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_commit_super(struct super_block *sb, int sync);
  66 static void ext4_mark_recovery_complete(struct super_block *sb,
  67                                         struct ext4_super_block *es);
  68 static void ext4_clear_journal_err(struct super_block *sb,
  69                                    struct ext4_super_block *es);
  70 static int ext4_sync_fs(struct super_block *sb, int wait);
  71 static const char *ext4_decode_error(struct super_block *sb, int errno,
  72                                      char nbuf[16]);
  73 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  75 static int ext4_unfreeze(struct super_block *sb);
  76 static void ext4_write_super(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  96 #else
  97 #define IS_EXT2_SB(sb) (0)
  98 #endif
  99
 100
 101 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 102 static struct file_system_type ext3_fs_type = {
 103         .owner          = THIS_MODULE,
 104         .name           = "ext3",
 105         .mount          = ext4_mount,
 106         .kill_sb        = kill_block_super,
 107         .fs_flags       = FS_REQUIRES_DEV,
 108 };
 109 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 110 #else
 111 #define IS_EXT3_SB(sb) (0)
 112 #endif
 113
 114 void *ext4_kvmalloc(size_t size, gfp_t flags)
 115 {
 116         void *ret;
 117
 118         ret = kmalloc(size, flags);
 119         if (!ret)
 120                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 121         return ret;
 122 }
 123
 124 void *ext4_kvzalloc(size_t size, gfp_t flags)
 125 {
 126         void *ret;
 127
 128         ret = kzalloc(size, flags);
 129         if (!ret)
 130                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 131         return ret;
 132 }
 133
 134 void ext4_kvfree(void *ptr)
 135 {
 136         if (is_vmalloc_addr(ptr))
 137                 vfree(ptr);
 138         else
 139                 kfree(ptr);
 140
 141 }
 142
 143 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 144                                struct ext4_group_desc *bg)
 145 {
 146         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 147                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 148                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 149 }
 150
 151 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 152                                struct ext4_group_desc *bg)
 153 {
 154         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 155                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 156                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 157 }
 158
 159 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 160                               struct ext4_group_desc *bg)
 161 {
 162         return le32_to_cpu(bg->bg_inode_table_lo) |
 163                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 164                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 165 }
 166
 167 __u32 ext4_free_group_clusters(struct super_block *sb,
 168                                struct ext4_group_desc *bg)
 169 {
 170         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 171                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 172                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 173 }
 174
 175 __u32 ext4_free_inodes_count(struct super_block *sb,
 176                               struct ext4_group_desc *bg)
 177 {
 178         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 179                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 180                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 181 }
 182
 183 __u32 ext4_used_dirs_count(struct super_block *sb,
 184                               struct ext4_group_desc *bg)
 185 {
 186         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 187                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 188                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 189 }
 190
 191 __u32 ext4_itable_unused_count(struct super_block *sb,
 192                               struct ext4_group_desc *bg)
 193 {
 194         return le16_to_cpu(bg->bg_itable_unused_lo) |
 195                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 196                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 197 }
 198
 199 void ext4_block_bitmap_set(struct super_block *sb,
 200                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 201 {
 202         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 203         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 204                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 205 }
 206
 207 void ext4_inode_bitmap_set(struct super_block *sb,
 208                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 209 {
 210         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 211         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 212                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 213 }
 214
 215 void ext4_inode_table_set(struct super_block *sb,
 216                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 217 {
 218         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 219         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 220                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 221 }
 222
 223 void ext4_free_group_clusters_set(struct super_block *sb,
 224                                   struct ext4_group_desc *bg, __u32 count)
 225 {
 226         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 227         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 228                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 229 }
 230
 231 void ext4_free_inodes_set(struct super_block *sb,
 232                           struct ext4_group_desc *bg, __u32 count)
 233 {
 234         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 235         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 236                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 237 }
 238
 239 void ext4_used_dirs_set(struct super_block *sb,
 240                           struct ext4_group_desc *bg, __u32 count)
 241 {
 242         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 243         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 244                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 245 }
 246
 247 void ext4_itable_unused_set(struct super_block *sb,
 248                           struct ext4_group_desc *bg, __u32 count)
 249 {
 250         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 251         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 252                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 253 }
 254
 255
 256 /* Just increment the non-pointer handle value */
 257 static handle_t *ext4_get_nojournal(void)
 258 {
 259         handle_t *handle = current->journal_info;
 260         unsigned long ref_cnt = (unsigned long)handle;
 261
 262         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 263
 264         ref_cnt++;
 265         handle = (handle_t *)ref_cnt;
 266
 267         current->journal_info = handle;
 268         return handle;
 269 }
 270
 271
 272 /* Decrement the non-pointer handle value */
 273 static void ext4_put_nojournal(handle_t *handle)
 274 {
 275         unsigned long ref_cnt = (unsigned long)handle;
 276
 277         BUG_ON(ref_cnt == 0);
 278
 279         ref_cnt--;
 280         handle = (handle_t *)ref_cnt;
 281
 282         current->journal_info = handle;
 283 }
 284
 285 /*
 286  * Wrappers for jbd2_journal_start/end.
 287  *
 288  * The only special thing we need to do here is to make sure that all
 289  * journal_end calls result in the superblock being marked dirty, so
 290  * that sync() will call the filesystem's write_super callback if
 291  * appropriate.
 292  *
 293  * To avoid j_barrier hold in userspace when a user calls freeze(),
 294  * ext4 prevents a new handle from being started by s_frozen, which
 295  * is in an upper layer.
 296  */
 297 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 298 {
 299         journal_t *journal;
 300         handle_t  *handle;
 301
 302         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 303         if (sb->s_flags & MS_RDONLY)
 304                 return ERR_PTR(-EROFS);
 305
 306         journal = EXT4_SB(sb)->s_journal;
 307         handle = ext4_journal_current_handle();
 308
 309         /*
 310          * If a handle has been started, it should be allowed to
 311          * finish, otherwise deadlock could happen between freeze
 312          * and others(e.g. truncate) due to the restart of the
 313          * journal handle if the filesystem is forzen and active
 314          * handles are not stopped.
 315          */
 316         if (!handle)
 317                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 318
 319         if (!journal)
 320                 return ext4_get_nojournal();
 321         /*
 322          * Special case here: if the journal has aborted behind our
 323          * backs (eg. EIO in the commit thread), then we still need to
 324          * take the FS itself readonly cleanly.
 325          */
 326         if (is_journal_aborted(journal)) {
 327                 ext4_abort(sb, "Detected aborted journal");
 328                 return ERR_PTR(-EROFS);
 329         }
 330         return jbd2_journal_start(journal, nblocks);
 331 }
 332
 333 /*
 334  * The only special thing we need to do here is to make sure that all
 335  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 336  * that sync() will call the filesystem's write_super callback if
 337  * appropriate.
 338  */
 339 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 340 {
 341         struct super_block *sb;
 342         int err;
 343         int rc;
 344
 345         if (!ext4_handle_valid(handle)) {
 346                 ext4_put_nojournal(handle);
 347                 return 0;
 348         }
 349         sb = handle->h_transaction->t_journal->j_private;
 350         err = handle->h_err;
 351         rc = jbd2_journal_stop(handle);
 352
 353         if (!err)
 354                 err = rc;
 355         if (err)
 356                 __ext4_std_error(sb, where, line, err);
 357         return err;
 358 }
 359
 360 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 361                                const char *err_fn, struct buffer_head *bh,
 362                                handle_t *handle, int err)
 363 {
 364         char nbuf[16];
 365         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 366
 367         BUG_ON(!ext4_handle_valid(handle));
 368
 369         if (bh)
 370                 BUFFER_TRACE(bh, "abort");
 371
 372         if (!handle->h_err)
 373                 handle->h_err = err;
 374
 375         if (is_handle_aborted(handle))
 376                 return;
 377
 378         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 379                caller, line, errstr, err_fn);
 380
 381         jbd2_journal_abort_handle(handle);
 382 }
 383
 384 static void __save_error_info(struct super_block *sb, const char *func,
 385                             unsigned int line)
 386 {
 387         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 388
 389         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 390         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 391         es->s_last_error_time = cpu_to_le32(get_seconds());
 392         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 393         es->s_last_error_line = cpu_to_le32(line);
 394         if (!es->s_first_error_time) {
 395                 es->s_first_error_time = es->s_last_error_time;
 396                 strncpy(es->s_first_error_func, func,
 397                         sizeof(es->s_first_error_func));
 398                 es->s_first_error_line = cpu_to_le32(line);
 399                 es->s_first_error_ino = es->s_last_error_ino;
 400                 es->s_first_error_block = es->s_last_error_block;
 401         }
 402         /*
 403          * Start the daily error reporting function if it hasn't been
 404          * started already
 405          */
 406         if (!es->s_error_count)
 407                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 408         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 409 }
 410
 411 static void save_error_info(struct super_block *sb, const char *func,
 412                             unsigned int line)
 413 {
 414         __save_error_info(sb, func, line);
 415         ext4_commit_super(sb, 1);
 416 }
 417
 418 /*
 419  * The del_gendisk() function uninitializes the disk-specific data
 420  * structures, including the bdi structure, without telling anyone
 421  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 422  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 423  * This is a kludge to prevent these oops until we can put in a proper
 424  * hook in del_gendisk() to inform the VFS and file system layers.
