "Journal does not support "
4066                                "requested data journaling mode");
4067                         goto failed_mount_wq;
4068                 }
4069         default:
4070                 break;
4071         }
4072         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4073
4074         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
4075
4076 no_journal:
4077         if (ext4_mballoc_ready) {
4078                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
4079                 if (!sbi->s_mb_cache) {
4080                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
4081                         goto failed_mount_wq;
4082                 }
4083         }
4084
4085         if ((DUMMY_ENCRYPTION_ENABLED(sbi) ||
4086              EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_ENCRYPT)) &&
4087             (blocksize != PAGE_CACHE_SIZE)) {
4088                 ext4_msg(sb, KERN_ERR,
4089                          "Unsupported blocksize for fs encryption");
4090                 goto failed_mount_wq;
4091         }
4092
4093         if (DUMMY_ENCRYPTION_ENABLED(sbi) &&
4094             !(sb->s_flags & MS_RDONLY) &&
4095             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_ENCRYPT)) {
4096                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_ENCRYPT);
4097                 ext4_commit_super(sb, 1);
4098         }
4099
4100         /*
4101          * Get the # of file system overhead blocks from the
4102          * superblock if present.
4103          */
4104         if (es->s_overhead_clusters)
4105                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
4106         else {
4107                 err = ext4_calculate_overhead(sb);
4108                 if (err)
4109                         goto failed_mount_wq;
4110         }
4111
4112         /*
4113          * The maximum number of concurrent works can be high and
4114          * concurrency isn't really necessary.  Limit it to 1.
4115          */
4116         EXT4_SB(sb)->rsv_conversion_wq =
4117                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4118         if (!EXT4_SB(sb)->rsv_conversion_wq) {
4119                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
4120                 ret = -ENOMEM;
4121                 goto failed_mount4;
4122         }
4123
4124         /*
4125          * The jbd2_journal_load will have done any necessary log recovery,
4126          * so we can safely mount the rest of the filesystem now.
4127          */
4128
4129         root = ext4_iget(sb, EXT4_ROOT_INO);
4130         if (IS_ERR(root)) {
4131                 ext4_msg(sb, KERN_ERR, "get root inode failed");
4132                 ret = PTR_ERR(root);
4133                 root = NULL;
4134                 goto failed_mount4;
4135         }
4136         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
4137                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
4138                 iput(root);
4139                 goto failed_mount4;
4140         }
4141         sb->s_root = d_make_root(root);
4142         if (!sb->s_root) {
4143                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
4144                 ret = -ENOMEM;
4145                 goto failed_mount4;
4146         }
4147
4148         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
4149                 sb->s_flags |= MS_RDONLY;
4150
4151         /* determine the minimum size of new large inodes, if present */
4152         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4153                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4154                                                      EXT4_GOOD_OLD_INODE_SIZE;
4155                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4156                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
4157                         if (sbi->s_want_extra_isize <
4158                             le16_to_cpu(es->s_want_extra_isize))
4159                                 sbi->s_want_extra_isize =
4160                                         le16_to_cpu(es->s_want_extra_isize);
4161                         if (sbi->s_want_extra_isize <
4162                             le16_to_cpu(es->s_min_extra_isize))
4163                                 sbi->s_want_extra_isize =
4164                                         le16_to_cpu(es->s_min_extra_isize);
4165                 }
4166         }
4167         /* Check if enough inode space is available */
4168         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4169                                                         sbi->s_inode_size) {
4170                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4171                                                        EXT4_GOOD_OLD_INODE_SIZE;
4172                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4173                          "available");
4174         }
4175
4176         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sb));
4177         if (err) {
4178                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
4179                          "reserved pool", ext4_calculate_resv_clusters(sb));
4180                 goto failed_mount4a;
4181         }
4182
4183         err = ext4_setup_system_zone(sb);
4184         if (err) {
4185                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4186                          "zone (%d)", err);
4187                 goto failed_mount4a;
4188         }
4189
4190         ext4_ext_init(sb);
4191         err = ext4_mb_init(sb);
4192         if (err) {
4193                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4194                          err);
4195                 goto failed_mount5;
4196         }
4197
4198         block = ext4_count_free_clusters(sb);
4199         ext4_free_blocks_count_set(sbi->s_es,
4200                                    EXT4_C2B(sbi, block));
4201         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
4202                                   GFP_KERNEL);
4203         if (!err) {
4204                 unsigned long freei = ext4_count_free_inodes(sb);
4205                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
4206                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
4207                                           GFP_KERNEL);
4208         }
4209         if (!err)
4210                 err = percpu_counter_init(&sbi->s_dirs_counter,
4211                                           ext4_count_dirs(sb), GFP_KERNEL);
4212         if (!err)
4213                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
4214                                           GFP_KERNEL);
4215         if (err) {
4216                 ext4_msg(sb, KERN_ERR, "insufficient memory");
4217                 goto failed_mount6;
4218         }
4219
4220         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
4221                 if (!ext4_fill_flex_info(sb)) {
4222                         ext4_msg(sb, KERN_ERR,
4223                                "unable to initialize "
4224                                "flex_bg meta info!");
4225                         goto failed_mount6;
4226                 }
4227
4228         err = ext4_register_li_request(sb, first_not_zeroed);
4229         if (err)
4230                 goto failed_mount6;
4231
4232         sbi->s_kobj.kset = ext4_kset;
4233         init_completion(&sbi->s_kobj_unregister);
4234         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4235                                    "%s", sb->s_id);
4236         if (err)
4237                 goto failed_mount7;
4238
4239 #ifdef CONFIG_QUOTA
4240         /* Enable quota usage during mount. */
4241         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4242             !(sb->s_flags & MS_RDONLY)) {
4243                 err = ext4_enable_quotas(sb);
4244                 if (err)
4245                         goto failed_mount8;
4246         }
4247 #endif  /* CONFIG_QUOTA */
4248
4249         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4250         ext4_orphan_cleanup(sb, es);
4251         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4252         if (needs_recovery) {
4253                 ext4_msg(sb, KERN_INFO, "recovery complete");
4254                 ext4_mark_recovery_complete(sb, es);
4255         }
4256         if (EXT4_SB(sb)->s_journal) {
4257                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4258                         descr = " journalled data mode";
4259                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4260                         descr = " ordered data mode";
4261                 else
4262                         descr = " writeback data mode";
4263         } else
