fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"
  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
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static const char *ext4_decode_error(struct super_block *sb, int errno,
  73                                      char nbuf[16]);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static void ext4_write_super(struct super_block *sb);
  78 static int ext4_freeze(struct super_block *sb);
  79 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  80                        const char *dev_name, void *data);
  81 static inline int ext2_feature_set_ok(struct super_block *sb);
  82 static inline int ext3_feature_set_ok(struct super_block *sb);
  83 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  84 static void ext4_destroy_lazyinit_thread(void);
  85 static void ext4_unregister_li_request(struct super_block *sb);
  86 static void ext4_clear_request_list(void);
  87
  88 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  89 static struct file_system_type ext2_fs_type = {
  90         .owner          = THIS_MODULE,
  91         .name           = "ext2",
  92         .mount          = ext4_mount,
  93         .kill_sb        = kill_block_super,
  94         .fs_flags       = FS_REQUIRES_DEV,
  95 };
  96 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  97 #else
  98 #define IS_EXT2_SB(sb) (0)
  99 #endif
 100
 101
 102 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 103 static struct file_system_type ext3_fs_type = {
 104         .owner          = THIS_MODULE,
 105         .name           = "ext3",
 106         .mount          = ext4_mount,
 107         .kill_sb        = kill_block_super,
 108         .fs_flags       = FS_REQUIRES_DEV,
 109 };
 110 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 111 #else
 112 #define IS_EXT3_SB(sb) (0)
 113 #endif
 114
 115 static int ext4_verify_csum_type(struct super_block *sb,
 116                                  struct ext4_super_block *es)
 117 {
 118         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 119                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 120                 return 1;
 121
 122         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 123 }
 124
 125 void *ext4_kvmalloc(size_t size, gfp_t flags)
 126 {
 127         void *ret;
 128
 129         ret = kmalloc(size, flags);
 130         if (!ret)
 131                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 132         return ret;
 133 }
 134
 135 void *ext4_kvzalloc(size_t size, gfp_t flags)
 136 {
 137         void *ret;
 138
 139         ret = kzalloc(size, flags);
 140         if (!ret)
 141                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 142         return ret;
 143 }
 144
 145 void ext4_kvfree(void *ptr)
 146 {
 147         if (is_vmalloc_addr(ptr))
 148                 vfree(ptr);
 149         else
 150                 kfree(ptr);
 151
 152 }
 153
 154 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 155                                struct ext4_group_desc *bg)
 156 {
 157         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 158                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 159                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 160 }
 161
 162 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 163                                struct ext4_group_desc *bg)
 164 {
 165         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 166                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 167                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 168 }
 169
 170 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 171                               struct ext4_group_desc *bg)
 172 {
 173         return le32_to_cpu(bg->bg_inode_table_lo) |
 174                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 175                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 176 }
 177
 178 __u32 ext4_free_group_clusters(struct super_block *sb,
 179                                struct ext4_group_desc *bg)
 180 {
 181         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 182                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 183                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 184 }
 185
 186 __u32 ext4_free_inodes_count(struct super_block *sb,
 187                               struct ext4_group_desc *bg)
 188 {
 189         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 190                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 191                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 192 }
 193
 194 __u32 ext4_used_dirs_count(struct super_block *sb,
 195                               struct ext4_group_desc *bg)
 196 {
 197         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 198                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 199                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 200 }
 201
 202 __u32 ext4_itable_unused_count(struct super_block *sb,
 203                               struct ext4_group_desc *bg)
 204 {
 205         return le16_to_cpu(bg->bg_itable_unused_lo) |
 206                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 207                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 208 }
 209
 210 void ext4_block_bitmap_set(struct super_block *sb,
 211                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 212 {
 213         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 214         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 215                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 216 }
 217
 218 void ext4_inode_bitmap_set(struct super_block *sb,
 219                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 220 {
 221         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 222         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 223                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 224 }
 225
 226 void ext4_inode_table_set(struct super_block *sb,
 227                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 228 {
 229         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 230         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 231                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 232 }
 233
 234 void ext4_free_group_clusters_set(struct super_block *sb,
 235                                   struct ext4_group_desc *bg, __u32 count)
 236 {
 237         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 238         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 239                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 240 }
 241
 242 void ext4_free_inodes_set(struct super_block *sb,
 243                           struct ext4_group_desc *bg, __u32 count)
 244 {
 245         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 246         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 247                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 248 }
 249
 250 void ext4_used_dirs_set(struct super_block *sb,
 251                           struct ext4_group_desc *bg, __u32 count)
 252 {
 253         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 254         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 255                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 256 }
 257
 258 void ext4_itable_unused_set(struct super_block *sb,
 259                           struct ext4_group_desc *bg, __u32 count)
 260 {
 261         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 262         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 263                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 264 }
 265
 266
 267 /* Just increment the non-pointer handle value */
 268 static handle_t *ext4_get_nojournal(void)
 269 {
 270         handle_t *handle = current->journal_info;
 271         unsigned long ref_cnt = (unsigned long)handle;
 272
 273         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 274
 275         ref_cnt++;
 276         handle = (handle_t *)ref_cnt;
 277
 278         current->journal_info = handle;
 279         return handle;
 280 }
 281
 282
 283 /* Decrement the non-pointer handle value */
 284 static void ext4_put_nojournal(handle_t *handle)
 285 {
 286         unsigned long ref_cnt = (unsigned long)handle;
 287
 288         BUG_ON(ref_cnt == 0);
 289
 290         ref_cnt--;
 291         handle = (handle_t *)ref_cnt;
 292
 293         current->journal_info = handle;
 294 }
 295
 296 /*
 297  * Wrappers for jbd2_journal_start/end.
 298  *
 299  * The only special thing we need to do here is to make sure that all
 300  * journal_end calls result in the superblock being marked dirty, so
 301  * that sync() will call the filesystem's write_super callback if
 302  * appropriate.
 303  *
 304  * To avoid j_barrier hold in userspace when a user calls freeze(),
 305  * ext4 prevents a new handle from being started by s_frozen, which
 306  * is in an upper layer.
 307  */
 308 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 309 {
 310         journal_t *journal;
 311         handle_t  *handle;
 312
 313         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 314         if (sb->s_flags & MS_RDONLY)
 315                 return ERR_PTR(-EROFS);
 316
 317         journal = EXT4_SB(sb)->s_journal;
 318         handle = ext4_journal_current_handle();
 319
 320         /*
 321          * If a handle has been started, it should be allowed to
 322          * finish, otherwise deadlock could happen between freeze
 323          * and others(e.g. truncate) due to the restart of the
 324          * journal handle if the filesystem is forzen and active
 325          * handles are not stopped.
 326          */
 327         if (!handle)
 328                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 329
 330         if (!journal)
 331                 return ext4_get_nojournal();
 332         /*
 333          * Special case here: if the journal has aborted behind our
 334          * backs (eg. EIO in the commit thread), then we still need to
 335          * take the FS itself readonly cleanly.
 336          */
 337         if (is_journal_aborted(journal)) {
 338                 ext4_abort(sb, "Detected aborted journal");
 339                 return ERR_PTR(-EROFS);
 340         }
 341         return jbd2_journal_start(journal, nblocks);
 342 }
 343
 344 /*
 345  * The only special thing we need to do here is to make sure that all
 346  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 347  * that sync() will call the filesystem's write_super callback if
 348  * appropriate.
 349  */
 350 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 351 {
 352         struct super_block *sb;
 353         int err;
 354         int rc;
 355
 356         if (!ext4_handle_valid(handle)) {
 357                 ext4_put_nojournal(handle);
 358                 return 0;
 359         }
 360         sb = handle->h_transaction->t_journal->j_private;
 361         err = handle->h_err;
 362         rc = jbd2_journal_stop(handle);
 363
 364         if (!err)
 365                 err = rc;
 366         if (err)
 367                 __ext4_std_error(sb, where, line, err);
 368         return err;
 369 }
 370
 371 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 372                                const char *err_fn, struct buffer_head *bh,
 373                                handle_t *handle, int err)
 374 {
 375         char nbuf[16];
 376         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 377
 378         BUG_ON(!ext4_handle_valid(handle));
 379
 380         if (bh)
 381                 BUFFER_TRACE(bh, "abort");
 382
 383         if (!handle->h_err)
 384                 handle->h_err = err;
 385
 386         if (is_handle_aborted(handle))
 387                 return;
 388
 389         printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
 390                caller, line, errstr, err_fn);
 391
 392         jbd2_journal_abort_handle(handle);
 393 }
 394
 395 static void __save_error_info(struct super_block *sb, const char *func,
 396                             unsigned int line)
 397 {
 398         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 399
 400         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 401         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 402         es->s_last_error_time = cpu_to_le32(get_seconds());
 403         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 404         es->s_last_error_line = cpu_to_le32(line);
 405         if (!es->s_first_error_time) {
 406                 es->s_first_error_time = es->s_last_error_time;
 407                 strncpy(es->s_first_error_func, func,
 408                         sizeof(es->s_first_error_func));
 409                 es->s_first_error_line = cpu_to_le32(line);
 410                 es->s_first_error_ino = es->s_last_error_ino;
 411                 es->s_first_error_block = es->s_last_error_block;
 412         }
 413         /*
 414          * Start the daily error reporting function if it hasn't been
 415          * started already
 416          */
 417         if (!es->s_error_count)
 418                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 419         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 420 }
 421
 422 static void save_error_info(struct super_block *sb, const char *func,
 423                             unsigned int line)
 424 {
 425         __save_error_info(sb, func, line);
 426         ext4_commit_super(sb, 1);
 427 }
 428
 429 /*
 430  * The del_gendisk() function uninitializes the disk-specific data
 431  * structures, including the bdi structure, without telling anyone
 432  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 433  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 434  * This is a kludge to prevent these oops until we can put in a proper
 435  * hook in del_gendisk() to inform the VFS and file system layers.
 436  */
 437 static int block_device_ejected(struct super_block *sb)
 438 {
 439         struct inode *bd_inode = sb->s_bdev->bd_inode;
 440         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 441
 442         return bdi->dev == NULL;
 443 }
 444
 445 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 446 {
 447         struct super_block              *sb = journal->j_private;
 448         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 449         int                             error = is_journal_aborted(journal);
 450         struct ext4_journal_cb_entry    *jce, *tmp;
 451
 452         spin_lock(&sbi->s_md_lock);
 453         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 454                 list_del_init(&jce->jce_list);
 455                 spin_unlock(&sbi->s_md_lock);
 456                 jce->jce_func(sb, jce, error);
 457                 spin_lock(&sbi->s_md_lock);
 458         }
 459         spin_unlock(&sbi->s_md_lock);
 460 }
 461
 462 /* Deal with the reporting of failure conditions on a filesystem such as
 463  * inconsistencies detected or read IO failures.
 464  *
 465  * On ext2, we can store the error state of the filesystem in the
 466  * superblock.  That is not possible on ext4, because we may have other
 467  * write ordering constraints on the superblock which prevent us from
 468  * writing it out straight away; and given that the journal is about to
 469  * be aborted, we can't rely on the current, or future, transactions to
 470  * write out the superblock safely.
 471  *
 472  * We'll just use the jbd2_journal_abort() error code to record an error in
 473  * the journal instead.  On recovery, the journal will complain about
 474  * that error until we've noted it down and cleared it.
 475  */
 476
 477 static void ext4_handle_error(struct super_block *sb)
 478 {
 479         if (sb->s_flags & MS_RDONLY)
 480                 return;
 481
 482         if (!test_opt(sb, ERRORS_CONT)) {
 483                 journal_t *journal = EXT4_SB(sb)->s_journal;
 484
 485                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 486                 if (journal)
 487                         jbd2_journal_abort(journal, -EIO);
 488         }
 489         if (test_opt(sb, ERRORS_RO)) {
 490                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 491                 sb->s_flags |= MS_RDONLY;
 492         }
 493         if (test_opt(sb, ERRORS_PANIC))
 494                 panic("EXT4-fs (device %s): panic forced after error\n",
 495                         sb->s_id);
 496 }
 497
 498 void __ext4_error(struct super_block *sb, const char *function,
 499                   unsigned int line, const char *fmt, ...)
 500 {
 501         struct va_format vaf;
 502         va_list args;
 503
 504         va_start(args, fmt);
 505         vaf.fmt = fmt;
 506         vaf.va = &args;
 507         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 508                sb->s_id, function, line, current->comm, &vaf);
 509         va_end(args);
 510
 511         ext4_handle_error(sb);
 512 }
 513
 514 void ext4_error_inode(struct inode *inode, const char *function,
 515                       unsigned int line, ext4_fsblk_t block,
 516                       const char *fmt, ...)
 517 {
 518         va_list args;
 519         struct va_format vaf;
 520         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 521
 522         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 523         es->s_last_error_block = cpu_to_le64(block);
 524         save_error_info(inode->i_sb, function, line);
 525         va_start(args, fmt);
 526         vaf.fmt = fmt;
 527         vaf.va = &args;
 528         if (block)
 529                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 530                        "inode #%lu: block %llu: comm %s: %pV\n",
 531                        inode->i_sb->s_id, function, line, inode->i_ino,
 532                        block, current->comm, &vaf);
 533         else
 534                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 535                        "inode #%lu: comm %s: %pV\n",
 536                        inode->i_sb->s_id, function, line, inode->i_ino,
 537                        current->comm, &vaf);
 538         va_end(args);
 539
 540         ext4_handle_error(inode->i_sb);
 541 }
 542
 543 void ext4_error_file(struct file *file, const char *function,
 544                      unsigned int line, ext4_fsblk_t block,
 545                      const char *fmt, ...)
 546 {
 547         va_list args;
 548         struct va_format vaf;
 549         struct ext4_super_block *es;
 550         struct inode *inode = file->f_dentry->d_inode;
 551         char pathname[80], *path;
 552
 553         es = EXT4_SB(inode->i_sb)->s_es;
 554         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 555         save_error_info(inode->i_sb, function, line);
 556         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 557         if (IS_ERR(path))
 558                 path = "(unknown)";
 559         va_start(args, fmt);
 560         vaf.fmt = fmt;
 561         vaf.va = &args;
 562         if (block)
 563                 printk(KERN_CRIT
 564                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 565                        "block %llu: comm %s: path %s: %pV\n",
 566                        inode->i_sb->s_id, function, line, inode->i_ino,
 567                        block, current->comm, path, &vaf);
 568         else
 569                 printk(KERN_CRIT
 570                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 571                        "comm %s: path %s: %pV\n",
 572                        inode->i_sb->s_id, function, line, inode->i_ino,
 573                        current->comm, path, &vaf);
 574         va_end(args);
 575
 576         ext4_handle_error(inode->i_sb);
 577 }
 578
 579 static const char *ext4_decode_error(struct super_block *sb, int errno,
 580                                      char nbuf[16])
 581 {
 582         char *errstr = NULL;
 583
 584         switch (errno) {
 585         case -EIO:
 586                 errstr = "IO failure";
 587                 break;
 588         case -ENOMEM:
 589                 errstr = "Out of memory";
 590                 break;
 591         case -EROFS:
 592                 if (!sb || (EXT4_SB(sb)->s_journal &&
 593                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 594                         errstr = "Journal has aborted";
 595                 else
 596                         errstr = "Readonly filesystem";
 597                 break;
 598         default:
 599                 /* If the caller passed in an extra buffer for unknown
 600                  * errors, textualise them now.  Else we just return
 601                  * NULL. */
 602                 if (nbuf) {
 603                         /* Check for truncated error codes... */
 604                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 605                                 errstr = nbuf;
 606                 }
 607                 break;
 608         }
 609
 610         return errstr;
 611 }
 612
 613 /* __ext4_std_error decodes expected errors from journaling functions
 614  * automatically and invokes the appropriate error response.  */
 615
 616 void __ext4_std_error(struct super_block *sb, const char *function,
 617                       unsigned int line, int errno)
 618 {
 619         char nbuf[16];
 620         const char *errstr;
 621
 622         /* Special case: if the error is EROFS, and we're not already
 623          * inside a transaction, then there's really no point in logging
 624          * an error. */
 625         if (errno == -EROFS && journal_current_handle() == NULL &&
 626             (sb->s_flags & MS_RDONLY))
 627                 return;
 628
 629         errstr = ext4_decode_error(sb, errno, nbuf);
 630         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 631                sb->s_id, function, line, errstr);
 632         save_error_info(sb, function, line);
 633
 634         ext4_handle_error(sb);
 635 }
 636
 637 /*
 638  * ext4_abort is a much stronger failure handler than ext4_error.  The
 639  * abort function may be used to deal with unrecoverable failures such
 640  * as journal IO errors or ENOMEM at a critical moment in log management.
 641  *
 642  * We unconditionally force the filesystem into an ABORT|READONLY state,
 643  * unless the error response on the fs has been set to panic in which
 644  * case we take the easy way out and panic immediately.
 645  */
 646
 647 void __ext4_abort(struct super_block *sb, const char *function,
 648                 unsigned int line, const char *fmt, ...)
 649 {
 650         va_list args;
 651
 652         save_error_info(sb, function, line);
 653         va_start(args, fmt);
 654         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 655                function, line);
 656         vprintk(fmt, args);
 657         printk("\n");
 658         va_end(args);
 659
 660         if ((sb->s_flags & MS_RDONLY) == 0) {
 661                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 662                 sb->s_flags |= MS_RDONLY;
 663                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 664                 if (EXT4_SB(sb)->s_journal)
 665                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 666                 save_error_info(sb, function, line);
 667         }
 668         if (test_opt(sb, ERRORS_PANIC))
 669                 panic("EXT4-fs panic from previous error\n");
 670 }
 671
 672 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 673 {
 674         struct va_format vaf;
 675         va_list args;
 676
 677         va_start(args, fmt);
 678         vaf.fmt = fmt;
 679         vaf.va = &args;
 680         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 681         va_end(args);
 682 }
 683
 684 void __ext4_warning(struct super_block *sb, const char *function,
 685                     unsigned int line, const char *fmt, ...)
 686 {
 687         struct va_format vaf;
 688         va_list args;
 689
 690         va_start(args, fmt);
 691         vaf.fmt = fmt;
 692         vaf.va = &args;
 693         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 694                sb->s_id, function, line, &vaf);
 695         va_end(args);
 696 }
 697
 698 void __ext4_grp_locked_error(const char *function, unsigned int line,
 699                              struct super_block *sb, ext4_group_t grp,
 700                              unsigned long ino, ext4_fsblk_t block,
 701                              const char *fmt, ...)
 702 __releases(bitlock)
 703 __acquires(bitlock)
 704 {
 705         struct va_format vaf;
 706         va_list args;
 707         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 708
 709         es->s_last_error_ino = cpu_to_le32(ino);
 710         es->s_last_error_block = cpu_to_le64(block);
 711         __save_error_info(sb, function, line);
 712
 713         va_start(args, fmt);
 714
 715         vaf.fmt = fmt;
 716         vaf.va = &args;
 717         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 718                sb->s_id, function, line, grp);
 719         if (ino)
 720                 printk(KERN_CONT "inode %lu: ", ino);
 721         if (block)
 722                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 723         printk(KERN_CONT "%pV\n", &vaf);
 724         va_end(args);
 725
 726         if (test_opt(sb, ERRORS_CONT)) {
 727                 ext4_commit_super(sb, 0);
 728                 return;
 729         }
 730
 731         ext4_unlock_group(sb, grp);
 732         ext4_handle_error(sb);
 733         /*
 734          * We only get here in the ERRORS_RO case; relocking the group
 735          * may be dangerous, but nothing bad will happen since the
 736          * filesystem will have already been marked read/only and the
 737          * journal has been aborted.  We return 1 as a hint to callers
 738          * who might what to use the return value from
 739          * ext4_grp_locked_error() to distinguish between the
 740          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 741          * aggressively from the ext4 function in question, with a
 742          * more appropriate error code.