 425  */
 426 static int block_device_ejected(struct super_block *sb)
 427 {
 428         struct inode *bd_inode = sb->s_bdev->bd_inode;
 429         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 430
 431         return bdi->dev == NULL;
 432 }
 433
 434 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 435 {
 436         struct super_block              *sb = journal->j_private;
 437         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 438         int                             error = is_journal_aborted(journal);
 439         struct ext4_journal_cb_entry    *jce, *tmp;
 440
 441         spin_lock(&sbi->s_md_lock);
 442         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 443                 list_del_init(&jce->jce_list);
 444                 spin_unlock(&sbi->s_md_lock);
 445                 jce->jce_func(sb, jce, error);
 446                 spin_lock(&sbi->s_md_lock);
 447         }
 448         spin_unlock(&sbi->s_md_lock);
 449 }
 450
 451 /* Deal with the reporting of failure conditions on a filesystem such as
 452  * inconsistencies detected or read IO failures.
 453  *
 454  * On ext2, we can store the error state of the filesystem in the
 455  * superblock.  That is not possible on ext4, because we may have other
 456  * write ordering constraints on the superblock which prevent us from
 457  * writing it out straight away; and given that the journal is about to
 458  * be aborted, we can't rely on the current, or future, transactions to
 459  * write out the superblock safely.
 460  *
 461  * We'll just use the jbd2_journal_abort() error code to record an error in
 462  * the journal instead.  On recovery, the journal will complain about
 463  * that error until we've noted it down and cleared it.
 464  */
 465
 466 static void ext4_handle_error(struct super_block *sb)
 467 {
 468         if (sb->s_flags & MS_RDONLY)
 469                 return;
 470
 471         if (!test_opt(sb, ERRORS_CONT)) {
 472                 journal_t *journal = EXT4_SB(sb)->s_journal;
 473
 474                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 475                 if (journal)
 476                         jbd2_journal_abort(journal, -EIO);
 477         }
 478         if (test_opt(sb, ERRORS_RO)) {
 479                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 480                 sb->s_flags |= MS_RDONLY;
 481         }
 482         if (test_opt(sb, ERRORS_PANIC))
 483                 panic("EXT4-fs (device %s): panic forced after error\n",
 484                         sb->s_id);
 485 }
 486
 487 void __ext4_error(struct super_block *sb, const char *function,
 488                   unsigned int line, const char *fmt, ...)
 489 {
 490         struct va_format vaf;
 491         va_list args;
 492
 493         va_start(args, fmt);
 494         vaf.fmt = fmt;
 495         vaf.va = &args;
 496         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 497                sb->s_id, function, line, current->comm, &vaf);
 498         va_end(args);
 499
 500         ext4_handle_error(sb);
 501 }
 502
 503 void ext4_error_inode(struct inode *inode, const char *function,
 504                       unsigned int line, ext4_fsblk_t block,
 505                       const char *fmt, ...)
 506 {
 507         va_list args;
 508         struct va_format vaf;
 509         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 510
 511         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 512         es->s_last_error_block = cpu_to_le64(block);
 513         save_error_info(inode->i_sb, function, line);
 514         va_start(args, fmt);
 515         vaf.fmt = fmt;
 516         vaf.va = &args;
 517         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 518                inode->i_sb->s_id, function, line, inode->i_ino);
 519         if (block)
 520                 printk(KERN_CONT "block %llu: ", block);
 521         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 522         va_end(args);
 523
 524         ext4_handle_error(inode->i_sb);
 525 }
 526
 527 void ext4_error_file(struct file *file, const char *function,
 528                      unsigned int line, ext4_fsblk_t block,
 529                      const char *fmt, ...)
 530 {
 531         va_list args;
 532         struct va_format vaf;
 533         struct ext4_super_block *es;
 534         struct inode *inode = file->f_dentry->d_inode;
 535         char pathname[80], *path;
 536
 537         es = EXT4_SB(inode->i_sb)->s_es;
 538         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 539         save_error_info(inode->i_sb, function, line);
 540         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 541         if (IS_ERR(path))
 542                 path = "(unknown)";
 543         printk(KERN_CRIT
 544                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 545                inode->i_sb->s_id, function, line, inode->i_ino);
 546         if (block)
 547                 printk(KERN_CONT "block %llu: ", block);
 548         va_start(args, fmt);
 549         vaf.fmt = fmt;
 550         vaf.va = &args;
 551         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 552         va_end(args);
 553
 554         ext4_handle_error(inode->i_sb);
 555 }
 556
 557 static const char *ext4_decode_error(struct super_block *sb, int errno,
 558                                      char nbuf[16])
 559 {
 560         char *errstr = NULL;
 561
 562         switch (errno) {
 563         case -EIO:
 564                 errstr = "IO failure";
 565                 break;
 566         case -ENOMEM:
 567                 errstr = "Out of memory";
 568                 break;
 569         case -EROFS:
 570                 if (!sb || (EXT4_SB(sb)->s_journal &&
 571                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 572                         errstr = "Journal has aborted";
 573                 else
 574                         errstr = "Readonly filesystem";
 575                 break;
 576         default:
 577                 /* If the caller passed in an extra buffer for unknown
 578                  * errors, textualise them now.  Else we just return
 579                  * NULL. */
 580                 if (nbuf) {
 581                         /* Check for truncated error codes... */
 582                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 583                                 errstr = nbuf;
 584                 }
 585                 break;
 586         }
 587
 588         return errstr;
 589 }
 590
 591 /* __ext4_std_error decodes expected errors from journaling functions
 592  * automatically and invokes the appropriate error response.  */
 593
 594 void __ext4_std_error(struct super_block *sb, const char *function,
 595                       unsigned int line, int errno)
 596 {
 597         char nbuf[16];
 598         const char *errstr;
 599
 600         /* Special case: if the error is EROFS, and we're not already
 601          * inside a transaction, then there's really no point in logging
 602          * an error. */
 603         if (errno == -EROFS && journal_current_handle() == NULL &&
 604             (sb->s_flags & MS_RDONLY))
 605                 return;
 606
 607         errstr = ext4_decode_error(sb, errno, nbuf);
 608         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 609                sb->s_id, function, line, errstr);
 610         save_error_info(sb, function, line);
 611
 612         ext4_handle_error(sb);
 613 }
 614
 615 /*
 616  * ext4_abort is a much stronger failure handler than ext4_error.  The
 617  * abort function may be used to deal with unrecoverable failures such
 618  * as journal IO errors or ENOMEM at a critical moment in log management.
 619  *
 620  * We unconditionally force the filesystem into an ABORT|READONLY state,
 621  * unless the error response on the fs has been set to panic in which
 622  * case we take the easy way out and panic immediately.
 623  */
 624
 625 void __ext4_abort(struct super_block *sb, const char *function,
 626                 unsigned int line, const char *fmt, ...)
 627 {
 628         va_list args;
 629
 630         save_error_info(sb, function, line);
 631         va_start(args, fmt);
 632         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 633                function, line);
 634         vprintk(fmt, args);
 635         printk("\n");
 636         va_end(args);
 637
 638         if ((sb->s_flags & MS_RDONLY) == 0) {
 639                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 640                 sb->s_flags |= MS_RDONLY;
 641                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 642                 if (EXT4_SB(sb)->s_journal)
 643                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 644                 save_error_info(sb, function, line);
 645         }
 646         if (test_opt(sb, ERRORS_PANIC))
 647                 panic("EXT4-fs panic from previous error\n");
 648 }
 649
 650 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 651 {
 652         struct va_format vaf;
 653         va_list args;
 654
 655         va_start(args, fmt);
 656         vaf.fmt = fmt;
 657         vaf.va = &args;
 658         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 659         va_end(args);
 660 }
 661
 662 void __ext4_warning(struct super_block *sb, const char *function,
 663                     unsigned int line, const char *fmt, ...)
 664 {
 665         struct va_format vaf;
 666         va_list args;
 667
 668         va_start(args, fmt);
 669         vaf.fmt = fmt;
 670         vaf.va = &args;
 671         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 672                sb->s_id, function, line, &vaf);
 673         va_end(args);
 674 }
 675
 676 void __ext4_grp_locked_error(const char *function, unsigned int line,
 677                              struct super_block *sb, ext4_group_t grp,
 678                              unsigned long ino, ext4_fsblk_t block,
 679                              const char *fmt, ...)
 680 __releases(bitlock)
 681 __acquires(bitlock)
 682 {
 683         struct va_format vaf;
 684         va_list args;
 685         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 686
 687         es->s_last_error_ino = cpu_to_le32(ino);
 688         es->s_last_error_block = cpu_to_le64(block);
 689         __save_error_info(sb, function, line);
 690
 691         va_start(args, fmt);
 692
 693         vaf.fmt = fmt;
 694         vaf.va = &args;
 695         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 696                sb->s_id, function, line, grp);
 697         if (ino)
 698                 printk(KERN_CONT "inode %lu: ", ino);
 699         if (block)
 700                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 701         printk(KERN_CONT "%pV\n", &vaf);
 702         va_end(args);
 703
 704         if (test_opt(sb, ERRORS_CONT)) {
 705                 ext4_commit_super(sb, 0);
 706                 return;
 707         }
 708
 709         ext4_unlock_group(sb, grp);
 710         ext4_handle_error(sb);
 711         /*
 712          * We only get here in the ERRORS_RO case; relocking the group
 713          * may be dangerous, but nothing bad will happen since the
 714          * filesystem will have already been marked read/only and the
 715          * journal has been aborted.  We return 1 as a hint to callers
 716          * who might what to use the return value from
 717          * ext4_grp_locked_error() to distinguish between the
 718          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 719          * aggressively from the ext4 function in question, with a
 720          * more appropriate error code.
