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 void *ext4_kvmalloc(size_t size, gfp_t flags)
 116 {
 117         void *ret;
 118
 119         ret = kmalloc(size, flags);
 120         if (!ret)
 121                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 122         return ret;
 123 }
 124
 125 void *ext4_kvzalloc(size_t size, gfp_t flags)
 126 {
 127         void *ret;
 128
 129         ret = kzalloc(size, flags);
 130         if (!ret)
 131                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 132         return ret;
 133 }
 134
 135 void ext4_kvfree(void *ptr)
 136 {
 137         if (is_vmalloc_addr(ptr))
 138                 vfree(ptr);
 139         else
 140                 kfree(ptr);
 141
 142 }
 143
 144 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 145                                struct ext4_group_desc *bg)
 146 {
 147         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 148                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 149                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 150 }
 151
 152 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 153                                struct ext4_group_desc *bg)
 154 {
 155         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 156                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 157                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 158 }
 159
 160 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 161                               struct ext4_group_desc *bg)
 162 {
 163         return le32_to_cpu(bg->bg_inode_table_lo) |
 164                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 165                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 166 }
 167
 168 __u32 ext4_free_group_clusters(struct super_block *sb,
 169                                struct ext4_group_desc *bg)
 170 {
 171         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 172                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 173                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 174 }
 175
 176 __u32 ext4_free_inodes_count(struct super_block *sb,
 177                               struct ext4_group_desc *bg)
 178 {
 179         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 180                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 181                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 182 }
 183
 184 __u32 ext4_used_dirs_count(struct super_block *sb,
 185                               struct ext4_group_desc *bg)
 186 {
 187         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 188                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 189                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 190 }
 191
 192 __u32 ext4_itable_unused_count(struct super_block *sb,
 193                               struct ext4_group_desc *bg)
 194 {
 195         return le16_to_cpu(bg->bg_itable_unused_lo) |
 196                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 197                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 198 }
 199
 200 void ext4_block_bitmap_set(struct super_block *sb,
 201                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 202 {
 203         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 204         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 205                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 206 }
 207
 208 void ext4_inode_bitmap_set(struct super_block *sb,
 209                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 210 {
 211         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 212         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 213                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 214 }
 215
 216 void ext4_inode_table_set(struct super_block *sb,
 217                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 218 {
 219         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 220         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 221                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 222 }
 223
 224 void ext4_free_group_clusters_set(struct super_block *sb,
 225                                   struct ext4_group_desc *bg, __u32 count)
 226 {
 227         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 228         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 229                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 230 }
 231
 232 void ext4_free_inodes_set(struct super_block *sb,
 233                           struct ext4_group_desc *bg, __u32 count)
 234 {
 235         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 236         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 237                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 238 }
 239
 240 void ext4_used_dirs_set(struct super_block *sb,
 241                           struct ext4_group_desc *bg, __u32 count)
 242 {
 243         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 244         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 245                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 246 }
 247
 248 void ext4_itable_unused_set(struct super_block *sb,
 249                           struct ext4_group_desc *bg, __u32 count)
 250 {
 251         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 252         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 253                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 254 }
 255
 256
 257 /* Just increment the non-pointer handle value */
 258 static handle_t *ext4_get_nojournal(void)
 259 {
 260         handle_t *handle = current->journal_info;
 261         unsigned long ref_cnt = (unsigned long)handle;
 262
 263         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 264
 265         ref_cnt++;
 266         handle = (handle_t *)ref_cnt;
 267
 268         current->journal_info = handle;
 269         return handle;
 270 }
 271
 272
 273 /* Decrement the non-pointer handle value */
 274 static void ext4_put_nojournal(handle_t *handle)
 275 {
 276         unsigned long ref_cnt = (unsigned long)handle;
 277
 278         BUG_ON(ref_cnt == 0);
 279
 280         ref_cnt--;
 281         handle = (handle_t *)ref_cnt;
 282
 283         current->journal_info = handle;
 284 }
 285
 286 /*
 287  * Wrappers for jbd2_journal_start/end.
 288  *
 289  * The only special thing we need to do here is to make sure that all
 290  * journal_end calls result in the superblock being marked dirty, so
 291  * that sync() will call the filesystem's write_super callback if
 292  * appropriate.
 293  *
 294  * To avoid j_barrier hold in userspace when a user calls freeze(),
 295  * ext4 prevents a new handle from being started by s_frozen, which
 296  * is in an upper layer.
 297  */
 298 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 299 {
 300         journal_t *journal;
 301         handle_t  *handle;
 302
 303         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 304         if (sb->s_flags & MS_RDONLY)
 305                 return ERR_PTR(-EROFS);
 306
 307         journal = EXT4_SB(sb)->s_journal;
 308         handle = ext4_journal_current_handle();
 309
 310         /*
 311          * If a handle has been started, it should be allowed to
 312          * finish, otherwise deadlock could happen between freeze
 313          * and others(e.g. truncate) due to the restart of the
 314          * journal handle if the filesystem is forzen and active
 315          * handles are not stopped.
 316          */
 317         if (!handle)
 318                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 319
 320         if (!journal)
 321                 return ext4_get_nojournal();
 322         /*
 323          * Special case here: if the journal has aborted behind our
 324          * backs (eg. EIO in the commit thread), then we still need to
 325          * take the FS itself readonly cleanly.
 326          */
 327         if (is_journal_aborted(journal)) {
 328                 ext4_abort(sb, "Detected aborted journal");
 329                 return ERR_PTR(-EROFS);
 330         }
 331         return jbd2_journal_start(journal, nblocks);
 332 }
 333
 334 /*
 335  * The only special thing we need to do here is to make sure that all
 336  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 337  * that sync() will call the filesystem's write_super callback if
 338  * appropriate.
 339  */
 340 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 341 {
 342         struct super_block *sb;
 343         int err;
 344         int rc;
 345
 346         if (!ext4_handle_valid(handle)) {
 347                 ext4_put_nojournal(handle);
 348                 return 0;
 349         }
 350         sb = handle->h_transaction->t_journal->j_private;
 351         err = handle->h_err;
 352         rc = jbd2_journal_stop(handle);
 353
 354         if (!err)
 355                 err = rc;
 356         if (err)
 357                 __ext4_std_error(sb, where, line, err);
 358         return err;
 359 }
 360
 361 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 362                                const char *err_fn, struct buffer_head *bh,
 363                                handle_t *handle, int err)
 364 {
 365         char nbuf[16];
 366         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 367
 368         BUG_ON(!ext4_handle_valid(handle));
 369
 370         if (bh)
 371                 BUFFER_TRACE(bh, "abort");
 372
 373         if (!handle->h_err)
 374                 handle->h_err = err;
 375
 376         if (is_handle_aborted(handle))
 377                 return;
 378
 379         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 380                caller, line, errstr, err_fn);
 381
 382         jbd2_journal_abort_handle(handle);
 383 }
 384
 385 static void __save_error_info(struct super_block *sb, const char *func,
 386                             unsigned int line)
 387 {
 388         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 389
 390         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 391         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 392         es->s_last_error_time = cpu_to_le32(get_seconds());
 393         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 394         es->s_last_error_line = cpu_to_le32(line);
 395         if (!es->s_first_error_time) {
 396                 es->s_first_error_time = es->s_last_error_time;
 397                 strncpy(es->s_first_error_func, func,
 398                         sizeof(es->s_first_error_func));
 399                 es->s_first_error_line = cpu_to_le32(line);
 400                 es->s_first_error_ino = es->s_last_error_ino;
 401                 es->s_first_error_block = es->s_last_error_block;
 402         }
 403         /*
 404          * Start the daily error reporting function if it hasn't been
 405          * started already
 406          */
 407         if (!es->s_error_count)
 408                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 409         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 410 }
 411
 412 static void save_error_info(struct super_block *sb, const char *func,
 413                             unsigned int line)
 414 {
 415         __save_error_info(sb, func, line);
 416         ext4_commit_super(sb, 1);
 417 }
 418
 419 /*
 420  * The del_gendisk() function uninitializes the disk-specific data
 421  * structures, including the bdi structure, without telling anyone
 422  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 423  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 424  * This is a kludge to prevent these oops until we can put in a proper
 425  * hook in del_gendisk() to inform the VFS and file system layers.
 426  */
 427 static int block_device_ejected(struct super_block *sb)
 428 {
 429         struct inode *bd_inode = sb->s_bdev->bd_inode;
 430         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 431
 432         return bdi->dev == NULL;
 433 }
 434
 435 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 436 {
 437         struct super_block              *sb = journal->j_private;
 438         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 439         int                             error = is_journal_aborted(journal);
 440         struct ext4_journal_cb_entry    *jce, *tmp;
 441
 442         spin_lock(&sbi->s_md_lock);
 443         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 444                 list_del_init(&jce->jce_list);
 445                 spin_unlock(&sbi->s_md_lock);
 446                 jce->jce_func(sb, jce, error);
 447                 spin_lock(&sbi->s_md_lock);
 448         }
 449         spin_unlock(&sbi->s_md_lock);
 450 }
 451
 452 /* Deal with the reporting of failure conditions on a filesystem such as
 453  * inconsistencies detected or read IO failures.
 454  *
 455  * On ext2, we can store the error state of the filesystem in the
 456  * superblock.  That is not possible on ext4, because we may have other
 457  * write ordering constraints on the superblock which prevent us from
 458  * writing it out straight away; and given that the journal is about to
 459  * be aborted, we can't rely on the current, or future, transactions to
 460  * write out the superblock safely.
 461  *
 462  * We'll just use the jbd2_journal_abort() error code to record an error in
 463  * the journal instead.  On recovery, the journal will complain about
 464  * that error until we've noted it down and cleared it.
 465  */
 466
 467 static void ext4_handle_error(struct super_block *sb)
 468 {
 469         if (sb->s_flags & MS_RDONLY)
 470                 return;
 471
 472         if (!test_opt(sb, ERRORS_CONT)) {
 473                 journal_t *journal = EXT4_SB(sb)->s_journal;
 474
 475                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 476                 if (journal)
 477                         jbd2_journal_abort(journal, -EIO);
 478         }
 479         if (test_opt(sb, ERRORS_RO)) {
 480                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 481                 sb->s_flags |= MS_RDONLY;
 482         }
 483         if (test_opt(sb, ERRORS_PANIC))
 484                 panic("EXT4-fs (device %s): panic forced after error\n",
 485                         sb->s_id);
 486 }
 487
 488 void __ext4_error(struct super_block *sb, const char *function,
 489                   unsigned int line, const char *fmt, ...)
 490 {
 491         struct va_format vaf;
 492         va_list args;
 493
 494         va_start(args, fmt);
 495         vaf.fmt = fmt;
 496         vaf.va = &args;
 497         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 498                sb->s_id, function, line, current->comm, &vaf);
 499         va_end(args);
 500
 501         ext4_handle_error(sb);
 502 }
 503
 504 void ext4_error_inode(struct inode *inode, const char *function,
 505                       unsigned int line, ext4_fsblk_t block,
 506                       const char *fmt, ...)
 507 {
 508         va_list args;
 509         struct va_format vaf;
 510         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 511
 512         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 513         es->s_last_error_block = cpu_to_le64(block);
 514         save_error_info(inode->i_sb, function, line);
 515         va_start(args, fmt);
 516         vaf.fmt = fmt;
 517         vaf.va = &args;
 518         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 519                inode->i_sb->s_id, function, line, inode->i_ino);
 520         if (block)
 521                 printk(KERN_CONT "block %llu: ", block);
 522         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 523         va_end(args);
 524
 525         ext4_handle_error(inode->i_sb);
 526 }
 527
 528 void ext4_error_file(struct file *file, const char *function,
 529                      unsigned int line, ext4_fsblk_t block,
 530                      const char *fmt, ...)
 531 {
 532         va_list args;
 533         struct va_format vaf;
 534         struct ext4_super_block *es;
 535         struct inode *inode = file->f_dentry->d_inode;
 536         char pathname[80], *path;
 537
 538         es = EXT4_SB(inode->i_sb)->s_es;
 539         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 540         save_error_info(inode->i_sb, function, line);
 541         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 542         if (IS_ERR(path))
 543                 path = "(unknown)";
 544         printk(KERN_CRIT
 545                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 546                inode->i_sb->s_id, function, line, inode->i_ino);
 547         if (block)
 548                 printk(KERN_CONT "block %llu: ", block);
 549         va_start(args, fmt);
 550         vaf.fmt = fmt;
 551         vaf.va = &args;
 552         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 553         va_end(args);
 554
 555         ext4_handle_error(inode->i_sb);
 556 }
 557
 558 static const char *ext4_decode_error(struct super_block *sb, int errno,
 559                                      char nbuf[16])
 560 {
 561         char *errstr = NULL;
 562
 563         switch (errno) {
 564         case -EIO:
 565                 errstr = "IO failure";
 566                 break;
 567         case -ENOMEM:
 568                 errstr = "Out of memory";
 569                 break;
 570         case -EROFS:
 571                 if (!sb || (EXT4_SB(sb)->s_journal &&
 572                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 573                         errstr = "Journal has aborted";
 574                 else
 575                         errstr = "Readonly filesystem";
 576                 break;
 577         default:
 578                 /* If the caller passed in an extra buffer for unknown
 579                  * errors, textualise them now.  Else we just return
 580                  * NULL. */
 581                 if (nbuf) {
 582                         /* Check for truncated error codes... */
 583                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 584                                 errstr = nbuf;
 585                 }
 586                 break;
 587         }
 588
 589         return errstr;
 590 }
 591
 592 /* __ext4_std_error decodes expected errors from journaling functions
 593  * automatically and invokes the appropriate error response.  */
 594
 595 void __ext4_std_error(struct super_block *sb, const char *function,
 596                       unsigned int line, int errno)
 597 {
 598         char nbuf[16];
 599         const char *errstr;
 600
 601         /* Special case: if the error is EROFS, and we're not already
 602          * inside a transaction, then there's really no point in logging
 603          * an error. */
 604         if (errno == -EROFS && journal_current_handle() == NULL &&
 605             (sb->s_flags & MS_RDONLY))
 606                 return;
 607
 608         errstr = ext4_decode_error(sb, errno, nbuf);
 609         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 610                sb->s_id, function, line, errstr);
 611         save_error_info(sb, function, line);
 612
 613         ext4_handle_error(sb);
 614 }
 615
 616 /*
 617  * ext4_abort is a much stronger failure handler than ext4_error.  The
 618  * abort function may be used to deal with unrecoverable failures such
 619  * as journal IO errors or ENOMEM at a critical moment in log management.
 620  *
 621  * We unconditionally force the filesystem into an ABORT|READONLY state,
 622  * unless the error response on the fs has been set to panic in which
 623  * case we take the easy way out and panic immediately.
 624  */
 625
 626 void __ext4_abort(struct super_block *sb, const char *function,
 627                 unsigned int line, const char *fmt, ...)
 628 {
 629         va_list args;
 630
 631         save_error_info(sb, function, line);
 632         va_start(args, fmt);
 633         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 634                function, line);
 635         vprintk(fmt, args);
 636         printk("\n");
 637         va_end(args);
 638
 639         if ((sb->s_flags & MS_RDONLY) == 0) {
 640                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 641                 sb->s_flags |= MS_RDONLY;
 642                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 643                 if (EXT4_SB(sb)->s_journal)
 644                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 645                 save_error_info(sb, function, line);
 646         }
 647         if (test_opt(sb, ERRORS_PANIC))
 648                 panic("EXT4-fs panic from previous error\n");
 649 }
 650
 651 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 652 {
 653         struct va_format vaf;
 654         va_list args;
 655
 656         va_start(args, fmt);
 657         vaf.fmt = fmt;
 658         vaf.va = &args;
 659         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 660         va_end(args);
 661 }
 662
 663 void __ext4_warning(struct super_block *sb, const char *function,
 664                     unsigned int line, const char *fmt, ...)
 665 {
 666         struct va_format vaf;
 667         va_list args;
 668
 669         va_start(args, fmt);
 670         vaf.fmt = fmt;
 671         vaf.va = &args;
 672         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 673                sb->s_id, function, line, &vaf);
 674         va_end(args);
 675 }
 676
 677 void __ext4_grp_locked_error(const char *function, unsigned int line,
 678                              struct super_block *sb, ext4_group_t grp,
 679                              unsigned long ino, ext4_fsblk_t block,
 680                              const char *fmt, ...)
 681 __releases(bitlock)
 682 __acquires(bitlock)
 683 {
 684         struct va_format vaf;
 685         va_list args;
 686         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 687
 688         es->s_last_error_ino = cpu_to_le32(ino);
 689         es->s_last_error_block = cpu_to_le64(block);
 690         __save_error_info(sb, function, line);
 691
 692         va_start(args, fmt);
 693
 694         vaf.fmt = fmt;
 695         vaf.va = &args;
 696         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 697                sb->s_id, function, line, grp);
 698         if (ino)
 699                 printk(KERN_CONT "inode %lu: ", ino);
 700         if (block)
 701                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 702         printk(KERN_CONT "%pV\n", &vaf);
 703         va_end(args);
 704
 705         if (test_opt(sb, ERRORS_CONT)) {
 706                 ext4_commit_super(sb, 0);
 707                 return;
 708         }
 709
 710         ext4_unlock_group(sb, grp);
 711         ext4_handle_error(sb);
 712         /*
 713          * We only get here in the ERRORS_RO case; relocking the group
 714          * may be dangerous, but nothing bad will happen since the
 715          * filesystem will have already been marked read/only and the
 716          * journal has been aborted.  We return 1 as a hint to callers
 717          * who might what to use the return value from
 718          * ext4_grp_locked_error() to distinguish between the
 719          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 720          * aggressively from the ext4 function in question, with a
 721          * more appropriate error code.
 722          */
 723         ext4_lock_group(sb, grp);
 724         return;
 725 }
 726
 727 void ext4_update_dynamic_rev(struct super_block *sb)
 728 {
 729         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 730
 731         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 732                 return;
 733
 734         ext4_warning(sb,
 735                      "updating to rev %d because of new feature flag, "
 736                      "running e2fsck is recommended",
 737                      EXT4_DYNAMIC_REV);
 738
 739         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 740         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 741         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 742         /* leave es->s_feature_*compat flags alone */
 743         /* es->s_uuid will be set by e2fsck if empty */
 744
 745         /*
 746          * The rest of the superblock fields should be zero, and if not it
 747          * means they are likely already in use, so leave them alone.  We
 748          * can leave it up to e2fsck to clean up any inconsistencies there.