4264                 descr = "out journal";
4265
4266         if (test_opt(sb, DISCARD)) {
4267                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4268                 if (!blk_queue_discard(q))
4269                         ext4_msg(sb, KERN_WARNING,
4270                                  "mounting with \"discard\" option, but "
4271                                  "the device does not support discard");
4272         }
4273
4274         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4275                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4276                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4277
4278         if (es->s_error_count)
4279                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4280
4281         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4282         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4283         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4284         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4285
4286         kfree(orig_data);
4287         return 0;
4288
4289 cantfind_ext4:
4290         if (!silent)
4291                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4292         goto failed_mount;
4293
4294 #ifdef CONFIG_QUOTA
4295 failed_mount8:
4296         kobject_del(&sbi->s_kobj);
4297 #endif
4298 failed_mount7:
4299         ext4_unregister_li_request(sb);
4300 failed_mount6:
4301         ext4_mb_release(sb);
4302         if (sbi->s_flex_groups)
4303                 kvfree(sbi->s_flex_groups);
4304         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4305         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4306         percpu_counter_destroy(&sbi->s_dirs_counter);
4307         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4308 failed_mount5:
4309         ext4_ext_release(sb);
4310         ext4_release_system_zone(sb);
4311 failed_mount4a:
4312         dput(sb->s_root);
4313         sb->s_root = NULL;
4314 failed_mount4:
4315         ext4_msg(sb, KERN_ERR, "mount failed");
4316         if (EXT4_SB(sb)->rsv_conversion_wq)
4317                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4318 failed_mount_wq:
4319         if (sbi->s_journal) {
4320                 jbd2_journal_destroy(sbi->s_journal);
4321                 sbi->s_journal = NULL;
4322         }
4323 failed_mount3a:
4324         ext4_es_unregister_shrinker(sbi);
4325 failed_mount3:
4326         del_timer_sync(&sbi->s_err_report);
4327         if (sbi->s_mmp_tsk)
4328                 kthread_stop(sbi->s_mmp_tsk);
4329 failed_mount2:
4330         for (i = 0; i < db_count; i++)
4331                 brelse(sbi->s_group_desc[i]);
4332         kvfree(sbi->s_group_desc);
4333 failed_mount:
4334         if (sbi->s_chksum_driver)
4335                 crypto_free_shash(sbi->s_chksum_driver);
4336         if (sbi->s_proc) {
4337                 remove_proc_entry("options", sbi->s_proc);
4338                 remove_proc_entry(sb->s_id, ext4_proc_root);
4339         }
4340 #ifdef CONFIG_QUOTA
4341         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4342                 kfree(sbi->s_qf_names[i]);
4343 #endif
4344         ext4_blkdev_remove(sbi);
4345         brelse(bh);
4346 out_fail:
4347         sb->s_fs_info = NULL;
4348         kfree(sbi->s_blockgroup_lock);
4349         kfree(sbi);
4350 out_free_orig:
4351         kfree(orig_data);
4352         return err ? err : ret;
4353 }
4354
4355 /*
4356  * Setup any per-fs journal parameters now.  We'll do this both on
4357  * initial mount, once the journal has been initialised but before we've
4358  * done any recovery; and again on any subsequent remount.
4359  */
4360 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4361 {
4362         struct ext4_sb_info *sbi = EXT4_SB(sb);
4363
4364         journal->j_commit_interval = sbi->s_commit_interval;
4365         journal->j_min_batch_time = sbi->s_min_batch_time;
4366         journal->j_max_batch_time = sbi->s_max_batch_time;
4367
4368         write_lock(&journal->j_state_lock);
4369         if (test_opt(sb, BARRIER))
4370                 journal->j_flags |= JBD2_BARRIER;
4371         else
4372                 journal->j_flags &= ~JBD2_BARRIER;
4373         if (test_opt(sb, DATA_ERR_ABORT))
4374                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4375         else
4376                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4377         write_unlock(&journal->j_state_lock);
4378 }
4379
4380 static journal_t *ext4_get_journal(struct super_block *sb,
4381                                    unsigned int journal_inum)
4382 {
4383         struct inode *journal_inode;
4384         journal_t *journal;
4385
4386         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4387
4388         /* First, test for the existence of a valid inode on disk.  Bad
4389          * things happen if we iget() an unused inode, as the subsequent
4390          * iput() will try to delete it. */
4391
4392         journal_inode = ext4_iget(sb, journal_inum);
4393         if (IS_ERR(journal_inode)) {
4394                 ext4_msg(sb, KERN_ERR, "no journal found");
4395                 return NULL;
4396         }
4397         if (!journal_inode->i_nlink) {
4398                 make_bad_inode(journal_inode);
4399                 iput(journal_inode);
4400                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4401                 return NULL;
4402         }
4403
4404         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4405                   journal_inode, journal_inode->i_size);
4406         if (!S_ISREG(journal_inode->i_mode)) {
4407                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4408                 iput(journal_inode);
4409                 return NULL;
4410         }
4411
4412         journal = jbd2_journal_init_inode(journal_inode);
4413         if (!journal) {
4414                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4415                 iput(journal_inode);
4416                 return NULL;
4417         }
4418         journal->j_private = sb;
4419         ext4_init_journal_params(sb, journal);
4420         return journal;
4421 }
4422
4423 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4424                                        dev_t j_dev)
4425 {
4426         struct buffer_head *bh;
4427         journal_t *journal;
4428         ext4_fsblk_t start;
4429         ext4_fsblk_t len;
4430         int hblock, blocksize;
4431         ext4_fsblk_t sb_block;
4432         unsigned long offset;
4433         struct ext4_super_block *es;
4434         struct block_device *bdev;
4435
4436         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4437
4438         bdev = ext4_blkdev_get(j_dev, sb);
4439         if (bdev == NULL)
4440                 return NULL;
4441
4442         blocksize = sb->s_blocksize;
4443         hblock = bdev_logical_block_size(bdev);
4444         if (blocksize < hblock) {
4445                 ext4_msg(sb, KERN_ERR,
4446                         "blocksize too small for journal device");
4447                 goto out_bdev;
4448         }
4449
4450         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4451         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4452         set_blocksize(bdev, blocksize);
4453         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4454                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4455                        "external journal");
4456                 goto out_bdev;
4457         }
4458
4459         es = (struct ext4_super_block *) (bh->b_data + offset);
4460         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4461             !(le32_to_cpu(es->s_feature_incompat) &
4462               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4463                 ext4_msg(sb, KERN_ERR, "external journal has "
4464                                         "bad superblock");
4465                 brelse(bh);
4466                 goto out_bdev;
4467         }
4468
4469         if ((le32_to_cpu(es->s_feature_ro_compat) &
4470              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4471             es->s_checksum != ext4_superblock_csum(sb, es)) {
4472                 ext4_msg(sb, KERN_ERR, "external journal has "
4473                                        "corrupt superblock");
4474                 brelse(bh);
4475                 goto out_bdev;
4476         }
4477
4478         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4479                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4480                 brelse(bh);
4481                 goto out_bdev;
4482         }
4483
4484         len = ext4_blocks_count(es);
4485         start = sb_block + 1;
4486         brelse(bh);     /* we're done with the superblock */