 721          */
 722         ext4_lock_group(sb, grp);
 723         return;
 724 }
 725
 726 void ext4_update_dynamic_rev(struct super_block *sb)
 727 {
 728         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 729
 730         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 731                 return;
 732
 733         ext4_warning(sb,
 734                      "updating to rev %d because of new feature flag, "
 735                      "running e2fsck is recommended",
 736                      EXT4_DYNAMIC_REV);
 737
 738         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 739         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 740         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 741         /* leave es->s_feature_*compat flags alone */
 742         /* es->s_uuid will be set by e2fsck if empty */
 743
 744         /*
 745          * The rest of the superblock fields should be zero, and if not it
 746          * means they are likely already in use, so leave them alone.  We
 747          * can leave it up to e2fsck to clean up any inconsistencies there.
 748          */
 749 }
 750
 751 /*
 752  * Open the external journal device
 753  */
 754 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 755 {
 756         struct block_device *bdev;
 757         char b[BDEVNAME_SIZE];
 758
 759         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 760         if (IS_ERR(bdev))
 761                 goto fail;
 762         return bdev;
 763
 764 fail:
 765         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 766                         __bdevname(dev, b), PTR_ERR(bdev));
 767         return NULL;
 768 }
 769
 770 /*
 771  * Release the journal device
 772  */
 773 static int ext4_blkdev_put(struct block_device *bdev)
 774 {
 775         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 776 }
 777
 778 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 779 {
 780         struct block_device *bdev;
 781         int ret = -ENODEV;
 782
 783         bdev = sbi->journal_bdev;
 784         if (bdev) {
 785                 ret = ext4_blkdev_put(bdev);
 786                 sbi->journal_bdev = NULL;
 787         }
 788         return ret;
 789 }
 790
 791 static inline struct inode *orphan_list_entry(struct list_head *l)
 792 {
 793         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 794 }
 795
 796 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 797 {
 798         struct list_head *l;
 799
 800         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 801                  le32_to_cpu(sbi->s_es->s_last_orphan));
 802
 803         printk(KERN_ERR "sb_info orphan list:\n");
 804         list_for_each(l, &sbi->s_orphan) {
 805                 struct inode *inode = orphan_list_entry(l);
 806                 printk(KERN_ERR "  "
 807                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 808                        inode->i_sb->s_id, inode->i_ino, inode,
 809                        inode->i_mode, inode->i_nlink,
 810                        NEXT_ORPHAN(inode));
 811         }
 812 }
 813
 814 static void ext4_put_super(struct super_block *sb)
 815 {
 816         struct ext4_sb_info *sbi = EXT4_SB(sb);
 817         struct ext4_super_block *es = sbi->s_es;
 818         int i, err;
 819
 820         ext4_unregister_li_request(sb);
 821         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 822
 823         flush_workqueue(sbi->dio_unwritten_wq);
 824         destroy_workqueue(sbi->dio_unwritten_wq);
 825
 826         lock_super(sb);
 827         if (sb->s_dirt)
 828                 ext4_commit_super(sb, 1);
 829
 830         if (sbi->s_journal) {
 831                 err = jbd2_journal_destroy(sbi->s_journal);
 832                 sbi->s_journal = NULL;
 833                 if (err < 0)
 834                         ext4_abort(sb, "Couldn't clean up the journal");
 835         }
 836
 837         del_timer(&sbi->s_err_report);
 838         ext4_release_system_zone(sb);
 839         ext4_mb_release(sb);
 840         ext4_ext_release(sb);
 841         ext4_xattr_put_super(sb);
 842
 843         if (!(sb->s_flags & MS_RDONLY)) {
 844                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 845                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 846                 ext4_commit_super(sb, 1);
 847         }
 848         if (sbi->s_proc) {
 849                 remove_proc_entry(sb->s_id, ext4_proc_root);
 850         }
 851         kobject_del(&sbi->s_kobj);
 852
 853         for (i = 0; i < sbi->s_gdb_count; i++)
 854                 brelse(sbi->s_group_desc[i]);
 855         ext4_kvfree(sbi->s_group_desc);
 856         ext4_kvfree(sbi->s_flex_groups);
 857         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 858         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 859         percpu_counter_destroy(&sbi->s_dirs_counter);
 860         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 861         brelse(sbi->s_sbh);
 862 #ifdef CONFIG_QUOTA
 863         for (i = 0; i < MAXQUOTAS; i++)
 864                 kfree(sbi->s_qf_names[i]);
 865 #endif
 866
 867         /* Debugging code just in case the in-memory inode orphan list
 868          * isn't empty.  The on-disk one can be non-empty if we've
 869          * detected an error and taken the fs readonly, but the
 870          * in-memory list had better be clean by this point. */
 871         if (!list_empty(&sbi->s_orphan))
 872                 dump_orphan_list(sb, sbi);
 873         J_ASSERT(list_empty(&sbi->s_orphan));
 874
 875         invalidate_bdev(sb->s_bdev);
 876         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 877                 /*
 878                  * Invalidate the journal device's buffers.  We don't want them
 879                  * floating about in memory - the physical journal device may
 880                  * hotswapped, and it breaks the `ro-after' testing code.
 881                  */
 882                 sync_blockdev(sbi->journal_bdev);
 883                 invalidate_bdev(sbi->journal_bdev);
 884                 ext4_blkdev_remove(sbi);
 885         }
 886         if (sbi->s_mmp_tsk)
 887                 kthread_stop(sbi->s_mmp_tsk);
 888         sb->s_fs_info = NULL;
 889         /*
 890          * Now that we are completely done shutting down the
 891          * superblock, we need to actually destroy the kobject.
 892          */
 893         unlock_super(sb);
 894         kobject_put(&sbi->s_kobj);
 895         wait_for_completion(&sbi->s_kobj_unregister);
 896         kfree(sbi->s_blockgroup_lock);
 897         kfree(sbi);
 898 }
 899
 900 static struct kmem_cache *ext4_inode_cachep;
 901
 902 /*
 903  * Called inside transaction, so use GFP_NOFS
 904  */
 905 static struct inode *ext4_alloc_inode(struct super_block *sb)
 906 {
 907         struct ext4_inode_info *ei;
 908
 909         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 910         if (!ei)
 911                 return NULL;
 912
 913         ei->vfs_inode.i_version = 1;
 914         ei->vfs_inode.i_data.writeback_index = 0;
 915         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 916         INIT_LIST_HEAD(&ei->i_prealloc_list);
 917         spin_lock_init(&ei->i_prealloc_lock);
 918         ei->i_reserved_data_blocks = 0;
 919         ei->i_reserved_meta_blocks = 0;
 920         ei->i_allocated_meta_blocks = 0;
 921         ei->i_da_metadata_calc_len = 0;
 922         spin_lock_init(&(ei->i_block_reservation_lock));
 923 #ifdef CONFIG_QUOTA
 924         ei->i_reserved_quota = 0;
 925 #endif
 926         ei->jinode = NULL;
 927         INIT_LIST_HEAD(&ei->i_completed_io_list);
 928         spin_lock_init(&ei->i_completed_io_lock);
 929         ei->cur_aio_dio = NULL;
 930         ei->i_sync_tid = 0;
 931         ei->i_datasync_tid = 0;
 932         atomic_set(&ei->i_ioend_count, 0);
 933         atomic_set(&ei->i_aiodio_unwritten, 0);
 934
 935         return &ei->vfs_inode;
 936 }
 937
 938 static int ext4_drop_inode(struct inode *inode)
 939 {
 940         int drop = generic_drop_inode(inode);
 941
 942         trace_ext4_drop_inode(inode, drop);
 943         return drop;
 944 }
 945
 946 static void ext4_i_callback(struct rcu_head *head)
 947 {
 948         struct inode *inode = container_of(head, struct inode, i_rcu);
 949         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 950 }
 951
 952 static void ext4_destroy_inode(struct inode *inode)
 953 {
 954         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 955                 ext4_msg(inode->i_sb, KERN_ERR,
 956                          "Inode %lu (%p): orphan list check failed!",
 957                          inode->i_ino, EXT4_I(inode));
 958                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 959                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 960                                 true);
 961                 dump_stack();
 962         }
 963         call_rcu(&inode->i_rcu, ext4_i_callback);
 964 }
 965
 966 static void init_once(void *foo)
 967 {
 968         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 969
 970         INIT_LIST_HEAD(&ei->i_orphan);
 971 #ifdef CONFIG_EXT4_FS_XATTR
 972         init_rwsem(&ei->xattr_sem);
 973 #endif
 974         init_rwsem(&ei->i_data_sem);
 975         inode_init_once(&ei->vfs_inode);
 976 }
 977
 978 static int init_inodecache(void)
 979 {
 980         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 981                                              sizeof(struct ext4_inode_info),
 982                                              0, (SLAB_RECLAIM_ACCOUNT|
 983                                                 SLAB_MEM_SPREAD),
 984                                              init_once);
 985         if (ext4_inode_cachep == NULL)
 986                 return -ENOMEM;
 987         return 0;
 988 }
 989
 990 static void destroy_inodecache(void)
 991 {
 992         kmem_cache_destroy(ext4_inode_cachep);
 993 }