 749          */
 750 }
 751
 752 /*
 753  * Open the external journal device
 754  */
 755 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 756 {
 757         struct block_device *bdev;
 758         char b[BDEVNAME_SIZE];
 759
 760         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 761         if (IS_ERR(bdev))
 762                 goto fail;
 763         return bdev;
 764
 765 fail:
 766         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 767                         __bdevname(dev, b), PTR_ERR(bdev));
 768         return NULL;
 769 }
 770
 771 /*
 772  * Release the journal device
 773  */
 774 static int ext4_blkdev_put(struct block_device *bdev)
 775 {
 776         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 777 }
 778
 779 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 780 {
 781         struct block_device *bdev;
 782         int ret = -ENODEV;
 783
 784         bdev = sbi->journal_bdev;
 785         if (bdev) {
 786                 ret = ext4_blkdev_put(bdev);
 787                 sbi->journal_bdev = NULL;
 788         }
 789         return ret;
 790 }
 791
 792 static inline struct inode *orphan_list_entry(struct list_head *l)
 793 {
 794         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 795 }
 796
 797 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 798 {
 799         struct list_head *l;
 800
 801         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 802                  le32_to_cpu(sbi->s_es->s_last_orphan));
 803
 804         printk(KERN_ERR "sb_info orphan list:\n");
 805         list_for_each(l, &sbi->s_orphan) {
 806                 struct inode *inode = orphan_list_entry(l);
 807                 printk(KERN_ERR "  "
 808                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 809                        inode->i_sb->s_id, inode->i_ino, inode,
 810                        inode->i_mode, inode->i_nlink,
 811                        NEXT_ORPHAN(inode));
 812         }
 813 }
 814
 815 static void ext4_put_super(struct super_block *sb)
 816 {
 817         struct ext4_sb_info *sbi = EXT4_SB(sb);
 818         struct ext4_super_block *es = sbi->s_es;
 819         int i, err;
 820
 821         ext4_unregister_li_request(sb);
 822         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 823
 824         flush_workqueue(sbi->dio_unwritten_wq);
 825         destroy_workqueue(sbi->dio_unwritten_wq);
 826
 827         lock_super(sb);
 828         if (sb->s_dirt)
 829                 ext4_commit_super(sb, 1);
 830
 831         if (sbi->s_journal) {
 832                 err = jbd2_journal_destroy(sbi->s_journal);
 833                 sbi->s_journal = NULL;
 834                 if (err < 0)
 835                         ext4_abort(sb, "Couldn't clean up the journal");
 836         }
 837
 838         del_timer(&sbi->s_err_report);
 839         ext4_release_system_zone(sb);
 840         ext4_mb_release(sb);
 841         ext4_ext_release(sb);
 842         ext4_xattr_put_super(sb);
 843
 844         if (!(sb->s_flags & MS_RDONLY)) {
 845                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 846                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 847                 ext4_commit_super(sb, 1);
 848         }
 849         if (sbi->s_proc) {
 850                 remove_proc_entry(sb->s_id, ext4_proc_root);
 851         }
 852         kobject_del(&sbi->s_kobj);
 853
 854         for (i = 0; i < sbi->s_gdb_count; i++)
 855                 brelse(sbi->s_group_desc[i]);
 856         ext4_kvfree(sbi->s_group_desc);
 857         ext4_kvfree(sbi->s_flex_groups);
 858         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 859         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 860         percpu_counter_destroy(&sbi->s_dirs_counter);
 861         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 862         brelse(sbi->s_sbh);
 863 #ifdef CONFIG_QUOTA
 864         for (i = 0; i < MAXQUOTAS; i++)
 865                 kfree(sbi->s_qf_names[i]);
 866 #endif
 867
 868         /* Debugging code just in case the in-memory inode orphan list
 869          * isn't empty.  The on-disk one can be non-empty if we've
 870          * detected an error and taken the fs readonly, but the
 871          * in-memory list had better be clean by this point. */
 872         if (!list_empty(&sbi->s_orphan))
 873                 dump_orphan_list(sb, sbi);
 874         J_ASSERT(list_empty(&sbi->s_orphan));
 875
 876         invalidate_bdev(sb->s_bdev);
 877         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 878                 /*
 879                  * Invalidate the journal device's buffers.  We don't want them
 880                  * floating about in memory - the physical journal device may
 881                  * hotswapped, and it breaks the `ro-after' testing code.
 882                  */
 883                 sync_blockdev(sbi->journal_bdev);
 884                 invalidate_bdev(sbi->journal_bdev);
 885                 ext4_blkdev_remove(sbi);
 886         }
 887         if (sbi->s_mmp_tsk)
 888                 kthread_stop(sbi->s_mmp_tsk);
 889         sb->s_fs_info = NULL;
 890         /*
 891          * Now that we are completely done shutting down the
 892          * superblock, we need to actually destroy the kobject.
 893          */
 894         unlock_super(sb);
 895         kobject_put(&sbi->s_kobj);
 896         wait_for_completion(&sbi->s_kobj_unregister);
 897         kfree(sbi->s_blockgroup_lock);
 898         kfree(sbi);
 899 }
 900
 901 static struct kmem_cache *ext4_inode_cachep;
 902
 903 /*
 904  * Called inside transaction, so use GFP_NOFS
 905  */
 906 static struct inode *ext4_alloc_inode(struct super_block *sb)
 907 {
 908         struct ext4_inode_info *ei;
 909
 910         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 911         if (!ei)
 912                 return NULL;
 913
 914         ei->vfs_inode.i_version = 1;
 915         ei->vfs_inode.i_data.writeback_index = 0;
 916         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 917         INIT_LIST_HEAD(&ei->i_prealloc_list);
 918         spin_lock_init(&ei->i_prealloc_lock);
 919         ei->i_reserved_data_blocks = 0;
 920         ei->i_reserved_meta_blocks = 0;
 921         ei->i_allocated_meta_blocks = 0;
 922         ei->i_da_metadata_calc_len = 0;
 923         spin_lock_init(&(ei->i_block_reservation_lock));
 924 #ifdef CONFIG_QUOTA
 925         ei->i_reserved_quota = 0;
 926 #endif
 927         ei->jinode = NULL;
 928         INIT_LIST_HEAD(&ei->i_completed_io_list);
 929         spin_lock_init(&ei->i_completed_io_lock);
 930         ei->cur_aio_dio = NULL;
 931         ei->i_sync_tid = 0;
 932         ei->i_datasync_tid = 0;
 933         atomic_set(&ei->i_ioend_count, 0);
 934         atomic_set(&ei->i_aiodio_unwritten, 0);
 935
 936         return &ei->vfs_inode;
 937 }
 938
 939 static int ext4_drop_inode(struct inode *inode)
 940 {
 941         int drop = generic_drop_inode(inode);
 942
 943         trace_ext4_drop_inode(inode, drop);
 944         return drop;
 945 }
 946
 947 static void ext4_i_callback(struct rcu_head *head)
 948 {
 949         struct inode *inode = container_of(head, struct inode, i_rcu);
 950         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 951 }
 952
 953 static void ext4_destroy_inode(struct inode *inode)
 954 {
 955         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 956                 ext4_msg(inode->i_sb, KERN_ERR,
 957                          "Inode %lu (%p): orphan list check failed!",
 958                          inode->i_ino, EXT4_I(inode));
 959                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 960                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 961                                 true);
 962                 dump_stack();
 963         }
 964         call_rcu(&inode->i_rcu, ext4_i_callback);
 965 }
 966
 967 static void init_once(void *foo)
 968 {
 969         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 970
 971         INIT_LIST_HEAD(&ei->i_orphan);
 972 #ifdef CONFIG_EXT4_FS_XATTR
 973         init_rwsem(&ei->xattr_sem);
 974 #endif
 975         init_rwsem(&ei->i_data_sem);
 976         inode_init_once(&ei->vfs_inode);
 977 }
 978
 979 static int init_inodecache(void)
 980 {
 981         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 982                                              sizeof(struct ext4_inode_info),
 983                                              0, (SLAB_RECLAIM_ACCOUNT|
 984                                                 SLAB_MEM_SPREAD),
 985                                              init_once);
 986         if (ext4_inode_cachep == NULL)
 987                 return -ENOMEM;
 988         return 0;
 989 }
 990
 991 static void destroy_inodecache(void)
 992 {
 993         kmem_cache_destroy(ext4_inode_cachep);
 994 }
 995
 996 void ext4_clear_inode(struct inode *inode)
 997 {
 998         invalidate_inode_buffers(inode);
 999         end_writeback(inode);
1000         dquot_drop(inode);
1001         ext4_discard_preallocations(inode);
1002         if (EXT4_I(inode)->jinode) {
1003                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1004                                                EXT4_I(inode)->jinode);
1005                 jbd2_free_inode(EXT4_I(inode)->jinode);
1006                 EXT4_I(inode)->jinode = NULL;
1007         }
1008 }
1009
1010 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1011                                         u64 ino, u32 generation)
1012 {
1013         struct inode *inode;
1014
1015         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1016                 return ERR_PTR(-ESTALE);
1017         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1018                 return ERR_PTR(-ESTALE);
1019
1020         /* iget isn't really right if the inode is currently unallocated!!
1021          *
1022          * ext4_read_inode will return a bad_inode if the inode had been
1023          * deleted, so we should be safe.
1024          *
1025          * Currently we don't know the generation for parent directory, so
1026          * a generation of 0 means "accept any"
1027          */
1028         inode = ext4_iget(sb, ino);
1029         if (IS_ERR(inode))
1030                 return ERR_CAST(inode);
1031         if (generation && inode->i_generation != generation) {
1032                 iput(inode);
1033                 return ERR_PTR(-ESTALE);
1034         }
1035
1036         return inode;
1037 }
1038
1039 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1040                                         int fh_len, int fh_type)
1041 {
1042         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1043                                     ext4_nfs_get_inode);
1044 }
1045
1046 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1047                                         int fh_len, int fh_type)
1048 {
1049         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1050                                     ext4_nfs_get_inode);
1051 }
1052
1053 /*
1054  * Try to release metadata pages (indirect blocks, directories) which are
1055  * mapped via the block device.  Since these pages could have journal heads
1056  * which would prevent try_to_free_buffers() from freeing them, we must use
1057  * jbd2 layer's try_to_free_buffers() function to release them.
1058  */
1059 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1060                                  gfp_t wait)
1061 {
1062         journal_t *journal = EXT4_SB(sb)->s_journal;
1063
1064         WARN_ON(PageChecked(page));
1065         if (!page_has_buffers(page))
1066                 return 0;
1067         if (journal)
1068                 return jbd2_journal_try_to_free_buffers(journal, page,
1069                                                         wait & ~__GFP_WAIT);
1070         return try_to_free_buffers(page);
1071 }
1072
1073 #ifdef CONFIG_QUOTA
1074 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1075 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1076
1077 static int ext4_write_dquot(struct dquot *dquot);
1078 static int ext4_acquire_dquot(struct dquot *dquot);
1079 static int ext4_release_dquot(struct dquot *dquot);
1080 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1081 static int ext4_write_info(struct super_block *sb, int type);
1082 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1083                          struct path *path);
1084 static int ext4_quota_off(struct super_block *sb, int type);
1085 static int ext4_quota_on_mount(struct super_block *sb, int type);
1086 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1087                                size_t len, loff_t off);
1088 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1089                                 const char *data, size_t len, loff_t off);
1090
1091 static const struct dquot_operations ext4_quota_operations = {
1092         .get_reserved_space = ext4_get_reserved_space,
1093         .write_dquot    = ext4_write_dquot,
1094         .acquire_dquot  = ext4_acquire_dquot,
1095         .release_dquot  = ext4_release_dquot,
1096         .mark_dirty     = ext4_mark_dquot_dirty,
1097         .write_info     = ext4_write_info,
1098         .alloc_dquot    = dquot_alloc,
1099         .destroy_dquot  = dquot_destroy,
1100 };
1101
1102 static const struct quotactl_ops ext4_qctl_operations = {
1103         .quota_on       = ext4_quota_on,
1104         .quota_off      = ext4_quota_off,
1105         .quota_sync     = dquot_quota_sync,
1106         .get_info       = dquot_get_dqinfo,
1107         .set_info       = dquot_set_dqinfo,
1108         .get_dqblk      = dquot_get_dqblk,
1109         .set_dqblk      = dquot_set_dqblk
1110 };
1111 #endif
1112
1113 static const struct super_operations ext4_sops = {
1114         .alloc_inode    = ext4_alloc_inode,
1115         .destroy_inode  = ext4_destroy_inode,
1116         .write_inode    = ext4_write_inode,
1117         .dirty_inode    = ext4_dirty_inode,
1118         .drop_inode     = ext4_drop_inode,
1119         .evict_inode    = ext4_evict_inode,
1120         .put_super      = ext4_put_super,
1121         .sync_fs        = ext4_sync_fs,
1122         .freeze_fs      = ext4_freeze,
1123         .unfreeze_fs    = ext4_unfreeze,
1124         .statfs         = ext4_statfs,
1125         .remount_fs     = ext4_remount,
1126         .show_options   = ext4_show_options,
1127 #ifdef CONFIG_QUOTA
1128         .quota_read     = ext4_quota_read,
1129         .quota_write    = ext4_quota_write,
1130 #endif
1131         .bdev_try_to_free_page = bdev_try_to_free_page,
1132 };
1133
1134 static const struct super_operations ext4_nojournal_sops = {
1135         .alloc_inode    = ext4_alloc_inode,
1136         .destroy_inode  = ext4_destroy_inode,
1137         .write_inode    = ext4_write_inode,
1138         .dirty_inode    = ext4_dirty_inode,
1139         .drop_inode     = ext4_drop_inode,
1140         .evict_inode    = ext4_evict_inode,
1141         .write_super    = ext4_write_super,
1142         .put_super      = ext4_put_super,
1143         .statfs         = ext4_statfs,
1144         .remount_fs     = ext4_remount,
1145         .show_options   = ext4_show_options,
1146 #ifdef CONFIG_QUOTA
1147         .quota_read     = ext4_quota_read,
1148         .quota_write    = ext4_quota_write,
1149 #endif
1150         .bdev_try_to_free_page = bdev_try_to_free_page,
1151 };
1152
1153 static const struct export_operations ext4_export_ops = {
1154         .fh_to_dentry = ext4_fh_to_dentry,
1155         .fh_to_parent = ext4_fh_to_parent,
1156         .get_parent = ext4_get_parent,
1157 };
1158
1159 enum {
1160         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1161         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1162         Opt_nouid32, Opt_debug, Opt_removed,
1163         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1164         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1165         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1166         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1167         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1168         Opt_data_err_abort, Opt_data_err_ignore,
1169         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1170         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1171         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1172         Opt_usrquota, Opt_grpquota, Opt_i_version,
1173         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1174         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1175         Opt_inode_readahead_blks, Opt_journal_ioprio,
1176         Opt_dioread_nolock, Opt_dioread_lock,
1177         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1178 };
1179
1180 static const match_table_t tokens = {
1181         {Opt_bsd_df, "bsddf"},
1182         {Opt_minix_df, "minixdf"},
1183         {Opt_grpid, "grpid"},
1184         {Opt_grpid, "bsdgroups"},
1185         {Opt_nogrpid, "nogrpid"},
1186         {Opt_nogrpid, "sysvgroups"},
1187         {Opt_resgid, "resgid=%u"},
1188         {Opt_resuid, "resuid=%u"},
1189         {Opt_sb, "sb=%u"},
1190         {Opt_err_cont, "errors=continue"},
1191         {Opt_err_panic, "errors=panic"},
1192         {Opt_err_ro, "errors=remount-ro"},
1193         {Opt_nouid32, "nouid32"},
1194         {Opt_debug, "debug"},
1195         {Opt_removed, "oldalloc"},
1196         {Opt_removed, "orlov"},
1197         {Opt_user_xattr, "user_xattr"},
1198         {Opt_nouser_xattr, "nouser_xattr"},
1199         {Opt_acl, "acl"},
1200         {Opt_noacl, "noacl"},
1201         {Opt_noload, "noload"},
1202         {Opt_noload, "norecovery"},
1203         {Opt_removed, "nobh"},
1204         {Opt_removed, "bh"},
1205         {Opt_commit, "commit=%u"},
1206         {Opt_min_batch_time, "min_batch_time=%u"},
1207         {Opt_max_batch_time, "max_batch_time=%u"},
1208         {Opt_journal_dev, "journal_dev=%u"},
1209         {Opt_journal_checksum, "journal_checksum"},
1210         {Opt_journal_async_commit, "journal_async_commit"},
1211         {Opt_abort, "abort"},
1212         {Opt_data_journal, "data=journal"},
1213         {Opt_data_ordered, "data=ordered"},
1214         {Opt_data_writeback, "data=writeback"},
1215         {Opt_data_err_abort, "data_err=abort"},
1216         {Opt_data_err_ignore, "data_err=ignore"},
1217         {Opt_offusrjquota, "usrjquota="},
1218         {Opt_usrjquota, "usrjquota=%s"},
1219         {Opt_offgrpjquota, "grpjquota="},
1220         {Opt_grpjquota, "grpjquota=%s"},
1221         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1222         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1223         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1224         {Opt_grpquota, "grpquota"},
1225         {Opt_noquota, "noquota"},
1226         {Opt_quota, "quota"},
1227         {Opt_usrquota, "usrquota"},
1228         {Opt_barrier, "barrier=%u"},
1229         {Opt_barrier, "barrier"},
1230         {Opt_nobarrier, "nobarrier"},
1231         {Opt_i_version, "i_version"},
1232         {Opt_stripe, "stripe=%u"},
1233         {Opt_delalloc, "delalloc"},
1234         {Opt_nodelalloc, "nodelalloc"},
1235         {Opt_mblk_io_submit, "mblk_io_submit"},
1236         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1237         {Opt_block_validity, "block_validity"},
1238         {Opt_noblock_validity, "noblock_validity"},
1239         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1240         {Opt_journal_ioprio, "journal_ioprio=%u"},
1241         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1242         {Opt_auto_da_alloc, "auto_da_alloc"},
1243         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1244         {Opt_dioread_nolock, "dioread_nolock"},
1245         {Opt_dioread_lock, "dioread_lock"},
1246         {Opt_discard, "discard"},
1247         {Opt_nodiscard, "nodiscard"},
1248         {Opt_init_itable, "init_itable=%u"},
1249         {Opt_init_itable, "init_itable"},
1250         {Opt_noinit_itable, "noinit_itable"},
1251         {Opt_err, NULL},
1252 };
1253
1254 static ext4_fsblk_t get_sb_block(void **data)
1255 {
1256         ext4_fsblk_t    sb_block;
1257         char            *options = (char *) *data;
1258
1259         if (!options || strncmp(options, "sb=", 3) != 0)
1260                 return 1;       /* Default location */
1261
1262         options += 3;
1263         /* TODO: use simple_strtoll with >32bit ext4 */
1264         sb_block = simple_strtoul(options, &options, 0);
1265         if (*options && *options != ',') {
1266                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1267                        (char *) *data);
1268                 return 1;
1269         }
1270         if (*options == ',')
1271                 options++;
1272         *data = (void *) options;
1273
1274         return sb_block;
1275 }
1276
1277 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1278 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1279         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1280
1281 #ifdef CONFIG_QUOTA