4487
4488         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4489                                         start, len, blocksize);
4490         if (!journal) {
4491                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4492                 goto out_bdev;
4493         }
4494         journal->j_private = sb;
4495         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4496         wait_on_buffer(journal->j_sb_buffer);
4497         if (!buffer_uptodate(journal->j_sb_buffer)) {
4498                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4499                 goto out_journal;
4500         }
4501         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4502                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4503                                         "user (unsupported) - %d",
4504                         be32_to_cpu(journal->j_superblock->s_nr_users));
4505                 goto out_journal;
4506         }
4507         EXT4_SB(sb)->journal_bdev = bdev;
4508         ext4_init_journal_params(sb, journal);
4509         return journal;
4510
4511 out_journal:
4512         jbd2_journal_destroy(journal);
4513 out_bdev:
4514         ext4_blkdev_put(bdev);
4515         return NULL;
4516 }
4517
4518 static int ext4_load_journal(struct super_block *sb,
4519                              struct ext4_super_block *es,
4520                              unsigned long journal_devnum)
4521 {
4522         journal_t *journal;
4523         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4524         dev_t journal_dev;
4525         int err = 0;
4526         int really_read_only;
4527
4528         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4529
4530         if (journal_devnum &&
4531             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4532                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4533                         "numbers have changed");
4534                 journal_dev = new_decode_dev(journal_devnum);
4535         } else
4536                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4537
4538         really_read_only = bdev_read_only(sb->s_bdev);
4539
4540         /*
4541          * Are we loading a blank journal or performing recovery after a
4542          * crash?  For recovery, we need to check in advance whether we
4543          * can get read-write access to the device.
4544          */
4545         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4546                 if (sb->s_flags & MS_RDONLY) {
4547                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4548                                         "required on readonly filesystem");
4549                         if (really_read_only) {
4550                                 ext4_msg(sb, KERN_ERR, "write access "
4551                                         "unavailable, cannot proceed");
4552                                 return -EROFS;
4553                         }
4554                         ext4_msg(sb, KERN_INFO, "write access will "
4555                                "be enabled during recovery");
4556                 }
4557         }
4558
4559         if (journal_inum && journal_dev) {
4560                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4561                        "and inode journals!");
4562                 return -EINVAL;
4563         }
4564
4565         if (journal_inum) {
4566                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4567                         return -EINVAL;
4568         } else {
4569                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4570                         return -EINVAL;
4571         }
4572
4573         if (!(journal->j_flags & JBD2_BARRIER))
4574                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4575
4576         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4577                 err = jbd2_journal_wipe(journal, !really_read_only);
4578         if (!err) {
4579                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4580                 if (save)
4581                         memcpy(save, ((char *) es) +
4582                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4583                 err = jbd2_journal_load(journal);
4584                 if (save)
4585                         memcpy(((char *) es) + EXT4_S_ERR_START,
4586                                save, EXT4_S_ERR_LEN);
4587                 kfree(save);
4588         }
4589
4590         if (err) {
4591                 ext4_msg(sb, KERN_ERR, "error loading journal");
4592                 jbd2_journal_destroy(journal);
4593                 return err;
4594         }
4595
4596         EXT4_SB(sb)->s_journal = journal;
4597         ext4_clear_journal_err(sb, es);
4598
4599         if (!really_read_only && journal_devnum &&
4600             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4601                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4602
4603                 /* Make sure we flush the recovery flag to disk. */
4604                 ext4_commit_super(sb, 1);
4605         }
4606
4607         return 0;
4608 }
4609
4610 static int ext4_commit_super(struct super_block *sb, int sync)
4611 {
4612         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4613         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4614         int error = 0;
4615
4616         if (!sbh)
4617                 return error;
4618         if (buffer_write_io_error(sbh)) {
4619                 /*
4620                  * Oh, dear.  A previous attempt to write the
4621                  * superblock failed.  This could happen because the
4622                  * USB device was yanked out.  Or it could happen to
4623                  * be a transient write error and maybe the block will
4624                  * be remapped.  Nothing we can do but to retry the
4625                  * write and hope for the best.
4626                  */
4627                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4628                        "superblock detected");
4629                 clear_buffer_write_io_error(sbh);
4630                 set_buffer_uptodate(sbh);
4631         }
4632         /*
4633          * If the file system is mounted read-only, don't update the
4634          * superblock write time.  This avoids updating the superblock
4635          * write time when we are mounting the root file system
4636          * read/only but we need to replay the journal; at that point,
4637          * for people who are east of GMT and who make their clock
4638          * tick in localtime for Windows bug-for-bug compatibility,
4639          * the clock is set in the future, and this will cause e2fsck
4640          * to complain and force a full file system check.
4641          */
4642         if (!(sb->s_flags & MS_RDONLY))
4643                 es->s_wtime = cpu_to_le32(get_seconds());
4644         if (sb->s_bdev->bd_part)
4645                 es->s_kbytes_written =
4646                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4647                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4648                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4649         else
4650                 es->s_kbytes_written =
4651                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4652         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4653                 ext4_free_blocks_count_set(es,
4654                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4655                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4656         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4657                 es->s_free_inodes_count =
4658                         cpu_to_le32(percpu_counter_sum_positive(
4659                                 &EXT4_SB(sb)->s_freeinodes_counter));
4660         BUFFER_TRACE(sbh, "marking dirty");
4661         ext4_superblock_csum_set(sb);
4662         mark_buffer_dirty(sbh);
4663         if (sync) {
4664                 error = sync_dirty_buffer(sbh);
4665                 if (error)
4666                         return error;
4667
4668                 error = buffer_write_io_error(sbh);
4669                 if (error) {
4670                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4671                                "superblock");
4672                         clear_buffer_write_io_error(sbh);
4673                         set_buffer_uptodate(sbh);
4674                 }
4675         }
4676         return error;
4677 }
4678
4679 /*
4680  * Have we just finished recovery?  If so, and if we are mounting (or
4681  * remounting) the filesystem readonly, then we will end up with a
4682  * consistent fs on disk.  Record that fact.