 994
 995 void ext4_clear_inode(struct inode *inode)
 996 {
 997         invalidate_inode_buffers(inode);
 998         end_writeback(inode);
 999         dquot_drop(inode);
1000         ext4_discard_preallocations(inode);
1001         if (EXT4_I(inode)->jinode) {
1002                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1003                                                EXT4_I(inode)->jinode);
1004                 jbd2_free_inode(EXT4_I(inode)->jinode);
1005                 EXT4_I(inode)->jinode = NULL;
1006         }
1007 }
1008
1009 static inline void ext4_show_quota_options(struct seq_file *seq,
1010                                            struct super_block *sb)
1011 {
1012 #if defined(CONFIG_QUOTA)
1013         struct ext4_sb_info *sbi = EXT4_SB(sb);
1014
1015         if (sbi->s_jquota_fmt) {
1016                 char *fmtname = "";
1017
1018                 switch (sbi->s_jquota_fmt) {
1019                 case QFMT_VFS_OLD:
1020                         fmtname = "vfsold";
1021                         break;
1022                 case QFMT_VFS_V0:
1023                         fmtname = "vfsv0";
1024                         break;
1025                 case QFMT_VFS_V1:
1026                         fmtname = "vfsv1";
1027                         break;
1028                 }
1029                 seq_printf(seq, ",jqfmt=%s", fmtname);
1030         }
1031
1032         if (sbi->s_qf_names[USRQUOTA])
1033                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1034
1035         if (sbi->s_qf_names[GRPQUOTA])
1036                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1037
1038         if (test_opt(sb, USRQUOTA))
1039                 seq_puts(seq, ",usrquota");
1040
1041         if (test_opt(sb, GRPQUOTA))
1042                 seq_puts(seq, ",grpquota");
1043 #endif
1044 }
1045
1046 /*
1047  * Show an option if
1048  *  - it's set to a non-default value OR
1049  *  - if the per-sb default is different from the global default
1050  */
1051 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1052 {
1053         int def_errors;
1054         unsigned long def_mount_opts;
1055         struct super_block *sb = root->d_sb;
1056         struct ext4_sb_info *sbi = EXT4_SB(sb);
1057         struct ext4_super_block *es = sbi->s_es;
1058
1059         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1060         def_errors     = le16_to_cpu(es->s_errors);
1061
1062         if (sbi->s_sb_block != 1)
1063                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1064         if (test_opt(sb, MINIX_DF))
1065                 seq_puts(seq, ",minixdf");
1066         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1067                 seq_puts(seq, ",grpid");
1068         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1069                 seq_puts(seq, ",nogrpid");
1070         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1071             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1072                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1073         }
1074         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1075             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1076                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1077         }
1078         if (test_opt(sb, ERRORS_RO)) {
1079                 if (def_errors == EXT4_ERRORS_PANIC ||
1080                     def_errors == EXT4_ERRORS_CONTINUE) {
1081                         seq_puts(seq, ",errors=remount-ro");
1082                 }
1083         }
1084         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1085                 seq_puts(seq, ",errors=continue");
1086         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1087                 seq_puts(seq, ",errors=panic");
1088         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1089                 seq_puts(seq, ",nouid32");
1090         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1091                 seq_puts(seq, ",debug");
1092 #ifdef CONFIG_EXT4_FS_XATTR
1093         if (test_opt(sb, XATTR_USER))
1094                 seq_puts(seq, ",user_xattr");
1095         if (!test_opt(sb, XATTR_USER))
1096                 seq_puts(seq, ",nouser_xattr");
1097 #endif
1098 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1099         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1100                 seq_puts(seq, ",acl");
1101         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1102                 seq_puts(seq, ",noacl");
1103 #endif
1104         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1105                 seq_printf(seq, ",commit=%u",
1106                            (unsigned) (sbi->s_commit_interval / HZ));
1107         }
1108         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1109                 seq_printf(seq, ",min_batch_time=%u",
1110                            (unsigned) sbi->s_min_batch_time);
1111         }
1112         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1113                 seq_printf(seq, ",max_batch_time=%u",
1114                            (unsigned) sbi->s_max_batch_time);
1115         }
1116
1117         /*
1118          * We're changing the default of barrier mount option, so
1119          * let's always display its mount state so it's clear what its
1120          * status is.
1121          */
1122         seq_puts(seq, ",barrier=");
1123         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1124         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1125                 seq_puts(seq, ",journal_async_commit");
1126         else if (test_opt(sb, JOURNAL_CHECKSUM))
1127                 seq_puts(seq, ",journal_checksum");
1128         if (sb->s_flags & MS_I_VERSION)
1129                 seq_puts(seq, ",i_version");
1130         if (!test_opt(sb, DELALLOC) &&
1131             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1132                 seq_puts(seq, ",nodelalloc");
1133
1134         if (!test_opt(sb, MBLK_IO_SUBMIT))
1135                 seq_puts(seq, ",nomblk_io_submit");
1136         if (sbi->s_stripe)
1137                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1138         /*
1139          * journal mode get enabled in different ways
1140          * So just print the value even if we didn't specify it
1141          */
1142         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1143                 seq_puts(seq, ",data=journal");
1144         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1145                 seq_puts(seq, ",data=ordered");
1146         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1147                 seq_puts(seq, ",data=writeback");
1148
1149         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1150                 seq_printf(seq, ",inode_readahead_blks=%u",
1151                            sbi->s_inode_readahead_blks);
1152
1153         if (test_opt(sb, DATA_ERR_ABORT))
1154                 seq_puts(seq, ",data_err=abort");
1155
1156         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1157                 seq_puts(seq, ",noauto_da_alloc");
1158
1159         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1160                 seq_puts(seq, ",discard");
1161
1162         if (test_opt(sb, NOLOAD))
1163                 seq_puts(seq, ",norecovery");
1164
1165         if (test_opt(sb, DIOREAD_NOLOCK))
1166                 seq_puts(seq, ",dioread_nolock");
1167
1168         if (test_opt(sb, BLOCK_VALIDITY) &&
1169             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1170                 seq_puts(seq, ",block_validity");
1171
1172         if (!test_opt(sb, INIT_INODE_TABLE))
1173                 seq_puts(seq, ",noinit_itable");
1174         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1175                 seq_printf(seq, ",init_itable=%u",
1176                            (unsigned) sbi->s_li_wait_mult);
1177
1178         ext4_show_quota_options(seq, sb);
1179
1180         return 0;
1181 }
1182
1183 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1184                                         u64 ino, u32 generation)
1185 {
1186         struct inode *inode;
1187
1188         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1189                 return ERR_PTR(-ESTALE);
1190         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1191                 return ERR_PTR(-ESTALE);
1192
1193         /* iget isn't really right if the inode is currently unallocated!!
1194          *
1195          * ext4_read_inode will return a bad_inode if the inode had been
1196          * deleted, so we should be safe.
1197          *
1198          * Currently we don't know the generation for parent directory, so
1199          * a generation of 0 means "accept any"
1200          */
1201         inode = ext4_iget(sb, ino);
1202         if (IS_ERR(inode))
1203                 return ERR_CAST(inode);
1204         if (generation && inode->i_generation != generation) {
1205                 iput(inode);
1206                 return ERR_PTR(-ESTALE);
1207         }
1208
1209         return inode;
1210 }
1211
1212 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1213                                         int fh_len, int fh_type)
1214 {
1215         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1216                                     ext4_nfs_get_inode);
1217 }
1218
1219 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1220                                         int fh_len, int fh_type)
1221 {
1222         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1223                                     ext4_nfs_get_inode);
1224 }
1225
1226 /*
1227  * Try to release metadata pages (indirect blocks, directories) which are
1228  * mapped via the block device.  Since these pages could have journal heads
1229  * which would prevent try_to_free_buffers() from freeing them, we must use
1230  * jbd2 layer's try_to_free_buffers() function to release them.