1282 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1283 {
1284         struct ext4_sb_info *sbi = EXT4_SB(sb);
1285         char *qname;
1286
1287         if (sb_any_quota_loaded(sb) &&
1288                 !sbi->s_qf_names[qtype]) {
1289                 ext4_msg(sb, KERN_ERR,
1290                         "Cannot change journaled "
1291                         "quota options when quota turned on");
1292                 return 0;
1293         }
1294         qname = match_strdup(args);
1295         if (!qname) {
1296                 ext4_msg(sb, KERN_ERR,
1297                         "Not enough memory for storing quotafile name");
1298                 return 0;
1299         }
1300         if (sbi->s_qf_names[qtype] &&
1301                 strcmp(sbi->s_qf_names[qtype], qname)) {
1302                 ext4_msg(sb, KERN_ERR,
1303                         "%s quota file already specified", QTYPE2NAME(qtype));
1304                 kfree(qname);
1305                 return 0;
1306         }
1307         sbi->s_qf_names[qtype] = qname;
1308         if (strchr(sbi->s_qf_names[qtype], '/')) {
1309                 ext4_msg(sb, KERN_ERR,
1310                         "quotafile must be on filesystem root");
1311                 kfree(sbi->s_qf_names[qtype]);
1312                 sbi->s_qf_names[qtype] = NULL;
1313                 return 0;
1314         }
1315         set_opt(sb, QUOTA);
1316         return 1;
1317 }
1318
1319 static int clear_qf_name(struct super_block *sb, int qtype)
1320 {
1321
1322         struct ext4_sb_info *sbi = EXT4_SB(sb);
1323
1324         if (sb_any_quota_loaded(sb) &&
1325                 sbi->s_qf_names[qtype]) {
1326                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1327                         " when quota turned on");
1328                 return 0;
1329         }
1330         /*
1331          * The space will be released later when all options are confirmed
1332          * to be correct
1333          */
1334         sbi->s_qf_names[qtype] = NULL;
1335         return 1;
1336 }
1337 #endif
1338
1339 #define MOPT_SET        0x0001
1340 #define MOPT_CLEAR      0x0002
1341 #define MOPT_NOSUPPORT  0x0004
1342 #define MOPT_EXPLICIT   0x0008
1343 #define MOPT_CLEAR_ERR  0x0010
1344 #define MOPT_GTE0       0x0020
1345 #ifdef CONFIG_QUOTA
1346 #define MOPT_Q          0
1347 #define MOPT_QFMT       0x0040
1348 #else
1349 #define MOPT_Q          MOPT_NOSUPPORT
1350 #define MOPT_QFMT       MOPT_NOSUPPORT
1351 #endif
1352 #define MOPT_DATAJ      0x0080
1353
1354 static const struct mount_opts {
1355         int     token;
1356         int     mount_opt;
1357         int     flags;
1358 } ext4_mount_opts[] = {
1359         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1360         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1361         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1362         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1363         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1364         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1365         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1366         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1367         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1368         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1369         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1370         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1371         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1372         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1373         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1374         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1375                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1376         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1377         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1378         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1379         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1380         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1381         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1382         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1383         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1384         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1385         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1386         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1387         {Opt_commit, 0, MOPT_GTE0},
1388         {Opt_max_batch_time, 0, MOPT_GTE0},
1389         {Opt_min_batch_time, 0, MOPT_GTE0},
1390         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1391         {Opt_init_itable, 0, MOPT_GTE0},
1392         {Opt_stripe, 0, MOPT_GTE0},
1393         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1394         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1395         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1396 #ifdef CONFIG_EXT4_FS_XATTR
1397         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1398         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1399 #else
1400         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1401         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1402 #endif
1403 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1404         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1405         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1406 #else
1407         {Opt_acl, 0, MOPT_NOSUPPORT},
1408         {Opt_noacl, 0, MOPT_NOSUPPORT},
1409 #endif
1410         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1411         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1412         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1413         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1414                                                         MOPT_SET | MOPT_Q},
1415         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1416                                                         MOPT_SET | MOPT_Q},
1417         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1418                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1419         {Opt_usrjquota, 0, MOPT_Q},
1420         {Opt_grpjquota, 0, MOPT_Q},
1421         {Opt_offusrjquota, 0, MOPT_Q},
1422         {Opt_offgrpjquota, 0, MOPT_Q},
1423         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1424         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1425         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1426         {Opt_err, 0, 0}
1427 };
1428
1429 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1430                             substring_t *args, unsigned long *journal_devnum,
1431                             unsigned int *journal_ioprio, int is_remount)
1432 {
1433         struct ext4_sb_info *sbi = EXT4_SB(sb);
1434         const struct mount_opts *m;
1435         int arg = 0;
1436
1437         if (args->from && match_int(args, &arg))
1438                 return -1;
1439         switch (token) {
1440         case Opt_sb:
1441                 return 1;       /* handled by get_sb_block() */
1442         case Opt_removed:
1443                 ext4_msg(sb, KERN_WARNING,
1444                          "Ignoring removed %s option", opt);
1445                 return 1;
1446         case Opt_resuid:
1447                 sbi->s_resuid = arg;
1448                 return 1;
1449         case Opt_resgid:
1450                 sbi->s_resgid = arg;
1451                 return 1;
1452         case Opt_abort:
1453                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1454                 return 1;
1455         case Opt_i_version:
1456                 sb->s_flags |= MS_I_VERSION;
1457                 return 1;
1458         case Opt_journal_dev:
1459                 if (is_remount) {
1460                         ext4_msg(sb, KERN_ERR,
1461                                  "Cannot specify journal on remount");
1462                         return -1;
1463                 }
1464                 *journal_devnum = arg;
1465                 return 1;
1466         case Opt_journal_ioprio:
1467                 if (arg < 0 || arg > 7)
1468                         return -1;
1469                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1470                 return 1;
1471         }
1472
1473         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1474                 if (token != m->token)
1475                         continue;
1476                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1477                         return -1;
1478                 if (m->flags & MOPT_EXPLICIT)
1479                         set_opt2(sb, EXPLICIT_DELALLOC);
1480                 if (m->flags & MOPT_CLEAR_ERR)
1481                         clear_opt(sb, ERRORS_MASK);
1482                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1483                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1484                                  "options when quota turned on");
1485                         return -1;
1486                 }
1487
1488                 if (m->flags & MOPT_NOSUPPORT) {
1489                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1490                 } else if (token == Opt_commit) {
1491                         if (arg == 0)
1492                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1493                         sbi->s_commit_interval = HZ * arg;
1494                 } else if (token == Opt_max_batch_time) {
1495                         if (arg == 0)
1496                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1497                         sbi->s_max_batch_time = arg;
1498                 } else if (token == Opt_min_batch_time) {
1499                         sbi->s_min_batch_time = arg;
1500                 } else if (token == Opt_inode_readahead_blks) {
1501                         if (arg > (1 << 30))
1502                                 return -1;
1503                         if (arg && !is_power_of_2(arg)) {
1504                                 ext4_msg(sb, KERN_ERR,
1505                                          "EXT4-fs: inode_readahead_blks"
1506                                          " must be a power of 2");
1507                                 return -1;
1508                         }
1509                         sbi->s_inode_readahead_blks = arg;
1510                 } else if (token == Opt_init_itable) {
1511                         set_opt(sb, INIT_INODE_TABLE);
1512                         if (!args->from)
1513                                 arg = EXT4_DEF_LI_WAIT_MULT;
1514                         sbi->s_li_wait_mult = arg;
1515                 } else if (token == Opt_stripe) {
1516                         sbi->s_stripe = arg;
1517                 } else if (m->flags & MOPT_DATAJ) {
1518                         if (is_remount) {
1519                                 if (!sbi->s_journal)
1520                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1521                                 else if (test_opt(sb, DATA_FLAGS) !=
1522                                          m->mount_opt) {
1523                                         ext4_msg(sb, KERN_ERR,
1524                                          "Cannot change data mode on remount");
1525                                         return -1;
1526                                 }
1527                         } else {
1528                                 clear_opt(sb, DATA_FLAGS);
1529                                 sbi->s_mount_opt |= m->mount_opt;
1530                         }
1531 #ifdef CONFIG_QUOTA
1532                 } else if (token == Opt_usrjquota) {
1533                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1534                                 return -1;
1535                 } else if (token == Opt_grpjquota) {
1536                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1537                                 return -1;
1538                 } else if (token == Opt_offusrjquota) {
1539                         if (!clear_qf_name(sb, USRQUOTA))
1540                                 return -1;
1541                 } else if (token == Opt_offgrpjquota) {
1542                         if (!clear_qf_name(sb, GRPQUOTA))
1543                                 return -1;
1544                 } else if (m->flags & MOPT_QFMT) {
1545                         if (sb_any_quota_loaded(sb) &&
1546                             sbi->s_jquota_fmt != m->mount_opt) {
1547                                 ext4_msg(sb, KERN_ERR, "Cannot "
1548                                          "change journaled quota options "
1549                                          "when quota turned on");
1550                                 return -1;
1551                         }
1552                         sbi->s_jquota_fmt = m->mount_opt;
1553 #endif
1554                 } else {
1555                         if (!args->from)
1556                                 arg = 1;
1557                         if (m->flags & MOPT_CLEAR)
1558                                 arg = !arg;
1559                         else if (unlikely(!(m->flags & MOPT_SET))) {
1560                                 ext4_msg(sb, KERN_WARNING,
1561                                          "buggy handling of option %s", opt);
1562                                 WARN_ON(1);
1563                                 return -1;
1564                         }
1565                         if (arg != 0)
1566                                 sbi->s_mount_opt |= m->mount_opt;
1567                         else
1568                                 sbi->s_mount_opt &= ~m->mount_opt;
1569                 }
1570                 return 1;
1571         }
1572         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1573                  "or missing value", opt);
1574         return -1;
1575 }
1576
1577 static int parse_options(char *options, struct super_block *sb,
1578                          unsigned long *journal_devnum,
1579                          unsigned int *journal_ioprio,
1580                          int is_remount)
1581 {
1582         struct ext4_sb_info *sbi = EXT4_SB(sb);
1583         char *p;
1584         substring_t args[MAX_OPT_ARGS];
1585         int token;
1586
1587         if (!options)
1588                 return 1;
1589
1590         while ((p = strsep(&options, ",")) != NULL) {
1591                 if (!*p)
1592                         continue;
1593                 /*
1594                  * Initialize args struct so we know whether arg was
1595                  * found; some options take optional arguments.
1596                  */
1597                 args[0].to = args[0].from = 0;
1598                 token = match_token(p, tokens, args);
1599                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1600                                      journal_ioprio, is_remount) < 0)
1601                         return 0;
1602         }
1603 #ifdef CONFIG_QUOTA
1604         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1605                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1606                         clear_opt(sb, USRQUOTA);
1607
1608                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1609                         clear_opt(sb, GRPQUOTA);
1610
1611                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1612                         ext4_msg(sb, KERN_ERR, "old and new quota "
1613                                         "format mixing");
1614                         return 0;
1615                 }
1616
1617                 if (!sbi->s_jquota_fmt) {
1618                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1619                                         "not specified");
1620                         return 0;
1621                 }
1622         } else {
1623                 if (sbi->s_jquota_fmt) {
1624                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1625                                         "specified with no journaling "
1626                                         "enabled");
1627                         return 0;
1628                 }
1629         }
1630 #endif
1631         return 1;
1632 }
1633
1634 static inline void ext4_show_quota_options(struct seq_file *seq,
1635                                            struct super_block *sb)
1636 {
1637 #if defined(CONFIG_QUOTA)
1638         struct ext4_sb_info *sbi = EXT4_SB(sb);
1639
1640         if (sbi->s_jquota_fmt) {
1641                 char *fmtname = "";
1642
1643                 switch (sbi->s_jquota_fmt) {
1644                 case QFMT_VFS_OLD:
1645                         fmtname = "vfsold";
1646                         break;
1647                 case QFMT_VFS_V0:
1648                         fmtname = "vfsv0";
1649                         break;
1650                 case QFMT_VFS_V1:
1651                         fmtname = "vfsv1";
1652                         break;
1653                 }
1654                 seq_printf(seq, ",jqfmt=%s", fmtname);
1655         }
1656
1657         if (sbi->s_qf_names[USRQUOTA])
1658                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1659
1660         if (sbi->s_qf_names[GRPQUOTA])
1661                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1662
1663         if (test_opt(sb, USRQUOTA))
1664                 seq_puts(seq, ",usrquota");
1665
1666         if (test_opt(sb, GRPQUOTA))
1667                 seq_puts(seq, ",grpquota");
1668 #endif
1669 }
1670
1671 /*
1672  * Show an option if
1673  *  - it's set to a non-default value OR
1674  *  - if the per-sb default is different from the global default
1675  */
1676 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1677 {
1678         int def_errors;
1679         unsigned long def_mount_opts;
1680         struct super_block *sb = root->d_sb;
1681         struct ext4_sb_info *sbi = EXT4_SB(sb);
1682         struct ext4_super_block *es = sbi->s_es;
1683
1684         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1685         def_errors     = le16_to_cpu(es->s_errors);
1686
1687         if (sbi->s_sb_block != 1)
1688                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1689         if (test_opt(sb, MINIX_DF))
1690                 seq_puts(seq, ",minixdf");
1691         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1692                 seq_puts(seq, ",grpid");
1693         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1694                 seq_puts(seq, ",nogrpid");
1695         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1696             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1697                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1698         }
1699         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1700             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1701                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1702         }
1703         if (test_opt(sb, ERRORS_RO)) {
1704                 if (def_errors == EXT4_ERRORS_PANIC ||
1705                     def_errors == EXT4_ERRORS_CONTINUE) {
1706                         seq_puts(seq, ",errors=remount-ro");
1707                 }
1708         }
1709         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1710                 seq_puts(seq, ",errors=continue");
1711         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1712                 seq_puts(seq, ",errors=panic");
1713         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1714                 seq_puts(seq, ",nouid32");
1715         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1716                 seq_puts(seq, ",debug");
1717 #ifdef CONFIG_EXT4_FS_XATTR
1718         if (test_opt(sb, XATTR_USER))
1719                 seq_puts(seq, ",user_xattr");
1720         if (!test_opt(sb, XATTR_USER))
1721                 seq_puts(seq, ",nouser_xattr");
1722 #endif
1723 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1724         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1725                 seq_puts(seq, ",acl");
1726         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1727                 seq_puts(seq, ",noacl");
1728 #endif
1729         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1730                 seq_printf(seq, ",commit=%u",
1731                            (unsigned) (sbi->s_commit_interval / HZ));
1732         }
1733         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1734                 seq_printf(seq, ",min_batch_time=%u",
1735                            (unsigned) sbi->s_min_batch_time);
1736         }
1737         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1738                 seq_printf(seq, ",max_batch_time=%u",
1739                            (unsigned) sbi->s_max_batch_time);
1740         }
1741
1742         /*
1743          * We're changing the default of barrier mount option, so
1744          * let's always display its mount state so it's clear what its
1745          * status is.