4683  */
4684 static void ext4_mark_recovery_complete(struct super_block *sb,
4685                                         struct ext4_super_block *es)
4686 {
4687         journal_t *journal = EXT4_SB(sb)->s_journal;
4688
4689         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4690                 BUG_ON(journal != NULL);
4691                 return;
4692         }
4693         jbd2_journal_lock_updates(journal);
4694         if (jbd2_journal_flush(journal) < 0)
4695                 goto out;
4696
4697         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4698             sb->s_flags & MS_RDONLY) {
4699                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4700                 ext4_commit_super(sb, 1);
4701         }
4702
4703 out:
4704         jbd2_journal_unlock_updates(journal);
4705 }
4706
4707 /*
4708  * If we are mounting (or read-write remounting) a filesystem whose journal
4709  * has recorded an error from a previous lifetime, move that error to the
4710  * main filesystem now.
4711  */
4712 static void ext4_clear_journal_err(struct super_block *sb,
4713                                    struct ext4_super_block *es)
4714 {
4715         journal_t *journal;
4716         int j_errno;
4717         const char *errstr;
4718
4719         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4720
4721         journal = EXT4_SB(sb)->s_journal;
4722
4723         /*
4724          * Now check for any error status which may have been recorded in the
4725          * journal by a prior ext4_error() or ext4_abort()
4726          */
4727
4728         j_errno = jbd2_journal_errno(journal);
4729         if (j_errno) {
4730                 char nbuf[16];
4731
4732                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4733                 ext4_warning(sb, "Filesystem error recorded "
4734                              "from previous mount: %s", errstr);
4735                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4736
4737                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4738                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4739                 ext4_commit_super(sb, 1);
4740
4741                 jbd2_journal_clear_err(journal);
4742                 jbd2_journal_update_sb_errno(journal);
4743         }
4744 }
4745
4746 /*
4747  * Force the running and committing transactions to commit,
4748  * and wait on the commit.
4749  */
4750 int ext4_force_commit(struct super_block *sb)
4751 {
4752         journal_t *journal;
4753
4754         if (sb->s_flags & MS_RDONLY)
4755                 return 0;
4756
4757         journal = EXT4_SB(sb)->s_journal;
4758         return ext4_journal_force_commit(journal);
4759 }
4760
4761 static int ext4_sync_fs(struct super_block *sb, int wait)
4762 {
4763         int ret = 0;
4764         tid_t target;
4765         bool needs_barrier = false;
4766         struct ext4_sb_info *sbi = EXT4_SB(sb);
4767
4768         trace_ext4_sync_fs(sb, wait);
4769         flush_workqueue(sbi->rsv_conversion_wq);
4770         /*
4771          * Writeback quota in non-journalled quota case - journalled quota has
4772          * no dirty dquots
4773          */
4774         dquot_writeback_dquots(sb, -1);
4775         /*
4776          * Data writeback is possible w/o journal transaction, so barrier must
4777          * being sent at the end of the function. But we can skip it if
4778          * transaction_commit will do it for us.
4779          */
4780         if (sbi->s_journal) {
4781                 target = jbd2_get_latest_transaction(sbi->s_journal);
4782                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4783                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4784                         needs_barrier = true;
4785
4786                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4787                         if (wait)
4788                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4789                                                            target);
4790                 }
4791         } else if (wait && test_opt(sb, BARRIER))
4792                 needs_barrier = true;
4793         if (needs_barrier) {
4794                 int err;
4795                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4796                 if (!ret)
4797                         ret = err;
4798         }
4799
4800         return ret;
4801 }
4802
4803 /*
4804  * LVM calls this function before a (read-only) snapshot is created.  This
4805  * gives us a chance to flush the journal completely and mark the fs clean.
4806  *
4807  * Note that only this function cannot bring a filesystem to be in a clean
4808  * state independently. It relies on upper layer to stop all data & metadata
4809  * modifications.
4810  */
4811 static int ext4_freeze(struct super_block *sb)
4812 {
4813         int error = 0;
4814         journal_t *journal;
4815
4816         if (sb->s_flags & MS_RDONLY)
4817                 return 0;
4818
4819         journal = EXT4_SB(sb)->s_journal;
4820
4821         if (journal) {
4822                 /* Now we set up the journal barrier. */
4823                 jbd2_journal_lock_updates(journal);
4824
4825                 /*
4826                  * Don't clear the needs_recovery flag if we failed to
4827                  * flush the journal.
4828                  */
4829                 error = jbd2_journal_flush(journal);
4830                 if (error < 0)
4831                         goto out;
4832         }
4833
4834         /* Journal blocked and flushed, clear needs_recovery flag. */
4835         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4836         error = ext4_commit_super(sb, 1);
4837 out:
4838         if (journal)
4839                 /* we rely on upper layer to stop further updates */
4840                 jbd2_journal_unlock_updates(journal);
4841         return error;
4842 }
4843
4844 /*
4845  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4846  * flag here, even though the filesystem is not technically dirty yet.