 743          */
 744         ext4_lock_group(sb, grp);
 745         return;
 746 }
 747
 748 void ext4_update_dynamic_rev(struct super_block *sb)
 749 {
 750         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 751
 752         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 753                 return;
 754
 755         ext4_warning(sb,
 756                      "updating to rev %d because of new feature flag, "
 757                      "running e2fsck is recommended",
 758                      EXT4_DYNAMIC_REV);
 759
 760         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 761         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 762         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 763         /* leave es->s_feature_*compat flags alone */
 764         /* es->s_uuid will be set by e2fsck if empty */
 765
 766         /*
 767          * The rest of the superblock fields should be zero, and if not it
 768          * means they are likely already in use, so leave them alone.  We
 769          * can leave it up to e2fsck to clean up any inconsistencies there.
 770          */
 771 }
 772
 773 /*
 774  * Open the external journal device
 775  */
 776 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 777 {
 778         struct block_device *bdev;
 779         char b[BDEVNAME_SIZE];
 780
 781         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 782         if (IS_ERR(bdev))
 783                 goto fail;
 784         return bdev;
 785
 786 fail:
 787         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 788                         __bdevname(dev, b), PTR_ERR(bdev));
 789         return NULL;
 790 }
 791
 792 /*
 793  * Release the journal device
 794  */
 795 static int ext4_blkdev_put(struct block_device *bdev)
 796 {
 797         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 798 }
 799
 800 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 801 {
 802         struct block_device *bdev;
 803         int ret = -ENODEV;
 804
 805         bdev = sbi->journal_bdev;
 806         if (bdev) {
 807                 ret = ext4_blkdev_put(bdev);
 808                 sbi->journal_bdev = NULL;
 809         }
 810         return ret;
 811 }
 812
 813 static inline struct inode *orphan_list_entry(struct list_head *l)
 814 {
 815         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 816 }
 817
 818 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 819 {
 820         struct list_head *l;
 821
 822         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 823                  le32_to_cpu(sbi->s_es->s_last_orphan));
 824
 825         printk(KERN_ERR "sb_info orphan list:\n");
 826         list_for_each(l, &sbi->s_orphan) {
 827                 struct inode *inode = orphan_list_entry(l);
 828                 printk(KERN_ERR "  "
 829                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 830                        inode->i_sb->s_id, inode->i_ino, inode,
 831                        inode->i_mode, inode->i_nlink,
 832                        NEXT_ORPHAN(inode));
 833         }
 834 }
 835
 836 static void ext4_put_super(struct super_block *sb)
 837 {
 838         struct ext4_sb_info *sbi = EXT4_SB(sb);
 839         struct ext4_super_block *es = sbi->s_es;
 840         int i, err;
 841
 842         ext4_unregister_li_request(sb);
 843         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 844
 845         flush_workqueue(sbi->dio_unwritten_wq);
 846         destroy_workqueue(sbi->dio_unwritten_wq);
 847
 848         lock_super(sb);
 849         if (sbi->s_journal) {
 850                 err = jbd2_journal_destroy(sbi->s_journal);
 851                 sbi->s_journal = NULL;
 852                 if (err < 0)
 853                         ext4_abort(sb, "Couldn't clean up the journal");
 854         }
 855
 856         del_timer(&sbi->s_err_report);
 857         ext4_release_system_zone(sb);
 858         ext4_mb_release(sb);
 859         ext4_ext_release(sb);
 860         ext4_xattr_put_super(sb);
 861
 862         if (!(sb->s_flags & MS_RDONLY)) {
 863                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 864                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 865         }
 866         if (sb->s_dirt || !(sb->s_flags & MS_RDONLY))
 867                 ext4_commit_super(sb, 1);
 868
 869         if (sbi->s_proc) {
 870                 remove_proc_entry("options", sbi->s_proc);
 871                 remove_proc_entry(sb->s_id, ext4_proc_root);
 872         }
 873         kobject_del(&sbi->s_kobj);
 874
 875         for (i = 0; i < sbi->s_gdb_count; i++)
 876                 brelse(sbi->s_group_desc[i]);
 877         ext4_kvfree(sbi->s_group_desc);
 878         ext4_kvfree(sbi->s_flex_groups);
 879         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 880         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 881         percpu_counter_destroy(&sbi->s_dirs_counter);
 882         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 883         brelse(sbi->s_sbh);
 884 #ifdef CONFIG_QUOTA
 885         for (i = 0; i < MAXQUOTAS; i++)
 886                 kfree(sbi->s_qf_names[i]);
 887 #endif
 888
 889         /* Debugging code just in case the in-memory inode orphan list
 890          * isn't empty.  The on-disk one can be non-empty if we've
 891          * detected an error and taken the fs readonly, but the
 892          * in-memory list had better be clean by this point. */
 893         if (!list_empty(&sbi->s_orphan))
 894                 dump_orphan_list(sb, sbi);
 895         J_ASSERT(list_empty(&sbi->s_orphan));
 896
 897         invalidate_bdev(sb->s_bdev);
 898         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 899                 /*
 900                  * Invalidate the journal device's buffers.  We don't want them
 901                  * floating about in memory - the physical journal device may
 902                  * hotswapped, and it breaks the `ro-after' testing code.
 903                  */
 904                 sync_blockdev(sbi->journal_bdev);
 905                 invalidate_bdev(sbi->journal_bdev);
 906                 ext4_blkdev_remove(sbi);
 907         }
 908         if (sbi->s_mmp_tsk)
 909                 kthread_stop(sbi->s_mmp_tsk);
 910         sb->s_fs_info = NULL;
 911         /*
 912          * Now that we are completely done shutting down the
 913          * superblock, we need to actually destroy the kobject.
 914          */
 915         unlock_super(sb);
 916         kobject_put(&sbi->s_kobj);
 917         wait_for_completion(&sbi->s_kobj_unregister);
 918         if (sbi->s_chksum_driver)
 919                 crypto_free_shash(sbi->s_chksum_driver);
 920         kfree(sbi->s_blockgroup_lock);
 921         kfree(sbi);
 922 }
 923
 924 static struct kmem_cache *ext4_inode_cachep;
 925
 926 /*
 927  * Called inside transaction, so use GFP_NOFS
 928  */
 929 static struct inode *ext4_alloc_inode(struct super_block *sb)
 930 {
 931         struct ext4_inode_info *ei;
 932
 933         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 934         if (!ei)
 935                 return NULL;
 936
 937         ei->vfs_inode.i_version = 1;
 938         ei->vfs_inode.i_data.writeback_index = 0;
 939         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 940         INIT_LIST_HEAD(&ei->i_prealloc_list);
 941         spin_lock_init(&ei->i_prealloc_lock);
 942         ei->i_reserved_data_blocks = 0;
 943         ei->i_reserved_meta_blocks = 0;
 944         ei->i_allocated_meta_blocks = 0;
 945         ei->i_da_metadata_calc_len = 0;
 946         spin_lock_init(&(ei->i_block_reservation_lock));
 947 #ifdef CONFIG_QUOTA
 948         ei->i_reserved_quota = 0;
 949 #endif
 950         ei->jinode = NULL;
 951         INIT_LIST_HEAD(&ei->i_completed_io_list);
 952         spin_lock_init(&ei->i_completed_io_lock);
 953         ei->cur_aio_dio = NULL;
 954         ei->i_sync_tid = 0;
 955         ei->i_datasync_tid = 0;
 956         atomic_set(&ei->i_ioend_count, 0);
 957         atomic_set(&ei->i_aiodio_unwritten, 0);
 958
 959         return &ei->vfs_inode;
 960 }
 961
 962 static int ext4_drop_inode(struct inode *inode)
 963 {
 964         int drop = generic_drop_inode(inode);
 965
 966         trace_ext4_drop_inode(inode, drop);
 967         return drop;
 968 }
 969
 970 static void ext4_i_callback(struct rcu_head *head)
 971 {
 972         struct inode *inode = container_of(head, struct inode, i_rcu);
 973         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 974 }
 975
 976 static void ext4_destroy_inode(struct inode *inode)
 977 {
 978         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 979                 ext4_msg(inode->i_sb, KERN_ERR,
 980                          "Inode %lu (%p): orphan list check failed!",
 981                          inode->i_ino, EXT4_I(inode));
 982                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 983                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 984                                 true);
 985                 dump_stack();
 986         }
 987         call_rcu(&inode->i_rcu, ext4_i_callback);
 988 }
 989
 990 static void init_once(void *foo)
 991 {
 992         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 993
 994         INIT_LIST_HEAD(&ei->i_orphan);
 995 #ifdef CONFIG_EXT4_FS_XATTR
 996         init_rwsem(&ei->xattr_sem);
 997 #endif
 998         init_rwsem(&ei->i_data_sem);
 999         inode_init_once(&ei->vfs_inode);
1000 }
1001
1002 static int init_inodecache(void)
1003 {
1004         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1005                                              sizeof(struct ext4_inode_info),
1006                                              0, (SLAB_RECLAIM_ACCOUNT|
1007                                                 SLAB_MEM_SPREAD),
1008                                              init_once);
1009         if (ext4_inode_cachep == NULL)
1010                 return -ENOMEM;
1011         return 0;
1012 }
1013
1014 static void destroy_inodecache(void)
1015 {
1016         kmem_cache_destroy(ext4_inode_cachep);
1017 }
1018
1019 void ext4_clear_inode(struct inode *inode)
1020 {
1021         invalidate_inode_buffers(inode);
1022         end_writeback(inode);
1023         dquot_drop(inode);
1024         ext4_discard_preallocations(inode);
1025         if (EXT4_I(inode)->jinode) {
1026                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1027                                                EXT4_I(inode)->jinode);
1028                 jbd2_free_inode(EXT4_I(inode)->jinode);
1029                 EXT4_I(inode)->jinode = NULL;
1030         }
1031 }
1032
1033 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1034                                         u64 ino, u32 generation)
1035 {
1036         struct inode *inode;
1037
1038         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1039                 return ERR_PTR(-ESTALE);
1040         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1041                 return ERR_PTR(-ESTALE);
1042
1043         /* iget isn't really right if the inode is currently unallocated!!
1044          *
1045          * ext4_read_inode will return a bad_inode if the inode had been
1046          * deleted, so we should be safe.
1047          *
1048          * Currently we don't know the generation for parent directory, so
1049          * a generation of 0 means "accept any"
1050          */
1051         inode = ext4_iget(sb, ino);
1052         if (IS_ERR(inode))
1053                 return ERR_CAST(inode);
1054         if (generation && inode->i_generation != generation) {
1055                 iput(inode);
1056                 return ERR_PTR(-ESTALE);
1057         }
1058
1059         return inode;
1060 }
1061
1062 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1063                                         int fh_len, int fh_type)
1064 {
1065         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1066                                     ext4_nfs_get_inode);
1067 }
1068
1069 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1070                                         int fh_len, int fh_type)
1071 {
1072         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1073                                     ext4_nfs_get_inode);
1074 }
1075
1076 /*
1077  * Try to release metadata pages (indirect blocks, directories) which are
1078  * mapped via the block device.  Since these pages could have journal heads
1079  * which would prevent try_to_free_buffers() from freeing them, we must use
1080  * jbd2 layer's try_to_free_buffers() function to release them.
1081  */
1082 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1083                                  gfp_t wait)
1084 {
1085         journal_t *journal = EXT4_SB(sb)->s_journal;
1086
1087         WARN_ON(PageChecked(page));
1088         if (!page_has_buffers(page))
1089                 return 0;
1090         if (journal)
1091                 return jbd2_journal_try_to_free_buffers(journal, page,
1092                                                         wait & ~__GFP_WAIT);
1093         return try_to_free_buffers(page);
1094 }
1095
1096 #ifdef CONFIG_QUOTA
1097 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1098 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1099
1100 static int ext4_write_dquot(struct dquot *dquot);
1101 static int ext4_acquire_dquot(struct dquot *dquot);
1102 static int ext4_release_dquot(struct dquot *dquot);
1103 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1104 static int ext4_write_info(struct super_block *sb, int type);
1105 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1106                          struct path *path);
1107 static int ext4_quota_off(struct super_block *sb, int type);
1108 static int ext4_quota_on_mount(struct super_block *sb, int type);
1109 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1110                                size_t len, loff_t off);
1111 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1112                                 const char *data, size_t len, loff_t off);
1113
1114 static const struct dquot_operations ext4_quota_operations = {
1115         .get_reserved_space = ext4_get_reserved_space,
1116         .write_dquot    = ext4_write_dquot,
1117         .acquire_dquot  = ext4_acquire_dquot,
1118         .release_dquot  = ext4_release_dquot,
1119         .mark_dirty     = ext4_mark_dquot_dirty,
1120         .write_info     = ext4_write_info,
1121         .alloc_dquot    = dquot_alloc,
1122         .destroy_dquot  = dquot_destroy,
1123 };
1124
1125 static const struct quotactl_ops ext4_qctl_operations = {
1126         .quota_on       = ext4_quota_on,
1127         .quota_off      = ext4_quota_off,
1128         .quota_sync     = dquot_quota_sync,
1129         .get_info       = dquot_get_dqinfo,
1130         .set_info       = dquot_set_dqinfo,
1131         .get_dqblk      = dquot_get_dqblk,
1132         .set_dqblk      = dquot_set_dqblk
1133 };
1134 #endif
1135
1136 static const struct super_operations ext4_sops = {
1137         .alloc_inode    = ext4_alloc_inode,
1138         .destroy_inode  = ext4_destroy_inode,
1139         .write_inode    = ext4_write_inode,
1140         .dirty_inode    = ext4_dirty_inode,
1141         .drop_inode     = ext4_drop_inode,
1142         .evict_inode    = ext4_evict_inode,
1143         .put_super      = ext4_put_super,
1144         .sync_fs        = ext4_sync_fs,
1145         .freeze_fs      = ext4_freeze,
1146         .unfreeze_fs    = ext4_unfreeze,
1147         .statfs         = ext4_statfs,
1148         .remount_fs     = ext4_remount,
1149         .show_options   = ext4_show_options,
1150 #ifdef CONFIG_QUOTA
1151         .quota_read     = ext4_quota_read,
1152         .quota_write    = ext4_quota_write,
1153 #endif
1154         .bdev_try_to_free_page = bdev_try_to_free_page,
1155 };
1156
1157 static const struct super_operations ext4_nojournal_sops = {
1158         .alloc_inode    = ext4_alloc_inode,
1159         .destroy_inode  = ext4_destroy_inode,
1160         .write_inode    = ext4_write_inode,
1161         .dirty_inode    = ext4_dirty_inode,
1162         .drop_inode     = ext4_drop_inode,
1163         .evict_inode    = ext4_evict_inode,
1164         .write_super    = ext4_write_super,
1165         .put_super      = ext4_put_super,
1166         .statfs         = ext4_statfs,
1167         .remount_fs     = ext4_remount,
1168         .show_options   = ext4_show_options,
1169 #ifdef CONFIG_QUOTA
1170         .quota_read     = ext4_quota_read,
1171         .quota_write    = ext4_quota_write,
1172 #endif
1173         .bdev_try_to_free_page = bdev_try_to_free_page,
1174 };
1175
1176 static const struct export_operations ext4_export_ops = {
1177         .fh_to_dentry = ext4_fh_to_dentry,
1178         .fh_to_parent = ext4_fh_to_parent,
1179         .get_parent = ext4_get_parent,
1180 };
1181
1182 enum {
1183         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1184         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1185         Opt_nouid32, Opt_debug, Opt_removed,
1186         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1187         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1188         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1189         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1190         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1191         Opt_data_err_abort, Opt_data_err_ignore,
1192         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1193         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1194         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1195         Opt_usrquota, Opt_grpquota, Opt_i_version,
1196         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1197         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1198         Opt_inode_readahead_blks, Opt_journal_ioprio,
1199         Opt_dioread_nolock, Opt_dioread_lock,
1200         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1201 };
1202
1203 static const match_table_t tokens = {
1204         {Opt_bsd_df, "bsddf"},
1205         {Opt_minix_df, "minixdf"},
1206         {Opt_grpid, "grpid"},
1207         {Opt_grpid, "bsdgroups"},
1208         {Opt_nogrpid, "nogrpid"},
1209         {Opt_nogrpid, "sysvgroups"},
1210         {Opt_resgid, "resgid=%u"},
1211         {Opt_resuid, "resuid=%u"},
1212         {Opt_sb, "sb=%u"},
1213         {Opt_err_cont, "errors=continue"},
1214         {Opt_err_panic, "errors=panic"},
1215         {Opt_err_ro, "errors=remount-ro"},
1216         {Opt_nouid32, "nouid32"},
1217         {Opt_debug, "debug"},
1218         {Opt_removed, "oldalloc"},
1219         {Opt_removed, "orlov"},
1220         {Opt_user_xattr, "user_xattr"},
1221         {Opt_nouser_xattr, "nouser_xattr"},
1222         {Opt_acl, "acl"},
1223         {Opt_noacl, "noacl"},
1224         {Opt_noload, "norecovery"},
1225         {Opt_noload, "noload"},
1226         {Opt_removed, "nobh"},
1227         {Opt_removed, "bh"},
1228         {Opt_commit, "commit=%u"},
1229         {Opt_min_batch_time, "min_batch_time=%u"},
1230         {Opt_max_batch_time, "max_batch_time=%u"},
1231         {Opt_journal_dev, "journal_dev=%u"},
1232         {Opt_journal_checksum, "journal_checksum"},
1233         {Opt_journal_async_commit, "journal_async_commit"},
1234         {Opt_abort, "abort"},
1235         {Opt_data_journal, "data=journal"},
1236         {Opt_data_ordered, "data=ordered"},
1237         {Opt_data_writeback, "data=writeback"},
1238         {Opt_data_err_abort, "data_err=abort"},
1239         {Opt_data_err_ignore, "data_err=ignore"},
1240         {Opt_offusrjquota, "usrjquota="},
1241         {Opt_usrjquota, "usrjquota=%s"},
1242         {Opt_offgrpjquota, "grpjquota="},
1243         {Opt_grpjquota, "grpjquota=%s"},
1244         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1245         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1246         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1247         {Opt_grpquota, "grpquota"},
1248         {Opt_noquota, "noquota"},
1249         {Opt_quota, "quota"},
1250         {Opt_usrquota, "usrquota"},
1251         {Opt_barrier, "barrier=%u"},
1252         {Opt_barrier, "barrier"},
1253         {Opt_nobarrier, "nobarrier"},
1254         {Opt_i_version, "i_version"},
1255         {Opt_stripe, "stripe=%u"},
1256         {Opt_delalloc, "delalloc"},
1257         {Opt_nodelalloc, "nodelalloc"},
1258         {Opt_mblk_io_submit, "mblk_io_submit"},
1259         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1260         {Opt_block_validity, "block_validity"},
1261         {Opt_noblock_validity, "noblock_validity"},
1262         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1263         {Opt_journal_ioprio, "journal_ioprio=%u"},
1264         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1265         {Opt_auto_da_alloc, "auto_da_alloc"},
1266         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1267         {Opt_dioread_nolock, "dioread_nolock"},
1268         {Opt_dioread_lock, "dioread_lock"},
1269         {Opt_discard, "discard"},
1270         {Opt_nodiscard, "nodiscard"},
1271         {Opt_init_itable, "init_itable=%u"},
1272         {Opt_init_itable, "init_itable"},
1273         {Opt_noinit_itable, "noinit_itable"},
1274         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1275         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1276         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1277         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1278         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1279         {Opt_err, NULL},
1280 };
1281
1282 static ext4_fsblk_t get_sb_block(void **data)
1283 {
1284         ext4_fsblk_t    sb_block;
1285         char            *options = (char *) *data;
1286
1287         if (!options || strncmp(options, "sb=", 3) != 0)
1288                 return 1;       /* Default location */
1289
1290         options += 3;
1291         /* TODO: use simple_strtoll with >32bit ext4 */
1292         sb_block = simple_strtoul(options, &options, 0);
1293         if (*options && *options != ',') {
1294                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1295                        (char *) *data);
1296                 return 1;
1297         }
1298         if (*options == ',')
1299                 options++;
1300         *data = (void *) options;
1301
1302         return sb_block;
1303 }
1304
1305 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1306 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1307         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1308
1309 #ifdef CONFIG_QUOTA
1310 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1311 {
1312         struct ext4_sb_info *sbi = EXT4_SB(sb);
1313         char *qname;
1314
1315         if (sb_any_quota_loaded(sb) &&
1316                 !sbi->s_qf_names[qtype]) {
1317                 ext4_msg(sb, KERN_ERR,
1318                         "Cannot change journaled "
1319                         "quota options when quota turned on");
1320                 return -1;
1321         }
1322         qname = match_strdup(args);
1323         if (!qname) {
1324                 ext4_msg(sb, KERN_ERR,
1325                         "Not enough memory for storing quotafile name");
1326                 return -1;
1327         }
1328         if (sbi->s_qf_names[qtype] &&
1329                 strcmp(sbi->s_qf_names[qtype], qname)) {
1330                 ext4_msg(sb, KERN_ERR,
1331                         "%s quota file already specified", QTYPE2NAME(qtype));
1332                 kfree(qname);
1333                 return -1;
1334         }
1335         sbi->s_qf_names[qtype] = qname;
1336         if (strchr(sbi->s_qf_names[qtype], '/')) {
1337                 ext4_msg(sb, KERN_ERR,
1338                         "quotafile must be on filesystem root");
1339                 kfree(sbi->s_qf_names[qtype]);
1340                 sbi->s_qf_names[qtype] = NULL;
1341                 return -1;
1342         }
1343         set_opt(sb, QUOTA);
1344         return 1;
1345 }
1346
1347 static int clear_qf_name(struct super_block *sb, int qtype)
1348 {
1349
1350         struct ext4_sb_info *sbi = EXT4_SB(sb);
1351
1352         if (sb_any_quota_loaded(sb) &&
1353                 sbi->s_qf_names[qtype]) {
1354                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1355                         " when quota turned on");
1356                 return -1;
1357         }
1358         /*
1359          * The space will be released later when all options are confirmed
1360          * to be correct
1361          */
1362         sbi->s_qf_names[qtype] = NULL;
1363         return 1;
1364 }
1365 #endif
1366
1367 #define MOPT_SET        0x0001
1368 #define MOPT_CLEAR      0x0002
1369 #define MOPT_NOSUPPORT  0x0004
1370 #define MOPT_EXPLICIT   0x0008
1371 #define MOPT_CLEAR_ERR  0x0010
1372 #define MOPT_GTE0       0x0020
1373 #ifdef CONFIG_QUOTA
1374 #define MOPT_Q          0
1375 #define MOPT_QFMT       0x0040
1376 #else
1377 #define MOPT_Q          MOPT_NOSUPPORT
1378 #define MOPT_QFMT       MOPT_NOSUPPORT
1379 #endif
1380 #define MOPT_DATAJ      0x0080
1381
1382 static const struct mount_opts {
1383         int     token;
1384         int     mount_opt;
1385         int     flags;
1386 } ext4_mount_opts[] = {
1387         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1388         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1389         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1390         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1391         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1392         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1393         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1394         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1395         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1396         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1397         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1398         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1399         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1400         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1401         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1402         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1403                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1404         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1405         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1406         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1407         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1408         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1409         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1410         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1411         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1412         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1413         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1414         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1415         {Opt_commit, 0, MOPT_GTE0},
1416         {Opt_max_batch_time, 0, MOPT_GTE0},
1417         {Opt_min_batch_time, 0, MOPT_GTE0},
1418         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1419         {Opt_init_itable, 0, MOPT_GTE0},
1420         {Opt_stripe, 0, MOPT_GTE0},
1421         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1422         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1423         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1424 #ifdef CONFIG_EXT4_FS_XATTR
1425         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1426         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1427 #else
1428         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1429         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1430 #endif