1746          */
1747         seq_puts(seq, ",barrier=");
1748         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1749         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1750                 seq_puts(seq, ",journal_async_commit");
1751         else if (test_opt(sb, JOURNAL_CHECKSUM))
1752                 seq_puts(seq, ",journal_checksum");
1753         if (sb->s_flags & MS_I_VERSION)
1754                 seq_puts(seq, ",i_version");
1755         if (!test_opt(sb, DELALLOC) &&
1756             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1757                 seq_puts(seq, ",nodelalloc");
1758
1759         if (!test_opt(sb, MBLK_IO_SUBMIT))
1760                 seq_puts(seq, ",nomblk_io_submit");
1761         if (sbi->s_stripe)
1762                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1763         /*
1764          * journal mode get enabled in different ways
1765          * So just print the value even if we didn't specify it
1766          */
1767         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1768                 seq_puts(seq, ",data=journal");
1769         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1770                 seq_puts(seq, ",data=ordered");
1771         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1772                 seq_puts(seq, ",data=writeback");
1773
1774         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1775                 seq_printf(seq, ",inode_readahead_blks=%u",
1776                            sbi->s_inode_readahead_blks);
1777
1778         if (test_opt(sb, DATA_ERR_ABORT))
1779                 seq_puts(seq, ",data_err=abort");
1780
1781         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1782                 seq_puts(seq, ",noauto_da_alloc");
1783
1784         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1785                 seq_puts(seq, ",discard");
1786
1787         if (test_opt(sb, NOLOAD))
1788                 seq_puts(seq, ",norecovery");
1789
1790         if (test_opt(sb, DIOREAD_NOLOCK))
1791                 seq_puts(seq, ",dioread_nolock");
1792
1793         if (test_opt(sb, BLOCK_VALIDITY) &&
1794             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1795                 seq_puts(seq, ",block_validity");
1796
1797         if (!test_opt(sb, INIT_INODE_TABLE))
1798                 seq_puts(seq, ",noinit_itable");
1799         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1800                 seq_printf(seq, ",init_itable=%u",
1801                            (unsigned) sbi->s_li_wait_mult);
1802
1803         ext4_show_quota_options(seq, sb);
1804
1805         return 0;
1806 }
1807
1808 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1809                             int read_only)
1810 {
1811         struct ext4_sb_info *sbi = EXT4_SB(sb);
1812         int res = 0;
1813
1814         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1815                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1816                          "forcing read-only mode");
1817                 res = MS_RDONLY;
1818         }
1819         if (read_only)
1820                 goto done;
1821         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1822                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1823                          "running e2fsck is recommended");
1824         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1825                 ext4_msg(sb, KERN_WARNING,
1826                          "warning: mounting fs with errors, "
1827                          "running e2fsck is recommended");
1828         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1829                  le16_to_cpu(es->s_mnt_count) >=
1830                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1831                 ext4_msg(sb, KERN_WARNING,
1832                          "warning: maximal mount count reached, "
1833                          "running e2fsck is recommended");
1834         else if (le32_to_cpu(es->s_checkinterval) &&
1835                 (le32_to_cpu(es->s_lastcheck) +
1836                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1837                 ext4_msg(sb, KERN_WARNING,
1838                          "warning: checktime reached, "
1839                          "running e2fsck is recommended");
1840         if (!sbi->s_journal)
1841                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1842         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1843                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1844         le16_add_cpu(&es->s_mnt_count, 1);
1845         es->s_mtime = cpu_to_le32(get_seconds());
1846         ext4_update_dynamic_rev(sb);
1847         if (sbi->s_journal)
1848                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1849
1850         ext4_commit_super(sb, 1);
1851 done:
1852         if (test_opt(sb, DEBUG))
1853                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1854                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1855                         sb->s_blocksize,
1856                         sbi->s_groups_count,
1857                         EXT4_BLOCKS_PER_GROUP(sb),
1858                         EXT4_INODES_PER_GROUP(sb),
1859                         sbi->s_mount_opt, sbi->s_mount_opt2);
1860
1861         cleancache_init_fs(sb);
1862         return res;
1863 }
1864
1865 static int ext4_fill_flex_info(struct super_block *sb)
1866 {
1867         struct ext4_sb_info *sbi = EXT4_SB(sb);
1868         struct ext4_group_desc *gdp = NULL;
1869         ext4_group_t flex_group_count;
1870         ext4_group_t flex_group;
1871         unsigned int groups_per_flex = 0;
1872         size_t size;
1873         int i;
1874
1875         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1876         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1877                 sbi->s_log_groups_per_flex = 0;
1878                 return 1;
1879         }
1880         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1881
1882         /* We allocate both existing and potentially added groups */
1883         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1884                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1885                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1886         size = flex_group_count * sizeof(struct flex_groups);
1887         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1888         if (sbi->s_flex_groups == NULL) {
1889                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1890                          flex_group_count);
1891                 goto failed;
1892         }
1893
1894         for (i = 0; i < sbi->s_groups_count; i++) {
1895                 gdp = ext4_get_group_desc(sb, i, NULL);
1896
1897                 flex_group = ext4_flex_group(sbi, i);
1898                 atomic_add(ext4_free_inodes_count(sb, gdp),
1899                            &sbi->s_flex_groups[flex_group].free_inodes);
1900                 atomic_add(ext4_free_group_clusters(sb, gdp),
1901                            &sbi->s_flex_groups[flex_group].free_clusters);
1902                 atomic_add(ext4_used_dirs_count(sb, gdp),
1903                            &sbi->s_flex_groups[flex_group].used_dirs);
1904         }
1905
1906         return 1;
1907 failed:
1908         return 0;
1909 }
1910
1911 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1912                             struct ext4_group_desc *gdp)
1913 {
1914         __u16 crc = 0;
1915
1916         if (sbi->s_es->s_feature_ro_compat &
1917             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1918                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1919                 __le32 le_group = cpu_to_le32(block_group);
1920
1921                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1922                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1923                 crc = crc16(crc, (__u8 *)gdp, offset);
1924                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1925                 /* for checksum of struct ext4_group_desc do the rest...*/
1926                 if ((sbi->s_es->s_feature_incompat &
1927                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1928                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1929                         crc = crc16(crc, (__u8 *)gdp + offset,
1930                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1931                                         offset);
1932         }
1933
1934         return cpu_to_le16(crc);
1935 }
1936
1937 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1938                                 struct ext4_group_desc *gdp)
1939 {
1940         if ((sbi->s_es->s_feature_ro_compat &
1941              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1942             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1943                 return 0;
1944
1945         return 1;
1946 }
1947
1948 /* Called at mount-time, super-block is locked */
1949 static int ext4_check_descriptors(struct super_block *sb,
1950                                   ext4_group_t *first_not_zeroed)
1951 {
1952         struct ext4_sb_info *sbi = EXT4_SB(sb);
1953         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1954         ext4_fsblk_t last_block;
1955         ext4_fsblk_t block_bitmap;
1956         ext4_fsblk_t inode_bitmap;
1957         ext4_fsblk_t inode_table;
1958         int flexbg_flag = 0;
1959         ext4_group_t i, grp = sbi->s_groups_count;
1960
1961         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1962                 flexbg_flag = 1;
1963
1964         ext4_debug("Checking group descriptors");
1965
1966         for (i = 0; i < sbi->s_groups_count; i++) {
1967                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1968
1969                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1970                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1971                 else
1972                         last_block = first_block +
1973                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1974
1975                 if ((grp == sbi->s_groups_count) &&
1976                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
1977                         grp = i;
1978
1979                 block_bitmap = ext4_block_bitmap(sb, gdp);
1980                 if (block_bitmap < first_block || block_bitmap > last_block) {
1981                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1982                                "Block bitmap for group %u not in group "
1983                                "(block %llu)!", i, block_bitmap);
1984                         return 0;
1985                 }
1986                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1987                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1988                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1989                                "Inode bitmap for group %u not in group "
1990                                "(block %llu)!", i, inode_bitmap);
1991                         return 0;
1992                 }
1993                 inode_table = ext4_inode_table(sb, gdp);
1994                 if (inode_table < first_block ||
1995                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1996                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1997                                "Inode table for group %u not in group "
1998                                "(block %llu)!", i, inode_table);
1999                         return 0;
2000                 }
2001                 ext4_lock_group(sb, i);
2002                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2003                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2004                                  "Checksum for group %u failed (%u!=%u)",
2005                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2006                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2007                         if (!(sb->s_flags & MS_RDONLY)) {
2008                                 ext4_unlock_group(sb, i);
2009                                 return 0;
2010                         }
2011                 }
2012                 ext4_unlock_group(sb, i);
2013                 if (!flexbg_flag)
2014                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2015         }
2016         if (NULL != first_not_zeroed)
2017                 *first_not_zeroed = grp;
2018
2019         ext4_free_blocks_count_set(sbi->s_es,
2020                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2021         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2022         return 1;
2023 }
2024
2025 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2026  * the superblock) which were deleted from all directories, but held open by
2027  * a process at the time of a crash.  We walk the list and try to delete these
2028  * inodes at recovery time (only with a read-write filesystem).
2029  *
2030  * In order to keep the orphan inode chain consistent during traversal (in
2031  * case of crash during recovery), we link each inode into the superblock
2032  * orphan list_head and handle it the same way as an inode deletion during
2033  * normal operation (which journals the operations for us).
2034  *
2035  * We only do an iget() and an iput() on each inode, which is very safe if we
2036  * accidentally point at an in-use or already deleted inode.  The worst that
2037  * can happen in this case is that we get a "bit already cleared" message from
2038  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2039  * e2fsck was run on this filesystem, and it must have already done the orphan
2040  * inode cleanup for us, so we can safely abort without any further action.
2041  */
2042 static void ext4_orphan_cleanup(struct super_block *sb,
2043                                 struct ext4_super_block *es)
2044 {
2045         unsigned int s_flags = sb->s_flags;
2046         int nr_orphans = 0, nr_truncates = 0;
2047 #ifdef CONFIG_QUOTA
2048         int i;
2049 #endif
2050         if (!es->s_last_orphan) {
2051                 jbd_debug(4, "no orphan inodes to clean up\n");
2052                 return;
2053         }
2054
2055         if (bdev_read_only(sb->s_bdev)) {
2056                 ext4_msg(sb, KERN_ERR, "write access "
2057                         "unavailable, skipping orphan cleanup");
2058                 return;
2059         }
2060
2061         /* Check if feature set would not allow a r/w mount */
2062         if (!ext4_feature_set_ok(sb, 0)) {
2063                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2064                          "unknown ROCOMPAT features");
2065                 return;
2066         }
2067
2068         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2069                 if (es->s_last_orphan)
2070                         jbd_debug(1, "Errors on filesystem, "
2071                                   "clearing orphan list.\n");
2072                 es->s_last_orphan = 0;
2073                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2074                 return;
2075         }
2076
2077         if (s_flags & MS_RDONLY) {
2078                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2079                 sb->s_flags &= ~MS_RDONLY;
2080         }
2081 #ifdef CONFIG_QUOTA
2082         /* Needed for iput() to work correctly and not trash data */
2083         sb->s_flags |= MS_ACTIVE;
2084         /* Turn on quotas so that they are updated correctly */
2085         for (i = 0; i < MAXQUOTAS; i++) {
2086                 if (EXT4_SB(sb)->s_qf_names[i]) {
2087                         int ret = ext4_quota_on_mount(sb, i);
2088                         if (ret < 0)
2089                                 ext4_msg(sb, KERN_ERR,
2090                                         "Cannot turn on journaled "
2091                                         "quota: error %d", ret);
2092                 }
2093         }
2094 #endif
2095
2096         while (es->s_last_orphan) {
2097                 struct inode *inode;
2098
2099                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2100                 if (IS_ERR(inode)) {
2101                         es->s_last_orphan = 0;
2102                         break;
2103                 }
2104
2105                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2106                 dquot_initialize(inode);
2107                 if (inode->i_nlink) {
2108                         ext4_msg(sb, KERN_DEBUG,
2109                                 "%s: truncating inode %lu to %lld bytes",
2110                                 __func__, inode->i_ino, inode->i_size);
2111                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2112                                   inode->i_ino, inode->i_size);
2113                         ext4_truncate(inode);
2114                         nr_truncates++;
2115                 } else {
2116                         ext4_msg(sb, KERN_DEBUG,
2117                                 "%s: deleting unreferenced inode %lu",
2118                                 __func__, inode->i_ino);
2119                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2120                                   inode->i_ino);
2121                         nr_orphans++;
2122                 }
2123                 iput(inode);  /* The delete magic happens here! */
2124         }
2125
2126 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2127
2128         if (nr_orphans)
2129                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2130                        PLURAL(nr_orphans));
2131         if (nr_truncates)
2132                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2133                        PLURAL(nr_truncates));
2134 #ifdef CONFIG_QUOTA
2135         /* Turn quotas off */
2136         for (i = 0; i < MAXQUOTAS; i++) {
2137                 if (sb_dqopt(sb)->files[i])
2138                         dquot_quota_off(sb, i);
2139         }
2140 #endif
2141         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2142 }
2143
2144 /*
2145  * Maximal extent format file size.
2146  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2147  * extent format containers, within a sector_t, and within i_blocks
2148  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2149  * so that won't be a limiting factor.
2150  *
2151  * However there is other limiting factor. We do store extents in the form
2152  * of starting block and length, hence the resulting length of the extent
2153  * covering maximum file size must fit into on-disk format containers as
2154  * well. Given that length is always by 1 unit bigger than max unit (because
2155  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2156  *
2157  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2158  */
2159 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2160 {
2161         loff_t res;
2162         loff_t upper_limit = MAX_LFS_FILESIZE;
2163
2164         /* small i_blocks in vfs inode? */
2165         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2166                 /*
2167                  * CONFIG_LBDAF is not enabled implies the inode
2168                  * i_block represent total blocks in 512 bytes
2169                  * 32 == size of vfs inode i_blocks * 8
2170                  */
2171                 upper_limit = (1LL << 32) - 1;
2172
2173                 /* total blocks in file system block size */
2174                 upper_limit >>= (blkbits - 9);
2175                 upper_limit <<= blkbits;
2176         }
2177
2178         /*
2179          * 32-bit extent-start container, ee_block. We lower the maxbytes
2180          * by one fs block, so ee_len can cover the extent of maximum file
2181          * size
2182          */
2183         res = (1LL << 32) - 1;
2184         res <<= blkbits;
2185
2186         /* Sanity check against vm- & vfs- imposed limits */
2187         if (res > upper_limit)
2188                 res = upper_limit;
2189
2190         return res;
2191 }
2192
2193 /*
2194  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2195  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2196  * We need to be 1 filesystem block less than the 2^48 sector limit.
2197  */
2198 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2199 {
2200         loff_t res = EXT4_NDIR_BLOCKS;
2201         int meta_blocks;
2202         loff_t upper_limit;
2203         /* This is calculated to be the largest file size for a dense, block
2204          * mapped file such that the file's total number of 512-byte sectors,
2205          * including data and all indirect blocks, does not exceed (2^48 - 1).
2206          *
2207          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2208          * number of 512-byte sectors of the file.
2209          */
2210
2211         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2212                 /*
2213                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2214                  * the inode i_block field represents total file blocks in
2215                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2216                  */
2217                 upper_limit = (1LL << 32) - 1;
2218
2219                 /* total blocks in file system block size */
2220                 upper_limit >>= (bits - 9);
2221
2222         } else {
2223                 /*
2224                  * We use 48 bit ext4_inode i_blocks
2225                  * With EXT4_HUGE_FILE_FL set the i_blocks
2226                  * represent total number of blocks in
2227                  * file system block size
2228                  */
2229                 upper_limit = (1LL << 48) - 1;
2230
2231         }
2232
2233         /* indirect blocks */
2234         meta_blocks = 1;
2235         /* double indirect blocks */
2236         meta_blocks += 1 + (1LL << (bits-2));
2237         /* tripple indirect blocks */
2238         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2239
2240         upper_limit -= meta_blocks;
2241         upper_limit <<= bits;
2242
2243         res += 1LL << (bits-2);
2244         res += 1LL << (2*(bits-2));
2245         res += 1LL << (3*(bits-2));
2246         res <<= bits;
2247         if (res > upper_limit)
2248                 res = upper_limit;
2249
2250         if (res > MAX_LFS_FILESIZE)
2251                 res = MAX_LFS_FILESIZE;
2252
2253         return res;
2254 }
2255
2256 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2257                                    ext4_fsblk_t logical_sb_block, int nr)
2258 {
2259         struct ext4_sb_info *sbi = EXT4_SB(sb);
2260         ext4_group_t bg, first_meta_bg;
2261         int has_super = 0;
2262
2263         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2264
2265         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2266             nr < first_meta_bg)
2267                 return logical_sb_block + nr + 1;
2268         bg = sbi->s_desc_per_block * nr;
2269         if (ext4_bg_has_super(sb, bg))
2270                 has_super = 1;
2271
2272         return (has_super + ext4_group_first_block_no(sb, bg));
2273 }
2274
2275 /**
2276  * ext4_get_stripe_size: Get the stripe size.
2277  * @sbi: In memory super block info
2278  *
2279  * If we have specified it via mount option, then
2280  * use the mount option value. If the value specified at mount time is
2281  * greater than the blocks per group use the super block value.
2282  * If the super block value is greater than blocks per group return 0.
2283  * Allocator needs it be less than blocks per group.
2284  *
2285  */
2286 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2287 {
2288         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2289         unsigned long stripe_width =
2290                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2291         int ret;
2292
2293         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2294                 ret = sbi->s_stripe;
2295         else if (stripe_width <= sbi->s_blocks_per_group)
2296                 ret = stripe_width;
2297         else if (stride <= sbi->s_blocks_per_group)
2298                 ret = stride;
2299         else
2300                 ret = 0;
2301
2302         /*
2303          * If the stripe width is 1, this makes no sense and
2304          * we set it to 0 to turn off stripe handling code.
2305          */
2306         if (ret <= 1)
2307                 ret = 0;
2308
2309         return ret;
2310 }
2311
2312 /* sysfs supprt */
2313
2314 struct ext4_attr {
2315         struct attribute attr;
2316         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2317         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2318                          const char *, size_t);
2319         int offset;
2320 };
2321
2322 static int parse_strtoul(const char *buf,
2323                 unsigned long max, unsigned long *value)
2324 {
2325         char *endp;
2326
2327         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2328         endp = skip_spaces(endp);
2329         if (*endp || *value > max)
2330                 return -EINVAL;
2331
2332         return 0;
2333 }
2334
2335 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2336                                               struct ext4_sb_info *sbi,
2337                                               char *buf)
2338 {
2339         return snprintf(buf, PAGE_SIZE, "%llu\n",
2340                 (s64) EXT4_C2B(sbi,
2341                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2342 }
2343
2344 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2345                                          struct ext4_sb_info *sbi, char *buf)
2346 {
2347         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2348
2349         if (!sb->s_bdev->bd_part)
2350                 return snprintf(buf, PAGE_SIZE, "0\n");
2351         return snprintf(buf, PAGE_SIZE, "%lu\n",
2352                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2353                          sbi->s_sectors_written_start) >> 1);
2354 }
2355
2356 static ssize_t lifetime_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, "%llu\n",
2364                         (unsigned long long)(sbi->s_kbytes_written +
2365                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2366                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2367 }
2368
2369 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2370                                       struct ext4_sb_info *sbi, char *buf)
2371 {
2372         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2373 }
2374
2375 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2376                                         struct ext4_sb_info *sbi, char *buf)
2377 {
2378         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
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_RO_ATTR(extent_cache_hits);
2439 EXT4_RO_ATTR(extent_cache_misses);
2440 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2441                  inode_readahead_blks_store, s_inode_readahead_blks);
2442 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2443 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2444 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2445 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2446 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2447 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2448 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2449 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2450
2451 static struct attribute *ext4_attrs[] = {
2452         ATTR_LIST(delayed_allocation_blocks),
2453         ATTR_LIST(session_write_kbytes),
2454         ATTR_LIST(lifetime_write_kbytes),
2455         ATTR_LIST(extent_cache_hits),
2456         ATTR_LIST(extent_cache_misses),
2457         ATTR_LIST(inode_readahead_blks),
2458         ATTR_LIST(inode_goal),
2459         ATTR_LIST(mb_stats),
2460         ATTR_LIST(mb_max_to_scan),
2461         ATTR_LIST(mb_min_to_scan),
2462         ATTR_LIST(mb_order2_req),
2463         ATTR_LIST(mb_stream_req),
2464         ATTR_LIST(mb_group_prealloc),
2465         ATTR_LIST(max_writeback_mb_bump),
2466         NULL,
2467 };
2468
2469 /* Features this copy of ext4 supports */
2470 EXT4_INFO_ATTR(lazy_itable_init);
2471 EXT4_INFO_ATTR(batched_discard);
2472
2473 static struct attribute *ext4_feat_attrs[] = {
2474         ATTR_LIST(lazy_itable_init),
2475         ATTR_LIST(batched_discard),
2476         NULL,
2477 };
2478
2479 static ssize_t ext4_attr_show(struct kobject *kobj,
2480                               struct attribute *attr, char *buf)
2481 {
2482         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2483                                                 s_kobj);
2484         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2485
2486         return a->show ? a->show(a, sbi, buf) : 0;
2487 }
2488
2489 static ssize_t ext4_attr_store(struct kobject *kobj,
2490                                struct attribute *attr,
2491                                const char *buf, size_t len)
2492 {
2493         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2494                                                 s_kobj);
2495         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2496
2497         return a->store ? a->store(a, sbi, buf, len) : 0;
2498 }
2499
2500 static void ext4_sb_release(struct kobject *kobj)
2501 {
2502         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2503                                                 s_kobj);
2504         complete(&sbi->s_kobj_unregister);
2505 }
2506
2507 static const struct sysfs_ops ext4_attr_ops = {
2508         .show   = ext4_attr_show,
2509         .store  = ext4_attr_store,
2510 };
2511
2512 static struct kobj_type ext4_ktype = {
2513         .default_attrs  = ext4_attrs,
2514         .sysfs_ops      = &ext4_attr_ops,
2515         .release        = ext4_sb_release,
2516 };
2517
2518 static void ext4_feat_release(struct kobject *kobj)
2519 {
2520         complete(&ext4_feat->f_kobj_unregister);
2521 }
2522
2523 static struct kobj_type ext4_feat_ktype = {
2524         .default_attrs  = ext4_feat_attrs,
2525         .sysfs_ops      = &ext4_attr_ops,
2526         .release        = ext4_feat_release,
2527 };
2528
2529 /*
2530  * Check whether this filesystem can be mounted based on
2531  * the features present and the RDONLY/RDWR mount requested.