4847  */
4848 static int ext4_unfreeze(struct super_block *sb)
4849 {
4850         if (sb->s_flags & MS_RDONLY)
4851                 return 0;
4852
4853         /* Reset the needs_recovery flag before the fs is unlocked. */
4854         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4855         ext4_commit_super(sb, 1);
4856         return 0;
4857 }
4858
4859 /*
4860  * Structure to save mount options for ext4_remount's benefit
4861  */
4862 struct ext4_mount_options {
4863         unsigned long s_mount_opt;
4864         unsigned long s_mount_opt2;
4865         kuid_t s_resuid;
4866         kgid_t s_resgid;
4867         unsigned long s_commit_interval;
4868         u32 s_min_batch_time, s_max_batch_time;
4869 #ifdef CONFIG_QUOTA
4870         int s_jquota_fmt;
4871         char *s_qf_names[EXT4_MAXQUOTAS];
4872 #endif
4873 };
4874
4875 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4876 {
4877         struct ext4_super_block *es;
4878         struct ext4_sb_info *sbi = EXT4_SB(sb);
4879         unsigned long old_sb_flags;
4880         struct ext4_mount_options old_opts;
4881         int enable_quota = 0;
4882         ext4_group_t g;
4883         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4884         int err = 0;
4885 #ifdef CONFIG_QUOTA
4886         int i, j;
4887 #endif
4888         char *orig_data = kstrdup(data, GFP_KERNEL);
4889
4890         /* Store the original options */
4891         old_sb_flags = sb->s_flags;
4892         old_opts.s_mount_opt = sbi->s_mount_opt;
4893         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4894         old_opts.s_resuid = sbi->s_resuid;
4895         old_opts.s_resgid = sbi->s_resgid;
4896         old_opts.s_commit_interval = sbi->s_commit_interval;
4897         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4898         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4899 #ifdef CONFIG_QUOTA
4900         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4901         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4902                 if (sbi->s_qf_names[i]) {
4903                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4904                                                          GFP_KERNEL);
4905                         if (!old_opts.s_qf_names[i]) {
4906                                 for (j = 0; j < i; j++)
4907                                         kfree(old_opts.s_qf_names[j]);
4908                                 kfree(orig_data);
4909                                 return -ENOMEM;
4910                         }
4911                 } else
4912                         old_opts.s_qf_names[i] = NULL;
4913 #endif
4914         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4915                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4916
4917         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4918                 err = -EINVAL;
4919                 goto restore_opts;
4920         }
4921
4922         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4923             test_opt(sb, JOURNAL_CHECKSUM)) {
4924                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4925                          "during remount not supported; ignoring");
4926                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4927         }
4928
4929         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4930                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4931                         ext4_msg(sb, KERN_ERR, "can't mount with "
4932                                  "both data=journal and delalloc");
4933                         err = -EINVAL;
4934                         goto restore_opts;
4935                 }
4936                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4937                         ext4_msg(sb, KERN_ERR, "can't mount with "
4938                                  "both data=journal and dioread_nolock");
4939                         err = -EINVAL;
4940                         goto restore_opts;
4941                 }
4942                 if (test_opt(sb, DAX)) {
4943                         ext4_msg(sb, KERN_ERR, "can't mount with "
4944                                  "both data=journal and dax");
4945                         err = -EINVAL;
4946                         goto restore_opts;
4947                 }
4948         }
4949
4950         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4951                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4952                         "dax flag with busy inodes while remounting");
4953                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4954         }
4955
4956         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4957                 ext4_abort(sb, "Abort forced by user");
4958
4959         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4960                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4961
4962         es = sbi->s_es;
4963
4964         if (sbi->s_journal) {
4965                 ext4_init_journal_params(sb, sbi->s_journal);
4966                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4967         }
4968
4969         if (*flags & MS_LAZYTIME)
4970                 sb->s_flags |= MS_LAZYTIME;
4971
4972         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4973                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4974                         err = -EROFS;
4975                         goto restore_opts;
4976                 }
4977
4978                 if (*flags & MS_RDONLY) {
4979                         err = sync_filesystem(sb);
4980                         if (err < 0)
4981                                 goto restore_opts;
4982                         err = dquot_suspend(sb, -1);
4983                         if (err < 0)
4984                                 goto restore_opts;
4985
4986                         /*
4987                          * First of all, the unconditional stuff we have to do
4988                          * to disable replay of the journal when we next remount
4989                          */
4990                         sb->s_flags |= MS_RDONLY;
4991
4992                         /*
4993                          * OK, test if we are remounting a valid rw partition
4994                          * readonly, and if so set the rdonly flag and then
4995                          * mark the partition as valid again.
4996                          */
4997                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4998                             (sbi->s_mount_state & EXT4_VALID_FS))
4999                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
5000
5001                         if (sbi->s_journal)
5002                                 ext4_mark_recovery_complete(sb, es);
5003                 } else {
5004                         /* Make sure we can mount this feature set readwrite */
5005                         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
5006                                         EXT4_FEATURE_RO_COMPAT_READONLY) ||
5007                             !ext4_feature_set_ok(sb, 0)) {
5008                                 err = -EROFS;
5009                                 goto restore_opts;
5010                         }
5011                         /*
5012                          * Make sure the group descriptor checksums
5013                          * are sane.  If they aren't, refuse to remount r/w.
5014                          */
5015                         for (g = 0; g < sbi->s_groups_count; g++) {
5016                                 struct ext4_group_desc *gdp =
5017                                         ext4_get_group_desc(sb, g, NULL);
5018
5019                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
5020                                         ext4_msg(sb, KERN_ERR,
5021                "ext4_remount: Checksum for group %u failed (%u!=%u)",
5022                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
5023                                                le16_to_cpu(gdp->bg_checksum));
5024                                         err = -EINVAL;
5025                                         goto restore_opts;
5026                                 }
5027                         }
5028
5029                         /*
5030                          * If we have an unprocessed orphan list hanging
5031                          * around from a previously readonly bdev mount,
5032                          * require a full umount/remount for now.
5033                          */
5034                         if (es->s_last_orphan) {
5035                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
5036                                        "remount RDWR because of unprocessed "
5037                                        "orphan inode list.  Please "
5038                                        "umount/remount instead");
5039                                 err = -EINVAL;
5040                                 goto restore_opts;
5041                         }
5042
5043                         /*
5044                          * Mounting a RDONLY partition read-write, so reread
5045                          * and store the current valid flag.  (It may have
5046                          * been changed by e2fsck since we originally mounted
5047                          * the partition.)