1231  */
1232 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1233                                  gfp_t wait)
1234 {
1235         journal_t *journal = EXT4_SB(sb)->s_journal;
1236
1237         WARN_ON(PageChecked(page));
1238         if (!page_has_buffers(page))
1239                 return 0;
1240         if (journal)
1241                 return jbd2_journal_try_to_free_buffers(journal, page,
1242                                                         wait & ~__GFP_WAIT);
1243         return try_to_free_buffers(page);
1244 }
1245
1246 #ifdef CONFIG_QUOTA
1247 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1248 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1249
1250 static int ext4_write_dquot(struct dquot *dquot);
1251 static int ext4_acquire_dquot(struct dquot *dquot);
1252 static int ext4_release_dquot(struct dquot *dquot);
1253 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1254 static int ext4_write_info(struct super_block *sb, int type);
1255 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1256                          struct path *path);
1257 static int ext4_quota_off(struct super_block *sb, int type);
1258 static int ext4_quota_on_mount(struct super_block *sb, int type);
1259 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1260                                size_t len, loff_t off);
1261 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1262                                 const char *data, size_t len, loff_t off);
1263
1264 static const struct dquot_operations ext4_quota_operations = {
1265         .get_reserved_space = ext4_get_reserved_space,
1266         .write_dquot    = ext4_write_dquot,
1267         .acquire_dquot  = ext4_acquire_dquot,
1268         .release_dquot  = ext4_release_dquot,
1269         .mark_dirty     = ext4_mark_dquot_dirty,
1270         .write_info     = ext4_write_info,
1271         .alloc_dquot    = dquot_alloc,
1272         .destroy_dquot  = dquot_destroy,
1273 };
1274
1275 static const struct quotactl_ops ext4_qctl_operations = {
1276         .quota_on       = ext4_quota_on,
1277         .quota_off      = ext4_quota_off,
1278         .quota_sync     = dquot_quota_sync,
1279         .get_info       = dquot_get_dqinfo,
1280         .set_info       = dquot_set_dqinfo,
1281         .get_dqblk      = dquot_get_dqblk,
1282         .set_dqblk      = dquot_set_dqblk
1283 };
1284 #endif
1285
1286 static const struct super_operations ext4_sops = {
1287         .alloc_inode    = ext4_alloc_inode,
1288         .destroy_inode  = ext4_destroy_inode,
1289         .write_inode    = ext4_write_inode,
1290         .dirty_inode    = ext4_dirty_inode,
1291         .drop_inode     = ext4_drop_inode,
1292         .evict_inode    = ext4_evict_inode,
1293         .put_super      = ext4_put_super,
1294         .sync_fs        = ext4_sync_fs,
1295         .freeze_fs      = ext4_freeze,
1296         .unfreeze_fs    = ext4_unfreeze,
1297         .statfs         = ext4_statfs,
1298         .remount_fs     = ext4_remount,
1299         .show_options   = ext4_show_options,
1300 #ifdef CONFIG_QUOTA
1301         .quota_read     = ext4_quota_read,
1302         .quota_write    = ext4_quota_write,
1303 #endif
1304         .bdev_try_to_free_page = bdev_try_to_free_page,
1305 };
1306
1307 static const struct super_operations ext4_nojournal_sops = {
1308         .alloc_inode    = ext4_alloc_inode,
1309         .destroy_inode  = ext4_destroy_inode,
1310         .write_inode    = ext4_write_inode,
1311         .dirty_inode    = ext4_dirty_inode,
1312         .drop_inode     = ext4_drop_inode,
1313         .evict_inode    = ext4_evict_inode,
1314         .write_super    = ext4_write_super,
1315         .put_super      = ext4_put_super,
1316         .statfs         = ext4_statfs,
1317         .remount_fs     = ext4_remount,
1318         .show_options   = ext4_show_options,
1319 #ifdef CONFIG_QUOTA
1320         .quota_read     = ext4_quota_read,
1321         .quota_write    = ext4_quota_write,
1322 #endif
1323         .bdev_try_to_free_page = bdev_try_to_free_page,
1324 };
1325
1326 static const struct export_operations ext4_export_ops = {
1327         .fh_to_dentry = ext4_fh_to_dentry,
1328         .fh_to_parent = ext4_fh_to_parent,
1329         .get_parent = ext4_get_parent,
1330 };
1331
1332 enum {
1333         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1334         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1335         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1336         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1337         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1338         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1339         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1340         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1341         Opt_data_err_abort, Opt_data_err_ignore,
1342         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1343         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1344         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1345         Opt_usrquota, Opt_grpquota, Opt_i_version,
1346         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1347         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1348         Opt_inode_readahead_blks, Opt_journal_ioprio,
1349         Opt_dioread_nolock, Opt_dioread_lock,
1350         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1351 };
1352
1353 static const match_table_t tokens = {
1354         {Opt_bsd_df, "bsddf"},
1355         {Opt_minix_df, "minixdf"},
1356         {Opt_grpid, "grpid"},
1357         {Opt_grpid, "bsdgroups"},
1358         {Opt_nogrpid, "nogrpid"},
1359         {Opt_nogrpid, "sysvgroups"},
1360         {Opt_resgid, "resgid=%u"},
1361         {Opt_resuid, "resuid=%u"},
1362         {Opt_sb, "sb=%u"},
1363         {Opt_err_cont, "errors=continue"},
1364         {Opt_err_panic, "errors=panic"},
1365         {Opt_err_ro, "errors=remount-ro"},
1366         {Opt_nouid32, "nouid32"},
1367         {Opt_debug, "debug"},
1368         {Opt_oldalloc, "oldalloc"},
1369         {Opt_orlov, "orlov"},
1370         {Opt_user_xattr, "user_xattr"},
1371         {Opt_nouser_xattr, "nouser_xattr"},
1372         {Opt_acl, "acl"},
1373         {Opt_noacl, "noacl"},
1374         {Opt_noload, "noload"},
1375         {Opt_noload, "norecovery"},
1376         {Opt_nobh, "nobh"},
1377         {Opt_bh, "bh"},
1378         {Opt_commit, "commit=%u"},
1379         {Opt_min_batch_time, "min_batch_time=%u"},
1380         {Opt_max_batch_time, "max_batch_time=%u"},
1381         {Opt_journal_dev, "journal_dev=%u"},
1382         {Opt_journal_checksum, "journal_checksum"},
1383         {Opt_journal_async_commit, "journal_async_commit"},
1384         {Opt_abort, "abort"},
1385         {Opt_data_journal, "data=journal"},
1386         {Opt_data_ordered, "data=ordered"},
1387         {Opt_data_writeback, "data=writeback"},
1388         {Opt_data_err_abort, "data_err=abort"},
1389         {Opt_data_err_ignore, "data_err=ignore"},
1390         {Opt_offusrjquota, "usrjquota="},
1391         {Opt_usrjquota, "usrjquota=%s"},
1392         {Opt_offgrpjquota, "grpjquota="},
1393         {Opt_grpjquota, "grpjquota=%s"},
1394         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1395         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1396         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1397         {Opt_grpquota, "grpquota"},
1398         {Opt_noquota, "noquota"},
1399         {Opt_quota, "quota"},
1400         {Opt_usrquota, "usrquota"},
1401         {Opt_barrier, "barrier=%u"},
1402         {Opt_barrier, "barrier"},
1403         {Opt_nobarrier, "nobarrier"},
1404         {Opt_i_version, "i_version"},
1405         {Opt_stripe, "stripe=%u"},
1406         {Opt_delalloc, "delalloc"},
1407         {Opt_nodelalloc, "nodelalloc"},
1408         {Opt_mblk_io_submit, "mblk_io_submit"},
1409         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1410         {Opt_block_validity, "block_validity"},
1411         {Opt_noblock_validity, "noblock_validity"},
1412         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1413         {Opt_journal_ioprio, "journal_ioprio=%u"},
1414         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1415         {Opt_auto_da_alloc, "auto_da_alloc"},
1416         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1417         {Opt_dioread_nolock, "dioread_nolock"},
1418         {Opt_dioread_lock, "dioread_lock"},
1419         {Opt_discard, "discard"},
1420         {Opt_nodiscard, "nodiscard"},
1421         {Opt_init_itable, "init_itable=%u"},
1422         {Opt_init_itable, "init_itable"},
1423         {Opt_noinit_itable, "noinit_itable"},
1424         {Opt_err, NULL},
1425 };
1426
1427 static ext4_fsblk_t get_sb_block(void **data)
1428 {
1429         ext4_fsblk_t    sb_block;
1430         char            *options = (char *) *data;
1431
1432         if (!options || strncmp(options, "sb=", 3) != 0)
1433                 return 1;       /* Default location */
1434
1435         options += 3;
1436         /* TODO: use simple_strtoll with >32bit ext4 */
1437         sb_block = simple_strtoul(options, &options, 0);
1438         if (*options && *options != ',') {
1439                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1440                        (char *) *data);
1441                 return 1;
1442         }
1443         if (*options == ',')
1444                 options++;
1445         *data = (void *) options;
1446
1447         return sb_block;
1448 }
1449
1450 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1451 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1452         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1453
1454 #ifdef CONFIG_QUOTA
1455 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1456 {
1457         struct ext4_sb_info *sbi = EXT4_SB(sb);
1458         char *qname;
1459
1460         if (sb_any_quota_loaded(sb) &&
1461                 !sbi->s_qf_names[qtype]) {
1462                 ext4_msg(sb, KERN_ERR,
1463                         "Cannot change journaled "
1464                         "quota options when quota turned on");
1465                 return 0;
1466         }
1467         qname = match_strdup(args);
1468         if (!qname) {
1469                 ext4_msg(sb, KERN_ERR,
1470                         "Not enough memory for storing quotafile name");
1471                 return 0;
1472         }
1473         if (sbi->s_qf_names[qtype] &&
1474                 strcmp(sbi->s_qf_names[qtype], qname)) {
1475                 ext4_msg(sb, KERN_ERR,
1476                         "%s quota file already specified", QTYPE2NAME(qtype));
1477                 kfree(qname);
1478                 return 0;
1479         }
1480         sbi->s_qf_names[qtype] = qname;
1481         if (strchr(sbi->s_qf_names[qtype], '/')) {
1482                 ext4_msg(sb, KERN_ERR,
1483                         "quotafile must be on filesystem root");
1484                 kfree(sbi->s_qf_names[qtype]);
1485                 sbi->s_qf_names[qtype] = NULL;
1486                 return 0;
1487         }
1488         set_opt(sb, QUOTA);
1489         return 1;
1490 }
1491
1492 static int clear_qf_name(struct super_block *sb, int qtype)
1493 {
1494
1495         struct ext4_sb_info *sbi = EXT4_SB(sb);
1496
1497         if (sb_any_quota_loaded(sb) &&
1498                 sbi->s_qf_names[qtype]) {
1499                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1500                         " when quota turned on");
1501                 return 0;
1502         }
1503         /*
1504          * The space will be released later when all options are confirmed
1505          * to be correct
1506          */
1507         sbi->s_qf_names[qtype] = NULL;
1508         return 1;
1509 }
1510 #endif
1511
1512 static int parse_options(char *options, struct super_block *sb,
1513                          unsigned long *journal_devnum,
1514                          unsigned int *journal_ioprio,
1515                          int is_remount)
1516 {
1517         struct ext4_sb_info *sbi = EXT4_SB(sb);
1518         char *p;
1519         substring_t args[MAX_OPT_ARGS];
1520         int data_opt = 0;
1521         int option;
1522 #ifdef CONFIG_QUOTA
1523         int qfmt;
1524 #endif
1525
1526         if (!options)
1527                 return 1;
1528
1529         while ((p = strsep(&options, ",")) != NULL) {
1530                 int token;
1531                 if (!*p)
1532                         continue;
1533
1534                 /*
1535                  * Initialize args struct so we know whether arg was
1536                  * found; some options take optional arguments.