2532  * Returns 1 if this filesystem can be mounted as requested,
2533  * 0 if it cannot be.
2534  */
2535 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2536 {
2537         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2538                 ext4_msg(sb, KERN_ERR,
2539                         "Couldn't mount because of "
2540                         "unsupported optional features (%x)",
2541                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2542                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2543                 return 0;
2544         }
2545
2546         if (readonly)
2547                 return 1;
2548
2549         /* Check that feature set is OK for a read-write mount */
2550         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2551                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2552                          "unsupported optional features (%x)",
2553                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2554                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2555                 return 0;
2556         }
2557         /*
2558          * Large file size enabled file system can only be mounted
2559          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2560          */
2561         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2562                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2563                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2564                                  "cannot be mounted RDWR without "
2565                                  "CONFIG_LBDAF");
2566                         return 0;
2567                 }
2568         }
2569         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2570             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2571                 ext4_msg(sb, KERN_ERR,
2572                          "Can't support bigalloc feature without "
2573                          "extents feature\n");
2574                 return 0;
2575         }
2576         return 1;
2577 }
2578
2579 /*
2580  * This function is called once a day if we have errors logged
2581  * on the file system
2582  */
2583 static void print_daily_error_info(unsigned long arg)
2584 {
2585         struct super_block *sb = (struct super_block *) arg;
2586         struct ext4_sb_info *sbi;
2587         struct ext4_super_block *es;
2588
2589         sbi = EXT4_SB(sb);
2590         es = sbi->s_es;
2591
2592         if (es->s_error_count)
2593                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2594                          le32_to_cpu(es->s_error_count));
2595         if (es->s_first_error_time) {
2596                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2597                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2598                        (int) sizeof(es->s_first_error_func),
2599                        es->s_first_error_func,
2600                        le32_to_cpu(es->s_first_error_line));
2601                 if (es->s_first_error_ino)
2602                         printk(": inode %u",
2603                                le32_to_cpu(es->s_first_error_ino));
2604                 if (es->s_first_error_block)
2605                         printk(": block %llu", (unsigned long long)
2606                                le64_to_cpu(es->s_first_error_block));
2607                 printk("\n");
2608         }
2609         if (es->s_last_error_time) {
2610                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2611                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2612                        (int) sizeof(es->s_last_error_func),
2613                        es->s_last_error_func,
2614                        le32_to_cpu(es->s_last_error_line));
2615                 if (es->s_last_error_ino)
2616                         printk(": inode %u",
2617                                le32_to_cpu(es->s_last_error_ino));
2618                 if (es->s_last_error_block)
2619                         printk(": block %llu", (unsigned long long)
2620                                le64_to_cpu(es->s_last_error_block));
2621                 printk("\n");
2622         }
2623         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2624 }
2625
2626 /* Find next suitable group and run ext4_init_inode_table */
2627 static int ext4_run_li_request(struct ext4_li_request *elr)
2628 {
2629         struct ext4_group_desc *gdp = NULL;
2630         ext4_group_t group, ngroups;
2631         struct super_block *sb;
2632         unsigned long timeout = 0;
2633         int ret = 0;
2634
2635         sb = elr->lr_super;
2636         ngroups = EXT4_SB(sb)->s_groups_count;
2637
2638         for (group = elr->lr_next_group; group < ngroups; group++) {
2639                 gdp = ext4_get_group_desc(sb, group, NULL);
2640                 if (!gdp) {
2641                         ret = 1;
2642                         break;
2643                 }
2644
2645                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2646                         break;
2647         }
2648
2649         if (group == ngroups)
2650                 ret = 1;
2651
2652         if (!ret) {
2653                 timeout = jiffies;
2654                 ret = ext4_init_inode_table(sb, group,
2655                                             elr->lr_timeout ? 0 : 1);
2656                 if (elr->lr_timeout == 0) {
2657                         timeout = (jiffies - timeout) *
2658                                   elr->lr_sbi->s_li_wait_mult;
2659                         elr->lr_timeout = timeout;
2660                 }
2661                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2662                 elr->lr_next_group = group + 1;
2663         }
2664
2665         return ret;
2666 }
2667
2668 /*
2669  * Remove lr_request from the list_request and free the
2670  * request structure. Should be called with li_list_mtx held
2671  */
2672 static void ext4_remove_li_request(struct ext4_li_request *elr)
2673 {
2674         struct ext4_sb_info *sbi;
2675
2676         if (!elr)
2677                 return;
2678
2679         sbi = elr->lr_sbi;
2680
2681         list_del(&elr->lr_request);
2682         sbi->s_li_request = NULL;
2683         kfree(elr);
2684 }
2685
2686 static void ext4_unregister_li_request(struct super_block *sb)
2687 {
2688         mutex_lock(&ext4_li_mtx);
2689         if (!ext4_li_info) {
2690                 mutex_unlock(&ext4_li_mtx);
2691                 return;
2692         }
2693
2694         mutex_lock(&ext4_li_info->li_list_mtx);
2695         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2696         mutex_unlock(&ext4_li_info->li_list_mtx);
2697         mutex_unlock(&ext4_li_mtx);
2698 }
2699
2700 static struct task_struct *ext4_lazyinit_task;
2701
2702 /*
2703  * This is the function where ext4lazyinit thread lives. It walks
2704  * through the request list searching for next scheduled filesystem.
2705  * When such a fs is found, run the lazy initialization request
2706  * (ext4_rn_li_request) and keep track of the time spend in this
2707  * function. Based on that time we compute next schedule time of
2708  * the request. When walking through the list is complete, compute
2709  * next waking time and put itself into sleep.
2710  */
2711 static int ext4_lazyinit_thread(void *arg)
2712 {
2713         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2714         struct list_head *pos, *n;
2715         struct ext4_li_request *elr;
2716         unsigned long next_wakeup, cur;
2717
2718         BUG_ON(NULL == eli);
2719
2720 cont_thread:
2721         while (true) {
2722                 next_wakeup = MAX_JIFFY_OFFSET;
2723
2724                 mutex_lock(&eli->li_list_mtx);
2725                 if (list_empty(&eli->li_request_list)) {
2726                         mutex_unlock(&eli->li_list_mtx);
2727                         goto exit_thread;
2728                 }
2729
2730                 list_for_each_safe(pos, n, &eli->li_request_list) {
2731                         elr = list_entry(pos, struct ext4_li_request,
2732                                          lr_request);
2733
2734                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2735                                 if (ext4_run_li_request(elr) != 0) {
2736                                         /* error, remove the lazy_init job */
2737                                         ext4_remove_li_request(elr);
2738                                         continue;
2739                                 }
2740                         }
2741
2742                         if (time_before(elr->lr_next_sched, next_wakeup))
2743                                 next_wakeup = elr->lr_next_sched;
2744                 }
2745                 mutex_unlock(&eli->li_list_mtx);
2746
2747                 try_to_freeze();
2748
2749                 cur = jiffies;
2750                 if ((time_after_eq(cur, next_wakeup)) ||
2751                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2752                         cond_resched();
2753                         continue;
2754                 }
2755
2756                 schedule_timeout_interruptible(next_wakeup - cur);
2757
2758                 if (kthread_should_stop()) {
2759                         ext4_clear_request_list();
2760                         goto exit_thread;
2761                 }
2762         }
2763
2764 exit_thread:
2765         /*
2766          * It looks like the request list is empty, but we need
2767          * to check it under the li_list_mtx lock, to prevent any
2768          * additions into it, and of course we should lock ext4_li_mtx
2769          * to atomically free the list and ext4_li_info, because at
2770          * this point another ext4 filesystem could be registering
2771          * new one.
2772          */
2773         mutex_lock(&ext4_li_mtx);
2774         mutex_lock(&eli->li_list_mtx);
2775         if (!list_empty(&eli->li_request_list)) {
2776                 mutex_unlock(&eli->li_list_mtx);
2777                 mutex_unlock(&ext4_li_mtx);
2778                 goto cont_thread;
2779         }
2780         mutex_unlock(&eli->li_list_mtx);
2781         kfree(ext4_li_info);
2782         ext4_li_info = NULL;
2783         mutex_unlock(&ext4_li_mtx);
2784
2785         return 0;
2786 }
2787
2788 static void ext4_clear_request_list(void)
2789 {
2790         struct list_head *pos, *n;
2791         struct ext4_li_request *elr;
2792
2793         mutex_lock(&ext4_li_info->li_list_mtx);
2794         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2795                 elr = list_entry(pos, struct ext4_li_request,
2796                                  lr_request);
2797                 ext4_remove_li_request(elr);
2798         }
2799         mutex_unlock(&ext4_li_info->li_list_mtx);
2800 }
2801
2802 static int ext4_run_lazyinit_thread(void)
2803 {
2804         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2805                                          ext4_li_info, "ext4lazyinit");
2806         if (IS_ERR(ext4_lazyinit_task)) {
2807                 int err = PTR_ERR(ext4_lazyinit_task);
2808                 ext4_clear_request_list();
2809                 kfree(ext4_li_info);
2810                 ext4_li_info = NULL;
2811                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2812                                  "initialization thread\n",
2813                                  err);
2814                 return err;
2815         }
2816         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2817         return 0;
2818 }
2819
2820 /*
2821  * Check whether it make sense to run itable init. thread or not.
2822  * If there is at least one uninitialized inode table, return
2823  * corresponding group number, else the loop goes through all
2824  * groups and return total number of groups.
2825  */
2826 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2827 {
2828         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2829         struct ext4_group_desc *gdp = NULL;
2830
2831         for (group = 0; group < ngroups; group++) {
2832                 gdp = ext4_get_group_desc(sb, group, NULL);
2833                 if (!gdp)
2834                         continue;
2835
2836                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2837                         break;
2838         }
2839
2840         return group;
2841 }
2842
2843 static int ext4_li_info_new(void)
2844 {
2845         struct ext4_lazy_init *eli = NULL;
2846
2847         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2848         if (!eli)
2849                 return -ENOMEM;
2850
2851         INIT_LIST_HEAD(&eli->li_request_list);
2852         mutex_init(&eli->li_list_mtx);
2853
2854         eli->li_state |= EXT4_LAZYINIT_QUIT;
2855
2856         ext4_li_info = eli;
2857
2858         return 0;
2859 }
2860
2861 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2862                                             ext4_group_t start)
2863 {
2864         struct ext4_sb_info *sbi = EXT4_SB(sb);
2865         struct ext4_li_request *elr;
2866         unsigned long rnd;
2867
2868         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2869         if (!elr)
2870                 return NULL;
2871
2872         elr->lr_super = sb;
2873         elr->lr_sbi = sbi;
2874         elr->lr_next_group = start;
2875
2876         /*
2877          * Randomize first schedule time of the request to
2878          * spread the inode table initialization requests
2879          * better.
2880          */
2881         get_random_bytes(&rnd, sizeof(rnd));
2882         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2883                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2884
2885         return elr;
2886 }
2887
2888 static int ext4_register_li_request(struct super_block *sb,
2889                                     ext4_group_t first_not_zeroed)
2890 {
2891         struct ext4_sb_info *sbi = EXT4_SB(sb);
2892         struct ext4_li_request *elr;
2893         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2894         int ret = 0;
2895
2896         if (sbi->s_li_request != NULL) {
2897                 /*
2898                  * Reset timeout so it can be computed again, because
2899                  * s_li_wait_mult might have changed.
2900                  */
2901                 sbi->s_li_request->lr_timeout = 0;
2902                 return 0;
2903         }
2904
2905         if (first_not_zeroed == ngroups ||
2906             (sb->s_flags & MS_RDONLY) ||
2907             !test_opt(sb, INIT_INODE_TABLE))
2908                 return 0;
2909
2910         elr = ext4_li_request_new(sb, first_not_zeroed);
2911         if (!elr)
2912                 return -ENOMEM;
2913
2914         mutex_lock(&ext4_li_mtx);
2915
2916         if (NULL == ext4_li_info) {
2917                 ret = ext4_li_info_new();
2918                 if (ret)
2919                         goto out;
2920         }
2921
2922         mutex_lock(&ext4_li_info->li_list_mtx);
2923         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2924         mutex_unlock(&ext4_li_info->li_list_mtx);
2925
2926         sbi->s_li_request = elr;
2927         /*
2928          * set elr to NULL here since it has been inserted to
2929          * the request_list and the removal and free of it is
2930          * handled by ext4_clear_request_list from now on.
2931          */
2932         elr = NULL;
2933
2934         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2935                 ret = ext4_run_lazyinit_thread();
2936                 if (ret)
2937                         goto out;
2938         }
2939 out:
2940         mutex_unlock(&ext4_li_mtx);
2941         if (ret)
2942                 kfree(elr);
2943         return ret;
2944 }
2945
2946 /*
2947  * We do not need to lock anything since this is called on
2948  * module unload.
2949  */
2950 static void ext4_destroy_lazyinit_thread(void)
2951 {
2952         /*
2953          * If thread exited earlier
2954          * there's nothing to be done.
2955          */
2956         if (!ext4_li_info || !ext4_lazyinit_task)
2957                 return;
2958
2959         kthread_stop(ext4_lazyinit_task);
2960 }
2961
2962 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2963 {
2964         char *orig_data = kstrdup(data, GFP_KERNEL);
2965         struct buffer_head *bh;
2966         struct ext4_super_block *es = NULL;
2967         struct ext4_sb_info *sbi;
2968         ext4_fsblk_t block;
2969         ext4_fsblk_t sb_block = get_sb_block(&data);
2970         ext4_fsblk_t logical_sb_block;
2971         unsigned long offset = 0;
2972         unsigned long journal_devnum = 0;
2973         unsigned long def_mount_opts;
2974         struct inode *root;
2975         char *cp;
2976         const char *descr;
2977         int ret = -ENOMEM;
2978         int blocksize, clustersize;
2979         unsigned int db_count;
2980         unsigned int i;
2981         int needs_recovery, has_huge_files, has_bigalloc;
2982         __u64 blocks_count;
2983         int err;
2984         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2985         ext4_group_t first_not_zeroed;
2986
2987         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2988         if (!sbi)
2989                 goto out_free_orig;
2990
2991         sbi->s_blockgroup_lock =
2992                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2993         if (!sbi->s_blockgroup_lock) {
2994                 kfree(sbi);
2995                 goto out_free_orig;
2996         }
2997         sb->s_fs_info = sbi;
2998         sbi->s_mount_opt = 0;
2999         sbi->s_resuid = EXT4_DEF_RESUID;
3000         sbi->s_resgid = EXT4_DEF_RESGID;
3001         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3002         sbi->s_sb_block = sb_block;
3003         if (sb->s_bdev->bd_part)
3004                 sbi->s_sectors_written_start =
3005                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3006
3007         /* Cleanup superblock name */
3008         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3009                 *cp = '!';
3010
3011         ret = -EINVAL;
3012         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3013         if (!blocksize) {
3014                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3015                 goto out_fail;
3016         }
3017
3018         /*
3019          * The ext4 superblock will not be buffer aligned for other than 1kB
3020          * block sizes.  We need to calculate the offset from buffer start.
3021          */
3022         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3023                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3024                 offset = do_div(logical_sb_block, blocksize);
3025         } else {
3026                 logical_sb_block = sb_block;
3027         }
3028
3029         if (!(bh = sb_bread(sb, logical_sb_block))) {
3030                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3031                 goto out_fail;
3032         }
3033         /*
3034          * Note: s_es must be initialized as soon as possible because
3035          *       some ext4 macro-instructions depend on its value
3036          */
3037         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3038         sbi->s_es = es;
3039         sb->s_magic = le16_to_cpu(es->s_magic);
3040         if (sb->s_magic != EXT4_SUPER_MAGIC)
3041                 goto cantfind_ext4;
3042         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3043
3044         /* Set defaults before we parse the mount options */
3045         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3046         set_opt(sb, INIT_INODE_TABLE);
3047         if (def_mount_opts & EXT4_DEFM_DEBUG)
3048                 set_opt(sb, DEBUG);
3049         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3050                 set_opt(sb, GRPID);
3051         if (def_mount_opts & EXT4_DEFM_UID16)
3052                 set_opt(sb, NO_UID32);
3053         /* xattr user namespace & acls are now defaulted on */
3054 #ifdef CONFIG_EXT4_FS_XATTR
3055         set_opt(sb, XATTR_USER);
3056 #endif
3057 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3058         set_opt(sb, POSIX_ACL);
3059 #endif
3060         set_opt(sb, MBLK_IO_SUBMIT);
3061         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3062                 set_opt(sb, JOURNAL_DATA);
3063         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3064                 set_opt(sb, ORDERED_DATA);
3065         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3066                 set_opt(sb, WRITEBACK_DATA);
3067
3068         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3069                 set_opt(sb, ERRORS_PANIC);
3070         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3071                 set_opt(sb, ERRORS_CONT);
3072         else
3073                 set_opt(sb, ERRORS_RO);
3074         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3075                 set_opt(sb, BLOCK_VALIDITY);
3076         if (def_mount_opts & EXT4_DEFM_DISCARD)
3077                 set_opt(sb, DISCARD);
3078
3079         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3080         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3081         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3082         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3083         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3084
3085         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3086                 set_opt(sb, BARRIER);
3087
3088         /*
3089          * enable delayed allocation by default
3090          * Use -o nodelalloc to turn it off
3091          */
3092         if (!IS_EXT3_SB(sb) &&
3093             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3094                 set_opt(sb, DELALLOC);
3095
3096         /*
3097          * set default s_li_wait_mult for lazyinit, for the case there is
3098          * no mount option specified.
3099          */
3100         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3101
3102         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3103                            &journal_devnum, &journal_ioprio, 0)) {
3104                 ext4_msg(sb, KERN_WARNING,
3105                          "failed to parse options in superblock: %s",
3106                          sbi->s_es->s_mount_opts);
3107         }
3108         if (!parse_options((char *) data, sb, &journal_devnum,
3109                            &journal_ioprio, 0))
3110                 goto failed_mount;
3111
3112         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3113                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3114                             "with data=journal disables delayed "
3115                             "allocation and O_DIRECT support!\n");
3116                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3117                         ext4_msg(sb, KERN_ERR, "can't mount with "
3118                                  "both data=journal and delalloc");
3119                         goto failed_mount;
3120                 }
3121                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3122                         ext4_msg(sb, KERN_ERR, "can't mount with "
3123                                  "both data=journal and delalloc");
3124                         goto failed_mount;
3125                 }
3126                 if (test_opt(sb, DELALLOC))
3127                         clear_opt(sb, DELALLOC);
3128         }
3129
3130         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3131         if (test_opt(sb, DIOREAD_NOLOCK)) {
3132                 if (blocksize < PAGE_SIZE) {
3133                         ext4_msg(sb, KERN_ERR, "can't mount with "
3134                                  "dioread_nolock if block size != PAGE_SIZE");
3135                         goto failed_mount;
3136                 }
3137         }
3138
3139         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3140                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3141
3142         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3143             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3144              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3145              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3146                 ext4_msg(sb, KERN_WARNING,
3147                        "feature flags set on rev 0 fs, "
3148                        "running e2fsck is recommended");
3149
3150         if (IS_EXT2_SB(sb)) {
3151                 if (ext2_feature_set_ok(sb))
3152                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3153                                  "using the ext4 subsystem");
3154                 else {
3155                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3156                                  "to feature incompatibilities");
3157                         goto failed_mount;
3158                 }
3159         }
3160
3161         if (IS_EXT3_SB(sb)) {
3162                 if (ext3_feature_set_ok(sb))
3163                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3164                                  "using the ext4 subsystem");
3165                 else {
3166                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3167                                  "to feature incompatibilities");
3168                         goto failed_mount;
3169                 }
3170         }
3171
3172         /*
3173          * Check feature flags regardless of the revision level, since we
3174          * previously didn't change the revision level when setting the flags,
3175          * so there is a chance incompat flags are set on a rev 0 filesystem.