1431 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1432         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1433         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1434 #else
1435         {Opt_acl, 0, MOPT_NOSUPPORT},
1436         {Opt_noacl, 0, MOPT_NOSUPPORT},
1437 #endif
1438         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1439         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1440         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1441         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1442                                                         MOPT_SET | MOPT_Q},
1443         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1444                                                         MOPT_SET | MOPT_Q},
1445         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1446                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1447         {Opt_usrjquota, 0, MOPT_Q},
1448         {Opt_grpjquota, 0, MOPT_Q},
1449         {Opt_offusrjquota, 0, MOPT_Q},
1450         {Opt_offgrpjquota, 0, MOPT_Q},
1451         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1452         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1453         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1454         {Opt_err, 0, 0}
1455 };
1456
1457 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1458                             substring_t *args, unsigned long *journal_devnum,
1459                             unsigned int *journal_ioprio, int is_remount)
1460 {
1461         struct ext4_sb_info *sbi = EXT4_SB(sb);
1462         const struct mount_opts *m;
1463         int arg = 0;
1464
1465 #ifdef CONFIG_QUOTA
1466         if (token == Opt_usrjquota)
1467                 return set_qf_name(sb, USRQUOTA, &args[0]);
1468         else if (token == Opt_grpjquota)
1469                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1470         else if (token == Opt_offusrjquota)
1471                 return clear_qf_name(sb, USRQUOTA);
1472         else if (token == Opt_offgrpjquota)
1473                 return clear_qf_name(sb, GRPQUOTA);
1474 #endif
1475         if (args->from && match_int(args, &arg))
1476                 return -1;
1477         switch (token) {
1478         case Opt_noacl:
1479         case Opt_nouser_xattr:
1480                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1481                 break;
1482         case Opt_sb:
1483                 return 1;       /* handled by get_sb_block() */
1484         case Opt_removed:
1485                 ext4_msg(sb, KERN_WARNING,
1486                          "Ignoring removed %s option", opt);
1487                 return 1;
1488         case Opt_resuid:
1489                 sbi->s_resuid = arg;
1490                 return 1;
1491         case Opt_resgid:
1492                 sbi->s_resgid = arg;
1493                 return 1;
1494         case Opt_abort:
1495                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1496                 return 1;
1497         case Opt_i_version:
1498                 sb->s_flags |= MS_I_VERSION;
1499                 return 1;
1500         case Opt_journal_dev:
1501                 if (is_remount) {
1502                         ext4_msg(sb, KERN_ERR,
1503                                  "Cannot specify journal on remount");
1504                         return -1;
1505                 }
1506                 *journal_devnum = arg;
1507                 return 1;
1508         case Opt_journal_ioprio:
1509                 if (arg < 0 || arg > 7)
1510                         return -1;
1511                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1512                 return 1;
1513         }
1514
1515         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1516                 if (token != m->token)
1517                         continue;
1518                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1519                         return -1;
1520                 if (m->flags & MOPT_EXPLICIT)
1521                         set_opt2(sb, EXPLICIT_DELALLOC);
1522                 if (m->flags & MOPT_CLEAR_ERR)
1523                         clear_opt(sb, ERRORS_MASK);
1524                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1525                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1526                                  "options when quota turned on");
1527                         return -1;
1528                 }
1529
1530                 if (m->flags & MOPT_NOSUPPORT) {
1531                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1532                 } else if (token == Opt_commit) {
1533                         if (arg == 0)
1534                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1535                         sbi->s_commit_interval = HZ * arg;
1536                 } else if (token == Opt_max_batch_time) {
1537                         if (arg == 0)
1538                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1539                         sbi->s_max_batch_time = arg;
1540                 } else if (token == Opt_min_batch_time) {
1541                         sbi->s_min_batch_time = arg;
1542                 } else if (token == Opt_inode_readahead_blks) {
1543                         if (arg > (1 << 30))
1544                                 return -1;
1545                         if (arg && !is_power_of_2(arg)) {
1546                                 ext4_msg(sb, KERN_ERR,
1547                                          "EXT4-fs: inode_readahead_blks"
1548                                          " must be a power of 2");
1549                                 return -1;
1550                         }
1551                         sbi->s_inode_readahead_blks = arg;
1552                 } else if (token == Opt_init_itable) {
1553                         set_opt(sb, INIT_INODE_TABLE);
1554                         if (!args->from)
1555                                 arg = EXT4_DEF_LI_WAIT_MULT;
1556                         sbi->s_li_wait_mult = arg;
1557                 } else if (token == Opt_stripe) {
1558                         sbi->s_stripe = arg;
1559                 } else if (m->flags & MOPT_DATAJ) {
1560                         if (is_remount) {
1561                                 if (!sbi->s_journal)
1562                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1563                                 else if (test_opt(sb, DATA_FLAGS) !=
1564                                          m->mount_opt) {
1565                                         ext4_msg(sb, KERN_ERR,
1566                                          "Cannot change data mode on remount");
1567                                         return -1;
1568                                 }
1569                         } else {
1570                                 clear_opt(sb, DATA_FLAGS);
1571                                 sbi->s_mount_opt |= m->mount_opt;
1572                         }
1573 #ifdef CONFIG_QUOTA
1574                 } else if (m->flags & MOPT_QFMT) {
1575                         if (sb_any_quota_loaded(sb) &&
1576                             sbi->s_jquota_fmt != m->mount_opt) {
1577                                 ext4_msg(sb, KERN_ERR, "Cannot "
1578                                          "change journaled quota options "
1579                                          "when quota turned on");
1580                                 return -1;
1581                         }
1582                         sbi->s_jquota_fmt = m->mount_opt;
1583 #endif
1584                 } else {
1585                         if (!args->from)
1586                                 arg = 1;
1587                         if (m->flags & MOPT_CLEAR)
1588                                 arg = !arg;
1589                         else if (unlikely(!(m->flags & MOPT_SET))) {
1590                                 ext4_msg(sb, KERN_WARNING,
1591                                          "buggy handling of option %s", opt);
1592                                 WARN_ON(1);
1593                                 return -1;
1594                         }
1595                         if (arg != 0)
1596                                 sbi->s_mount_opt |= m->mount_opt;
1597                         else
1598                                 sbi->s_mount_opt &= ~m->mount_opt;
1599                 }
1600                 return 1;
1601         }
1602         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1603                  "or missing value", opt);
1604         return -1;
1605 }
1606
1607 static int parse_options(char *options, struct super_block *sb,
1608                          unsigned long *journal_devnum,
1609                          unsigned int *journal_ioprio,
1610                          int is_remount)
1611 {
1612 #ifdef CONFIG_QUOTA
1613         struct ext4_sb_info *sbi = EXT4_SB(sb);
1614 #endif
1615         char *p;
1616         substring_t args[MAX_OPT_ARGS];
1617         int token;
1618
1619         if (!options)
1620                 return 1;
1621
1622         while ((p = strsep(&options, ",")) != NULL) {
1623                 if (!*p)
1624                         continue;
1625                 /*
1626                  * Initialize args struct so we know whether arg was
1627                  * found; some options take optional arguments.
1628                  */
1629                 args[0].to = args[0].from = 0;
1630                 token = match_token(p, tokens, args);
1631                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1632                                      journal_ioprio, is_remount) < 0)
1633                         return 0;
1634         }
1635 #ifdef CONFIG_QUOTA
1636         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1637                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1638                         clear_opt(sb, USRQUOTA);
1639
1640                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1641                         clear_opt(sb, GRPQUOTA);
1642
1643                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1644                         ext4_msg(sb, KERN_ERR, "old and new quota "
1645                                         "format mixing");
1646                         return 0;
1647                 }
1648
1649                 if (!sbi->s_jquota_fmt) {
1650                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1651                                         "not specified");
1652                         return 0;
1653                 }
1654         } else {
1655                 if (sbi->s_jquota_fmt) {
1656                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1657                                         "specified with no journaling "
1658                                         "enabled");
1659                         return 0;
1660                 }
1661         }
1662 #endif
1663         return 1;
1664 }
1665
1666 static inline void ext4_show_quota_options(struct seq_file *seq,
1667                                            struct super_block *sb)
1668 {
1669 #if defined(CONFIG_QUOTA)
1670         struct ext4_sb_info *sbi = EXT4_SB(sb);
1671
1672         if (sbi->s_jquota_fmt) {
1673                 char *fmtname = "";
1674
1675                 switch (sbi->s_jquota_fmt) {
1676                 case QFMT_VFS_OLD:
1677                         fmtname = "vfsold";
1678                         break;
1679                 case QFMT_VFS_V0:
1680                         fmtname = "vfsv0";
1681                         break;
1682                 case QFMT_VFS_V1:
1683                         fmtname = "vfsv1";
1684                         break;
1685                 }
1686                 seq_printf(seq, ",jqfmt=%s", fmtname);
1687         }
1688
1689         if (sbi->s_qf_names[USRQUOTA])
1690                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1691
1692         if (sbi->s_qf_names[GRPQUOTA])
1693                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1694
1695         if (test_opt(sb, USRQUOTA))
1696                 seq_puts(seq, ",usrquota");
1697
1698         if (test_opt(sb, GRPQUOTA))
1699                 seq_puts(seq, ",grpquota");
1700 #endif
1701 }
1702
1703 static const char *token2str(int token)
1704 {
1705         static const struct match_token *t;
1706
1707         for (t = tokens; t->token != Opt_err; t++)
1708                 if (t->token == token && !strchr(t->pattern, '='))
1709                         break;
1710         return t->pattern;
1711 }
1712
1713 /*
1714  * Show an option if
1715  *  - it's set to a non-default value OR
1716  *  - if the per-sb default is different from the global default
1717  */
1718 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1719                               int nodefs)
1720 {
1721         struct ext4_sb_info *sbi = EXT4_SB(sb);
1722         struct ext4_super_block *es = sbi->s_es;
1723         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1724         const struct mount_opts *m;
1725         char sep = nodefs ? '\n' : ',';
1726
1727 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1728 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1729
1730         if (sbi->s_sb_block != 1)
1731                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1732
1733         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1734                 int want_set = m->flags & MOPT_SET;
1735                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1736                     (m->flags & MOPT_CLEAR_ERR))
1737                         continue;
1738                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1739                         continue; /* skip if same as the default */
1740                 if ((want_set &&
1741                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1742                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1743                         continue; /* select Opt_noFoo vs Opt_Foo */
1744                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1745         }
1746
1747         if (nodefs || sbi->s_resuid != EXT4_DEF_RESUID ||
1748             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1749                 SEQ_OPTS_PRINT("resuid=%u", sbi->s_resuid);
1750         if (nodefs || sbi->s_resgid != EXT4_DEF_RESGID ||
1751             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1752                 SEQ_OPTS_PRINT("resgid=%u", sbi->s_resgid);
1753         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1754         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1755                 SEQ_OPTS_PUTS("errors=remount-ro");
1756         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1757                 SEQ_OPTS_PUTS("errors=continue");
1758         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1759                 SEQ_OPTS_PUTS("errors=panic");
1760         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1761                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1762         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1763                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1764         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1765                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1766         if (sb->s_flags & MS_I_VERSION)
1767                 SEQ_OPTS_PUTS("i_version");
1768         if (nodefs || sbi->s_stripe)
1769                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1770         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1771                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1772                         SEQ_OPTS_PUTS("data=journal");
1773                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1774                         SEQ_OPTS_PUTS("data=ordered");
1775                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1776                         SEQ_OPTS_PUTS("data=writeback");
1777         }
1778         if (nodefs ||
1779             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1780                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1781                                sbi->s_inode_readahead_blks);
1782
1783         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1784                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1785                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1786
1787         ext4_show_quota_options(seq, sb);
1788         return 0;
1789 }
1790
1791 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1792 {
1793         return _ext4_show_options(seq, root->d_sb, 0);
1794 }
1795
1796 static int options_seq_show(struct seq_file *seq, void *offset)
1797 {
1798         struct super_block *sb = seq->private;
1799         int rc;
1800
1801         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1802         rc = _ext4_show_options(seq, sb, 1);
1803         seq_puts(seq, "\n");
1804         return rc;
1805 }
1806
1807 static int options_open_fs(struct inode *inode, struct file *file)
1808 {
1809         return single_open(file, options_seq_show, PDE(inode)->data);
1810 }
1811
1812 static const struct file_operations ext4_seq_options_fops = {
1813         .owner = THIS_MODULE,
1814         .open = options_open_fs,
1815         .read = seq_read,
1816         .llseek = seq_lseek,
1817         .release = single_release,
1818 };
1819
1820 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1821                             int read_only)
1822 {
1823         struct ext4_sb_info *sbi = EXT4_SB(sb);
1824         int res = 0;
1825
1826         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1827                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1828                          "forcing read-only mode");
1829                 res = MS_RDONLY;
1830         }
1831         if (read_only)
1832                 goto done;
1833         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1834                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1835                          "running e2fsck is recommended");
1836         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1837                 ext4_msg(sb, KERN_WARNING,
1838                          "warning: mounting fs with errors, "
1839                          "running e2fsck is recommended");
1840         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1841                  le16_to_cpu(es->s_mnt_count) >=
1842                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1843                 ext4_msg(sb, KERN_WARNING,
1844                          "warning: maximal mount count reached, "
1845                          "running e2fsck is recommended");
1846         else if (le32_to_cpu(es->s_checkinterval) &&
1847                 (le32_to_cpu(es->s_lastcheck) +
1848                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1849                 ext4_msg(sb, KERN_WARNING,
1850                          "warning: checktime reached, "
1851                          "running e2fsck is recommended");
1852         if (!sbi->s_journal)
1853                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1854         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1855                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1856         le16_add_cpu(&es->s_mnt_count, 1);
1857         es->s_mtime = cpu_to_le32(get_seconds());
1858         ext4_update_dynamic_rev(sb);
1859         if (sbi->s_journal)
1860                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1861
1862         ext4_commit_super(sb, 1);
1863 done:
1864         if (test_opt(sb, DEBUG))
1865                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1866                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1867                         sb->s_blocksize,
1868                         sbi->s_groups_count,
1869                         EXT4_BLOCKS_PER_GROUP(sb),
1870                         EXT4_INODES_PER_GROUP(sb),
1871                         sbi->s_mount_opt, sbi->s_mount_opt2);
1872
1873         cleancache_init_fs(sb);
1874         return res;
1875 }
1876
1877 static int ext4_fill_flex_info(struct super_block *sb)
1878 {
1879         struct ext4_sb_info *sbi = EXT4_SB(sb);
1880         struct ext4_group_desc *gdp = NULL;
1881         ext4_group_t flex_group_count;
1882         ext4_group_t flex_group;
1883         unsigned int groups_per_flex = 0;
1884         size_t size;
1885         int i;
1886
1887         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1888         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1889                 sbi->s_log_groups_per_flex = 0;
1890                 return 1;
1891         }
1892         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1893
1894         /* We allocate both existing and potentially added groups */
1895         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1896                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1897                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1898         size = flex_group_count * sizeof(struct flex_groups);
1899         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1900         if (sbi->s_flex_groups == NULL) {
1901                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1902                          flex_group_count);
1903                 goto failed;
1904         }
1905
1906         for (i = 0; i < sbi->s_groups_count; i++) {
1907                 gdp = ext4_get_group_desc(sb, i, NULL);
1908
1909                 flex_group = ext4_flex_group(sbi, i);
1910                 atomic_add(ext4_free_inodes_count(sb, gdp),
1911                            &sbi->s_flex_groups[flex_group].free_inodes);
1912                 atomic_add(ext4_free_group_clusters(sb, gdp),
1913                            &sbi->s_flex_groups[flex_group].free_clusters);
1914                 atomic_add(ext4_used_dirs_count(sb, gdp),
1915                            &sbi->s_flex_groups[flex_group].used_dirs);
1916         }
1917
1918         return 1;
1919 failed:
1920         return 0;
1921 }
1922
1923 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1924                             struct ext4_group_desc *gdp)
1925 {
1926         __u16 crc = 0;
1927
1928         if (sbi->s_es->s_feature_ro_compat &
1929             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1930                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1931                 __le32 le_group = cpu_to_le32(block_group);
1932
1933                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1934                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1935                 crc = crc16(crc, (__u8 *)gdp, offset);
1936                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1937                 /* for checksum of struct ext4_group_desc do the rest...*/
1938                 if ((sbi->s_es->s_feature_incompat &
1939                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1940                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1941                         crc = crc16(crc, (__u8 *)gdp + offset,
1942                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1943                                         offset);
1944         }
1945
1946         return cpu_to_le16(crc);
1947 }
1948
1949 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1950                                 struct ext4_group_desc *gdp)
1951 {
1952         if ((sbi->s_es->s_feature_ro_compat &
1953              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1954             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1955                 return 0;
1956
1957         return 1;
1958 }
1959
1960 /* Called at mount-time, super-block is locked */
1961 static int ext4_check_descriptors(struct super_block *sb,
1962                                   ext4_group_t *first_not_zeroed)
1963 {
1964         struct ext4_sb_info *sbi = EXT4_SB(sb);
1965         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1966         ext4_fsblk_t last_block;
1967         ext4_fsblk_t block_bitmap;
1968         ext4_fsblk_t inode_bitmap;
1969         ext4_fsblk_t inode_table;
1970         int flexbg_flag = 0;
1971         ext4_group_t i, grp = sbi->s_groups_count;
1972
1973         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1974                 flexbg_flag = 1;
1975
1976         ext4_debug("Checking group descriptors");
1977
1978         for (i = 0; i < sbi->s_groups_count; i++) {
1979                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1980
1981                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1982                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1983                 else
1984                         last_block = first_block +
1985                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1986
1987                 if ((grp == sbi->s_groups_count) &&
1988                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
1989                         grp = i;
1990
1991                 block_bitmap = ext4_block_bitmap(sb, gdp);
1992                 if (block_bitmap < first_block || block_bitmap > last_block) {
1993                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1994                                "Block bitmap for group %u not in group "
1995                                "(block %llu)!", i, block_bitmap);
1996                         return 0;
1997                 }
1998                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1999                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2000                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2001                                "Inode bitmap for group %u not in group "
2002                                "(block %llu)!", i, inode_bitmap);
2003                         return 0;
2004                 }
2005                 inode_table = ext4_inode_table(sb, gdp);
2006                 if (inode_table < first_block ||
2007                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2008                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2009                                "Inode table for group %u not in group "
2010                                "(block %llu)!", i, inode_table);
2011                         return 0;
2012                 }
2013                 ext4_lock_group(sb, i);
2014                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2015                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2016                                  "Checksum for group %u failed (%u!=%u)",
2017                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2018                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2019                         if (!(sb->s_flags & MS_RDONLY)) {
2020                                 ext4_unlock_group(sb, i);
2021                                 return 0;
2022                         }
2023                 }
2024                 ext4_unlock_group(sb, i);
2025                 if (!flexbg_flag)
2026                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2027         }
2028         if (NULL != first_not_zeroed)
2029                 *first_not_zeroed = grp;
2030
2031         ext4_free_blocks_count_set(sbi->s_es,
2032                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2033         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2034         return 1;
2035 }
2036
2037 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2038  * the superblock) which were deleted from all directories, but held open by
2039  * a process at the time of a crash.  We walk the list and try to delete these
2040  * inodes at recovery time (only with a read-write filesystem).