1537                  */
1538                 args[0].to = args[0].from = NULL;
1539                 token = match_token(p, tokens, args);
1540                 switch (token) {
1541                 case Opt_bsd_df:
1542                         clear_opt(sb, MINIX_DF);
1543                         break;
1544                 case Opt_minix_df:
1545                         set_opt(sb, MINIX_DF);
1546                         break;
1547                 case Opt_grpid:
1548                         set_opt(sb, GRPID);
1549                         break;
1550                 case Opt_nogrpid:
1551                         clear_opt(sb, GRPID);
1552                         break;
1553                 case Opt_resuid:
1554                         if (match_int(&args[0], &option))
1555                                 return 0;
1556                         sbi->s_resuid = option;
1557                         break;
1558                 case Opt_resgid:
1559                         if (match_int(&args[0], &option))
1560                                 return 0;
1561                         sbi->s_resgid = option;
1562                         break;
1563                 case Opt_sb:
1564                         /* handled by get_sb_block() instead of here */
1565                         /* *sb_block = match_int(&args[0]); */
1566                         break;
1567                 case Opt_err_panic:
1568                         clear_opt(sb, ERRORS_CONT);
1569                         clear_opt(sb, ERRORS_RO);
1570                         set_opt(sb, ERRORS_PANIC);
1571                         break;
1572                 case Opt_err_ro:
1573                         clear_opt(sb, ERRORS_CONT);
1574                         clear_opt(sb, ERRORS_PANIC);
1575                         set_opt(sb, ERRORS_RO);
1576                         break;
1577                 case Opt_err_cont:
1578                         clear_opt(sb, ERRORS_RO);
1579                         clear_opt(sb, ERRORS_PANIC);
1580                         set_opt(sb, ERRORS_CONT);
1581                         break;
1582                 case Opt_nouid32:
1583                         set_opt(sb, NO_UID32);
1584                         break;
1585                 case Opt_debug:
1586                         set_opt(sb, DEBUG);
1587                         break;
1588                 case Opt_oldalloc:
1589                         ext4_msg(sb, KERN_WARNING,
1590                                  "Ignoring deprecated oldalloc option");
1591                         break;
1592                 case Opt_orlov:
1593                         ext4_msg(sb, KERN_WARNING,
1594                                  "Ignoring deprecated orlov option");
1595                         break;
1596 #ifdef CONFIG_EXT4_FS_XATTR
1597                 case Opt_user_xattr:
1598                         set_opt(sb, XATTR_USER);
1599                         break;
1600                 case Opt_nouser_xattr:
1601                         clear_opt(sb, XATTR_USER);
1602                         break;
1603 #else
1604                 case Opt_user_xattr:
1605                 case Opt_nouser_xattr:
1606                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1607                         break;
1608 #endif
1609 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1610                 case Opt_acl:
1611                         set_opt(sb, POSIX_ACL);
1612                         break;
1613                 case Opt_noacl:
1614                         clear_opt(sb, POSIX_ACL);
1615                         break;
1616 #else
1617                 case Opt_acl:
1618                 case Opt_noacl:
1619                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1620                         break;
1621 #endif
1622                 case Opt_journal_dev:
1623                         if (is_remount) {
1624                                 ext4_msg(sb, KERN_ERR,
1625                                         "Cannot specify journal on remount");
1626                                 return 0;
1627                         }
1628                         if (match_int(&args[0], &option))
1629                                 return 0;
1630                         *journal_devnum = option;
1631                         break;
1632                 case Opt_journal_checksum:
1633                         set_opt(sb, JOURNAL_CHECKSUM);
1634                         break;
1635                 case Opt_journal_async_commit:
1636                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1637                         set_opt(sb, JOURNAL_CHECKSUM);
1638                         break;
1639                 case Opt_noload:
1640                         set_opt(sb, NOLOAD);
1641                         break;
1642                 case Opt_commit:
1643                         if (match_int(&args[0], &option))
1644                                 return 0;
1645                         if (option < 0)
1646                                 return 0;
1647                         if (option == 0)
1648                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1649                         sbi->s_commit_interval = HZ * option;
1650                         break;
1651                 case Opt_max_batch_time:
1652                         if (match_int(&args[0], &option))
1653                                 return 0;
1654                         if (option < 0)
1655                                 return 0;
1656                         if (option == 0)
1657                                 option = EXT4_DEF_MAX_BATCH_TIME;
1658                         sbi->s_max_batch_time = option;
1659                         break;
1660                 case Opt_min_batch_time:
1661                         if (match_int(&args[0], &option))
1662                                 return 0;
1663                         if (option < 0)
1664                                 return 0;
1665                         sbi->s_min_batch_time = option;
1666                         break;
1667                 case Opt_data_journal:
1668                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1669                         goto datacheck;
1670                 case Opt_data_ordered:
1671                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1672                         goto datacheck;
1673                 case Opt_data_writeback:
1674                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1675                 datacheck:
1676                         if (is_remount) {
1677                                 if (!sbi->s_journal)
1678                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1679                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1680                                         ext4_msg(sb, KERN_ERR,
1681                                                 "Cannot change data mode on remount");
1682                                         return 0;
1683                                 }
1684                         } else {
1685                                 clear_opt(sb, DATA_FLAGS);
1686                                 sbi->s_mount_opt |= data_opt;
1687                         }
1688                         break;
1689                 case Opt_data_err_abort:
1690                         set_opt(sb, DATA_ERR_ABORT);
1691                         break;
1692                 case Opt_data_err_ignore:
1693                         clear_opt(sb, DATA_ERR_ABORT);
1694                         break;
1695 #ifdef CONFIG_QUOTA
1696                 case Opt_usrjquota:
1697                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1698                                 return 0;
1699                         break;
1700                 case Opt_grpjquota:
1701                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1702                                 return 0;
1703                         break;
1704                 case Opt_offusrjquota:
1705                         if (!clear_qf_name(sb, USRQUOTA))
1706                                 return 0;
1707                         break;
1708                 case Opt_offgrpjquota:
1709                         if (!clear_qf_name(sb, GRPQUOTA))
1710                                 return 0;
1711                         break;
1712
1713                 case Opt_jqfmt_vfsold:
1714                         qfmt = QFMT_VFS_OLD;
1715                         goto set_qf_format;
1716                 case Opt_jqfmt_vfsv0:
1717                         qfmt = QFMT_VFS_V0;
1718                         goto set_qf_format;
1719                 case Opt_jqfmt_vfsv1:
1720                         qfmt = QFMT_VFS_V1;
1721 set_qf_format:
1722                         if (sb_any_quota_loaded(sb) &&
1723                             sbi->s_jquota_fmt != qfmt) {
1724                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1725                                         "journaled quota options when "
1726                                         "quota turned on");
1727                                 return 0;
1728                         }
1729                         sbi->s_jquota_fmt = qfmt;
1730                         break;
1731                 case Opt_quota:
1732                 case Opt_usrquota:
1733                         set_opt(sb, QUOTA);
1734                         set_opt(sb, USRQUOTA);
1735                         break;
1736                 case Opt_grpquota:
1737                         set_opt(sb, QUOTA);
1738                         set_opt(sb, GRPQUOTA);
1739                         break;
1740                 case Opt_noquota:
1741                         if (sb_any_quota_loaded(sb)) {
1742                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1743                                         "options when quota turned on");
1744                                 return 0;
1745                         }
1746                         clear_opt(sb, QUOTA);
1747                         clear_opt(sb, USRQUOTA);
1748                         clear_opt(sb, GRPQUOTA);
1749                         break;
1750 #else
1751                 case Opt_quota:
1752                 case Opt_usrquota:
1753                 case Opt_grpquota:
1754                         ext4_msg(sb, KERN_ERR,
1755                                 "quota options not supported");
1756                         break;
1757                 case Opt_usrjquota:
1758                 case Opt_grpjquota:
1759                 case Opt_offusrjquota:
1760                 case Opt_offgrpjquota:
1761                 case Opt_jqfmt_vfsold:
1762                 case Opt_jqfmt_vfsv0:
1763                 case Opt_jqfmt_vfsv1:
1764                         ext4_msg(sb, KERN_ERR,
1765                                 "journaled quota options not supported");
1766                         break;
1767                 case Opt_noquota:
1768                         break;
1769 #endif
1770                 case Opt_abort:
1771                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1772                         break;
1773                 case Opt_nobarrier:
1774                         clear_opt(sb, BARRIER);
1775                         break;
1776                 case Opt_barrier:
1777                         if (args[0].from) {
1778                                 if (match_int(&args[0], &option))
1779                                         return 0;
1780                         } else
1781                                 option = 1;     /* No argument, default to 1 */
1782                         if (option)
1783                                 set_opt(sb, BARRIER);
1784                         else
1785                                 clear_opt(sb, BARRIER);
1786                         break;
1787                 case Opt_nobh:
1788                         ext4_msg(sb, KERN_WARNING,
1789                                  "Ignoring deprecated nobh option");
1790                         break;
1791                 case Opt_bh:
1792                         ext4_msg(sb, KERN_WARNING,
1793                                  "Ignoring deprecated bh option");
1794                         break;
1795                 case Opt_i_version:
1796                         sb->s_flags |= MS_I_VERSION;
1797                         break;
1798                 case Opt_nodelalloc:
1799                         clear_opt(sb, DELALLOC);
1800                         clear_opt2(sb, EXPLICIT_DELALLOC);
1801                         break;
1802                 case Opt_mblk_io_submit:
1803                         set_opt(sb, MBLK_IO_SUBMIT);
1804                         break;
1805                 case Opt_nomblk_io_submit:
1806                         clear_opt(sb, MBLK_IO_SUBMIT);
1807                         break;
1808                 case Opt_stripe:
1809                         if (match_int(&args[0], &option))
1810                                 return 0;
1811                         if (option < 0)
1812                                 return 0;
1813                         sbi->s_stripe = option;
1814                         break;
1815                 case Opt_delalloc:
1816                         set_opt(sb, DELALLOC);
1817                         set_opt2(sb, EXPLICIT_DELALLOC);
1818                         break;
1819                 case Opt_block_validity:
1820                         set_opt(sb, BLOCK_VALIDITY);
1821                         break;
1822                 case Opt_noblock_validity:
1823                         clear_opt(sb, BLOCK_VALIDITY);
1824                         break;
1825                 case Opt_inode_readahead_blks:
1826                         if (match_int(&args[0], &option))
1827                                 return 0;
1828                         if (option < 0 || option > (1 << 30))
1829                                 return 0;
1830                         if (option && !is_power_of_2(option)) {
1831                                 ext4_msg(sb, KERN_ERR,
1832                                          "EXT4-fs: inode_readahead_blks"
1833                                          " must be a power of 2");
1834                                 return 0;
1835                         }
1836                         sbi->s_inode_readahead_blks = option;
1837                         break;
1838                 case Opt_journal_ioprio:
1839                         if (match_int(&args[0], &option))
1840                                 return 0;
1841                         if (option < 0 || option > 7)
1842                                 break;
1843                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1844                                                             option);
1845                         break;
1846                 case Opt_noauto_da_alloc:
1847                         set_opt(sb, NO_AUTO_DA_ALLOC);
1848                         break;
1849                 case Opt_auto_da_alloc:
1850                         if (args[0].from) {
1851                                 if (match_int(&args[0], &option))
1852                                         return 0;
1853                         } else
1854                                 option = 1;     /* No argument, default to 1 */
1855                         if (option)
1856                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1857                         else
1858                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1859                         break;
1860                 case Opt_discard:
1861                         set_opt(sb, DISCARD);
1862                         break;
1863                 case Opt_nodiscard:
1864                         clear_opt(sb, DISCARD);
1865                         break;
1866                 case Opt_dioread_nolock:
1867                         set_opt(sb, DIOREAD_NOLOCK);
1868                         break;
1869                 case Opt_dioread_lock:
1870                         clear_opt(sb, DIOREAD_NOLOCK);
1871                         break;
1872                 case Opt_init_itable:
1873                         set_opt(sb, INIT_INODE_TABLE);
1874                         if (args[0].from) {
1875                                 if (match_int(&args[0], &option))
1876                                         return 0;
1877                         } else
1878                                 option = EXT4_DEF_LI_WAIT_MULT;
1879                         if (option < 0)
1880                                 return 0;
1881                         sbi->s_li_wait_mult = option;
1882                         break;
1883                 case Opt_noinit_itable:
1884                         clear_opt(sb, INIT_INODE_TABLE);
1885                         break;
1886                 default:
1887                         ext4_msg(sb, KERN_ERR,
1888                                "Unrecognized mount option \"%s\" "
1889                                "or missing value", p);
1890                         return 0;
1891                 }
1892         }
1893 #ifdef CONFIG_QUOTA
1894         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1895                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1896                         clear_opt(sb, USRQUOTA);
1897
1898                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1899                         clear_opt(sb, GRPQUOTA);
1900
1901                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1902                         ext4_msg(sb, KERN_ERR, "old and new quota "
1903                                         "format mixing");
1904                         return 0;
1905                 }
1906
1907                 if (!sbi->s_jquota_fmt) {
1908                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1909                                         "not specified");
1910                         return 0;
1911                 }
1912         } else {
1913                 if (sbi->s_jquota_fmt) {
1914                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1915                                         "specified with no journaling "
1916                                         "enabled");
1917                         return 0;
1918                 }
1919         }
1920 #endif
1921         return 1;
1922 }
1923
1924 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1925                             int read_only)
1926 {
1927         struct ext4_sb_info *sbi = EXT4_SB(sb);
1928         int res = 0;
1929
1930         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1931                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1932                          "forcing read-only mode");
1933                 res = MS_RDONLY;
1934         }
1935         if (read_only)
1936                 goto done;
1937         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1938                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1939                          "running e2fsck is recommended");
1940         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1941                 ext4_msg(sb, KERN_WARNING,
1942                          "warning: mounting fs with errors, "
1943                          "running e2fsck is recommended");
1944         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1945                  le16_to_cpu(es->s_mnt_count) >=
1946                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1947                 ext4_msg(sb, KERN_WARNING,
1948                          "warning: maximal mount count reached, "
1949                          "running e2fsck is recommended");
1950         else if (le32_to_cpu(es->s_checkinterval) &&
1951                 (le32_to_cpu(es->s_lastcheck) +
1952                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1953                 ext4_msg(sb, KERN_WARNING,
1954                          "warning: checktime reached, "
1955                          "running e2fsck is recommended");
1956         if (!sbi->s_journal)
1957                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1958         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1959                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1960         le16_add_cpu(&es->s_mnt_count, 1);
1961         es->s_mtime = cpu_to_le32(get_seconds());
1962         ext4_update_dynamic_rev(sb);
1963         if (sbi->s_journal)
1964                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1965
1966         ext4_commit_super(sb, 1);
1967 done:
1968         if (test_opt(sb, DEBUG))
1969                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1970                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1971                         sb->s_blocksize,
1972                         sbi->s_groups_count,
1973                         EXT4_BLOCKS_PER_GROUP(sb),
1974                         EXT4_INODES_PER_GROUP(sb),
1975                         sbi->s_mount_opt, sbi->s_mount_opt2);
1976
1977         cleancache_init_fs(sb);
1978         return res;
1979 }
1980
1981 static int ext4_fill_flex_info(struct super_block *sb)
1982 {
1983         struct ext4_sb_info *sbi = EXT4_SB(sb);
1984         struct ext4_group_desc *gdp = NULL;
1985         ext4_group_t flex_group_count;
1986         ext4_group_t flex_group;
1987         unsigned int groups_per_flex = 0;
1988         size_t size;
1989         int i;
1990
1991         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1992         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1993                 sbi->s_log_groups_per_flex = 0;
1994                 return 1;
1995         }
1996         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1997
1998         /* We allocate both existing and potentially added groups */
1999         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2000                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2001                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2002         size = flex_group_count * sizeof(struct flex_groups);
2003         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2004         if (sbi->s_flex_groups == NULL) {
2005                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2006                          flex_group_count);
2007                 goto failed;
2008         }
2009
2010         for (i = 0; i < sbi->s_groups_count; i++) {
2011                 gdp = ext4_get_group_desc(sb, i, NULL);
2012
2013                 flex_group = ext4_flex_group(sbi, i);
2014                 atomic_add(ext4_free_inodes_count(sb, gdp),
2015                            &sbi->s_flex_groups[flex_group].free_inodes);
2016                 atomic_add(ext4_free_group_clusters(sb, gdp),
2017                            &sbi->s_flex_groups[flex_group].free_clusters);
2018                 atomic_add(ext4_used_dirs_count(sb, gdp),
2019                            &sbi->s_flex_groups[flex_group].used_dirs);
2020         }
2021
2022         return 1;
2023 failed:
2024         return 0;
2025 }
2026
2027 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2028                             struct ext4_group_desc *gdp)
2029 {
2030         __u16 crc = 0;
2031
2032         if (sbi->s_es->s_feature_ro_compat &
2033             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2034                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2035                 __le32 le_group = cpu_to_le32(block_group);
2036
2037                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2038                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2039                 crc = crc16(crc, (__u8 *)gdp, offset);
2040                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2041                 /* for checksum of struct ext4_group_desc do the rest...*/
2042                 if ((sbi->s_es->s_feature_incompat &
2043                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2044                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2045                         crc = crc16(crc, (__u8 *)gdp + offset,
2046                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2047                                         offset);
2048         }
2049
2050         return cpu_to_le16(crc);
2051 }
2052
2053 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2054                                 struct ext4_group_desc *gdp)
2055 {
2056         if ((sbi->s_es->s_feature_ro_compat &
2057              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2058             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2059                 return 0;
2060
2061         return 1;
2062 }
2063
2064 /* Called at mount-time, super-block is locked */
2065 static int ext4_check_descriptors(struct super_block *sb,
2066                                   ext4_group_t *first_not_zeroed)
2067 {
2068         struct ext4_sb_info *sbi = EXT4_SB(sb);
2069         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2070         ext4_fsblk_t last_block;
2071         ext4_fsblk_t block_bitmap;
2072         ext4_fsblk_t inode_bitmap;
2073         ext4_fsblk_t inode_table;
2074         int flexbg_flag = 0;
2075         ext4_group_t i, grp = sbi->s_groups_count;
2076
2077         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2078                 flexbg_flag = 1;
2079
2080         ext4_debug("Checking group descriptors");
2081
2082         for (i = 0; i < sbi->s_groups_count; i++) {
2083                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2084
2085                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2086                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2087                 else
2088                         last_block = first_block +
2089                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2090
2091                 if ((grp == sbi->s_groups_count) &&
2092                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2093                         grp = i;
2094
2095                 block_bitmap = ext4_block_bitmap(sb, gdp);
2096                 if (block_bitmap < first_block || block_bitmap > last_block) {
2097                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2098                                "Block bitmap for group %u not in group "
2099                                "(block %llu)!", i, block_bitmap);
2100                         return 0;
2101                 }
2102                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2103                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2104                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2105                                "Inode bitmap for group %u not in group "
2106                                "(block %llu)!", i, inode_bitmap);
2107                         return 0;
2108                 }
2109                 inode_table = ext4_inode_table(sb, gdp);
2110                 if (inode_table < first_block ||
2111                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2112                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2113                                "Inode table for group %u not in group "
2114                                "(block %llu)!", i, inode_table);
2115                         return 0;
2116                 }
2117                 ext4_lock_group(sb, i);
2118                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2119                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2120                                  "Checksum for group %u failed (%u!=%u)",
2121                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2122                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2123                         if (!(sb->s_flags & MS_RDONLY)) {
2124                                 ext4_unlock_group(sb, i);
2125                                 return 0;
2126                         }
2127                 }
2128                 ext4_unlock_group(sb, i);
2129                 if (!flexbg_flag)
2130                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2131         }
2132         if (NULL != first_not_zeroed)
2133                 *first_not_zeroed = grp;
2134
2135         ext4_free_blocks_count_set(sbi->s_es,
2136                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2137         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2138         return 1;
2139 }
2140
2141 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2142  * the superblock) which were deleted from all directories, but held open by
2143  * a process at the time of a crash.  We walk the list and try to delete these
2144  * inodes at recovery time (only with a read-write filesystem).