3176          */
3177         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3178                 goto failed_mount;
3179
3180         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3181             blocksize > EXT4_MAX_BLOCK_SIZE) {
3182                 ext4_msg(sb, KERN_ERR,
3183                        "Unsupported filesystem blocksize %d", blocksize);
3184                 goto failed_mount;
3185         }
3186
3187         if (sb->s_blocksize != blocksize) {
3188                 /* Validate the filesystem blocksize */
3189                 if (!sb_set_blocksize(sb, blocksize)) {
3190                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3191                                         blocksize);
3192                         goto failed_mount;
3193                 }
3194
3195                 brelse(bh);
3196                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3197                 offset = do_div(logical_sb_block, blocksize);
3198                 bh = sb_bread(sb, logical_sb_block);
3199                 if (!bh) {
3200                         ext4_msg(sb, KERN_ERR,
3201                                "Can't read superblock on 2nd try");
3202                         goto failed_mount;
3203                 }
3204                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3205                 sbi->s_es = es;
3206                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3207                         ext4_msg(sb, KERN_ERR,
3208                                "Magic mismatch, very weird!");
3209                         goto failed_mount;
3210                 }
3211         }
3212
3213         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3214                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3215         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3216                                                       has_huge_files);
3217         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3218
3219         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3220                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3221                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3222         } else {
3223                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3224                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3225                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3226                     (!is_power_of_2(sbi->s_inode_size)) ||
3227                     (sbi->s_inode_size > blocksize)) {
3228                         ext4_msg(sb, KERN_ERR,
3229                                "unsupported inode size: %d",
3230                                sbi->s_inode_size);
3231                         goto failed_mount;
3232                 }
3233                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3234                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3235         }
3236
3237         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3238         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3239                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3240                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3241                     !is_power_of_2(sbi->s_desc_size)) {
3242                         ext4_msg(sb, KERN_ERR,
3243                                "unsupported descriptor size %lu",
3244                                sbi->s_desc_size);
3245                         goto failed_mount;
3246                 }
3247         } else
3248                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3249
3250         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3251         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3252         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3253                 goto cantfind_ext4;
3254
3255         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3256         if (sbi->s_inodes_per_block == 0)
3257                 goto cantfind_ext4;
3258         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3259                                         sbi->s_inodes_per_block;
3260         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3261         sbi->s_sbh = bh;
3262         sbi->s_mount_state = le16_to_cpu(es->s_state);
3263         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3264         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3265
3266         for (i = 0; i < 4; i++)
3267                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3268         sbi->s_def_hash_version = es->s_def_hash_version;
3269         i = le32_to_cpu(es->s_flags);
3270         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3271                 sbi->s_hash_unsigned = 3;
3272         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3273 #ifdef __CHAR_UNSIGNED__
3274                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3275                 sbi->s_hash_unsigned = 3;
3276 #else
3277                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3278 #endif
3279                 sb->s_dirt = 1;
3280         }
3281
3282         /* Handle clustersize */
3283         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3284         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3285                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3286         if (has_bigalloc) {
3287                 if (clustersize < blocksize) {
3288                         ext4_msg(sb, KERN_ERR,
3289                                  "cluster size (%d) smaller than "
3290                                  "block size (%d)", clustersize, blocksize);
3291                         goto failed_mount;
3292                 }
3293                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3294                         le32_to_cpu(es->s_log_block_size);
3295                 sbi->s_clusters_per_group =
3296                         le32_to_cpu(es->s_clusters_per_group);
3297                 if (sbi->s_clusters_per_group > blocksize * 8) {
3298                         ext4_msg(sb, KERN_ERR,
3299                                  "#clusters per group too big: %lu",
3300                                  sbi->s_clusters_per_group);
3301                         goto failed_mount;
3302                 }
3303                 if (sbi->s_blocks_per_group !=
3304                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3305                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3306                                  "clusters per group (%lu) inconsistent",
3307                                  sbi->s_blocks_per_group,
3308                                  sbi->s_clusters_per_group);
3309                         goto failed_mount;
3310                 }
3311         } else {
3312                 if (clustersize != blocksize) {
3313                         ext4_warning(sb, "fragment/cluster size (%d) != "
3314                                      "block size (%d)", clustersize,
3315                                      blocksize);
3316                         clustersize = blocksize;
3317                 }
3318                 if (sbi->s_blocks_per_group > blocksize * 8) {
3319                         ext4_msg(sb, KERN_ERR,
3320                                  "#blocks per group too big: %lu",
3321                                  sbi->s_blocks_per_group);
3322                         goto failed_mount;
3323                 }
3324                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3325                 sbi->s_cluster_bits = 0;
3326         }
3327         sbi->s_cluster_ratio = clustersize / blocksize;
3328
3329         if (sbi->s_inodes_per_group > blocksize * 8) {
3330                 ext4_msg(sb, KERN_ERR,
3331                        "#inodes per group too big: %lu",
3332                        sbi->s_inodes_per_group);
3333                 goto failed_mount;
3334         }
3335
3336         /*
3337          * Test whether we have more sectors than will fit in sector_t,
3338          * and whether the max offset is addressable by the page cache.
3339          */
3340         err = generic_check_addressable(sb->s_blocksize_bits,
3341                                         ext4_blocks_count(es));
3342         if (err) {
3343                 ext4_msg(sb, KERN_ERR, "filesystem"
3344                          " too large to mount safely on this system");
3345                 if (sizeof(sector_t) < 8)
3346                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3347                 ret = err;
3348                 goto failed_mount;
3349         }
3350
3351         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3352                 goto cantfind_ext4;
3353
3354         /* check blocks count against device size */
3355         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3356         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3357                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3358                        "exceeds size of device (%llu blocks)",
3359                        ext4_blocks_count(es), blocks_count);
3360                 goto failed_mount;
3361         }
3362
3363         /*
3364          * It makes no sense for the first data block to be beyond the end
3365          * of the filesystem.
3366          */
3367         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3368                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3369                          "block %u is beyond end of filesystem (%llu)",
3370                          le32_to_cpu(es->s_first_data_block),
3371                          ext4_blocks_count(es));
3372                 goto failed_mount;
3373         }
3374         blocks_count = (ext4_blocks_count(es) -
3375                         le32_to_cpu(es->s_first_data_block) +
3376                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3377         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3378         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3379                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3380                        "(block count %llu, first data block %u, "
3381                        "blocks per group %lu)", sbi->s_groups_count,
3382                        ext4_blocks_count(es),
3383                        le32_to_cpu(es->s_first_data_block),
3384                        EXT4_BLOCKS_PER_GROUP(sb));
3385                 goto failed_mount;
3386         }
3387         sbi->s_groups_count = blocks_count;
3388         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3389                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3390         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3391                    EXT4_DESC_PER_BLOCK(sb);
3392         sbi->s_group_desc = ext4_kvmalloc(db_count *
3393                                           sizeof(struct buffer_head *),
3394                                           GFP_KERNEL);
3395         if (sbi->s_group_desc == NULL) {
3396                 ext4_msg(sb, KERN_ERR, "not enough memory");
3397                 goto failed_mount;
3398         }
3399
3400         if (ext4_proc_root)
3401                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3402
3403         bgl_lock_init(sbi->s_blockgroup_lock);
3404
3405         for (i = 0; i < db_count; i++) {
3406                 block = descriptor_loc(sb, logical_sb_block, i);
3407                 sbi->s_group_desc[i] = sb_bread(sb, block);
3408                 if (!sbi->s_group_desc[i]) {
3409                         ext4_msg(sb, KERN_ERR,
3410                                "can't read group descriptor %d", i);
3411                         db_count = i;
3412                         goto failed_mount2;
3413                 }
3414         }
3415         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3416                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3417                 goto failed_mount2;
3418         }
3419         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3420                 if (!ext4_fill_flex_info(sb)) {
3421                         ext4_msg(sb, KERN_ERR,
3422                                "unable to initialize "
3423                                "flex_bg meta info!");
3424                         goto failed_mount2;
3425                 }
3426
3427         sbi->s_gdb_count = db_count;
3428         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3429         spin_lock_init(&sbi->s_next_gen_lock);
3430
3431         init_timer(&sbi->s_err_report);
3432         sbi->s_err_report.function = print_daily_error_info;
3433         sbi->s_err_report.data = (unsigned long) sb;
3434
3435         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3436                         ext4_count_free_clusters(sb));
3437         if (!err) {
3438                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3439                                 ext4_count_free_inodes(sb));
3440         }
3441         if (!err) {
3442                 err = percpu_counter_init(&sbi->s_dirs_counter,
3443                                 ext4_count_dirs(sb));
3444         }
3445         if (!err) {
3446                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3447         }
3448         if (err) {
3449                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3450                 goto failed_mount3;
3451         }
3452
3453         sbi->s_stripe = ext4_get_stripe_size(sbi);
3454         sbi->s_max_writeback_mb_bump = 128;
3455
3456         /*
3457          * set up enough so that it can read an inode
3458          */
3459         if (!test_opt(sb, NOLOAD) &&
3460             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3461                 sb->s_op = &ext4_sops;
3462         else
3463                 sb->s_op = &ext4_nojournal_sops;
3464         sb->s_export_op = &ext4_export_ops;
3465         sb->s_xattr = ext4_xattr_handlers;
3466 #ifdef CONFIG_QUOTA
3467         sb->s_qcop = &ext4_qctl_operations;
3468         sb->dq_op = &ext4_quota_operations;
3469 #endif
3470         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3471
3472         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3473         mutex_init(&sbi->s_orphan_lock);
3474         sbi->s_resize_flags = 0;
3475
3476         sb->s_root = NULL;
3477
3478         needs_recovery = (es->s_last_orphan != 0 ||
3479                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3480                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3481
3482         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3483             !(sb->s_flags & MS_RDONLY))
3484                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3485                         goto failed_mount3;
3486
3487         /*
3488          * The first inode we look at is the journal inode.  Don't try
3489          * root first: it may be modified in the journal!
3490          */
3491         if (!test_opt(sb, NOLOAD) &&
3492             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3493                 if (ext4_load_journal(sb, es, journal_devnum))
3494                         goto failed_mount3;
3495         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3496               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3497                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3498                        "suppressed and not mounted read-only");
3499                 goto failed_mount_wq;
3500         } else {
3501                 clear_opt(sb, DATA_FLAGS);
3502                 sbi->s_journal = NULL;
3503                 needs_recovery = 0;
3504                 goto no_journal;
3505         }
3506
3507         if (ext4_blocks_count(es) > 0xffffffffULL &&
3508             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3509                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3510                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3511                 goto failed_mount_wq;
3512         }
3513
3514         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3515                 jbd2_journal_set_features(sbi->s_journal,
3516                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3517                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3518         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3519                 jbd2_journal_set_features(sbi->s_journal,
3520                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3521                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3522                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3523         } else {
3524                 jbd2_journal_clear_features(sbi->s_journal,
3525                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3526                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3527         }
3528
3529         /* We have now updated the journal if required, so we can
3530          * validate the data journaling mode. */
3531         switch (test_opt(sb, DATA_FLAGS)) {
3532         case 0:
3533                 /* No mode set, assume a default based on the journal
3534                  * capabilities: ORDERED_DATA if the journal can
3535                  * cope, else JOURNAL_DATA
3536                  */
3537                 if (jbd2_journal_check_available_features
3538                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3539                         set_opt(sb, ORDERED_DATA);
3540                 else
3541                         set_opt(sb, JOURNAL_DATA);
3542                 break;
3543
3544         case EXT4_MOUNT_ORDERED_DATA:
3545         case EXT4_MOUNT_WRITEBACK_DATA:
3546                 if (!jbd2_journal_check_available_features
3547                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3548                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3549                                "requested data journaling mode");
3550                         goto failed_mount_wq;
3551                 }
3552         default:
3553                 break;
3554         }
3555         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3556
3557         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3558
3559         /*
3560          * The journal may have updated the bg summary counts, so we
3561          * need to update the global counters.
3562          */
3563         percpu_counter_set(&sbi->s_freeclusters_counter,
3564                            ext4_count_free_clusters(sb));
3565         percpu_counter_set(&sbi->s_freeinodes_counter,
3566                            ext4_count_free_inodes(sb));
3567         percpu_counter_set(&sbi->s_dirs_counter,
3568                            ext4_count_dirs(sb));
3569         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3570
3571 no_journal:
3572         /*
3573          * The maximum number of concurrent works can be high and
3574          * concurrency isn't really necessary.  Limit it to 1.
3575          */
3576         EXT4_SB(sb)->dio_unwritten_wq =
3577                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3578         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3579                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3580                 goto failed_mount_wq;
3581         }
3582
3583         /*
3584          * The jbd2_journal_load will have done any necessary log recovery,
3585          * so we can safely mount the rest of the filesystem now.
3586          */
3587
3588         root = ext4_iget(sb, EXT4_ROOT_INO);
3589         if (IS_ERR(root)) {
3590                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3591                 ret = PTR_ERR(root);
3592                 root = NULL;
3593                 goto failed_mount4;
3594         }
3595         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3596                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3597                 iput(root);
3598                 goto failed_mount4;
3599         }
3600         sb->s_root = d_alloc_root(root);
3601         if (!sb->s_root) {
3602                 iput(root);
3603                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3604                 ret = -ENOMEM;
3605                 goto failed_mount4;
3606         }
3607
3608         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3609
3610         /* determine the minimum size of new large inodes, if present */
3611         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3612                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3613                                                      EXT4_GOOD_OLD_INODE_SIZE;
3614                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3615                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3616                         if (sbi->s_want_extra_isize <
3617                             le16_to_cpu(es->s_want_extra_isize))
3618                                 sbi->s_want_extra_isize =
3619                                         le16_to_cpu(es->s_want_extra_isize);
3620                         if (sbi->s_want_extra_isize <
3621                             le16_to_cpu(es->s_min_extra_isize))
3622                                 sbi->s_want_extra_isize =
3623                                         le16_to_cpu(es->s_min_extra_isize);
3624                 }
3625         }
3626         /* Check if enough inode space is available */
3627         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3628                                                         sbi->s_inode_size) {
3629                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3630                                                        EXT4_GOOD_OLD_INODE_SIZE;
3631                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3632                          "available");
3633         }
3634
3635         err = ext4_setup_system_zone(sb);
3636         if (err) {
3637                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3638                          "zone (%d)", err);
3639                 goto failed_mount4a;
3640         }
3641
3642         ext4_ext_init(sb);
3643         err = ext4_mb_init(sb, needs_recovery);
3644         if (err) {
3645                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3646                          err);
3647                 goto failed_mount5;
3648         }
3649
3650         err = ext4_register_li_request(sb, first_not_zeroed);
3651         if (err)
3652                 goto failed_mount6;
3653
3654         sbi->s_kobj.kset = ext4_kset;
3655         init_completion(&sbi->s_kobj_unregister);
3656         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3657                                    "%s", sb->s_id);
3658         if (err)
3659                 goto failed_mount7;
3660
3661         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3662         ext4_orphan_cleanup(sb, es);
3663         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3664         if (needs_recovery) {
3665                 ext4_msg(sb, KERN_INFO, "recovery complete");
3666                 ext4_mark_recovery_complete(sb, es);
3667         }
3668         if (EXT4_SB(sb)->s_journal) {
3669                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3670                         descr = " journalled data mode";
3671                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3672                         descr = " ordered data mode";
3673                 else
3674                         descr = " writeback data mode";
3675         } else
3676                 descr = "out journal";
3677
3678         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3679                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3680                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3681
3682         if (es->s_error_count)
3683                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3684
3685         kfree(orig_data);
3686         return 0;
3687
3688 cantfind_ext4:
3689         if (!silent)
3690                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3691         goto failed_mount;
3692
3693 failed_mount7:
3694         ext4_unregister_li_request(sb);
3695 failed_mount6:
3696         ext4_mb_release(sb);
3697 failed_mount5:
3698         ext4_ext_release(sb);
3699         ext4_release_system_zone(sb);
3700 failed_mount4a:
3701         dput(sb->s_root);
3702         sb->s_root = NULL;
3703 failed_mount4:
3704         ext4_msg(sb, KERN_ERR, "mount failed");
3705         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3706 failed_mount_wq:
3707         if (sbi->s_journal) {
3708                 jbd2_journal_destroy(sbi->s_journal);
3709                 sbi->s_journal = NULL;
3710         }
3711 failed_mount3:
3712         del_timer(&sbi->s_err_report);
3713         if (sbi->s_flex_groups)
3714                 ext4_kvfree(sbi->s_flex_groups);
3715         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3716         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3717         percpu_counter_destroy(&sbi->s_dirs_counter);
3718         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3719         if (sbi->s_mmp_tsk)
3720                 kthread_stop(sbi->s_mmp_tsk);
3721 failed_mount2:
3722         for (i = 0; i < db_count; i++)
3723                 brelse(sbi->s_group_desc[i]);
3724         ext4_kvfree(sbi->s_group_desc);
3725 failed_mount:
3726         if (sbi->s_proc) {
3727                 remove_proc_entry(sb->s_id, ext4_proc_root);
3728         }
3729 #ifdef CONFIG_QUOTA
3730         for (i = 0; i < MAXQUOTAS; i++)
3731                 kfree(sbi->s_qf_names[i]);
3732 #endif
3733         ext4_blkdev_remove(sbi);
3734         brelse(bh);
3735 out_fail:
3736         sb->s_fs_info = NULL;
3737         kfree(sbi->s_blockgroup_lock);
3738         kfree(sbi);
3739 out_free_orig:
3740         kfree(orig_data);
3741         return ret;
3742 }
3743
3744 /*
3745  * Setup any per-fs journal parameters now.  We'll do this both on
3746  * initial mount, once the journal has been initialised but before we've
3747  * done any recovery; and again on any subsequent remount.