5048                          */
5049                         if (sbi->s_journal)
5050                                 ext4_clear_journal_err(sb, es);
5051                         sbi->s_mount_state = le16_to_cpu(es->s_state);
5052                         if (!ext4_setup_super(sb, es, 0))
5053                                 sb->s_flags &= ~MS_RDONLY;
5054                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
5055                                                      EXT4_FEATURE_INCOMPAT_MMP))
5056                                 if (ext4_multi_mount_protect(sb,
5057                                                 le64_to_cpu(es->s_mmp_block))) {
5058                                         err = -EROFS;
5059                                         goto restore_opts;
5060                                 }
5061                         enable_quota = 1;
5062                 }
5063         }
5064
5065         /*
5066          * Reinitialize lazy itable initialization thread based on
5067          * current settings
5068          */
5069         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
5070                 ext4_unregister_li_request(sb);
5071         else {
5072                 ext4_group_t first_not_zeroed;
5073                 first_not_zeroed = ext4_has_uninit_itable(sb);
5074                 ext4_register_li_request(sb, first_not_zeroed);
5075         }
5076
5077         ext4_setup_system_zone(sb);
5078         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
5079                 ext4_commit_super(sb, 1);
5080
5081 #ifdef CONFIG_QUOTA
5082         /* Release old quota file names */
5083         for (i = 0; i < EXT4_MAXQUOTAS; i++)
5084                 kfree(old_opts.s_qf_names[i]);
5085         if (enable_quota) {
5086                 if (sb_any_quota_suspended(sb))
5087                         dquot_resume(sb, -1);
5088                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
5089                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
5090                         err = ext4_enable_quotas(sb);
5091                         if (err)
5092                                 goto restore_opts;
5093                 }
5094         }
5095 #endif
5096
5097         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
5098         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
5099         kfree(orig_data);
5100         return 0;
5101
5102 restore_opts:
5103         sb->s_flags = old_sb_flags;
5104         sbi->s_mount_opt = old_opts.s_mount_opt;
5105         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
5106         sbi->s_resuid = old_opts.s_resuid;
5107         sbi->s_resgid = old_opts.s_resgid;
5108         sbi->s_commit_interval = old_opts.s_commit_interval;
5109         sbi->s_min_batch_time = old_opts.s_min_batch_time;
5110         sbi->s_max_batch_time = old_opts.s_max_batch_time;
5111 #ifdef CONFIG_QUOTA
5112         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
5113         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
5114                 kfree(sbi->s_qf_names[i]);
5115                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
5116         }
5117 #endif
5118         kfree(orig_data);
5119         return err;
5120 }
5121
5122 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
5123 {
5124         struct super_block *sb = dentry->d_sb;
5125         struct ext4_sb_info *sbi = EXT4_SB(sb);
5126         struct ext4_super_block *es = sbi->s_es;
5127         ext4_fsblk_t overhead = 0, resv_blocks;
5128         u64 fsid;
5129         s64 bfree;
5130         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
5131
5132         if (!test_opt(sb, MINIX_DF))
5133                 overhead = sbi->s_overhead;
5134
5135         buf->f_type = EXT4_SUPER_MAGIC;
5136         buf->f_bsize = sb->s_blocksize;
5137         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
5138         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
5139                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
5140         /* prevent underflow in case that few free space is available */
5141         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
5142         buf->f_bavail = buf->f_bfree -
5143                         (ext4_r_blocks_count(es) + resv_blocks);
5144         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
5145                 buf->f_bavail = 0;
5146         buf->f_files = le32_to_cpu(es->s_inodes_count);
5147         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
5148         buf->f_namelen = EXT4_NAME_LEN;
5149         fsid = le64_to_cpup((void *)es->s_uuid) ^
5150                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
5151         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
5152         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
5153
5154         return 0;
5155 }
5156
5157 /* Helper function for writing quotas on sync - we need to start transaction
5158  * before quota file is locked for write. Otherwise the are possible deadlocks:
5159  * Process 1                         Process 2
5160  * ext4_create()                     quota_sync()
5161  *   jbd2_journal_start()                  write_dquot()
5162  *   dquot_initialize()                         down(dqio_mutex)
5163  *     down(dqio_mutex)                    jbd2_journal_start()
5164  *
5165  */
5166
5167 #ifdef CONFIG_QUOTA
5168
5169 static inline struct inode *dquot_to_inode(struct dquot *dquot)
5170 {
5171         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
5172 }
5173
5174 static int ext4_write_dquot(struct dquot *dquot)
5175 {
5176         int ret, err;
5177         handle_t *handle;
5178         struct inode *inode;
5179
5180         inode = dquot_to_inode(dquot);
5181         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
5182                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
5183         if (IS_ERR(handle))
5184                 return PTR_ERR(handle);
5185         ret = dquot_commit(dquot);
5186         err = ext4_journal_stop(handle);
5187         if (!ret)
5188                 ret = err;
5189         return ret;
5190 }
5191
5192 static int ext4_acquire_dquot(struct dquot *dquot)
5193 {
5194         int ret, err;
5195         handle_t *handle;
5196
5197         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5198                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
5199         if (IS_ERR(handle))
5200                 return PTR_ERR(handle);
5201         ret = dquot_acquire(dquot);
5202         err = ext4_journal_stop(handle);
5203         if (!ret)
5204                 ret = err;
5205         return ret;
5206 }
5207
5208 static int ext4_release_dquot(struct dquot *dquot)
5209 {
5210         int ret, err;
5211         handle_t *handle;
5212
5213         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
5214                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
5215         if (IS_ERR(handle)) {
5216                 /* Release dquot anyway to avoid endless cycle in dqput() */
5217                 dquot_release(dquot);
5218                 return PTR_ERR(handle);
5219         }
5220         ret = dquot_release(dquot);
5221         err = ext4_journal_stop(handle);
5222         if (!ret)
5223                 ret = err;
5224         return ret;
5225 }
5226
5227 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5228 {
5229         struct super_block *sb = dquot->dq_sb;
5230         struct ext4_sb_info *sbi = EXT4_SB(sb);
5231
5232         /* Are we journaling quotas? */
5233         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
5234             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5235                 dquot_mark_dquot_dirty(dquot);
5236                 return ext4_write_dquot(dquot);
5237         } else {
5238                 return dquot_mark_dquot_dirty(dquot);
5239         }
5240 }
5241
5242 static int ext4_write_info(struct super_block *sb, int type)
5243 {
5244         int ret, err;
5245         handle_t *handle;
5246
5247         /* Data block + inode block */
5248         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5249         if (IS_ERR(handle))
5250                 return PTR_ERR(handle);
5251         ret = dquot_commit_info(sb, type);
5252         err = ext4_journal_stop(handle);
5253         if (!ret)
5254                 ret = err;
5255         return ret;
5256 }
5257
5258 /*
5259  * Turn on quotas during mount time - we need to find
5260  * the quota file and such...