2041  *
2042  * In order to keep the orphan inode chain consistent during traversal (in
2043  * case of crash during recovery), we link each inode into the superblock
2044  * orphan list_head and handle it the same way as an inode deletion during
2045  * normal operation (which journals the operations for us).
2046  *
2047  * We only do an iget() and an iput() on each inode, which is very safe if we
2048  * accidentally point at an in-use or already deleted inode.  The worst that
2049  * can happen in this case is that we get a "bit already cleared" message from
2050  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2051  * e2fsck was run on this filesystem, and it must have already done the orphan
2052  * inode cleanup for us, so we can safely abort without any further action.
2053  */
2054 static void ext4_orphan_cleanup(struct super_block *sb,
2055                                 struct ext4_super_block *es)
2056 {
2057         unsigned int s_flags = sb->s_flags;
2058         int nr_orphans = 0, nr_truncates = 0;
2059 #ifdef CONFIG_QUOTA
2060         int i;
2061 #endif
2062         if (!es->s_last_orphan) {
2063                 jbd_debug(4, "no orphan inodes to clean up\n");
2064                 return;
2065         }
2066
2067         if (bdev_read_only(sb->s_bdev)) {
2068                 ext4_msg(sb, KERN_ERR, "write access "
2069                         "unavailable, skipping orphan cleanup");
2070                 return;
2071         }
2072
2073         /* Check if feature set would not allow a r/w mount */
2074         if (!ext4_feature_set_ok(sb, 0)) {
2075                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2076                          "unknown ROCOMPAT features");
2077                 return;
2078         }
2079
2080         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2081                 if (es->s_last_orphan)
2082                         jbd_debug(1, "Errors on filesystem, "
2083                                   "clearing orphan list.\n");
2084                 es->s_last_orphan = 0;
2085                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2086                 return;
2087         }
2088
2089         if (s_flags & MS_RDONLY) {
2090                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2091                 sb->s_flags &= ~MS_RDONLY;
2092         }
2093 #ifdef CONFIG_QUOTA
2094         /* Needed for iput() to work correctly and not trash data */
2095         sb->s_flags |= MS_ACTIVE;
2096         /* Turn on quotas so that they are updated correctly */
2097         for (i = 0; i < MAXQUOTAS; i++) {
2098                 if (EXT4_SB(sb)->s_qf_names[i]) {
2099                         int ret = ext4_quota_on_mount(sb, i);
2100                         if (ret < 0)
2101                                 ext4_msg(sb, KERN_ERR,
2102                                         "Cannot turn on journaled "
2103                                         "quota: error %d", ret);
2104                 }
2105         }
2106 #endif
2107
2108         while (es->s_last_orphan) {
2109                 struct inode *inode;
2110
2111                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2112                 if (IS_ERR(inode)) {
2113                         es->s_last_orphan = 0;
2114                         break;
2115                 }
2116
2117                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2118                 dquot_initialize(inode);
2119                 if (inode->i_nlink) {
2120                         ext4_msg(sb, KERN_DEBUG,
2121                                 "%s: truncating inode %lu to %lld bytes",
2122                                 __func__, inode->i_ino, inode->i_size);
2123                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2124                                   inode->i_ino, inode->i_size);
2125                         ext4_truncate(inode);
2126                         nr_truncates++;
2127                 } else {
2128                         ext4_msg(sb, KERN_DEBUG,
2129                                 "%s: deleting unreferenced inode %lu",
2130                                 __func__, inode->i_ino);
2131                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2132                                   inode->i_ino);
2133                         nr_orphans++;
2134                 }
2135                 iput(inode);  /* The delete magic happens here! */
2136         }
2137
2138 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2139
2140         if (nr_orphans)
2141                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2142                        PLURAL(nr_orphans));
2143         if (nr_truncates)
2144                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2145                        PLURAL(nr_truncates));
2146 #ifdef CONFIG_QUOTA
2147         /* Turn quotas off */
2148         for (i = 0; i < MAXQUOTAS; i++) {
2149                 if (sb_dqopt(sb)->files[i])
2150                         dquot_quota_off(sb, i);
2151         }
2152 #endif
2153         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2154 }
2155
2156 /*
2157  * Maximal extent format file size.
2158  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2159  * extent format containers, within a sector_t, and within i_blocks
2160  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2161  * so that won't be a limiting factor.
2162  *
2163  * However there is other limiting factor. We do store extents in the form
2164  * of starting block and length, hence the resulting length of the extent
2165  * covering maximum file size must fit into on-disk format containers as
2166  * well. Given that length is always by 1 unit bigger than max unit (because
2167  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2168  *
2169  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2170  */
2171 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2172 {
2173         loff_t res;
2174         loff_t upper_limit = MAX_LFS_FILESIZE;
2175
2176         /* small i_blocks in vfs inode? */
2177         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2178                 /*
2179                  * CONFIG_LBDAF is not enabled implies the inode
2180                  * i_block represent total blocks in 512 bytes
2181                  * 32 == size of vfs inode i_blocks * 8
2182                  */
2183                 upper_limit = (1LL << 32) - 1;
2184
2185                 /* total blocks in file system block size */
2186                 upper_limit >>= (blkbits - 9);
2187                 upper_limit <<= blkbits;
2188         }
2189
2190         /*
2191          * 32-bit extent-start container, ee_block. We lower the maxbytes
2192          * by one fs block, so ee_len can cover the extent of maximum file
2193          * size
2194          */
2195         res = (1LL << 32) - 1;
2196         res <<= blkbits;
2197
2198         /* Sanity check against vm- & vfs- imposed limits */
2199         if (res > upper_limit)
2200                 res = upper_limit;
2201
2202         return res;
2203 }
2204
2205 /*
2206  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2207  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2208  * We need to be 1 filesystem block less than the 2^48 sector limit.
2209  */
2210 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2211 {
2212         loff_t res = EXT4_NDIR_BLOCKS;
2213         int meta_blocks;
2214         loff_t upper_limit;
2215         /* This is calculated to be the largest file size for a dense, block
2216          * mapped file such that the file's total number of 512-byte sectors,
2217          * including data and all indirect blocks, does not exceed (2^48 - 1).
2218          *
2219          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2220          * number of 512-byte sectors of the file.
2221          */
2222
2223         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2224                 /*
2225                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2226                  * the inode i_block field represents total file blocks in
2227                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2228                  */
2229                 upper_limit = (1LL << 32) - 1;
2230
2231                 /* total blocks in file system block size */
2232                 upper_limit >>= (bits - 9);
2233
2234         } else {
2235                 /*
2236                  * We use 48 bit ext4_inode i_blocks
2237                  * With EXT4_HUGE_FILE_FL set the i_blocks
2238                  * represent total number of blocks in
2239                  * file system block size
2240                  */
2241                 upper_limit = (1LL << 48) - 1;
2242
2243         }
2244
2245         /* indirect blocks */
2246         meta_blocks = 1;
2247         /* double indirect blocks */
2248         meta_blocks += 1 + (1LL << (bits-2));
2249         /* tripple indirect blocks */
2250         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2251
2252         upper_limit -= meta_blocks;
2253         upper_limit <<= bits;
2254
2255         res += 1LL << (bits-2);
2256         res += 1LL << (2*(bits-2));
2257         res += 1LL << (3*(bits-2));
2258         res <<= bits;
2259         if (res > upper_limit)
2260                 res = upper_limit;
2261
2262         if (res > MAX_LFS_FILESIZE)
2263                 res = MAX_LFS_FILESIZE;
2264
2265         return res;
2266 }
2267
2268 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2269                                    ext4_fsblk_t logical_sb_block, int nr)
2270 {
2271         struct ext4_sb_info *sbi = EXT4_SB(sb);
2272         ext4_group_t bg, first_meta_bg;
2273         int has_super = 0;
2274
2275         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2276
2277         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2278             nr < first_meta_bg)
2279                 return logical_sb_block + nr + 1;
2280         bg = sbi->s_desc_per_block * nr;
2281         if (ext4_bg_has_super(sb, bg))
2282                 has_super = 1;
2283
2284         return (has_super + ext4_group_first_block_no(sb, bg));
2285 }
2286
2287 /**
2288  * ext4_get_stripe_size: Get the stripe size.
2289  * @sbi: In memory super block info
2290  *
2291  * If we have specified it via mount option, then
2292  * use the mount option value. If the value specified at mount time is
2293  * greater than the blocks per group use the super block value.
2294  * If the super block value is greater than blocks per group return 0.
2295  * Allocator needs it be less than blocks per group.
2296  *
2297  */
2298 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2299 {
2300         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2301         unsigned long stripe_width =
2302                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2303         int ret;
2304
2305         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2306                 ret = sbi->s_stripe;
2307         else if (stripe_width <= sbi->s_blocks_per_group)
2308                 ret = stripe_width;
2309         else if (stride <= sbi->s_blocks_per_group)
2310                 ret = stride;
2311         else
2312                 ret = 0;
2313
2314         /*
2315          * If the stripe width is 1, this makes no sense and
2316          * we set it to 0 to turn off stripe handling code.
2317          */
2318         if (ret <= 1)
2319                 ret = 0;
2320
2321         return ret;
2322 }
2323
2324 /* sysfs supprt */
2325
2326 struct ext4_attr {
2327         struct attribute attr;
2328         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2329         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2330                          const char *, size_t);
2331         int offset;
2332 };
2333
2334 static int parse_strtoul(const char *buf,
2335                 unsigned long max, unsigned long *value)
2336 {
2337         char *endp;
2338
2339         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2340         endp = skip_spaces(endp);
2341         if (*endp || *value > max)
2342                 return -EINVAL;
2343
2344         return 0;
2345 }
2346
2347 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2348                                               struct ext4_sb_info *sbi,
2349                                               char *buf)
2350 {
2351         return snprintf(buf, PAGE_SIZE, "%llu\n",
2352                 (s64) EXT4_C2B(sbi,
2353                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2354 }
2355
2356 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2357                                          struct ext4_sb_info *sbi, char *buf)
2358 {
2359         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2360
2361         if (!sb->s_bdev->bd_part)
2362                 return snprintf(buf, PAGE_SIZE, "0\n");
2363         return snprintf(buf, PAGE_SIZE, "%lu\n",
2364                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2365                          sbi->s_sectors_written_start) >> 1);
2366 }
2367
2368 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2369                                           struct ext4_sb_info *sbi, char *buf)
2370 {
2371         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2372
2373         if (!sb->s_bdev->bd_part)
2374                 return snprintf(buf, PAGE_SIZE, "0\n");
2375         return snprintf(buf, PAGE_SIZE, "%llu\n",
2376                         (unsigned long long)(sbi->s_kbytes_written +
2377                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2378                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2379 }
2380
2381 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2382                                           struct ext4_sb_info *sbi,
2383                                           const char *buf, size_t count)
2384 {
2385         unsigned long t;
2386
2387         if (parse_strtoul(buf, 0x40000000, &t))
2388                 return -EINVAL;
2389
2390         if (t && !is_power_of_2(t))
2391                 return -EINVAL;
2392
2393         sbi->s_inode_readahead_blks = t;
2394         return count;
2395 }
2396
2397 static ssize_t sbi_ui_show(struct ext4_attr *a,
2398                            struct ext4_sb_info *sbi, char *buf)
2399 {
2400         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2401
2402         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2403 }
2404
2405 static ssize_t sbi_ui_store(struct ext4_attr *a,
2406                             struct ext4_sb_info *sbi,
2407                             const char *buf, size_t count)
2408 {
2409         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2410         unsigned long t;
2411
2412         if (parse_strtoul(buf, 0xffffffff, &t))
2413                 return -EINVAL;
2414         *ui = t;
2415         return count;
2416 }
2417
2418 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2419 static struct ext4_attr ext4_attr_##_name = {                   \
2420         .attr = {.name = __stringify(_name), .mode = _mode },   \
2421         .show   = _show,                                        \
2422         .store  = _store,                                       \
2423         .offset = offsetof(struct ext4_sb_info, _elname),       \
2424 }
2425 #define EXT4_ATTR(name, mode, show, store) \
2426 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2427
2428 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2429 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2430 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2431 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2432         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2433 #define ATTR_LIST(name) &ext4_attr_##name.attr
2434
2435 EXT4_RO_ATTR(delayed_allocation_blocks);
2436 EXT4_RO_ATTR(session_write_kbytes);
2437 EXT4_RO_ATTR(lifetime_write_kbytes);
2438 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2439                  inode_readahead_blks_store, s_inode_readahead_blks);
2440 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2441 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2442 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2443 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2444 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2445 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2446 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2447 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2448
2449 static struct attribute *ext4_attrs[] = {
2450         ATTR_LIST(delayed_allocation_blocks),
2451         ATTR_LIST(session_write_kbytes),
2452         ATTR_LIST(lifetime_write_kbytes),
2453         ATTR_LIST(inode_readahead_blks),
2454         ATTR_LIST(inode_goal),
2455         ATTR_LIST(mb_stats),
2456         ATTR_LIST(mb_max_to_scan),
2457         ATTR_LIST(mb_min_to_scan),
2458         ATTR_LIST(mb_order2_req),
2459         ATTR_LIST(mb_stream_req),
2460         ATTR_LIST(mb_group_prealloc),
2461         ATTR_LIST(max_writeback_mb_bump),
2462         NULL,
2463 };
2464
2465 /* Features this copy of ext4 supports */
2466 EXT4_INFO_ATTR(lazy_itable_init);
2467 EXT4_INFO_ATTR(batched_discard);
2468
2469 static struct attribute *ext4_feat_attrs[] = {
2470         ATTR_LIST(lazy_itable_init),
2471         ATTR_LIST(batched_discard),
2472         NULL,
2473 };
2474
2475 static ssize_t ext4_attr_show(struct kobject *kobj,
2476                               struct attribute *attr, char *buf)
2477 {
2478         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2479                                                 s_kobj);
2480         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2481
2482         return a->show ? a->show(a, sbi, buf) : 0;
2483 }
2484
2485 static ssize_t ext4_attr_store(struct kobject *kobj,
2486                                struct attribute *attr,
2487                                const char *buf, size_t len)
2488 {
2489         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2490                                                 s_kobj);
2491         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2492
2493         return a->store ? a->store(a, sbi, buf, len) : 0;
2494 }
2495
2496 static void ext4_sb_release(struct kobject *kobj)
2497 {
2498         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2499                                                 s_kobj);
2500         complete(&sbi->s_kobj_unregister);
2501 }
2502
2503 static const struct sysfs_ops ext4_attr_ops = {
2504         .show   = ext4_attr_show,
2505         .store  = ext4_attr_store,
2506 };
2507
2508 static struct kobj_type ext4_ktype = {
2509         .default_attrs  = ext4_attrs,
2510         .sysfs_ops      = &ext4_attr_ops,
2511         .release        = ext4_sb_release,
2512 };
2513
2514 static void ext4_feat_release(struct kobject *kobj)
2515 {
2516         complete(&ext4_feat->f_kobj_unregister);
2517 }
2518
2519 static struct kobj_type ext4_feat_ktype = {
2520         .default_attrs  = ext4_feat_attrs,
2521         .sysfs_ops      = &ext4_attr_ops,
2522         .release        = ext4_feat_release,
2523 };
2524
2525 /*
2526  * Check whether this filesystem can be mounted based on
2527  * the features present and the RDONLY/RDWR mount requested.