3748  */
3749 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3750 {
3751         struct ext4_sb_info *sbi = EXT4_SB(sb);
3752
3753         journal->j_commit_interval = sbi->s_commit_interval;
3754         journal->j_min_batch_time = sbi->s_min_batch_time;
3755         journal->j_max_batch_time = sbi->s_max_batch_time;
3756
3757         write_lock(&journal->j_state_lock);
3758         if (test_opt(sb, BARRIER))
3759                 journal->j_flags |= JBD2_BARRIER;
3760         else
3761                 journal->j_flags &= ~JBD2_BARRIER;
3762         if (test_opt(sb, DATA_ERR_ABORT))
3763                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3764         else
3765                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3766         write_unlock(&journal->j_state_lock);
3767 }
3768
3769 static journal_t *ext4_get_journal(struct super_block *sb,
3770                                    unsigned int journal_inum)
3771 {
3772         struct inode *journal_inode;
3773         journal_t *journal;
3774
3775         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3776
3777         /* First, test for the existence of a valid inode on disk.  Bad
3778          * things happen if we iget() an unused inode, as the subsequent
3779          * iput() will try to delete it. */
3780
3781         journal_inode = ext4_iget(sb, journal_inum);
3782         if (IS_ERR(journal_inode)) {
3783                 ext4_msg(sb, KERN_ERR, "no journal found");
3784                 return NULL;
3785         }
3786         if (!journal_inode->i_nlink) {
3787                 make_bad_inode(journal_inode);
3788                 iput(journal_inode);
3789                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3790                 return NULL;
3791         }
3792
3793         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3794                   journal_inode, journal_inode->i_size);
3795         if (!S_ISREG(journal_inode->i_mode)) {
3796                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3797                 iput(journal_inode);
3798                 return NULL;
3799         }
3800
3801         journal = jbd2_journal_init_inode(journal_inode);
3802         if (!journal) {
3803                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3804                 iput(journal_inode);
3805                 return NULL;
3806         }
3807         journal->j_private = sb;
3808         ext4_init_journal_params(sb, journal);
3809         return journal;
3810 }
3811
3812 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3813                                        dev_t j_dev)
3814 {
3815         struct buffer_head *bh;
3816         journal_t *journal;
3817         ext4_fsblk_t start;
3818         ext4_fsblk_t len;
3819         int hblock, blocksize;
3820         ext4_fsblk_t sb_block;
3821         unsigned long offset;
3822         struct ext4_super_block *es;
3823         struct block_device *bdev;
3824
3825         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3826
3827         bdev = ext4_blkdev_get(j_dev, sb);
3828         if (bdev == NULL)
3829                 return NULL;
3830
3831         blocksize = sb->s_blocksize;
3832         hblock = bdev_logical_block_size(bdev);
3833         if (blocksize < hblock) {
3834                 ext4_msg(sb, KERN_ERR,
3835                         "blocksize too small for journal device");
3836                 goto out_bdev;
3837         }
3838
3839         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3840         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3841         set_blocksize(bdev, blocksize);
3842         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3843                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3844                        "external journal");
3845                 goto out_bdev;
3846         }
3847
3848         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3849         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3850             !(le32_to_cpu(es->s_feature_incompat) &
3851               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3852                 ext4_msg(sb, KERN_ERR, "external journal has "
3853                                         "bad superblock");
3854                 brelse(bh);
3855                 goto out_bdev;
3856         }
3857
3858         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3859                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3860                 brelse(bh);
3861                 goto out_bdev;
3862         }
3863
3864         len = ext4_blocks_count(es);
3865         start = sb_block + 1;
3866         brelse(bh);     /* we're done with the superblock */
3867
3868         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3869                                         start, len, blocksize);
3870         if (!journal) {
3871                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3872                 goto out_bdev;
3873         }
3874         journal->j_private = sb;
3875         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3876         wait_on_buffer(journal->j_sb_buffer);
3877         if (!buffer_uptodate(journal->j_sb_buffer)) {
3878                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3879                 goto out_journal;
3880         }
3881         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3882                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3883                                         "user (unsupported) - %d",
3884                         be32_to_cpu(journal->j_superblock->s_nr_users));
3885                 goto out_journal;
3886         }
3887         EXT4_SB(sb)->journal_bdev = bdev;
3888         ext4_init_journal_params(sb, journal);
3889         return journal;
3890
3891 out_journal:
3892         jbd2_journal_destroy(journal);
3893 out_bdev:
3894         ext4_blkdev_put(bdev);
3895         return NULL;
3896 }
3897
3898 static int ext4_load_journal(struct super_block *sb,
3899                              struct ext4_super_block *es,
3900                              unsigned long journal_devnum)
3901 {
3902         journal_t *journal;
3903         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3904         dev_t journal_dev;
3905         int err = 0;
3906         int really_read_only;
3907
3908         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3909
3910         if (journal_devnum &&
3911             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3912                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3913                         "numbers have changed");
3914                 journal_dev = new_decode_dev(journal_devnum);
3915         } else
3916                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3917
3918         really_read_only = bdev_read_only(sb->s_bdev);
3919
3920         /*
3921          * Are we loading a blank journal or performing recovery after a
3922          * crash?  For recovery, we need to check in advance whether we
3923          * can get read-write access to the device.
3924          */
3925         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3926                 if (sb->s_flags & MS_RDONLY) {
3927                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3928                                         "required on readonly filesystem");
3929                         if (really_read_only) {
3930                                 ext4_msg(sb, KERN_ERR, "write access "
3931                                         "unavailable, cannot proceed");
3932                                 return -EROFS;
3933                         }
3934                         ext4_msg(sb, KERN_INFO, "write access will "
3935                                "be enabled during recovery");
3936                 }
3937         }
3938
3939         if (journal_inum && journal_dev) {
3940                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3941                        "and inode journals!");
3942                 return -EINVAL;
3943         }
3944
3945         if (journal_inum) {
3946                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3947                         return -EINVAL;
3948         } else {
3949                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3950                         return -EINVAL;
3951         }
3952
3953         if (!(journal->j_flags & JBD2_BARRIER))
3954                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3955
3956         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3957                 err = jbd2_journal_wipe(journal, !really_read_only);
3958         if (!err) {
3959                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3960                 if (save)
3961                         memcpy(save, ((char *) es) +
3962                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3963                 err = jbd2_journal_load(journal);
3964                 if (save)
3965                         memcpy(((char *) es) + EXT4_S_ERR_START,
3966                                save, EXT4_S_ERR_LEN);
3967                 kfree(save);
3968         }
3969
3970         if (err) {
3971                 ext4_msg(sb, KERN_ERR, "error loading journal");
3972                 jbd2_journal_destroy(journal);
3973                 return err;
3974         }
3975
3976         EXT4_SB(sb)->s_journal = journal;
3977         ext4_clear_journal_err(sb, es);
3978
3979         if (!really_read_only && journal_devnum &&
3980             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3981                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3982
3983                 /* Make sure we flush the recovery flag to disk. */
3984                 ext4_commit_super(sb, 1);
3985         }
3986
3987         return 0;
3988 }
3989
3990 static int ext4_commit_super(struct super_block *sb, int sync)
3991 {
3992         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3993         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3994         int error = 0;
3995
3996         if (!sbh || block_device_ejected(sb))
3997                 return error;
3998         if (buffer_write_io_error(sbh)) {
3999                 /*
4000                  * Oh, dear.  A previous attempt to write the
4001                  * superblock failed.  This could happen because the
4002                  * USB device was yanked out.  Or it could happen to
4003                  * be a transient write error and maybe the block will
4004                  * be remapped.  Nothing we can do but to retry the
4005                  * write and hope for the best.
4006                  */
4007                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4008                        "superblock detected");
4009                 clear_buffer_write_io_error(sbh);
4010                 set_buffer_uptodate(sbh);
4011         }
4012         /*
4013          * If the file system is mounted read-only, don't update the
4014          * superblock write time.  This avoids updating the superblock
4015          * write time when we are mounting the root file system
4016          * read/only but we need to replay the journal; at that point,
4017          * for people who are east of GMT and who make their clock
4018          * tick in localtime for Windows bug-for-bug compatibility,
4019          * the clock is set in the future, and this will cause e2fsck
4020          * to complain and force a full file system check.
4021          */
4022         if (!(sb->s_flags & MS_RDONLY))
4023                 es->s_wtime = cpu_to_le32(get_seconds());
4024         if (sb->s_bdev->bd_part)
4025                 es->s_kbytes_written =
4026                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4027                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4028                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4029         else
4030                 es->s_kbytes_written =
4031                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4032         ext4_free_blocks_count_set(es,
4033                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4034                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4035         es->s_free_inodes_count =
4036                 cpu_to_le32(percpu_counter_sum_positive(
4037                                 &EXT4_SB(sb)->s_freeinodes_counter));
4038         sb->s_dirt = 0;
4039         BUFFER_TRACE(sbh, "marking dirty");
4040         mark_buffer_dirty(sbh);
4041         if (sync) {
4042                 error = sync_dirty_buffer(sbh);
4043                 if (error)
4044                         return error;
4045
4046                 error = buffer_write_io_error(sbh);
4047                 if (error) {
4048                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4049                                "superblock");
4050                         clear_buffer_write_io_error(sbh);
4051                         set_buffer_uptodate(sbh);
4052                 }
4053         }
4054         return error;
4055 }
4056
4057 /*
4058  * Have we just finished recovery?  If so, and if we are mounting (or
4059  * remounting) the filesystem readonly, then we will end up with a
4060  * consistent fs on disk.  Record that fact.
4061  */
4062 static void ext4_mark_recovery_complete(struct super_block *sb,
4063                                         struct ext4_super_block *es)
4064 {
4065         journal_t *journal = EXT4_SB(sb)->s_journal;
4066
4067         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4068                 BUG_ON(journal != NULL);
4069                 return;
4070         }
4071         jbd2_journal_lock_updates(journal);
4072         if (jbd2_journal_flush(journal) < 0)
4073                 goto out;
4074
4075         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4076             sb->s_flags & MS_RDONLY) {
4077                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4078                 ext4_commit_super(sb, 1);
4079         }
4080
4081 out:
4082         jbd2_journal_unlock_updates(journal);
4083 }
4084
4085 /*
4086  * If we are mounting (or read-write remounting) a filesystem whose journal
4087  * has recorded an error from a previous lifetime, move that error to the
4088  * main filesystem now.
4089  */
4090 static void ext4_clear_journal_err(struct super_block *sb,
4091                                    struct ext4_super_block *es)
4092 {
4093         journal_t *journal;
4094         int j_errno;
4095         const char *errstr;
4096
4097         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4098
4099         journal = EXT4_SB(sb)->s_journal;
4100
4101         /*
4102          * Now check for any error status which may have been recorded in the
4103          * journal by a prior ext4_error() or ext4_abort()
4104          */
4105
4106         j_errno = jbd2_journal_errno(journal);
4107         if (j_errno) {
4108                 char nbuf[16];
4109
4110                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4111                 ext4_warning(sb, "Filesystem error recorded "
4112                              "from previous mount: %s", errstr);
4113                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4114
4115                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4116                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4117                 ext4_commit_super(sb, 1);
4118
4119                 jbd2_journal_clear_err(journal);
4120         }
4121 }
4122
4123 /*
4124  * Force the running and committing transactions to commit,
4125  * and wait on the commit.
4126  */
4127 int ext4_force_commit(struct super_block *sb)
4128 {
4129         journal_t *journal;
4130         int ret = 0;
4131
4132         if (sb->s_flags & MS_RDONLY)
4133                 return 0;
4134
4135         journal = EXT4_SB(sb)->s_journal;
4136         if (journal) {
4137                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4138                 ret = ext4_journal_force_commit(journal);
4139         }
4140
4141         return ret;
4142 }
4143
4144 static void ext4_write_super(struct super_block *sb)
4145 {
4146         lock_super(sb);
4147         ext4_commit_super(sb, 1);
4148         unlock_super(sb);
4149 }
4150
4151 static int ext4_sync_fs(struct super_block *sb, int wait)
4152 {
4153         int ret = 0;
4154         tid_t target;
4155         struct ext4_sb_info *sbi = EXT4_SB(sb);
4156
4157         trace_ext4_sync_fs(sb, wait);
4158         flush_workqueue(sbi->dio_unwritten_wq);
4159         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4160                 if (wait)
4161                         jbd2_log_wait_commit(sbi->s_journal, target);
4162         }
4163         return ret;
4164 }
4165
4166 /*
4167  * LVM calls this function before a (read-only) snapshot is created.  This
4168  * gives us a chance to flush the journal completely and mark the fs clean.
4169  *
4170  * Note that only this function cannot bring a filesystem to be in a clean
4171  * state independently, because ext4 prevents a new handle from being started
4172  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4173  * the upper layer.
4174  */
4175 static int ext4_freeze(struct super_block *sb)
4176 {
4177         int error = 0;
4178         journal_t *journal;
4179
4180         if (sb->s_flags & MS_RDONLY)
4181                 return 0;
4182
4183         journal = EXT4_SB(sb)->s_journal;
4184
4185         /* Now we set up the journal barrier. */
4186         jbd2_journal_lock_updates(journal);
4187
4188         /*
4189          * Don't clear the needs_recovery flag if we failed to flush
4190          * the journal.
4191          */
4192         error = jbd2_journal_flush(journal);
4193         if (error < 0)
4194                 goto out;
4195
4196         /* Journal blocked and flushed, clear needs_recovery flag. */
4197         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4198         error = ext4_commit_super(sb, 1);
4199 out:
4200         /* we rely on s_frozen to stop further updates */
4201         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4202         return error;
4203 }
4204
4205 /*
4206  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4207  * flag here, even though the filesystem is not technically dirty yet.
4208  */
4209 static int ext4_unfreeze(struct super_block *sb)
4210 {
4211         if (sb->s_flags & MS_RDONLY)
4212                 return 0;
4213
4214         lock_super(sb);
4215         /* Reset the needs_recovery flag before the fs is unlocked. */
4216         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4217         ext4_commit_super(sb, 1);
4218         unlock_super(sb);
4219         return 0;
4220 }
4221
4222 /*
4223  * Structure to save mount options for ext4_remount's benefit
4224  */
4225 struct ext4_mount_options {
4226         unsigned long s_mount_opt;
4227         unsigned long s_mount_opt2;
4228         uid_t s_resuid;
4229         gid_t s_resgid;
4230         unsigned long s_commit_interval;
4231         u32 s_min_batch_time, s_max_batch_time;
4232 #ifdef CONFIG_QUOTA
4233         int s_jquota_fmt;
4234         char *s_qf_names[MAXQUOTAS];
4235 #endif
4236 };
4237
4238 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4239 {
4240         struct ext4_super_block *es;
4241         struct ext4_sb_info *sbi = EXT4_SB(sb);
4242         unsigned long old_sb_flags;
4243         struct ext4_mount_options old_opts;
4244         int enable_quota = 0;
4245         ext4_group_t g;
4246         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4247         int err = 0;
4248 #ifdef CONFIG_QUOTA
4249         int i;
4250 #endif
4251         char *orig_data = kstrdup(data, GFP_KERNEL);
4252
4253         /* Store the original options */
4254         lock_super(sb);
4255         old_sb_flags = sb->s_flags;
4256         old_opts.s_mount_opt = sbi->s_mount_opt;
4257         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4258         old_opts.s_resuid = sbi->s_resuid;
4259         old_opts.s_resgid = sbi->s_resgid;
4260         old_opts.s_commit_interval = sbi->s_commit_interval;
4261         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4262         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4263 #ifdef CONFIG_QUOTA
4264         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4265         for (i = 0; i < MAXQUOTAS; i++)
4266                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4267 #endif
4268         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4269                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4270
4271         /*
4272          * Allow the "check" option to be passed as a remount option.
4273          */
4274         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4275                 err = -EINVAL;
4276                 goto restore_opts;
4277         }
4278
4279         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4280                 ext4_abort(sb, "Abort forced by user");
4281
4282         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4283                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4284
4285         es = sbi->s_es;
4286
4287         if (sbi->s_journal) {
4288                 ext4_init_journal_params(sb, sbi->s_journal);
4289                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4290         }
4291
4292         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4293                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4294                         err = -EROFS;
4295                         goto restore_opts;
4296                 }
4297
4298                 if (*flags & MS_RDONLY) {
4299                         err = dquot_suspend(sb, -1);
4300                         if (err < 0)
4301                                 goto restore_opts;
4302
4303                         /*
4304                          * First of all, the unconditional stuff we have to do
4305                          * to disable replay of the journal when we next remount
4306                          */
4307                         sb->s_flags |= MS_RDONLY;
4308
4309                         /*
4310                          * OK, test if we are remounting a valid rw partition
4311                          * readonly, and if so set the rdonly flag and then
4312                          * mark the partition as valid again.
4313                          */
4314                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4315                             (sbi->s_mount_state & EXT4_VALID_FS))
4316                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4317
4318                         if (sbi->s_journal)
4319                                 ext4_mark_recovery_complete(sb, es);
4320                 } else {
4321                         /* Make sure we can mount this feature set readwrite */
4322                         if (!ext4_feature_set_ok(sb, 0)) {
4323                                 err = -EROFS;
4324                                 goto restore_opts;
4325                         }
4326                         /*
4327                          * Make sure the group descriptor checksums
4328                          * are sane.  If they aren't, refuse to remount r/w.
4329                          */
4330                         for (g = 0; g < sbi->s_groups_count; g++) {
4331                                 struct ext4_group_desc *gdp =
4332                                         ext4_get_group_desc(sb, g, NULL);
4333
4334                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4335                                         ext4_msg(sb, KERN_ERR,
4336                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4337                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4338                                                le16_to_cpu(gdp->bg_checksum));
4339                                         err = -EINVAL;
4340                                         goto restore_opts;
4341                                 }
4342                         }
4343
4344                         /*
4345                          * If we have an unprocessed orphan list hanging
4346                          * around from a previously readonly bdev mount,
4347                          * require a full umount/remount for now.
4348                          */
4349                         if (es->s_last_orphan) {
4350                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4351                                        "remount RDWR because of unprocessed "
4352                                        "orphan inode list.  Please "
4353                                        "umount/remount instead");
4354                                 err = -EINVAL;
4355                                 goto restore_opts;
4356                         }
4357
4358                         /*
4359                          * Mounting a RDONLY partition read-write, so reread
4360                          * and store the current valid flag.  (It may have
4361                          * been changed by e2fsck since we originally mounted
4362                          * the partition.)