5261  */
5262 static int ext4_quota_on_mount(struct super_block *sb, int type)
5263 {
5264         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5265                                         EXT4_SB(sb)->s_jquota_fmt, type);
5266 }
5267
5268 /*
5269  * Standard function to be called on quota_on
5270  */
5271 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5272                          struct path *path)
5273 {
5274         int err;
5275
5276         if (!test_opt(sb, QUOTA))
5277                 return -EINVAL;
5278
5279         /* Quotafile not on the same filesystem? */
5280         if (path->dentry->d_sb != sb)
5281                 return -EXDEV;
5282         /* Journaling quota? */
5283         if (EXT4_SB(sb)->s_qf_names[type]) {
5284                 /* Quotafile not in fs root? */
5285                 if (path->dentry->d_parent != sb->s_root)
5286                         ext4_msg(sb, KERN_WARNING,
5287                                 "Quota file not on filesystem root. "
5288                                 "Journaled quota will not work");
5289         }
5290
5291         /*
5292          * When we journal data on quota file, we have to flush journal to see
5293          * all updates to the file when we bypass pagecache...
5294          */
5295         if (EXT4_SB(sb)->s_journal &&
5296             ext4_should_journal_data(d_inode(path->dentry))) {
5297                 /*
5298                  * We don't need to lock updates but journal_flush() could
5299                  * otherwise be livelocked...
5300                  */
5301                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5302                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5303                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5304                 if (err)
5305                         return err;
5306         }
5307
5308         return dquot_quota_on(sb, type, format_id, path);
5309 }
5310
5311 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5312                              unsigned int flags)
5313 {
5314         int err;
5315         struct inode *qf_inode;
5316         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5317                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5318                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5319         };
5320
5321         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5322
5323         if (!qf_inums[type])
5324                 return -EPERM;
5325
5326         qf_inode = ext4_iget(sb, qf_inums[type]);
5327         if (IS_ERR(qf_inode)) {
5328                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5329                 return PTR_ERR(qf_inode);
5330         }
5331
5332         /* Don't account quota for quota files to avoid recursion */
5333         qf_inode->i_flags |= S_NOQUOTA;
5334         err = dquot_enable(qf_inode, type, format_id, flags);
5335         iput(qf_inode);
5336
5337         return err;
5338 }
5339
5340 /* Enable usage tracking for all quota types. */
5341 static int ext4_enable_quotas(struct super_block *sb)
5342 {
5343         int type, err = 0;
5344         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5345                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5346                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5347         };
5348
5349         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5350         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5351                 if (qf_inums[type]) {
5352                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5353                                                 DQUOT_USAGE_ENABLED);
5354                         if (err) {
5355                                 ext4_warning(sb,
5356                                         "Failed to enable quota tracking "
5357                                         "(type=%d, err=%d). Please run "
5358                                         "e2fsck to fix.", type, err);
5359                                 return err;
5360                         }
5361                 }
5362         }
5363         return 0;
5364 }
5365
5366 static int ext4_quota_off(struct super_block *sb, int type)
5367 {
5368         struct inode *inode = sb_dqopt(sb)->files[type];
5369         handle_t *handle;
5370
5371         /* Force all delayed allocation blocks to be allocated.
5372          * Caller already holds s_umount sem */
5373         if (test_opt(sb, DELALLOC))
5374                 sync_filesystem(sb);
5375
5376         if (!inode)
5377                 goto out;
5378
5379         /* Update modification times of quota files when userspace can
5380          * start looking at them */
5381         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5382         if (IS_ERR(handle))
5383                 goto out;
5384         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5385         ext4_mark_inode_dirty(handle, inode);
5386         ext4_journal_stop(handle);
5387
5388 out:
5389         return dquot_quota_off(sb, type);
5390 }
5391
5392 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5393  * acquiring the locks... As quota files are never truncated and quota code
5394  * itself serializes the operations (and no one else should touch the files)
5395  * we don't have to be afraid of races */
5396 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5397                                size_t len, loff_t off)
5398 {
5399         struct inode *inode = sb_dqopt(sb)->files[type];
5400         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5401         int offset = off & (sb->s_blocksize - 1);
5402         int tocopy;
5403         size_t toread;
5404         struct buffer_head *bh;
5405         loff_t i_size = i_size_read(inode);
5406
5407         if (off > i_size)
5408                 return 0;
5409         if (off+len > i_size)
5410                 len = i_size-off;
5411         toread = len;
5412         while (toread > 0) {
5413                 tocopy = sb->s_blocksize - offset < toread ?
5414                                 sb->s_blocksize - offset : toread;
5415                 bh = ext4_bread(NULL, inode, blk, 0);
5416                 if (IS_ERR(bh))
5417                         return PTR_ERR(bh);
5418                 if (!bh)        /* A hole? */
5419                         memset(data, 0, tocopy);
5420                 else
5421                         memcpy(data, bh->b_data+offset, tocopy);
5422                 brelse(bh);
5423                 offset = 0;
5424                 toread -= tocopy;
5425                 data += tocopy;
5426                 blk++;
5427         }
5428         return len;
5429 }
5430
5431 /* Write to quotafile (we know the transaction is already started and has
5432  * enough credits) */
5433 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5434                                 const char *data, size_t len, loff_t off)
5435 {
5436         struct inode *inode = sb_dqopt(sb)->files[type];
5437         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5438         int err, offset = off & (sb->s_blocksize - 1);
5439         int retries = 0;
5440         struct buffer_head *bh;
5441         handle_t *handle = journal_current_handle();
5442
5443         if (EXT4_SB(sb)->s_journal && !handle) {
5444                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5445                         " cancelled because transaction is not started",
5446                         (unsigned long long)off, (unsigned long long)len);
5447                 return -EIO;
5448         }
5449         /*
5450          * Since we account only one data block in transaction credits,
5451          * then it is impossible to cross a block boundary.