2528  * Returns 1 if this filesystem can be mounted as requested,
2529  * 0 if it cannot be.
2530  */
2531 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2532 {
2533         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2534                 ext4_msg(sb, KERN_ERR,
2535                         "Couldn't mount because of "
2536                         "unsupported optional features (%x)",
2537                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2538                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2539                 return 0;
2540         }
2541
2542         if (readonly)
2543                 return 1;
2544
2545         /* Check that feature set is OK for a read-write mount */
2546         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2547                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2548                          "unsupported optional features (%x)",
2549                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2550                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2551                 return 0;
2552         }
2553         /*
2554          * Large file size enabled file system can only be mounted
2555          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2556          */
2557         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2558                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2559                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2560                                  "cannot be mounted RDWR without "
2561                                  "CONFIG_LBDAF");
2562                         return 0;
2563                 }
2564         }
2565         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2566             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2567                 ext4_msg(sb, KERN_ERR,
2568                          "Can't support bigalloc feature without "
2569                          "extents feature\n");
2570                 return 0;
2571         }
2572         return 1;
2573 }
2574
2575 /*
2576  * This function is called once a day if we have errors logged
2577  * on the file system
2578  */
2579 static void print_daily_error_info(unsigned long arg)
2580 {
2581         struct super_block *sb = (struct super_block *) arg;
2582         struct ext4_sb_info *sbi;
2583         struct ext4_super_block *es;
2584
2585         sbi = EXT4_SB(sb);
2586         es = sbi->s_es;
2587
2588         if (es->s_error_count)
2589                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2590                          le32_to_cpu(es->s_error_count));
2591         if (es->s_first_error_time) {
2592                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2593                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2594                        (int) sizeof(es->s_first_error_func),
2595                        es->s_first_error_func,
2596                        le32_to_cpu(es->s_first_error_line));
2597                 if (es->s_first_error_ino)
2598                         printk(": inode %u",
2599                                le32_to_cpu(es->s_first_error_ino));
2600                 if (es->s_first_error_block)
2601                         printk(": block %llu", (unsigned long long)
2602                                le64_to_cpu(es->s_first_error_block));
2603                 printk("\n");
2604         }
2605         if (es->s_last_error_time) {
2606                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2607                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2608                        (int) sizeof(es->s_last_error_func),
2609                        es->s_last_error_func,
2610                        le32_to_cpu(es->s_last_error_line));
2611                 if (es->s_last_error_ino)
2612                         printk(": inode %u",
2613                                le32_to_cpu(es->s_last_error_ino));
2614                 if (es->s_last_error_block)
2615                         printk(": block %llu", (unsigned long long)
2616                                le64_to_cpu(es->s_last_error_block));
2617                 printk("\n");
2618         }
2619         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2620 }
2621
2622 /* Find next suitable group and run ext4_init_inode_table */
2623 static int ext4_run_li_request(struct ext4_li_request *elr)
2624 {
2625         struct ext4_group_desc *gdp = NULL;
2626         ext4_group_t group, ngroups;
2627         struct super_block *sb;
2628         unsigned long timeout = 0;
2629         int ret = 0;
2630
2631         sb = elr->lr_super;
2632         ngroups = EXT4_SB(sb)->s_groups_count;
2633
2634         for (group = elr->lr_next_group; group < ngroups; group++) {
2635                 gdp = ext4_get_group_desc(sb, group, NULL);
2636                 if (!gdp) {
2637                         ret = 1;
2638                         break;
2639                 }
2640
2641                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2642                         break;
2643         }
2644
2645         if (group == ngroups)
2646                 ret = 1;
2647
2648         if (!ret) {
2649                 timeout = jiffies;
2650                 ret = ext4_init_inode_table(sb, group,
2651                                             elr->lr_timeout ? 0 : 1);
2652                 if (elr->lr_timeout == 0) {
2653                         timeout = (jiffies - timeout) *
2654                                   elr->lr_sbi->s_li_wait_mult;
2655                         elr->lr_timeout = timeout;
2656                 }
2657                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2658                 elr->lr_next_group = group + 1;
2659         }
2660
2661         return ret;
2662 }
2663
2664 /*
2665  * Remove lr_request from the list_request and free the
2666  * request structure. Should be called with li_list_mtx held
2667  */
2668 static void ext4_remove_li_request(struct ext4_li_request *elr)
2669 {
2670         struct ext4_sb_info *sbi;
2671
2672         if (!elr)
2673                 return;
2674
2675         sbi = elr->lr_sbi;
2676
2677         list_del(&elr->lr_request);
2678         sbi->s_li_request = NULL;
2679         kfree(elr);
2680 }
2681
2682 static void ext4_unregister_li_request(struct super_block *sb)
2683 {
2684         mutex_lock(&ext4_li_mtx);
2685         if (!ext4_li_info) {
2686                 mutex_unlock(&ext4_li_mtx);
2687                 return;
2688         }
2689
2690         mutex_lock(&ext4_li_info->li_list_mtx);
2691         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2692         mutex_unlock(&ext4_li_info->li_list_mtx);
2693         mutex_unlock(&ext4_li_mtx);
2694 }
2695
2696 static struct task_struct *ext4_lazyinit_task;
2697
2698 /*
2699  * This is the function where ext4lazyinit thread lives. It walks
2700  * through the request list searching for next scheduled filesystem.
2701  * When such a fs is found, run the lazy initialization request
2702  * (ext4_rn_li_request) and keep track of the time spend in this
2703  * function. Based on that time we compute next schedule time of
2704  * the request. When walking through the list is complete, compute
2705  * next waking time and put itself into sleep.
2706  */
2707 static int ext4_lazyinit_thread(void *arg)
2708 {
2709         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2710         struct list_head *pos, *n;
2711         struct ext4_li_request *elr;
2712         unsigned long next_wakeup, cur;
2713
2714         BUG_ON(NULL == eli);
2715
2716 cont_thread:
2717         while (true) {
2718                 next_wakeup = MAX_JIFFY_OFFSET;
2719
2720                 mutex_lock(&eli->li_list_mtx);
2721                 if (list_empty(&eli->li_request_list)) {
2722                         mutex_unlock(&eli->li_list_mtx);
2723                         goto exit_thread;
2724                 }
2725
2726                 list_for_each_safe(pos, n, &eli->li_request_list) {
2727                         elr = list_entry(pos, struct ext4_li_request,
2728                                          lr_request);
2729
2730                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2731                                 if (ext4_run_li_request(elr) != 0) {
2732                                         /* error, remove the lazy_init job */
2733                                         ext4_remove_li_request(elr);
2734                                         continue;
2735                                 }
2736                         }
2737
2738                         if (time_before(elr->lr_next_sched, next_wakeup))
2739                                 next_wakeup = elr->lr_next_sched;
2740                 }
2741                 mutex_unlock(&eli->li_list_mtx);
2742
2743                 try_to_freeze();
2744
2745                 cur = jiffies;
2746                 if ((time_after_eq(cur, next_wakeup)) ||
2747                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2748                         cond_resched();
2749                         continue;
2750                 }
2751
2752                 schedule_timeout_interruptible(next_wakeup - cur);
2753
2754                 if (kthread_should_stop()) {
2755                         ext4_clear_request_list();
2756                         goto exit_thread;
2757                 }
2758         }
2759
2760 exit_thread:
2761         /*
2762          * It looks like the request list is empty, but we need
2763          * to check it under the li_list_mtx lock, to prevent any
2764          * additions into it, and of course we should lock ext4_li_mtx
2765          * to atomically free the list and ext4_li_info, because at
2766          * this point another ext4 filesystem could be registering
2767          * new one.
2768          */
2769         mutex_lock(&ext4_li_mtx);
2770         mutex_lock(&eli->li_list_mtx);
2771         if (!list_empty(&eli->li_request_list)) {
2772                 mutex_unlock(&eli->li_list_mtx);
2773                 mutex_unlock(&ext4_li_mtx);
2774                 goto cont_thread;
2775         }
2776         mutex_unlock(&eli->li_list_mtx);
2777         kfree(ext4_li_info);
2778         ext4_li_info = NULL;
2779         mutex_unlock(&ext4_li_mtx);
2780
2781         return 0;
2782 }
2783
2784 static void ext4_clear_request_list(void)
2785 {
2786         struct list_head *pos, *n;
2787         struct ext4_li_request *elr;
2788
2789         mutex_lock(&ext4_li_info->li_list_mtx);
2790         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2791                 elr = list_entry(pos, struct ext4_li_request,
2792                                  lr_request);
2793                 ext4_remove_li_request(elr);
2794         }
2795         mutex_unlock(&ext4_li_info->li_list_mtx);
2796 }
2797
2798 static int ext4_run_lazyinit_thread(void)
2799 {
2800         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2801                                          ext4_li_info, "ext4lazyinit");
2802         if (IS_ERR(ext4_lazyinit_task)) {
2803                 int err = PTR_ERR(ext4_lazyinit_task);
2804                 ext4_clear_request_list();
2805                 kfree(ext4_li_info);
2806                 ext4_li_info = NULL;
2807                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2808                                  "initialization thread\n",
2809                                  err);
2810                 return err;
2811         }
2812         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2813         return 0;
2814 }
2815
2816 /*
2817  * Check whether it make sense to run itable init. thread or not.
2818  * If there is at least one uninitialized inode table, return
2819  * corresponding group number, else the loop goes through all
2820  * groups and return total number of groups.
2821  */
2822 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2823 {
2824         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2825         struct ext4_group_desc *gdp = NULL;
2826
2827         for (group = 0; group < ngroups; group++) {
2828                 gdp = ext4_get_group_desc(sb, group, NULL);
2829                 if (!gdp)
2830                         continue;
2831
2832                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2833                         break;
2834         }
2835
2836         return group;
2837 }
2838
2839 static int ext4_li_info_new(void)
2840 {
2841         struct ext4_lazy_init *eli = NULL;
2842
2843         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2844         if (!eli)
2845                 return -ENOMEM;
2846
2847         INIT_LIST_HEAD(&eli->li_request_list);
2848         mutex_init(&eli->li_list_mtx);
2849
2850         eli->li_state |= EXT4_LAZYINIT_QUIT;
2851
2852         ext4_li_info = eli;
2853
2854         return 0;
2855 }
2856
2857 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2858                                             ext4_group_t start)
2859 {
2860         struct ext4_sb_info *sbi = EXT4_SB(sb);
2861         struct ext4_li_request *elr;
2862         unsigned long rnd;
2863
2864         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2865         if (!elr)
2866                 return NULL;
2867
2868         elr->lr_super = sb;
2869         elr->lr_sbi = sbi;
2870         elr->lr_next_group = start;
2871
2872         /*
2873          * Randomize first schedule time of the request to
2874          * spread the inode table initialization requests
2875          * better.
2876          */
2877         get_random_bytes(&rnd, sizeof(rnd));
2878         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2879                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2880
2881         return elr;
2882 }
2883
2884 static int ext4_register_li_request(struct super_block *sb,
2885                                     ext4_group_t first_not_zeroed)
2886 {
2887         struct ext4_sb_info *sbi = EXT4_SB(sb);
2888         struct ext4_li_request *elr;
2889         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2890         int ret = 0;
2891
2892         if (sbi->s_li_request != NULL) {
2893                 /*
2894                  * Reset timeout so it can be computed again, because
2895                  * s_li_wait_mult might have changed.
2896                  */
2897                 sbi->s_li_request->lr_timeout = 0;
2898                 return 0;
2899         }
2900
2901         if (first_not_zeroed == ngroups ||
2902             (sb->s_flags & MS_RDONLY) ||
2903             !test_opt(sb, INIT_INODE_TABLE))
2904                 return 0;
2905
2906         elr = ext4_li_request_new(sb, first_not_zeroed);
2907         if (!elr)
2908                 return -ENOMEM;
2909
2910         mutex_lock(&ext4_li_mtx);
2911
2912         if (NULL == ext4_li_info) {
2913                 ret = ext4_li_info_new();
2914                 if (ret)
2915                         goto out;
2916         }
2917
2918         mutex_lock(&ext4_li_info->li_list_mtx);
2919         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2920         mutex_unlock(&ext4_li_info->li_list_mtx);
2921
2922         sbi->s_li_request = elr;
2923         /*
2924          * set elr to NULL here since it has been inserted to
2925          * the request_list and the removal and free of it is
2926          * handled by ext4_clear_request_list from now on.
2927          */
2928         elr = NULL;
2929
2930         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2931                 ret = ext4_run_lazyinit_thread();
2932                 if (ret)
2933                         goto out;
2934         }
2935 out:
2936         mutex_unlock(&ext4_li_mtx);
2937         if (ret)
2938                 kfree(elr);
2939         return ret;
2940 }
2941
2942 /*
2943  * We do not need to lock anything since this is called on
2944  * module unload.
2945  */
2946 static void ext4_destroy_lazyinit_thread(void)
2947 {
2948         /*
2949          * If thread exited earlier
2950          * there's nothing to be done.
2951          */
2952         if (!ext4_li_info || !ext4_lazyinit_task)
2953                 return;
2954
2955         kthread_stop(ext4_lazyinit_task);
2956 }
2957
2958 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2959 {
2960         char *orig_data = kstrdup(data, GFP_KERNEL);
2961         struct buffer_head *bh;
2962         struct ext4_super_block *es = NULL;
2963         struct ext4_sb_info *sbi;
2964         ext4_fsblk_t block;
2965         ext4_fsblk_t sb_block = get_sb_block(&data);
2966         ext4_fsblk_t logical_sb_block;
2967         unsigned long offset = 0;
2968         unsigned long journal_devnum = 0;
2969         unsigned long def_mount_opts;
2970         struct inode *root;
2971         char *cp;
2972         const char *descr;
2973         int ret = -ENOMEM;
2974         int blocksize, clustersize;
2975         unsigned int db_count;
2976         unsigned int i;
2977         int needs_recovery, has_huge_files, has_bigalloc;
2978         __u64 blocks_count;
2979         int err;
2980         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2981         ext4_group_t first_not_zeroed;
2982
2983         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2984         if (!sbi)
2985                 goto out_free_orig;
2986
2987         sbi->s_blockgroup_lock =
2988                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2989         if (!sbi->s_blockgroup_lock) {
2990                 kfree(sbi);
2991                 goto out_free_orig;
2992         }
2993         sb->s_fs_info = sbi;
2994         sbi->s_mount_opt = 0;
2995         sbi->s_resuid = EXT4_DEF_RESUID;
2996         sbi->s_resgid = EXT4_DEF_RESGID;
2997         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
2998         sbi->s_sb_block = sb_block;
2999         if (sb->s_bdev->bd_part)
3000                 sbi->s_sectors_written_start =
3001                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3002
3003         /* Cleanup superblock name */
3004         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3005                 *cp = '!';
3006
3007         ret = -EINVAL;
3008         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3009         if (!blocksize) {
3010                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3011                 goto out_fail;
3012         }
3013
3014         /*
3015          * The ext4 superblock will not be buffer aligned for other than 1kB
3016          * block sizes.  We need to calculate the offset from buffer start.
3017          */
3018         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3019                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3020                 offset = do_div(logical_sb_block, blocksize);
3021         } else {
3022                 logical_sb_block = sb_block;
3023         }
3024
3025         if (!(bh = sb_bread(sb, logical_sb_block))) {
3026                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3027                 goto out_fail;
3028         }
3029         /*
3030          * Note: s_es must be initialized as soon as possible because
3031          *       some ext4 macro-instructions depend on its value
3032          */
3033         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3034         sbi->s_es = es;
3035         sb->s_magic = le16_to_cpu(es->s_magic);
3036         if (sb->s_magic != EXT4_SUPER_MAGIC)
3037                 goto cantfind_ext4;
3038         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3039
3040         /* Check for a known checksum algorithm */
3041         if (!ext4_verify_csum_type(sb, es)) {
3042                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3043                          "unknown checksum algorithm.");
3044                 silent = 1;
3045                 goto cantfind_ext4;
3046         }
3047
3048         /* Load the checksum driver */
3049         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3050                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3051                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3052                 if (IS_ERR(sbi->s_chksum_driver)) {
3053                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3054                         ret = PTR_ERR(sbi->s_chksum_driver);
3055                         sbi->s_chksum_driver = NULL;
3056                         goto failed_mount;
3057                 }
3058         }
3059
3060         /* Set defaults before we parse the mount options */
3061         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3062         set_opt(sb, INIT_INODE_TABLE);
3063         if (def_mount_opts & EXT4_DEFM_DEBUG)
3064                 set_opt(sb, DEBUG);
3065         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3066                 set_opt(sb, GRPID);
3067         if (def_mount_opts & EXT4_DEFM_UID16)
3068                 set_opt(sb, NO_UID32);
3069         /* xattr user namespace & acls are now defaulted on */
3070 #ifdef CONFIG_EXT4_FS_XATTR
3071         set_opt(sb, XATTR_USER);
3072 #endif
3073 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3074         set_opt(sb, POSIX_ACL);
3075 #endif
3076         set_opt(sb, MBLK_IO_SUBMIT);
3077         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3078                 set_opt(sb, JOURNAL_DATA);
3079         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3080                 set_opt(sb, ORDERED_DATA);
3081         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3082                 set_opt(sb, WRITEBACK_DATA);
3083
3084         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3085                 set_opt(sb, ERRORS_PANIC);
3086         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3087                 set_opt(sb, ERRORS_CONT);
3088         else
3089                 set_opt(sb, ERRORS_RO);
3090         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3091                 set_opt(sb, BLOCK_VALIDITY);
3092         if (def_mount_opts & EXT4_DEFM_DISCARD)
3093                 set_opt(sb, DISCARD);
3094
3095         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3096         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3097         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3098         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3099         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3100
3101         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3102                 set_opt(sb, BARRIER);
3103
3104         /*
3105          * enable delayed allocation by default
3106          * Use -o nodelalloc to turn it off
3107          */
3108         if (!IS_EXT3_SB(sb) &&
3109             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3110                 set_opt(sb, DELALLOC);
3111
3112         /*
3113          * set default s_li_wait_mult for lazyinit, for the case there is
3114          * no mount option specified.