4363                          */
4364                         if (sbi->s_journal)
4365                                 ext4_clear_journal_err(sb, es);
4366                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4367                         if (!ext4_setup_super(sb, es, 0))
4368                                 sb->s_flags &= ~MS_RDONLY;
4369                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4370                                                      EXT4_FEATURE_INCOMPAT_MMP))
4371                                 if (ext4_multi_mount_protect(sb,
4372                                                 le64_to_cpu(es->s_mmp_block))) {
4373                                         err = -EROFS;
4374                                         goto restore_opts;
4375                                 }
4376                         enable_quota = 1;
4377                 }
4378         }
4379
4380         /*
4381          * Reinitialize lazy itable initialization thread based on
4382          * current settings
4383          */
4384         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4385                 ext4_unregister_li_request(sb);
4386         else {
4387                 ext4_group_t first_not_zeroed;
4388                 first_not_zeroed = ext4_has_uninit_itable(sb);
4389                 ext4_register_li_request(sb, first_not_zeroed);
4390         }
4391
4392         ext4_setup_system_zone(sb);
4393         if (sbi->s_journal == NULL)
4394                 ext4_commit_super(sb, 1);
4395
4396 #ifdef CONFIG_QUOTA
4397         /* Release old quota file names */
4398         for (i = 0; i < MAXQUOTAS; i++)
4399                 if (old_opts.s_qf_names[i] &&
4400                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4401                         kfree(old_opts.s_qf_names[i]);
4402 #endif
4403         unlock_super(sb);
4404         if (enable_quota)
4405                 dquot_resume(sb, -1);
4406
4407         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4408         kfree(orig_data);
4409         return 0;
4410
4411 restore_opts:
4412         sb->s_flags = old_sb_flags;
4413         sbi->s_mount_opt = old_opts.s_mount_opt;
4414         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4415         sbi->s_resuid = old_opts.s_resuid;
4416         sbi->s_resgid = old_opts.s_resgid;
4417         sbi->s_commit_interval = old_opts.s_commit_interval;
4418         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4419         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4420 #ifdef CONFIG_QUOTA
4421         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4422         for (i = 0; i < MAXQUOTAS; i++) {
4423                 if (sbi->s_qf_names[i] &&
4424                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4425                         kfree(sbi->s_qf_names[i]);
4426                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4427         }
4428 #endif
4429         unlock_super(sb);
4430         kfree(orig_data);
4431         return err;
4432 }
4433
4434 /*
4435  * Note: calculating the overhead so we can be compatible with
4436  * historical BSD practice is quite difficult in the face of
4437  * clusters/bigalloc.  This is because multiple metadata blocks from
4438  * different block group can end up in the same allocation cluster.
4439  * Calculating the exact overhead in the face of clustered allocation
4440  * requires either O(all block bitmaps) in memory or O(number of block
4441  * groups**2) in time.  We will still calculate the superblock for
4442  * older file systems --- and if we come across with a bigalloc file
4443  * system with zero in s_overhead_clusters the estimate will be close to
4444  * correct especially for very large cluster sizes --- but for newer
4445  * file systems, it's better to calculate this figure once at mkfs
4446  * time, and store it in the superblock.  If the superblock value is
4447  * present (even for non-bigalloc file systems), we will use it.
4448  */
4449 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4450 {
4451         struct super_block *sb = dentry->d_sb;
4452         struct ext4_sb_info *sbi = EXT4_SB(sb);
4453         struct ext4_super_block *es = sbi->s_es;
4454         struct ext4_group_desc *gdp;
4455         u64 fsid;
4456         s64 bfree;
4457
4458         if (test_opt(sb, MINIX_DF)) {
4459                 sbi->s_overhead_last = 0;
4460         } else if (es->s_overhead_clusters) {
4461                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4462         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4463                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4464                 ext4_fsblk_t overhead = 0;
4465
4466                 /*
4467                  * Compute the overhead (FS structures).  This is constant
4468                  * for a given filesystem unless the number of block groups
4469                  * changes so we cache the previous value until it does.
4470                  */
4471
4472                 /*
4473                  * All of the blocks before first_data_block are
4474                  * overhead
4475                  */
4476                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4477
4478                 /*
4479                  * Add the overhead found in each block group
4480                  */
4481                 for (i = 0; i < ngroups; i++) {
4482                         gdp = ext4_get_group_desc(sb, i, NULL);
4483                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4484                         cond_resched();
4485                 }
4486                 sbi->s_overhead_last = overhead;
4487                 smp_wmb();
4488                 sbi->s_blocks_last = ext4_blocks_count(es);
4489         }
4490
4491         buf->f_type = EXT4_SUPER_MAGIC;
4492         buf->f_bsize = sb->s_blocksize;
4493         buf->f_blocks = (ext4_blocks_count(es) -
4494                          EXT4_C2B(sbi, sbi->s_overhead_last));
4495         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4496                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4497         /* prevent underflow in case that few free space is available */
4498         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4499         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4500         if (buf->f_bfree < ext4_r_blocks_count(es))
4501                 buf->f_bavail = 0;
4502         buf->f_files = le32_to_cpu(es->s_inodes_count);
4503         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4504         buf->f_namelen = EXT4_NAME_LEN;
4505         fsid = le64_to_cpup((void *)es->s_uuid) ^
4506                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4507         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4508         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4509
4510         return 0;
4511 }
4512
4513 /* Helper function for writing quotas on sync - we need to start transaction
4514  * before quota file is locked for write. Otherwise the are possible deadlocks:
4515  * Process 1                         Process 2
4516  * ext4_create()                     quota_sync()
4517  *   jbd2_journal_start()                  write_dquot()
4518  *   dquot_initialize()                         down(dqio_mutex)
4519  *     down(dqio_mutex)                    jbd2_journal_start()
4520  *
4521  */
4522
4523 #ifdef CONFIG_QUOTA
4524
4525 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4526 {
4527         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4528 }
4529
4530 static int ext4_write_dquot(struct dquot *dquot)
4531 {
4532         int ret, err;
4533         handle_t *handle;
4534         struct inode *inode;
4535
4536         inode = dquot_to_inode(dquot);
4537         handle = ext4_journal_start(inode,
4538                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4539         if (IS_ERR(handle))
4540                 return PTR_ERR(handle);
4541         ret = dquot_commit(dquot);
4542         err = ext4_journal_stop(handle);
4543         if (!ret)
4544                 ret = err;
4545         return ret;
4546 }
4547
4548 static int ext4_acquire_dquot(struct dquot *dquot)
4549 {
4550         int ret, err;
4551         handle_t *handle;
4552
4553         handle = ext4_journal_start(dquot_to_inode(dquot),
4554                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4555         if (IS_ERR(handle))
4556                 return PTR_ERR(handle);
4557         ret = dquot_acquire(dquot);
4558         err = ext4_journal_stop(handle);
4559         if (!ret)
4560                 ret = err;
4561         return ret;
4562 }
4563
4564 static int ext4_release_dquot(struct dquot *dquot)
4565 {
4566         int ret, err;
4567         handle_t *handle;
4568
4569         handle = ext4_journal_start(dquot_to_inode(dquot),
4570                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4571         if (IS_ERR(handle)) {
4572                 /* Release dquot anyway to avoid endless cycle in dqput() */
4573                 dquot_release(dquot);
4574                 return PTR_ERR(handle);
4575         }
4576         ret = dquot_release(dquot);
4577         err = ext4_journal_stop(handle);
4578         if (!ret)
4579                 ret = err;
4580         return ret;
4581 }
4582
4583 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4584 {
4585         /* Are we journaling quotas? */
4586         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4587             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4588                 dquot_mark_dquot_dirty(dquot);
4589                 return ext4_write_dquot(dquot);
4590         } else {
4591                 return dquot_mark_dquot_dirty(dquot);
4592         }
4593 }
4594
4595 static int ext4_write_info(struct super_block *sb, int type)
4596 {
4597         int ret, err;
4598         handle_t *handle;
4599
4600         /* Data block + inode block */
4601         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4602         if (IS_ERR(handle))
4603                 return PTR_ERR(handle);
4604         ret = dquot_commit_info(sb, type);
4605         err = ext4_journal_stop(handle);
4606         if (!ret)
4607                 ret = err;
4608         return ret;
4609 }
4610
4611 /*
4612  * Turn on quotas during mount time - we need to find
4613  * the quota file and such...
4614  */
4615 static int ext4_quota_on_mount(struct super_block *sb, int type)
4616 {
4617         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4618                                         EXT4_SB(sb)->s_jquota_fmt, type);
4619 }
4620
4621 /*
4622  * Standard function to be called on quota_on
4623  */
4624 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4625                          struct path *path)
4626 {
4627         int err;
4628
4629         if (!test_opt(sb, QUOTA))
4630                 return -EINVAL;
4631
4632         /* Quotafile not on the same filesystem? */
4633         if (path->dentry->d_sb != sb)
4634                 return -EXDEV;
4635         /* Journaling quota? */
4636         if (EXT4_SB(sb)->s_qf_names[type]) {
4637                 /* Quotafile not in fs root? */
4638                 if (path->dentry->d_parent != sb->s_root)
4639                         ext4_msg(sb, KERN_WARNING,
4640                                 "Quota file not on filesystem root. "
4641                                 "Journaled quota will not work");
4642         }
4643
4644         /*
4645          * When we journal data on quota file, we have to flush journal to see
4646          * all updates to the file when we bypass pagecache...
4647          */
4648         if (EXT4_SB(sb)->s_journal &&
4649             ext4_should_journal_data(path->dentry->d_inode)) {
4650                 /*
4651                  * We don't need to lock updates but journal_flush() could
4652                  * otherwise be livelocked...
4653                  */
4654                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4655                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4656                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4657                 if (err)
4658                         return err;
4659         }
4660
4661         return dquot_quota_on(sb, type, format_id, path);
4662 }
4663
4664 static int ext4_quota_off(struct super_block *sb, int type)
4665 {
4666         struct inode *inode = sb_dqopt(sb)->files[type];
4667         handle_t *handle;
4668
4669         /* Force all delayed allocation blocks to be allocated.
4670          * Caller already holds s_umount sem */
4671         if (test_opt(sb, DELALLOC))
4672                 sync_filesystem(sb);
4673
4674         if (!inode)
4675                 goto out;
4676
4677         /* Update modification times of quota files when userspace can
4678          * start looking at them */
4679         handle = ext4_journal_start(inode, 1);
4680         if (IS_ERR(handle))
4681                 goto out;
4682         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4683         ext4_mark_inode_dirty(handle, inode);
4684         ext4_journal_stop(handle);
4685
4686 out:
4687         return dquot_quota_off(sb, type);
4688 }
4689
4690 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4691  * acquiring the locks... As quota files are never truncated and quota code
4692  * itself serializes the operations (and no one else should touch the files)
4693  * we don't have to be afraid of races */
4694 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4695                                size_t len, loff_t off)
4696 {
4697         struct inode *inode = sb_dqopt(sb)->files[type];
4698         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4699         int err = 0;
4700         int offset = off & (sb->s_blocksize - 1);
4701         int tocopy;
4702         size_t toread;
4703         struct buffer_head *bh;
4704         loff_t i_size = i_size_read(inode);
4705
4706         if (off > i_size)
4707                 return 0;
4708         if (off+len > i_size)
4709                 len = i_size-off;
4710         toread = len;
4711         while (toread > 0) {
4712                 tocopy = sb->s_blocksize - offset < toread ?
4713                                 sb->s_blocksize - offset : toread;
4714                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4715                 if (err)
4716                         return err;
4717                 if (!bh)        /* A hole? */
4718                         memset(data, 0, tocopy);
4719                 else
4720                         memcpy(data, bh->b_data+offset, tocopy);
4721                 brelse(bh);
4722                 offset = 0;
4723                 toread -= tocopy;
4724                 data += tocopy;
4725                 blk++;
4726         }
4727         return len;
4728 }
4729
4730 /* Write to quotafile (we know the transaction is already started and has
4731  * enough credits) */
4732 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4733                                 const char *data, size_t len, loff_t off)
4734 {
4735         struct inode *inode = sb_dqopt(sb)->files[type];
4736         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4737         int err = 0;
4738         int offset = off & (sb->s_blocksize - 1);
4739         struct buffer_head *bh;
4740         handle_t *handle = journal_current_handle();
4741
4742         if (EXT4_SB(sb)->s_journal && !handle) {
4743                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4744                         " cancelled because transaction is not started",
4745                         (unsigned long long)off, (unsigned long long)len);
4746                 return -EIO;
4747         }
4748         /*
4749          * Since we account only one data block in transaction credits,
4750          * then it is impossible to cross a block boundary.
4751          */
4752         if (sb->s_blocksize - offset < len) {
4753                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4754                         " cancelled because not block aligned",
4755                         (unsigned long long)off, (unsigned long long)len);
4756                 return -EIO;
4757         }
4758
4759         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4760         bh = ext4_bread(handle, inode, blk, 1, &err);
4761         if (!bh)
4762                 goto out;
4763         err = ext4_journal_get_write_access(handle, bh);
4764         if (err) {
4765                 brelse(bh);
4766                 goto out;
4767         }
4768         lock_buffer(bh);
4769         memcpy(bh->b_data+offset, data, len);
4770         flush_dcache_page(bh->b_page);
4771         unlock_buffer(bh);
4772         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4773         brelse(bh);
4774 out:
4775         if (err) {
4776                 mutex_unlock(&inode->i_mutex);
4777                 return err;
4778         }
4779         if (inode->i_size < off + len) {
4780                 i_size_write(inode, off + len);
4781                 EXT4_I(inode)->i_disksize = inode->i_size;
4782                 ext4_mark_inode_dirty(handle, inode);
4783         }
4784         mutex_unlock(&inode->i_mutex);
4785         return len;
4786 }
4787
4788 #endif
4789
4790 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4791                        const char *dev_name, void *data)
4792 {
4793         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4794 }
4795
4796 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4797 static inline void register_as_ext2(void)
4798 {
4799         int err = register_filesystem(&ext2_fs_type);
4800         if (err)
4801                 printk(KERN_WARNING
4802                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4803 }
4804
4805 static inline void unregister_as_ext2(void)
4806 {
4807         unregister_filesystem(&ext2_fs_type);
4808 }
4809
4810 static inline int ext2_feature_set_ok(struct super_block *sb)
4811 {
4812         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4813                 return 0;
4814         if (sb->s_flags & MS_RDONLY)
4815                 return 1;
4816         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4817                 return 0;
4818         return 1;
4819 }
4820 MODULE_ALIAS("ext2");
4821 #else
4822 static inline void register_as_ext2(void) { }
4823 static inline void unregister_as_ext2(void) { }
4824 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4825 #endif
4826
4827 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4828 static inline void register_as_ext3(void)
4829 {
4830         int err = register_filesystem(&ext3_fs_type);
4831         if (err)
4832                 printk(KERN_WARNING
4833                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4834 }
4835
4836 static inline void unregister_as_ext3(void)
4837 {
4838         unregister_filesystem(&ext3_fs_type);
4839 }
4840
4841 static inline int ext3_feature_set_ok(struct super_block *sb)
4842 {
4843         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4844                 return 0;
4845         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4846                 return 0;
4847         if (sb->s_flags & MS_RDONLY)
4848                 return 1;
4849         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4850                 return 0;
4851         return 1;
4852 }
4853 MODULE_ALIAS("ext3");
4854 #else
4855 static inline void register_as_ext3(void) { }
4856 static inline void unregister_as_ext3(void) { }
4857 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4858 #endif
4859
4860 static struct file_system_type ext4_fs_type = {
4861         .owner          = THIS_MODULE,
4862         .name           = "ext4",
4863         .mount          = ext4_mount,
4864         .kill_sb        = kill_block_super,
4865         .fs_flags       = FS_REQUIRES_DEV,
4866 };
4867
4868 static int __init ext4_init_feat_adverts(void)
4869 {
4870         struct ext4_features *ef;
4871         int ret = -ENOMEM;
4872
4873         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4874         if (!ef)
4875                 goto out;
4876
4877         ef->f_kobj.kset = ext4_kset;
4878         init_completion(&ef->f_kobj_unregister);
4879         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4880                                    "features");
4881         if (ret) {
4882                 kfree(ef);
4883                 goto out;
4884         }
4885
4886         ext4_feat = ef;
4887         ret = 0;
4888 out:
4889         return ret;
4890 }
4891
4892 static void ext4_exit_feat_adverts(void)
4893 {
4894         kobject_put(&ext4_feat->f_kobj);
4895         wait_for_completion(&ext4_feat->f_kobj_unregister);
4896         kfree(ext4_feat);
4897 }
4898
4899 /* Shared across all ext4 file systems */
4900 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4901 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4902
4903 static int __init ext4_init_fs(void)
4904 {
4905         int i, err;
4906
4907         ext4_check_flag_values();
4908
4909         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4910                 mutex_init(&ext4__aio_mutex[i]);
4911                 init_waitqueue_head(&ext4__ioend_wq[i]);
4912         }
4913
4914         err = ext4_init_pageio();
4915         if (err)
4916                 return err;
4917         err = ext4_init_system_zone();
4918         if (err)
4919                 goto out6;
4920         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4921         if (!ext4_kset)
4922                 goto out5;
4923         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4924
4925         err = ext4_init_feat_adverts();
4926         if (err)
4927                 goto out4;
4928
4929         err = ext4_init_mballoc();
4930         if (err)
4931                 goto out3;
4932
4933         err = ext4_init_xattr();
4934         if (err)
4935                 goto out2;
4936         err = init_inodecache();
4937         if (err)
4938                 goto out1;
4939         register_as_ext3();
4940         register_as_ext2();
4941         err = register_filesystem(&ext4_fs_type);
4942         if (err)
4943                 goto out;
4944
4945         ext4_li_info = NULL;
4946         mutex_init(&ext4_li_mtx);
4947         return 0;
4948 out:
4949         unregister_as_ext2();
4950         unregister_as_ext3();
4951         destroy_inodecache();
4952 out1:
4953         ext4_exit_xattr();
4954 out2:
4955         ext4_exit_mballoc();
4956 out3:
4957         ext4_exit_feat_adverts();
4958 out4:
4959         if (ext4_proc_root)
4960                 remove_proc_entry("fs/ext4", NULL);
4961         kset_unregister(ext4_kset);
4962 out5:
4963         ext4_exit_system_zone();
4964 out6:
4965         ext4_exit_pageio();
4966         return err;
4967 }
4968
4969 static void __exit ext4_exit_fs(void)
4970 {
4971         ext4_destroy_lazyinit_thread();
4972         unregister_as_ext2();
4973         unregister_as_ext3();
4974         unregister_filesystem(&ext4_fs_type);
4975         destroy_inodecache();
4976         ext4_exit_xattr();
4977         ext4_exit_mballoc();
4978         ext4_exit_feat_adverts();
4979         remove_proc_entry("fs/ext4", NULL);
4980         kset_unregister(ext4_kset);
4981         ext4_exit_system_zone();
4982         ext4_exit_pageio();
4983 }
4984
4985 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4986 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4987 MODULE_LICENSE("GPL");
4988 module_init(ext4_init_fs)
4989 module_exit(ext4_exit_fs)