5452          */
5453         if (sb->s_blocksize - offset < len) {
5454                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5455                         " cancelled because not block aligned",
5456                         (unsigned long long)off, (unsigned long long)len);
5457                 return -EIO;
5458         }
5459
5460         do {
5461                 bh = ext4_bread(handle, inode, blk,
5462                                 EXT4_GET_BLOCKS_CREATE |
5463                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5464         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5465                  ext4_should_retry_alloc(inode->i_sb, &retries));
5466         if (IS_ERR(bh))
5467                 return PTR_ERR(bh);
5468         if (!bh)
5469                 goto out;
5470         BUFFER_TRACE(bh, "get write access");
5471         err = ext4_journal_get_write_access(handle, bh);
5472         if (err) {
5473                 brelse(bh);
5474                 return err;
5475         }
5476         lock_buffer(bh);
5477         memcpy(bh->b_data+offset, data, len);
5478         flush_dcache_page(bh->b_page);
5479         unlock_buffer(bh);
5480         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5481         brelse(bh);
5482 out:
5483         if (inode->i_size < off + len) {
5484                 i_size_write(inode, off + len);
5485                 EXT4_I(inode)->i_disksize = inode->i_size;
5486                 ext4_mark_inode_dirty(handle, inode);
5487         }
5488         return len;
5489 }
5490
5491 #endif
5492
5493 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5494                        const char *dev_name, void *data)
5495 {
5496         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5497 }
5498
5499 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5500 static inline void register_as_ext2(void)
5501 {
5502         int err = register_filesystem(&ext2_fs_type);
5503         if (err)
5504                 printk(KERN_WARNING
5505                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5506 }
5507
5508 static inline void unregister_as_ext2(void)
5509 {
5510         unregister_filesystem(&ext2_fs_type);
5511 }
5512
5513 static inline int ext2_feature_set_ok(struct super_block *sb)
5514 {
5515         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5516                 return 0;
5517         if (sb->s_flags & MS_RDONLY)
5518                 return 1;
5519         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5520                 return 0;
5521         return 1;
5522 }
5523 #else
5524 static inline void register_as_ext2(void) { }
5525 static inline void unregister_as_ext2(void) { }
5526 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5527 #endif
5528
5529 static inline void register_as_ext3(void)
5530 {
5531         int err = register_filesystem(&ext3_fs_type);
5532         if (err)
5533                 printk(KERN_WARNING
5534                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5535 }
5536
5537 static inline void unregister_as_ext3(void)
5538 {
5539         unregister_filesystem(&ext3_fs_type);
5540 }
5541
5542 static inline int ext3_feature_set_ok(struct super_block *sb)
5543 {
5544         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5545                 return 0;
5546         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5547                 return 0;
5548         if (sb->s_flags & MS_RDONLY)
5549                 return 1;
5550         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5551                 return 0;
5552         return 1;
5553 }
5554
5555 static struct file_system_type ext4_fs_type = {
5556         .owner          = THIS_MODULE,
5557         .name           = "ext4",
5558         .mount          = ext4_mount,
5559         .kill_sb        = kill_block_super,
5560         .fs_flags       = FS_REQUIRES_DEV,
5561 };
5562 MODULE_ALIAS_FS("ext4");
5563
5564 static int __init ext4_init_feat_adverts(void)
5565 {
5566         struct ext4_features *ef;
5567         int ret = -ENOMEM;
5568
5569         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5570         if (!ef)
5571                 goto out;
5572
5573         ef->f_kobj.kset = ext4_kset;
5574         init_completion(&ef->f_kobj_unregister);
5575         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5576                                    "features");
5577         if (ret) {
5578                 kfree(ef);
5579                 goto out;
5580         }
5581
5582         ext4_feat = ef;
5583         ret = 0;
5584 out:
5585         return ret;
5586 }
5587
5588 static void ext4_exit_feat_adverts(void)
5589 {
5590         kobject_put(&ext4_feat->f_kobj);
5591         wait_for_completion(&ext4_feat->f_kobj_unregister);
5592         kfree(ext4_feat);
5593 }
5594
5595 /* Shared across all ext4 file systems */
5596 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5597 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5598
5599 static int __init ext4_init_fs(void)
5600 {
5601         int i, err;
5602
5603         ext4_li_info = NULL;
5604         mutex_init(&ext4_li_mtx);
5605
5606         /* Build-time check for flags consistency */
5607         ext4_check_flag_values();
5608
5609         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5610                 mutex_init(&ext4__aio_mutex[i]);
5611                 init_waitqueue_head(&ext4__ioend_wq[i]);
5612         }
5613
5614         err = ext4_init_es();
5615         if (err)
5616                 return err;
5617
5618         err = ext4_init_pageio();
5619         if (err)
5620                 goto out7;
5621
5622         err = ext4_init_system_zone();
5623         if (err)
5624                 goto out6;
5625         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5626         if (!ext4_kset) {
5627                 err = -ENOMEM;
5628                 goto out5;
5629         }
5630         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5631
5632         err = ext4_init_feat_adverts();
5633         if (err)
5634                 goto out4;
5635
5636         err = ext4_init_mballoc();
5637         if (err)
5638                 goto out2;
5639         else
5640                 ext4_mballoc_ready = 1;
5641         err = init_inodecache();
5642         if (err)
5643                 goto out1;
5644         register_as_ext3();
5645         register_as_ext2();
5646         err = register_filesystem(&ext4_fs_type);
5647         if (err)
5648                 goto out;
5649
5650         return 0;
5651 out:
5652         unregister_as_ext2();
5653         unregister_as_ext3();
5654         destroy_inodecache();
5655 out1:
5656         ext4_mballoc_ready = 0;
5657         ext4_exit_mballoc();
5658 out2:
5659         ext4_exit_feat_adverts();
5660 out4:
5661         if (ext4_proc_root)
5662                 remove_proc_entry("fs/ext4", NULL);
5663         kset_unregister(ext4_kset);
5664 out5:
5665         ext4_exit_system_zone();
5666 out6:
5667         ext4_exit_pageio();
5668 out7:
5669         ext4_exit_es();
5670
5671         return err;
5672 }
5673
5674 static void __exit ext4_exit_fs(void)
5675 {
5676         ext4_exit_crypto();
5677         ext4_destroy_lazyinit_thread();
5678         unregister_as_ext2();
5679         unregister_as_ext3();
5680         unregister_filesystem(&ext4_fs_type);
5681         destroy_inodecache();
5682         ext4_exit_mballoc();
5683         ext4_exit_feat_adverts();
5684         remove_proc_entry("fs/ext4", NULL);
5685         kset_unregister(ext4_kset);
5686         ext4_exit_system_zone();
5687         ext4_exit_pageio();
5688         ext4_exit_es();
5689 }
5690
5691 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5692 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5693 MODULE_LICENSE("GPL");
5694 module_init(ext4_init_fs)
5695 module_exit(ext4_exit_fs)