3115          */
3116         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3117
3118         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3119                            &journal_devnum, &journal_ioprio, 0)) {
3120                 ext4_msg(sb, KERN_WARNING,
3121                          "failed to parse options in superblock: %s",
3122                          sbi->s_es->s_mount_opts);
3123         }
3124         sbi->s_def_mount_opt = sbi->s_mount_opt;
3125         if (!parse_options((char *) data, sb, &journal_devnum,
3126                            &journal_ioprio, 0))
3127                 goto failed_mount;
3128
3129         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3130                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3131                             "with data=journal disables delayed "
3132                             "allocation and O_DIRECT support!\n");
3133                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3134                         ext4_msg(sb, KERN_ERR, "can't mount with "
3135                                  "both data=journal and delalloc");
3136                         goto failed_mount;
3137                 }
3138                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3139                         ext4_msg(sb, KERN_ERR, "can't mount with "
3140                                  "both data=journal and delalloc");
3141                         goto failed_mount;
3142                 }
3143                 if (test_opt(sb, DELALLOC))
3144                         clear_opt(sb, DELALLOC);
3145         }
3146
3147         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3148         if (test_opt(sb, DIOREAD_NOLOCK)) {
3149                 if (blocksize < PAGE_SIZE) {
3150                         ext4_msg(sb, KERN_ERR, "can't mount with "
3151                                  "dioread_nolock if block size != PAGE_SIZE");
3152                         goto failed_mount;
3153                 }
3154         }
3155
3156         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3157                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3158
3159         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3160             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3161              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3162              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3163                 ext4_msg(sb, KERN_WARNING,
3164                        "feature flags set on rev 0 fs, "
3165                        "running e2fsck is recommended");
3166
3167         if (IS_EXT2_SB(sb)) {
3168                 if (ext2_feature_set_ok(sb))
3169                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3170                                  "using the ext4 subsystem");
3171                 else {
3172                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3173                                  "to feature incompatibilities");
3174                         goto failed_mount;
3175                 }
3176         }
3177
3178         if (IS_EXT3_SB(sb)) {
3179                 if (ext3_feature_set_ok(sb))
3180                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3181                                  "using the ext4 subsystem");
3182                 else {
3183                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3184                                  "to feature incompatibilities");
3185                         goto failed_mount;
3186                 }
3187         }
3188
3189         /*
3190          * Check feature flags regardless of the revision level, since we
3191          * previously didn't change the revision level when setting the flags,
3192          * so there is a chance incompat flags are set on a rev 0 filesystem.
3193          */
3194         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3195                 goto failed_mount;
3196
3197         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3198             blocksize > EXT4_MAX_BLOCK_SIZE) {
3199                 ext4_msg(sb, KERN_ERR,
3200                        "Unsupported filesystem blocksize %d", blocksize);
3201                 goto failed_mount;
3202         }
3203
3204         if (sb->s_blocksize != blocksize) {
3205                 /* Validate the filesystem blocksize */
3206                 if (!sb_set_blocksize(sb, blocksize)) {
3207                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3208                                         blocksize);
3209                         goto failed_mount;
3210                 }
3211
3212                 brelse(bh);
3213                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3214                 offset = do_div(logical_sb_block, blocksize);
3215                 bh = sb_bread(sb, logical_sb_block);
3216                 if (!bh) {
3217                         ext4_msg(sb, KERN_ERR,
3218                                "Can't read superblock on 2nd try");
3219                         goto failed_mount;
3220                 }
3221                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3222                 sbi->s_es = es;
3223                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3224                         ext4_msg(sb, KERN_ERR,
3225                                "Magic mismatch, very weird!");
3226                         goto failed_mount;
3227                 }
3228         }
3229
3230         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3231                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3232         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3233                                                       has_huge_files);
3234         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3235
3236         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3237                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3238                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3239         } else {
3240                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3241                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3242                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3243                     (!is_power_of_2(sbi->s_inode_size)) ||
3244                     (sbi->s_inode_size > blocksize)) {
3245                         ext4_msg(sb, KERN_ERR,
3246                                "unsupported inode size: %d",
3247                                sbi->s_inode_size);
3248                         goto failed_mount;
3249                 }
3250                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3251                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3252         }
3253
3254         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3255         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3256                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3257                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3258                     !is_power_of_2(sbi->s_desc_size)) {
3259                         ext4_msg(sb, KERN_ERR,
3260                                "unsupported descriptor size %lu",
3261                                sbi->s_desc_size);
3262                         goto failed_mount;
3263                 }
3264         } else
3265                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3266
3267         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3268         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3269         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3270                 goto cantfind_ext4;
3271
3272         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3273         if (sbi->s_inodes_per_block == 0)
3274                 goto cantfind_ext4;
3275         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3276                                         sbi->s_inodes_per_block;
3277         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3278         sbi->s_sbh = bh;
3279         sbi->s_mount_state = le16_to_cpu(es->s_state);
3280         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3281         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3282
3283         for (i = 0; i < 4; i++)
3284                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3285         sbi->s_def_hash_version = es->s_def_hash_version;
3286         i = le32_to_cpu(es->s_flags);
3287         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3288                 sbi->s_hash_unsigned = 3;
3289         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3290 #ifdef __CHAR_UNSIGNED__
3291                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3292                 sbi->s_hash_unsigned = 3;
3293 #else
3294                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3295 #endif
3296         }
3297
3298         /* Handle clustersize */
3299         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3300         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3301                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3302         if (has_bigalloc) {
3303                 if (clustersize < blocksize) {
3304                         ext4_msg(sb, KERN_ERR,
3305                                  "cluster size (%d) smaller than "
3306                                  "block size (%d)", clustersize, blocksize);
3307                         goto failed_mount;
3308                 }
3309                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3310                         le32_to_cpu(es->s_log_block_size);
3311                 sbi->s_clusters_per_group =
3312                         le32_to_cpu(es->s_clusters_per_group);
3313                 if (sbi->s_clusters_per_group > blocksize * 8) {
3314                         ext4_msg(sb, KERN_ERR,
3315                                  "#clusters per group too big: %lu",
3316                                  sbi->s_clusters_per_group);
3317                         goto failed_mount;
3318                 }
3319                 if (sbi->s_blocks_per_group !=
3320                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3321                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3322                                  "clusters per group (%lu) inconsistent",
3323                                  sbi->s_blocks_per_group,
3324                                  sbi->s_clusters_per_group);
3325                         goto failed_mount;
3326                 }
3327         } else {
3328                 if (clustersize != blocksize) {
3329                         ext4_warning(sb, "fragment/cluster size (%d) != "
3330                                      "block size (%d)", clustersize,
3331                                      blocksize);
3332                         clustersize = blocksize;
3333                 }
3334                 if (sbi->s_blocks_per_group > blocksize * 8) {
3335                         ext4_msg(sb, KERN_ERR,
3336                                  "#blocks per group too big: %lu",
3337                                  sbi->s_blocks_per_group);
3338                         goto failed_mount;
3339                 }
3340                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3341                 sbi->s_cluster_bits = 0;
3342         }
3343         sbi->s_cluster_ratio = clustersize / blocksize;
3344
3345         if (sbi->s_inodes_per_group > blocksize * 8) {
3346                 ext4_msg(sb, KERN_ERR,
3347                        "#inodes per group too big: %lu",
3348                        sbi->s_inodes_per_group);
3349                 goto failed_mount;
3350         }
3351
3352         /*
3353          * Test whether we have more sectors than will fit in sector_t,
3354          * and whether the max offset is addressable by the page cache.
3355          */
3356         err = generic_check_addressable(sb->s_blocksize_bits,
3357                                         ext4_blocks_count(es));
3358         if (err) {
3359                 ext4_msg(sb, KERN_ERR, "filesystem"
3360                          " too large to mount safely on this system");
3361                 if (sizeof(sector_t) < 8)
3362                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3363                 ret = err;
3364                 goto failed_mount;
3365         }
3366
3367         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3368                 goto cantfind_ext4;
3369
3370         /* check blocks count against device size */
3371         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3372         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3373                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3374                        "exceeds size of device (%llu blocks)",
3375                        ext4_blocks_count(es), blocks_count);
3376                 goto failed_mount;
3377         }
3378
3379         /*
3380          * It makes no sense for the first data block to be beyond the end
3381          * of the filesystem.
3382          */
3383         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3384                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3385                          "block %u is beyond end of filesystem (%llu)",
3386                          le32_to_cpu(es->s_first_data_block),
3387                          ext4_blocks_count(es));
3388                 goto failed_mount;
3389         }
3390         blocks_count = (ext4_blocks_count(es) -
3391                         le32_to_cpu(es->s_first_data_block) +
3392                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3393         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3394         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3395                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3396                        "(block count %llu, first data block %u, "
3397                        "blocks per group %lu)", sbi->s_groups_count,
3398                        ext4_blocks_count(es),
3399                        le32_to_cpu(es->s_first_data_block),
3400                        EXT4_BLOCKS_PER_GROUP(sb));
3401                 goto failed_mount;
3402         }
3403         sbi->s_groups_count = blocks_count;
3404         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3405                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3406         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3407                    EXT4_DESC_PER_BLOCK(sb);
3408         sbi->s_group_desc = ext4_kvmalloc(db_count *
3409                                           sizeof(struct buffer_head *),
3410                                           GFP_KERNEL);
3411         if (sbi->s_group_desc == NULL) {
3412                 ext4_msg(sb, KERN_ERR, "not enough memory");
3413                 goto failed_mount;
3414         }
3415
3416         if (ext4_proc_root)
3417                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3418
3419         if (sbi->s_proc)
3420                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3421                                  &ext4_seq_options_fops, sb);
3422
3423         bgl_lock_init(sbi->s_blockgroup_lock);
3424
3425         for (i = 0; i < db_count; i++) {
3426                 block = descriptor_loc(sb, logical_sb_block, i);
3427                 sbi->s_group_desc[i] = sb_bread(sb, block);
3428                 if (!sbi->s_group_desc[i]) {
3429                         ext4_msg(sb, KERN_ERR,
3430                                "can't read group descriptor %d", i);
3431                         db_count = i;
3432                         goto failed_mount2;
3433                 }
3434         }
3435         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3436                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3437                 goto failed_mount2;
3438         }
3439         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3440                 if (!ext4_fill_flex_info(sb)) {
3441                         ext4_msg(sb, KERN_ERR,
3442                                "unable to initialize "
3443                                "flex_bg meta info!");
3444                         goto failed_mount2;
3445                 }
3446
3447         sbi->s_gdb_count = db_count;
3448         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3449         spin_lock_init(&sbi->s_next_gen_lock);
3450
3451         init_timer(&sbi->s_err_report);
3452         sbi->s_err_report.function = print_daily_error_info;
3453         sbi->s_err_report.data = (unsigned long) sb;
3454
3455         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3456                         ext4_count_free_clusters(sb));
3457         if (!err) {
3458                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3459                                 ext4_count_free_inodes(sb));
3460         }
3461         if (!err) {
3462                 err = percpu_counter_init(&sbi->s_dirs_counter,
3463                                 ext4_count_dirs(sb));
3464         }
3465         if (!err) {
3466                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3467         }
3468         if (err) {
3469                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3470                 goto failed_mount3;
3471         }
3472
3473         sbi->s_stripe = ext4_get_stripe_size(sbi);
3474         sbi->s_max_writeback_mb_bump = 128;
3475
3476         /*
3477          * set up enough so that it can read an inode
3478          */
3479         if (!test_opt(sb, NOLOAD) &&
3480             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3481                 sb->s_op = &ext4_sops;
3482         else
3483                 sb->s_op = &ext4_nojournal_sops;
3484         sb->s_export_op = &ext4_export_ops;
3485         sb->s_xattr = ext4_xattr_handlers;
3486 #ifdef CONFIG_QUOTA
3487         sb->s_qcop = &ext4_qctl_operations;
3488         sb->dq_op = &ext4_quota_operations;
3489 #endif
3490         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3491
3492         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3493         mutex_init(&sbi->s_orphan_lock);
3494         sbi->s_resize_flags = 0;
3495
3496         sb->s_root = NULL;
3497
3498         needs_recovery = (es->s_last_orphan != 0 ||
3499                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3500                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3501
3502         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3503             !(sb->s_flags & MS_RDONLY))
3504                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3505                         goto failed_mount3;
3506
3507         /*
3508          * The first inode we look at is the journal inode.  Don't try
3509          * root first: it may be modified in the journal!
3510          */
3511         if (!test_opt(sb, NOLOAD) &&
3512             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3513                 if (ext4_load_journal(sb, es, journal_devnum))
3514                         goto failed_mount3;
3515         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3516               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3517                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3518                        "suppressed and not mounted read-only");
3519                 goto failed_mount_wq;
3520         } else {
3521                 clear_opt(sb, DATA_FLAGS);
3522                 sbi->s_journal = NULL;
3523                 needs_recovery = 0;
3524                 goto no_journal;
3525         }
3526
3527         if (ext4_blocks_count(es) > 0xffffffffULL &&
3528             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3529                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3530                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3531                 goto failed_mount_wq;
3532         }
3533
3534         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3535                 jbd2_journal_set_features(sbi->s_journal,
3536                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3537                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3538         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3539                 jbd2_journal_set_features(sbi->s_journal,
3540                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3541                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3542                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3543         } else {
3544                 jbd2_journal_clear_features(sbi->s_journal,
3545                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3546                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3547         }
3548
3549         /* We have now updated the journal if required, so we can
3550          * validate the data journaling mode. */
3551         switch (test_opt(sb, DATA_FLAGS)) {
3552         case 0:
3553                 /* No mode set, assume a default based on the journal
3554                  * capabilities: ORDERED_DATA if the journal can
3555                  * cope, else JOURNAL_DATA
3556                  */
3557                 if (jbd2_journal_check_available_features
3558                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3559                         set_opt(sb, ORDERED_DATA);
3560                 else
3561                         set_opt(sb, JOURNAL_DATA);
3562                 break;
3563
3564         case EXT4_MOUNT_ORDERED_DATA:
3565         case EXT4_MOUNT_WRITEBACK_DATA:
3566                 if (!jbd2_journal_check_available_features
3567                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3568                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3569                                "requested data journaling mode");
3570                         goto failed_mount_wq;
3571                 }
3572         default:
3573                 break;
3574         }
3575         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3576
3577         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3578
3579         /*
3580          * The journal may have updated the bg summary counts, so we
3581          * need to update the global counters.
3582          */
3583         percpu_counter_set(&sbi->s_freeclusters_counter,
3584                            ext4_count_free_clusters(sb));
3585         percpu_counter_set(&sbi->s_freeinodes_counter,
3586                            ext4_count_free_inodes(sb));
3587         percpu_counter_set(&sbi->s_dirs_counter,
3588                            ext4_count_dirs(sb));
3589         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3590
3591 no_journal:
3592         /*
3593          * The maximum number of concurrent works can be high and
3594          * concurrency isn't really necessary.  Limit it to 1.
3595          */
3596         EXT4_SB(sb)->dio_unwritten_wq =
3597                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3598         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3599                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3600                 goto failed_mount_wq;
3601         }
3602
3603         /*
3604          * The jbd2_journal_load will have done any necessary log recovery,
3605          * so we can safely mount the rest of the filesystem now.
3606          */
3607
3608         root = ext4_iget(sb, EXT4_ROOT_INO);
3609         if (IS_ERR(root)) {
3610                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3611                 ret = PTR_ERR(root);
3612                 root = NULL;
3613                 goto failed_mount4;
3614         }
3615         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3616                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3617                 iput(root);
3618                 goto failed_mount4;
3619         }
3620         sb->s_root = d_make_root(root);
3621         if (!sb->s_root) {
3622                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3623                 ret = -ENOMEM;
3624                 goto failed_mount4;
3625         }
3626
3627         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3628
3629         /* determine the minimum size of new large inodes, if present */
3630         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3631                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3632                                                      EXT4_GOOD_OLD_INODE_SIZE;
3633                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3634                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3635                         if (sbi->s_want_extra_isize <
3636                             le16_to_cpu(es->s_want_extra_isize))
3637                                 sbi->s_want_extra_isize =
3638                                         le16_to_cpu(es->s_want_extra_isize);
3639                         if (sbi->s_want_extra_isize <
3640                             le16_to_cpu(es->s_min_extra_isize))
3641                                 sbi->s_want_extra_isize =
3642                                         le16_to_cpu(es->s_min_extra_isize);
3643                 }
3644         }
3645         /* Check if enough inode space is available */
3646         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3647                                                         sbi->s_inode_size) {
3648                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3649                                                        EXT4_GOOD_OLD_INODE_SIZE;
3650                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3651                          "available");
3652         }
3653
3654         err = ext4_setup_system_zone(sb);
3655         if (err) {
3656                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3657                          "zone (%d)", err);
3658                 goto failed_mount4a;
3659         }
3660
3661         ext4_ext_init(sb);
3662         err = ext4_mb_init(sb, needs_recovery);
3663         if (err) {
3664                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3665                          err);
3666                 goto failed_mount5;
3667         }
3668
3669         err = ext4_register_li_request(sb, first_not_zeroed);
3670         if (err)
3671                 goto failed_mount6;
3672
3673         sbi->s_kobj.kset = ext4_kset;
3674         init_completion(&sbi->s_kobj_unregister);
3675         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3676                                    "%s", sb->s_id);
3677         if (err)
3678                 goto failed_mount7;
3679
3680         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3681         ext4_orphan_cleanup(sb, es);
3682         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3683         if (needs_recovery) {
3684                 ext4_msg(sb, KERN_INFO, "recovery complete");
3685                 ext4_mark_recovery_complete(sb, es);
3686         }
3687         if (EXT4_SB(sb)->s_journal) {
3688                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3689                         descr = " journalled data mode";
3690                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3691                         descr = " ordered data mode";
3692                 else
3693                         descr = " writeback data mode";
3694         } else
3695                 descr = "out journal";
3696
3697         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3698                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3699                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3700
3701         if (es->s_error_count)
3702                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3703
3704         kfree(orig_data);
3705         return 0;
3706
3707 cantfind_ext4:
3708         if (!silent)
3709                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3710         goto failed_mount;
3711
3712 failed_mount7:
3713         ext4_unregister_li_request(sb);
3714 failed_mount6:
3715         ext4_mb_release(sb);
3716 failed_mount5:
3717         ext4_ext_release(sb);
3718         ext4_release_system_zone(sb);
3719 failed_mount4a:
3720         dput(sb->s_root);
3721         sb->s_root = NULL;
3722 failed_mount4:
3723         ext4_msg(sb, KERN_ERR, "mount failed");
3724         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3725 failed_mount_wq:
3726         if (sbi->s_journal) {
3727                 jbd2_journal_destroy(sbi->s_journal);
3728                 sbi->s_journal = NULL;
3729         }
3730 failed_mount3:
3731         del_timer(&sbi->s_err_report);
3732         if (sbi->s_flex_groups)
3733                 ext4_kvfree(sbi->s_flex_groups);
3734         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3735         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3736         percpu_counter_destroy(&sbi->s_dirs_counter);
3737         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3738         if (sbi->s_mmp_tsk)
3739                 kthread_stop(sbi->s_mmp_tsk);
3740 failed_mount2:
3741         for (i = 0; i < db_count; i++)
3742                 brelse(sbi->s_group_desc[i]);
3743         ext4_kvfree(sbi->s_group_desc);
3744 failed_mount:
3745         if (sbi->s_chksum_driver)
3746                 crypto_free_shash(sbi->s_chksum_driver);
3747         if (sbi->s_proc) {
3748                 remove_proc_entry("options", sbi->s_proc);
3749                 remove_proc_entry(sb->s_id, ext4_proc_root);
3750         }
3751 #ifdef CONFIG_QUOTA
3752         for (i = 0; i < MAXQUOTAS; i++)
3753                 kfree(sbi->s_qf_names[i]);
3754 #endif
3755         ext4_blkdev_remove(sbi);
3756         brelse(bh);
3757 out_fail:
3758         sb->s_fs_info = NULL;
3759         kfree(sbi->s_blockgroup_lock);
3760         kfree(sbi);
3761 out_free_orig:
3762         kfree(orig_data);
3763         return ret;
3764 }
3765
3766 /*
3767  * Setup any per-fs journal parameters now.  We'll do this both on
3768  * initial mount, once the journal has been initialised but before we've
3769  * done any recovery; and again on any subsequent remount.
3770  */
3771 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3772 {
3773         struct ext4_sb_info *sbi = EXT4_SB(sb);
3774
3775         journal->j_commit_interval = sbi->s_commit_interval;
3776         journal->j_min_batch_time = sbi->s_min_batch_time;
3777         journal->j_max_batch_time = sbi->s_max_batch_time;
3778
3779         write_lock(&journal->j_state_lock);
3780         if (test_opt(sb, BARRIER))
3781                 journal->j_flags |= JBD2_BARRIER;
3782         else
3783                 journal->j_flags &= ~JBD2_BARRIER;
3784         if (test_opt(sb, DATA_ERR_ABORT))
3785                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3786         else
3787                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3788         write_unlock(&journal->j_state_lock);
3789 }
3790
3791 static journal_t *ext4_get_journal(struct super_block *sb,
3792                                    unsigned int journal_inum)
3793 {
3794         struct inode *journal_inode;
3795         journal_t *journal;
3796
3797         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3798
3799         /* First, test for the existence of a valid inode on disk.  Bad
3800          * things happen if we iget() an unused inode, as the subsequent
3801          * iput() will try to delete it. */
3802
3803         journal_inode = ext4_iget(sb, journal_inum);
3804         if (IS_ERR(journal_inode)) {
3805                 ext4_msg(sb, KERN_ERR, "no journal found");
3806                 return NULL;
3807         }
3808         if (!journal_inode->i_nlink) {
3809                 make_bad_inode(journal_inode);
3810                 iput(journal_inode);
3811                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3812                 return NULL;
3813         }
3814
3815         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3816                   journal_inode, journal_inode->i_size);
3817         if (!S_ISREG(journal_inode->i_mode)) {
3818                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3819                 iput(journal_inode);
3820                 return NULL;
3821         }
3822
3823         journal = jbd2_journal_init_inode(journal_inode);
3824         if (!journal) {
3825                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3826                 iput(journal_inode);
3827                 return NULL;
3828         }
3829         journal->j_private = sb;
3830         ext4_init_journal_params(sb, journal);
3831         return journal;
3832 }
3833
3834 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3835                                        dev_t j_dev)
3836 {
3837         struct buffer_head *bh;
3838         journal_t *journal;
3839         ext4_fsblk_t start;
3840         ext4_fsblk_t len;
3841         int hblock, blocksize;
3842         ext4_fsblk_t sb_block;
3843         unsigned long offset;
3844         struct ext4_super_block *es;
3845         struct block_device *bdev;
3846
3847         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3848
3849         bdev = ext4_blkdev_get(j_dev, sb);
3850         if (bdev == NULL)
3851                 return NULL;
3852
3853         blocksize = sb->s_blocksize;
3854         hblock = bdev_logical_block_size(bdev);
3855         if (blocksize < hblock) {
3856                 ext4_msg(sb, KERN_ERR,
3857                         "blocksize too small for journal device");
3858                 goto out_bdev;
3859         }
3860
3861         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3862         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3863         set_blocksize(bdev, blocksize);
3864         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3865                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3866                        "external journal");
3867                 goto out_bdev;
3868         }
3869
3870         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3871         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3872             !(le32_to_cpu(es->s_feature_incompat) &
3873               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3874                 ext4_msg(sb, KERN_ERR, "external journal has "
3875                                         "bad superblock");
3876                 brelse(bh);
3877                 goto out_bdev;
3878         }
3879
3880         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3881                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3882                 brelse(bh);
3883                 goto out_bdev;
3884         }
3885
3886         len = ext4_blocks_count(es);
3887         start = sb_block + 1;
3888         brelse(bh);     /* we're done with the superblock */
3889
3890         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3891                                         start, len, blocksize);
3892         if (!journal) {
3893                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3894                 goto out_bdev;
3895         }
3896         journal->j_private = sb;
3897         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3898         wait_on_buffer(journal->j_sb_buffer);
3899         if (!buffer_uptodate(journal->j_sb_buffer)) {
3900                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3901                 goto out_journal;
3902         }
3903         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3904                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3905                                         "user (unsupported) - %d",
3906                         be32_to_cpu(journal->j_superblock->s_nr_users));
3907                 goto out_journal;
3908         }
3909         EXT4_SB(sb)->journal_bdev = bdev;
3910         ext4_init_journal_params(sb, journal);
3911         return journal;
3912
3913 out_journal:
3914         jbd2_journal_destroy(journal);
3915 out_bdev:
3916         ext4_blkdev_put(bdev);
3917         return NULL;
3918 }
3919
3920 static int ext4_load_journal(struct super_block *sb,
3921                              struct ext4_super_block *es,
3922                              unsigned long journal_devnum)
3923 {
3924         journal_t *journal;
3925         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3926         dev_t journal_dev;
3927         int err = 0;
3928         int really_read_only;
3929
3930         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3931
3932         if (journal_devnum &&
3933             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3934                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3935                         "numbers have changed");
3936                 journal_dev = new_decode_dev(journal_devnum);
3937         } else
3938                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3939
3940         really_read_only = bdev_read_only(sb->s_bdev);
3941
3942         /*
3943          * Are we loading a blank journal or performing recovery after a
3944          * crash?  For recovery, we need to check in advance whether we
3945          * can get read-write access to the device.
3946          */
3947         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3948                 if (sb->s_flags & MS_RDONLY) {
3949                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3950                                         "required on readonly filesystem");
3951                         if (really_read_only) {
3952                                 ext4_msg(sb, KERN_ERR, "write access "
3953                                         "unavailable, cannot proceed");
3954                                 return -EROFS;
3955                         }
3956                         ext4_msg(sb, KERN_INFO, "write access will "
3957                                "be enabled during recovery");
3958                 }
3959         }
3960
3961         if (journal_inum && journal_dev) {
3962                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3963                        "and inode journals!");
3964                 return -EINVAL;
3965         }
3966
3967         if (journal_inum) {
3968                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3969                         return -EINVAL;
3970         } else {
3971                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3972                         return -EINVAL;
3973         }
3974
3975         if (!(journal->j_flags & JBD2_BARRIER))
3976                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3977
3978         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3979                 err = jbd2_journal_wipe(journal, !really_read_only);
3980         if (!err) {
3981                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3982                 if (save)
3983                         memcpy(save, ((char *) es) +
3984                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3985                 err = jbd2_journal_load(journal);
3986                 if (save)
3987                         memcpy(((char *) es) + EXT4_S_ERR_START,
3988                                save, EXT4_S_ERR_LEN);
3989                 kfree(save);
3990         }
3991
3992         if (err) {
3993                 ext4_msg(sb, KERN_ERR, "error loading journal");
3994                 jbd2_journal_destroy(journal);
3995                 return err;
3996         }
3997
3998         EXT4_SB(sb)->s_journal = journal;
3999         ext4_clear_journal_err(sb, es);
4000
4001         if (!really_read_only && journal_devnum &&
4002             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4003                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4004
4005                 /* Make sure we flush the recovery flag to disk. */
4006                 ext4_commit_super(sb, 1);
4007         }
4008
4009         return 0;
4010 }
4011
4012 static int ext4_commit_super(struct super_block *sb, int sync)
4013 {
4014         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4015         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4016         int error = 0;
4017
4018         if (!sbh || block_device_ejected(sb))
4019                 return error;
4020         if (buffer_write_io_error(sbh)) {
4021                 /*
4022                  * Oh, dear.  A previous attempt to write the
4023                  * superblock failed.  This could happen because the
4024                  * USB device was yanked out.  Or it could happen to
4025                  * be a transient write error and maybe the block will
4026                  * be remapped.  Nothing we can do but to retry the
4027                  * write and hope for the best.
4028                  */
4029                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4030                        "superblock detected");
4031                 clear_buffer_write_io_error(sbh);
4032                 set_buffer_uptodate(sbh);
4033         }
4034         /*
4035          * If the file system is mounted read-only, don't update the
4036          * superblock write time.  This avoids updating the superblock
4037          * write time when we are mounting the root file system
4038          * read/only but we need to replay the journal; at that point,
4039          * for people who are east of GMT and who make their clock
4040          * tick in localtime for Windows bug-for-bug compatibility,
4041          * the clock is set in the future, and this will cause e2fsck
4042          * to complain and force a full file system check.
4043          */
4044         if (!(sb->s_flags & MS_RDONLY))
4045                 es->s_wtime = cpu_to_le32(get_seconds());
4046         if (sb->s_bdev->bd_part)
4047                 es->s_kbytes_written =
4048                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4049                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4050                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4051         else
4052                 es->s_kbytes_written =
4053                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4054         ext4_free_blocks_count_set(es,
4055                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4056                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4057         es->s_free_inodes_count =
4058                 cpu_to_le32(percpu_counter_sum_positive(
4059                                 &EXT4_SB(sb)->s_freeinodes_counter));
4060         sb->s_dirt = 0;
4061         BUFFER_TRACE(sbh, "marking dirty");
4062         mark_buffer_dirty(sbh);
4063         if (sync) {
4064                 error = sync_dirty_buffer(sbh);
4065                 if (error)
4066                         return error;
4067
4068                 error = buffer_write_io_error(sbh);
4069                 if (error) {
4070                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4071                                "superblock");
4072                         clear_buffer_write_io_error(sbh);
4073                         set_buffer_uptodate(sbh);
4074                 }
4075         }
4076         return error;
4077 }
4078
4079 /*
4080  * Have we just finished recovery?  If so, and if we are mounting (or
4081  * remounting) the filesystem readonly, then we will end up with a
4082  * consistent fs on disk.  Record that fact.
4083  */
4084 static void ext4_mark_recovery_complete(struct super_block *sb,
4085                                         struct ext4_super_block *es)
4086 {
4087         journal_t *journal = EXT4_SB(sb)->s_journal;
4088
4089         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4090                 BUG_ON(journal != NULL);
4091                 return;
4092         }
4093         jbd2_journal_lock_updates(journal);
4094         if (jbd2_journal_flush(journal) < 0)
4095                 goto out;
4096
4097         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4098             sb->s_flags & MS_RDONLY) {
4099                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4100                 ext4_commit_super(sb, 1);
4101         }
4102
4103 out:
4104         jbd2_journal_unlock_updates(journal);
4105 }
4106
4107 /*
4108  * If we are mounting (or read-write remounting) a filesystem whose journal
4109  * has recorded an error from a previous lifetime, move that error to the
4110  * main filesystem now.
4111  */
4112 static void ext4_clear_journal_err(struct super_block *sb,
4113                                    struct ext4_super_block *es)
4114 {
4115         journal_t *journal;
4116         int j_errno;
4117         const char *errstr;
4118
4119         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4120
4121         journal = EXT4_SB(sb)->s_journal;
4122
4123         /*
4124          * Now check for any error status which may have been recorded in the
4125          * journal by a prior ext4_error() or ext4_abort()
4126          */
4127
4128         j_errno = jbd2_journal_errno(journal);
4129         if (j_errno) {
4130                 char nbuf[16];
4131
4132                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4133                 ext4_warning(sb, "Filesystem error recorded "
4134                              "from previous mount: %s", errstr);
4135                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4136
4137                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4138                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4139                 ext4_commit_super(sb, 1);
4140
4141                 jbd2_journal_clear_err(journal);
4142         }
4143 }
4144
4145 /*
4146  * Force the running and committing transactions to commit,
4147  * and wait on the commit.
4148  */
4149 int ext4_force_commit(struct super_block *sb)
4150 {
4151         journal_t *journal;
4152         int ret = 0;
4153
4154         if (sb->s_flags & MS_RDONLY)
4155                 return 0;
4156
4157         journal = EXT4_SB(sb)->s_journal;
4158         if (journal) {
4159                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4160                 ret = ext4_journal_force_commit(journal);
4161         }
4162
4163         return ret;
4164 }
4165
4166 static void ext4_write_super(struct super_block *sb)
4167 {
4168         lock_super(sb);
4169         ext4_commit_super(sb, 1);
4170         unlock_super(sb);
4171 }
4172
4173 static int ext4_sync_fs(struct super_block *sb, int wait)
4174 {
4175         int ret = 0;
4176         tid_t target;
4177         struct ext4_sb_info *sbi = EXT4_SB(sb);
4178
4179         trace_ext4_sync_fs(sb, wait);
4180         flush_workqueue(sbi->dio_unwritten_wq);
4181         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4182                 if (wait)
4183                         jbd2_log_wait_commit(sbi->s_journal, target);
4184         }
4185         return ret;
4186 }
4187
4188 /*
4189  * LVM calls this function before a (read-only) snapshot is created.  This
4190  * gives us a chance to flush the journal completely and mark the fs clean.
4191  *
4192  * Note that only this function cannot bring a filesystem to be in a clean
4193  * state independently, because ext4 prevents a new handle from being started
4194  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4195  * the upper layer.
4196  */
4197 static int ext4_freeze(struct super_block *sb)
4198 {
4199         int error = 0;
4200         journal_t *journal;
4201
4202         if (sb->s_flags & MS_RDONLY)
4203                 return 0;
4204
4205         journal = EXT4_SB(sb)->s_journal;
4206
4207         /* Now we set up the journal barrier. */
4208         jbd2_journal_lock_updates(journal);
4209
4210         /*
4211          * Don't clear the needs_recovery flag if we failed to flush
4212          * the journal.
4213          */
4214         error = jbd2_journal_flush(journal);
4215         if (error < 0)
4216                 goto out;
4217
4218         /* Journal blocked and flushed, clear needs_recovery flag. */
4219         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4220         error = ext4_commit_super(sb, 1);
4221 out:
4222         /* we rely on s_frozen to stop further updates */
4223         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4224         return error;
4225 }
4226
4227 /*
4228  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4229  * flag here, even though the filesystem is not technically dirty yet.
4230  */
4231 static int ext4_unfreeze(struct super_block *sb)
4232 {
4233         if (sb->s_flags & MS_RDONLY)
4234                 return 0;
4235
4236         lock_super(sb);
4237         /* Reset the needs_recovery flag before the fs is unlocked. */
4238         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4239         ext4_commit_super(sb, 1);
4240         unlock_super(sb);
4241         return 0;
4242 }
4243
4244 /*
4245  * Structure to save mount options for ext4_remount's benefit
4246  */
4247 struct ext4_mount_options {
4248         unsigned long s_mount_opt;
4249         unsigned long s_mount_opt2;
4250         uid_t s_resuid;
4251         gid_t s_resgid;
4252         unsigned long s_commit_interval;
4253         u32 s_min_batch_time, s_max_batch_time;
4254 #ifdef CONFIG_QUOTA
4255         int s_jquota_fmt;
4256         char *s_qf_names[MAXQUOTAS];
4257 #endif
4258 };
4259
4260 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4261 {
4262         struct ext4_super_block *es;
4263         struct ext4_sb_info *sbi = EXT4_SB(sb);
4264         unsigned long old_sb_flags;
4265         struct ext4_mount_options old_opts;
4266         int enable_quota = 0;
4267         ext4_group_t g;
4268         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4269         int err = 0;
4270 #ifdef CONFIG_QUOTA
4271         int i;
4272 #endif
4273         char *orig_data = kstrdup(data, GFP_KERNEL);
4274
4275         /* Store the original options */
4276         lock_super(sb);
4277         old_sb_flags = sb->s_flags;
4278         old_opts.s_mount_opt = sbi->s_mount_opt;
4279         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4280         old_opts.s_resuid = sbi->s_resuid;
4281         old_opts.s_resgid = sbi->s_resgid;
4282         old_opts.s_commit_interval = sbi->s_commit_interval;
4283         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4284         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4285 #ifdef CONFIG_QUOTA
4286         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4287         for (i = 0; i < MAXQUOTAS; i++)
4288                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4289 #endif
4290         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4291                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4292
4293         /*
4294          * Allow the "check" option to be passed as a remount option.
4295          */
4296         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4297                 err = -EINVAL;
4298                 goto restore_opts;
4299         }
4300
4301         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4302                 ext4_abort(sb, "Abort forced by user");
4303
4304         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4305                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4306
4307         es = sbi->s_es;
4308
4309         if (sbi->s_journal) {
4310                 ext4_init_journal_params(sb, sbi->s_journal);
4311                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4312         }
4313
4314         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4315                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4316                         err = -EROFS;
4317                         goto restore_opts;
4318                 }
4319
4320                 if (*flags & MS_RDONLY) {
4321                         err = dquot_suspend(sb, -1);
4322                         if (err < 0)
4323                                 goto restore_opts;
4324
4325                         /*
4326                          * First of all, the unconditional stuff we have to do
4327                          * to disable replay of the journal when we next remount
4328                          */
4329                         sb->s_flags |= MS_RDONLY;
4330
4331                         /*
4332                          * OK, test if we are remounting a valid rw partition
4333                          * readonly, and if so set the rdonly flag and then
4334                          * mark the partition as valid again.
4335                          */
4336                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4337                             (sbi->s_mount_state & EXT4_VALID_FS))
4338                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4339
4340                         if (sbi->s_journal)
4341                                 ext4_mark_recovery_complete(sb, es);
4342                 } else {
4343                         /* Make sure we can mount this feature set readwrite */
4344                         if (!ext4_feature_set_ok(sb, 0)) {
4345                                 err = -EROFS;
4346                                 goto restore_opts;
4347                         }
4348                         /*
4349                          * Make sure the group descriptor checksums
4350                          * are sane.  If they aren't, refuse to remount r/w.
4351                          */
4352                         for (g = 0; g < sbi->s_groups_count; g++) {
4353                                 struct ext4_group_desc *gdp =
4354                                         ext4_get_group_desc(sb, g, NULL);
4355
4356                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4357                                         ext4_msg(sb, KERN_ERR,
4358                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4359                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4360                                                le16_to_cpu(gdp->bg_checksum));
4361                                         err = -EINVAL;
4362                                         goto restore_opts;
4363                                 }
4364                         }
4365
4366                         /*
4367                          * If we have an unprocessed orphan list hanging
4368                          * around from a previously readonly bdev mount,
4369                          * require a full umount/remount for now.
4370                          */
4371                         if (es->s_last_orphan) {
4372                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4373                                        "remount RDWR because of unprocessed "
4374                                        "orphan inode list.  Please "
4375                                        "umount/remount instead");
4376                                 err = -EINVAL;
4377                                 goto restore_opts;
4378                         }
4379
4380                         /*
4381                          * Mounting a RDONLY partition read-write, so reread
4382                          * and store the current valid flag.  (It may have
4383                          * been changed by e2fsck since we originally mounted
4384                          * the partition.)
4385                          */
4386                         if (sbi->s_journal)
4387                                 ext4_clear_journal_err(sb, es);
4388                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4389                         if (!ext4_setup_super(sb, es, 0))
4390                                 sb->s_flags &= ~MS_RDONLY;
4391                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4392                                                      EXT4_FEATURE_INCOMPAT_MMP))
4393                                 if (ext4_multi_mount_protect(sb,
4394                                                 le64_to_cpu(es->s_mmp_block))) {
4395                                         err = -EROFS;
4396                                         goto restore_opts;
4397                                 }
4398                         enable_quota = 1;
4399                 }
4400         }
4401
4402         /*
4403          * Reinitialize lazy itable initialization thread based on
4404          * current settings
4405          */
4406         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4407                 ext4_unregister_li_request(sb);
4408         else {
4409                 ext4_group_t first_not_zeroed;
4410                 first_not_zeroed = ext4_has_uninit_itable(sb);
4411                 ext4_register_li_request(sb, first_not_zeroed);
4412         }
4413
4414         ext4_setup_system_zone(sb);
4415         if (sbi->s_journal == NULL)
4416                 ext4_commit_super(sb, 1);
4417
4418 #ifdef CONFIG_QUOTA
4419         /* Release old quota file names */
4420         for (i = 0; i < MAXQUOTAS; i++)
4421                 if (old_opts.s_qf_names[i] &&
4422                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4423                         kfree(old_opts.s_qf_names[i]);
4424 #endif
4425         unlock_super(sb);
4426         if (enable_quota)
4427                 dquot_resume(sb, -1);
4428
4429         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4430         kfree(orig_data);
4431         return 0;
4432
4433 restore_opts:
4434         sb->s_flags = old_sb_flags;
4435         sbi->s_mount_opt = old_opts.s_mount_opt;
4436         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4437         sbi->s_resuid = old_opts.s_resuid;
4438         sbi->s_resgid = old_opts.s_resgid;
4439         sbi->s_commit_interval = old_opts.s_commit_interval;
4440         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4441         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4442 #ifdef CONFIG_QUOTA
4443         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4444         for (i = 0; i < MAXQUOTAS; i++) {
4445                 if (sbi->s_qf_names[i] &&
4446                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4447                         kfree(sbi->s_qf_names[i]);
4448                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4449         }
4450 #endif
4451         unlock_super(sb);
4452         kfree(orig_data);
4453         return err;
4454 }
4455
4456 /*
4457  * Note: calculating the overhead so we can be compatible with
4458  * historical BSD practice is quite difficult in the face of
4459  * clusters/bigalloc.  This is because multiple metadata blocks from
4460  * different block group can end up in the same allocation cluster.
4461  * Calculating the exact overhead in the face of clustered allocation
4462  * requires either O(all block bitmaps) in memory or O(number of block
4463  * groups**2) in time.  We will still calculate the superblock for
4464  * older file systems --- and if we come across with a bigalloc file
4465  * system with zero in s_overhead_clusters the estimate will be close to
4466  * correct especially for very large cluster sizes --- but for newer
4467  * file systems, it's better to calculate this figure once at mkfs
4468  * time, and store it in the superblock.  If the superblock value is
4469  * present (even for non-bigalloc file systems), we will use it.
4470  */
4471 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4472 {
4473         struct super_block *sb = dentry->d_sb;
4474         struct ext4_sb_info *sbi = EXT4_SB(sb);
4475         struct ext4_super_block *es = sbi->s_es;
4476         struct ext4_group_desc *gdp;
4477         u64 fsid;
4478         s64 bfree;
4479
4480         if (test_opt(sb, MINIX_DF)) {
4481                 sbi->s_overhead_last = 0;
4482         } else if (es->s_overhead_clusters) {
4483                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4484         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4485                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4486                 ext4_fsblk_t overhead = 0;
4487
4488                 /*
4489                  * Compute the overhead (FS structures).  This is constant
4490                  * for a given filesystem unless the number of block groups
4491                  * changes so we cache the previous value until it does.
4492                  */
4493
4494                 /*
4495                  * All of the blocks before first_data_block are
4496                  * overhead
4497                  */
4498                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4499
4500                 /*
4501                  * Add the overhead found in each block group
4502                  */
4503                 for (i = 0; i < ngroups; i++) {
4504                         gdp = ext4_get_group_desc(sb, i, NULL);
4505                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4506                         cond_resched();
4507                 }
4508                 sbi->s_overhead_last = overhead;
4509                 smp_wmb();
4510                 sbi->s_blocks_last = ext4_blocks_count(es);
4511         }
4512
4513         buf->f_type = EXT4_SUPER_MAGIC;
4514         buf->f_bsize = sb->s_blocksize;
4515         buf->f_blocks = (ext4_blocks_count(es) -
4516                          EXT4_C2B(sbi, sbi->s_overhead_last));
4517         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4518                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4519         /* prevent underflow in case that few free space is available */
4520         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4521         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4522         if (buf->f_bfree < ext4_r_blocks_count(es))
4523                 buf->f_bavail = 0;
4524         buf->f_files = le32_to_cpu(es->s_inodes_count);
4525         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4526         buf->f_namelen = EXT4_NAME_LEN;
4527         fsid = le64_to_cpup((void *)es->s_uuid) ^
4528                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4529         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4530         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4531
4532         return 0;
4533 }
4534
4535 /* Helper function for writing quotas on sync - we need to start transaction
4536  * before quota file is locked for write. Otherwise the are possible deadlocks:
4537  * Process 1                         Process 2
4538  * ext4_create()                     quota_sync()
4539  *   jbd2_journal_start()                  write_dquot()
4540  *   dquot_initialize()                         down(dqio_mutex)
4541  *     down(dqio_mutex)                    jbd2_journal_start()
4542  *
4543  */
4544
4545 #ifdef CONFIG_QUOTA
4546
4547 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4548 {
4549         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4550 }
4551
4552 static int ext4_write_dquot(struct dquot *dquot)
4553 {
4554         int ret, err;
4555         handle_t *handle;
4556         struct inode *inode;
4557
4558         inode = dquot_to_inode(dquot);
4559         handle = ext4_journal_start(inode,
4560                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4561         if (IS_ERR(handle))
4562                 return PTR_ERR(handle);
4563         ret = dquot_commit(dquot);
4564         err = ext4_journal_stop(handle);
4565         if (!ret)
4566                 ret = err;
4567         return ret;
4568 }
4569
4570 static int ext4_acquire_dquot(struct dquot *dquot)
4571 {
4572         int ret, err;
4573         handle_t *handle;
4574
4575         handle = ext4_journal_start(dquot_to_inode(dquot),
4576                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4577         if (IS_ERR(handle))
4578                 return PTR_ERR(handle);
4579         ret = dquot_acquire(dquot);
4580         err = ext4_journal_stop(handle);
4581         if (!ret)
4582                 ret = err;
4583         return ret;
4584 }
4585
4586 static int ext4_release_dquot(struct dquot *dquot)
4587 {
4588         int ret, err;
4589         handle_t *handle;
4590
4591         handle = ext4_journal_start(dquot_to_inode(dquot),
4592                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4593         if (IS_ERR(handle)) {
4594                 /* Release dquot anyway to avoid endless cycle in dqput() */
4595                 dquot_release(dquot);
4596                 return PTR_ERR(handle);
4597         }
4598         ret = dquot_release(dquot);
4599         err = ext4_journal_stop(handle);
4600         if (!ret)
4601                 ret = err;
4602         return ret;
4603 }
4604
4605 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4606 {
4607         /* Are we journaling quotas? */
4608         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4609             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4610                 dquot_mark_dquot_dirty(dquot);
4611                 return ext4_write_dquot(dquot);
4612         } else {
4613                 return dquot_mark_dquot_dirty(dquot);
4614         }
4615 }
4616
4617 static int ext4_write_info(struct super_block *sb, int type)
4618 {
4619         int ret, err;
4620         handle_t *handle;
4621
4622         /* Data block + inode block */
4623         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4624         if (IS_ERR(handle))
4625                 return PTR_ERR(handle);
4626         ret = dquot_commit_info(sb, type);
4627         err = ext4_journal_stop(handle);
4628         if (!ret)
4629                 ret = err;
4630         return ret;
4631 }
4632
4633 /*
4634  * Turn on quotas during mount time - we need to find
4635  * the quota file and such...
4636  */
4637 static int ext4_quota_on_mount(struct super_block *sb, int type)
4638 {
4639         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4640                                         EXT4_SB(sb)->s_jquota_fmt, type);
4641 }
4642
4643 /*
4644  * Standard function to be called on quota_on
4645  */
4646 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4647                          struct path *path)
4648 {
4649         int err;
4650
4651         if (!test_opt(sb, QUOTA))
4652                 return -EINVAL;
4653
4654         /* Quotafile not on the same filesystem? */
4655         if (path->dentry->d_sb != sb)
4656                 return -EXDEV;
4657         /* Journaling quota? */
4658         if (EXT4_SB(sb)->s_qf_names[type]) {
4659                 /* Quotafile not in fs root? */
4660                 if (path->dentry->d_parent != sb->s_root)
4661                         ext4_msg(sb, KERN_WARNING,
4662                                 "Quota file not on filesystem root. "
4663                                 "Journaled quota will not work");
4664         }
4665
4666         /*
4667          * When we journal data on quota file, we have to flush journal to see
4668          * all updates to the file when we bypass pagecache...
4669          */
4670         if (EXT4_SB(sb)->s_journal &&
4671             ext4_should_journal_data(path->dentry->d_inode)) {
4672                 /*
4673                  * We don't need to lock updates but journal_flush() could
4674                  * otherwise be livelocked...
4675                  */
4676                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4677                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4678                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4679                 if (err)
4680                         return err;
4681         }
4682
4683         return dquot_quota_on(sb, type, format_id, path);
4684 }
4685
4686 static int ext4_quota_off(struct super_block *sb, int type)
4687 {
4688         struct inode *inode = sb_dqopt(sb)->files[type];
4689         handle_t *handle;
4690
4691         /* Force all delayed allocation blocks to be allocated.
4692          * Caller already holds s_umount sem */
4693         if (test_opt(sb, DELALLOC))
4694                 sync_filesystem(sb);
4695
4696         if (!inode)
4697                 goto out;
4698
4699         /* Update modification times of quota files when userspace can
4700          * start looking at them */
4701         handle = ext4_journal_start(inode, 1);
4702         if (IS_ERR(handle))
4703                 goto out;
4704         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4705         ext4_mark_inode_dirty(handle, inode);
4706         ext4_journal_stop(handle);
4707
4708 out:
4709         return dquot_quota_off(sb, type);
4710 }
4711
4712 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4713  * acquiring the locks... As quota files are never truncated and quota code
4714  * itself serializes the operations (and no one else should touch the files)
4715  * we don't have to be afraid of races */
4716 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4717                                size_t len, loff_t off)
4718 {
4719         struct inode *inode = sb_dqopt(sb)->files[type];
4720         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4721         int err = 0;
4722         int offset = off & (sb->s_blocksize - 1);
4723         int tocopy;
4724         size_t toread;
4725         struct buffer_head *bh;
4726         loff_t i_size = i_size_read(inode);
4727
4728         if (off > i_size)
4729                 return 0;
4730         if (off+len > i_size)
4731                 len = i_size-off;
4732         toread = len;
4733         while (toread > 0) {
4734                 tocopy = sb->s_blocksize - offset < toread ?
4735                                 sb->s_blocksize - offset : toread;
4736                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4737                 if (err)
4738                         return err;
4739                 if (!bh)        /* A hole? */
4740                         memset(data, 0, tocopy);
4741                 else
4742                         memcpy(data, bh->b_data+offset, tocopy);
4743                 brelse(bh);
4744                 offset = 0;
4745                 toread -= tocopy;
4746                 data += tocopy;
4747                 blk++;
4748         }
4749         return len;
4750 }
4751
4752 /* Write to quotafile (we know the transaction is already started and has
4753  * enough credits) */
4754 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4755                                 const char *data, size_t len, loff_t off)
4756 {
4757         struct inode *inode = sb_dqopt(sb)->files[type];
4758         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4759         int err = 0;
4760         int offset = off & (sb->s_blocksize - 1);
4761         struct buffer_head *bh;
4762         handle_t *handle = journal_current_handle();
4763
4764         if (EXT4_SB(sb)->s_journal && !handle) {
4765                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4766                         " cancelled because transaction is not started",
4767                         (unsigned long long)off, (unsigned long long)len);
4768                 return -EIO;
4769         }
4770         /*
4771          * Since we account only one data block in transaction credits,
4772          * then it is impossible to cross a block boundary.
4773          */
4774         if (sb->s_blocksize - offset < len) {
4775                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4776                         " cancelled because not block aligned",
4777                         (unsigned long long)off, (unsigned long long)len);
4778                 return -EIO;
4779         }
4780
4781         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4782         bh = ext4_bread(handle, inode, blk, 1, &err);
4783         if (!bh)
4784                 goto out;
4785         err = ext4_journal_get_write_access(handle, bh);
4786         if (err) {
4787                 brelse(bh);
4788                 goto out;
4789         }
4790         lock_buffer(bh);
4791         memcpy(bh->b_data+offset, data, len);
4792         flush_dcache_page(bh->b_page);
4793         unlock_buffer(bh);
4794         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4795         brelse(bh);
4796 out:
4797         if (err) {
4798                 mutex_unlock(&inode->i_mutex);
4799                 return err;
4800         }
4801         if (inode->i_size < off + len) {
4802                 i_size_write(inode, off + len);
4803                 EXT4_I(inode)->i_disksize = inode->i_size;
4804                 ext4_mark_inode_dirty(handle, inode);
4805         }
4806         mutex_unlock(&inode->i_mutex);
4807         return len;
4808 }
4809
4810 #endif
4811
4812 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4813                        const char *dev_name, void *data)
4814 {
4815         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4816 }
4817
4818 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4819 static inline void register_as_ext2(void)
4820 {
4821         int err = register_filesystem(&ext2_fs_type);
4822         if (err)
4823                 printk(KERN_WARNING
4824                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4825 }
4826
4827 static inline void unregister_as_ext2(void)
4828 {
4829         unregister_filesystem(&ext2_fs_type);
4830 }
4831
4832 static inline int ext2_feature_set_ok(struct super_block *sb)
4833 {
4834         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4835                 return 0;
4836         if (sb->s_flags & MS_RDONLY)
4837                 return 1;
4838         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4839                 return 0;
4840         return 1;
4841 }
4842 MODULE_ALIAS("ext2");
4843 #else
4844 static inline void register_as_ext2(void) { }
4845 static inline void unregister_as_ext2(void) { }
4846 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4847 #endif
4848
4849 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4850 static inline void register_as_ext3(void)
4851 {
4852         int err = register_filesystem(&ext3_fs_type);
4853         if (err)
4854                 printk(KERN_WARNING
4855                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4856 }
4857
4858 static inline void unregister_as_ext3(void)
4859 {
4860         unregister_filesystem(&ext3_fs_type);
4861 }
4862
4863 static inline int ext3_feature_set_ok(struct super_block *sb)
4864 {
4865         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4866                 return 0;
4867         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4868                 return 0;
4869         if (sb->s_flags & MS_RDONLY)
4870                 return 1;
4871         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4872                 return 0;
4873         return 1;
4874 }
4875 MODULE_ALIAS("ext3");
4876 #else
4877 static inline void register_as_ext3(void) { }
4878 static inline void unregister_as_ext3(void) { }
4879 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4880 #endif
4881
4882 static struct file_system_type ext4_fs_type = {
4883         .owner          = THIS_MODULE,
4884         .name           = "ext4",
4885         .mount          = ext4_mount,
4886         .kill_sb        = kill_block_super,
4887         .fs_flags       = FS_REQUIRES_DEV,
4888 };
4889
4890 static int __init ext4_init_feat_adverts(void)
4891 {
4892         struct ext4_features *ef;
4893         int ret = -ENOMEM;
4894
4895         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4896         if (!ef)
4897                 goto out;
4898
4899         ef->f_kobj.kset = ext4_kset;
4900         init_completion(&ef->f_kobj_unregister);
4901         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4902                                    "features");
4903         if (ret) {
4904                 kfree(ef);
4905                 goto out;
4906         }
4907
4908         ext4_feat = ef;
4909         ret = 0;
4910 out:
4911         return ret;
4912 }
4913
4914 static void ext4_exit_feat_adverts(void)
4915 {
4916         kobject_put(&ext4_feat->f_kobj);
4917         wait_for_completion(&ext4_feat->f_kobj_unregister);
4918         kfree(ext4_feat);
4919 }
4920
4921 /* Shared across all ext4 file systems */
4922 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4923 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4924
4925 static int __init ext4_init_fs(void)
4926 {
4927         int i, err;
4928
4929         ext4_li_info = NULL;
4930         mutex_init(&ext4_li_mtx);
4931
4932         ext4_check_flag_values();
4933
4934         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4935                 mutex_init(&ext4__aio_mutex[i]);
4936                 init_waitqueue_head(&ext4__ioend_wq[i]);
4937         }
4938
4939         err = ext4_init_pageio();
4940         if (err)
4941                 return err;
4942         err = ext4_init_system_zone();
4943         if (err)
4944                 goto out6;
4945         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4946         if (!ext4_kset)
4947                 goto out5;
4948         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4949
4950         err = ext4_init_feat_adverts();
4951         if (err)
4952                 goto out4;
4953
4954         err = ext4_init_mballoc();
4955         if (err)
4956                 goto out3;
4957
4958         err = ext4_init_xattr();
4959         if (err)
4960                 goto out2;
4961         err = init_inodecache();
4962         if (err)
4963                 goto out1;
4964         register_as_ext3();
4965         register_as_ext2();
4966         err = register_filesystem(&ext4_fs_type);
4967         if (err)
4968                 goto out;
4969
4970         return 0;
4971 out:
4972         unregister_as_ext2();
4973         unregister_as_ext3();
4974         destroy_inodecache();
4975 out1:
4976         ext4_exit_xattr();
4977 out2:
4978         ext4_exit_mballoc();
4979 out3:
4980         ext4_exit_feat_adverts();
4981 out4:
4982         if (ext4_proc_root)
4983                 remove_proc_entry("fs/ext4", NULL);
4984         kset_unregister(ext4_kset);
4985 out5:
4986         ext4_exit_system_zone();
4987 out6:
4988         ext4_exit_pageio();
4989         return err;
4990 }
4991
4992 static void __exit ext4_exit_fs(void)
4993 {
4994         ext4_destroy_lazyinit_thread();
4995         unregister_as_ext2();
4996         unregister_as_ext3();
4997         unregister_filesystem(&ext4_fs_type);
4998         destroy_inodecache();
4999         ext4_exit_xattr();
5000         ext4_exit_mballoc();
5001         ext4_exit_feat_adverts();
5002         remove_proc_entry("fs/ext4", NULL);
5003         kset_unregister(ext4_kset);
5004         ext4_exit_system_zone();
5005         ext4_exit_pageio();
5006 }
5007
5008 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5009 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5010 MODULE_LICENSE("GPL");
5011 module_init(ext4_init_fs)
5012 module_exit(ext4_exit_fs)