2145  *
2146  * In order to keep the orphan inode chain consistent during traversal (in
2147  * case of crash during recovery), we link each inode into the superblock
2148  * orphan list_head and handle it the same way as an inode deletion during
2149  * normal operation (which journals the operations for us).
2150  *
2151  * We only do an iget() and an iput() on each inode, which is very safe if we
2152  * accidentally point at an in-use or already deleted inode.  The worst that
2153  * can happen in this case is that we get a "bit already cleared" message from
2154  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2155  * e2fsck was run on this filesystem, and it must have already done the orphan
2156  * inode cleanup for us, so we can safely abort without any further action.
2157  */
2158 static void ext4_orphan_cleanup(struct super_block *sb,
2159                                 struct ext4_super_block *es)
2160 {
2161         unsigned int s_flags = sb->s_flags;
2162         int nr_orphans = 0, nr_truncates = 0;
2163 #ifdef CONFIG_QUOTA
2164         int i;
2165 #endif
2166         if (!es->s_last_orphan) {
2167                 jbd_debug(4, "no orphan inodes to clean up\n");
2168                 return;
2169         }
2170
2171         if (bdev_read_only(sb->s_bdev)) {
2172                 ext4_msg(sb, KERN_ERR, "write access "
2173                         "unavailable, skipping orphan cleanup");
2174                 return;
2175         }
2176
2177         /* Check if feature set would not allow a r/w mount */
2178         if (!ext4_feature_set_ok(sb, 0)) {
2179                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2180                          "unknown ROCOMPAT features");
2181                 return;
2182         }
2183
2184         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2185                 if (es->s_last_orphan)
2186                         jbd_debug(1, "Errors on filesystem, "
2187                                   "clearing orphan list.\n");
2188                 es->s_last_orphan = 0;
2189                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2190                 return;
2191         }
2192
2193         if (s_flags & MS_RDONLY) {
2194                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2195                 sb->s_flags &= ~MS_RDONLY;
2196         }
2197 #ifdef CONFIG_QUOTA
2198         /* Needed for iput() to work correctly and not trash data */
2199         sb->s_flags |= MS_ACTIVE;
2200         /* Turn on quotas so that they are updated correctly */
2201         for (i = 0; i < MAXQUOTAS; i++) {
2202                 if (EXT4_SB(sb)->s_qf_names[i]) {
2203                         int ret = ext4_quota_on_mount(sb, i);
2204                         if (ret < 0)
2205                                 ext4_msg(sb, KERN_ERR,
2206                                         "Cannot turn on journaled "
2207                                         "quota: error %d", ret);
2208                 }
2209         }
2210 #endif
2211
2212         while (es->s_last_orphan) {
2213                 struct inode *inode;
2214
2215                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2216                 if (IS_ERR(inode)) {
2217                         es->s_last_orphan = 0;
2218                         break;
2219                 }
2220
2221                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2222                 dquot_initialize(inode);
2223                 if (inode->i_nlink) {
2224                         ext4_msg(sb, KERN_DEBUG,
2225                                 "%s: truncating inode %lu to %lld bytes",
2226                                 __func__, inode->i_ino, inode->i_size);
2227                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2228                                   inode->i_ino, inode->i_size);
2229                         ext4_truncate(inode);
2230                         nr_truncates++;
2231                 } else {
2232                         ext4_msg(sb, KERN_DEBUG,
2233                                 "%s: deleting unreferenced inode %lu",
2234                                 __func__, inode->i_ino);
2235                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2236                                   inode->i_ino);
2237                         nr_orphans++;
2238                 }
2239                 iput(inode);  /* The delete magic happens here! */
2240         }
2241
2242 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2243
2244         if (nr_orphans)
2245                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2246                        PLURAL(nr_orphans));
2247         if (nr_truncates)
2248                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2249                        PLURAL(nr_truncates));
2250 #ifdef CONFIG_QUOTA
2251         /* Turn quotas off */
2252         for (i = 0; i < MAXQUOTAS; i++) {
2253                 if (sb_dqopt(sb)->files[i])
2254                         dquot_quota_off(sb, i);
2255         }
2256 #endif
2257         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2258 }
2259
2260 /*
2261  * Maximal extent format file size.
2262  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2263  * extent format containers, within a sector_t, and within i_blocks
2264  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2265  * so that won't be a limiting factor.
2266  *
2267  * However there is other limiting factor. We do store extents in the form
2268  * of starting block and length, hence the resulting length of the extent
2269  * covering maximum file size must fit into on-disk format containers as
2270  * well. Given that length is always by 1 unit bigger than max unit (because
2271  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2272  *
2273  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2274  */
2275 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2276 {
2277         loff_t res;
2278         loff_t upper_limit = MAX_LFS_FILESIZE;
2279
2280         /* small i_blocks in vfs inode? */
2281         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2282                 /*
2283                  * CONFIG_LBDAF is not enabled implies the inode
2284                  * i_block represent total blocks in 512 bytes
2285                  * 32 == size of vfs inode i_blocks * 8
2286                  */
2287                 upper_limit = (1LL << 32) - 1;
2288
2289                 /* total blocks in file system block size */
2290                 upper_limit >>= (blkbits - 9);
2291                 upper_limit <<= blkbits;
2292         }
2293
2294         /*
2295          * 32-bit extent-start container, ee_block. We lower the maxbytes
2296          * by one fs block, so ee_len can cover the extent of maximum file
2297          * size
2298          */
2299         res = (1LL << 32) - 1;
2300         res <<= blkbits;
2301
2302         /* Sanity check against vm- & vfs- imposed limits */
2303         if (res > upper_limit)
2304                 res = upper_limit;
2305
2306         return res;
2307 }
2308
2309 /*
2310  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2311  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2312  * We need to be 1 filesystem block less than the 2^48 sector limit.
2313  */
2314 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2315 {
2316         loff_t res = EXT4_NDIR_BLOCKS;
2317         int meta_blocks;
2318         loff_t upper_limit;
2319         /* This is calculated to be the largest file size for a dense, block
2320          * mapped file such that the file's total number of 512-byte sectors,
2321          * including data and all indirect blocks, does not exceed (2^48 - 1).
2322          *
2323          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2324          * number of 512-byte sectors of the file.
2325          */
2326
2327         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2328                 /*
2329                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2330                  * the inode i_block field represents total file blocks in
2331                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2332                  */
2333                 upper_limit = (1LL << 32) - 1;
2334
2335                 /* total blocks in file system block size */
2336                 upper_limit >>= (bits - 9);
2337
2338         } else {
2339                 /*
2340                  * We use 48 bit ext4_inode i_blocks
2341                  * With EXT4_HUGE_FILE_FL set the i_blocks
2342                  * represent total number of blocks in
2343                  * file system block size
2344                  */
2345                 upper_limit = (1LL << 48) - 1;
2346
2347         }
2348
2349         /* indirect blocks */
2350         meta_blocks = 1;
2351         /* double indirect blocks */
2352         meta_blocks += 1 + (1LL << (bits-2));
2353         /* tripple indirect blocks */
2354         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2355
2356         upper_limit -= meta_blocks;
2357         upper_limit <<= bits;
2358
2359         res += 1LL << (bits-2);
2360         res += 1LL << (2*(bits-2));
2361         res += 1LL << (3*(bits-2));
2362         res <<= bits;
2363         if (res > upper_limit)
2364                 res = upper_limit;
2365
2366         if (res > MAX_LFS_FILESIZE)
2367                 res = MAX_LFS_FILESIZE;
2368
2369         return res;
2370 }
2371
2372 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2373                                    ext4_fsblk_t logical_sb_block, int nr)
2374 {
2375         struct ext4_sb_info *sbi = EXT4_SB(sb);
2376         ext4_group_t bg, first_meta_bg;
2377         int has_super = 0;
2378
2379         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2380
2381         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2382             nr < first_meta_bg)
2383                 return logical_sb_block + nr + 1;
2384         bg = sbi->s_desc_per_block * nr;
2385         if (ext4_bg_has_super(sb, bg))
2386                 has_super = 1;
2387
2388         return (has_super + ext4_group_first_block_no(sb, bg));
2389 }
2390
2391 /**
2392  * ext4_get_stripe_size: Get the stripe size.
2393  * @sbi: In memory super block info
2394  *
2395  * If we have specified it via mount option, then
2396  * use the mount option value. If the value specified at mount time is
2397  * greater than the blocks per group use the super block value.
2398  * If the super block value is greater than blocks per group return 0.
2399  * Allocator needs it be less than blocks per group.
2400  *
2401  */
2402 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2403 {
2404         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2405         unsigned long stripe_width =
2406                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2407         int ret;
2408
2409         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2410                 ret = sbi->s_stripe;
2411         else if (stripe_width <= sbi->s_blocks_per_group)
2412                 ret = stripe_width;
2413         else if (stride <= sbi->s_blocks_per_group)
2414                 ret = stride;
2415         else
2416                 ret = 0;
2417
2418         /*
2419          * If the stripe width is 1, this makes no sense and
2420          * we set it to 0 to turn off stripe handling code.
2421          */
2422         if (ret <= 1)
2423                 ret = 0;
2424
2425         return ret;
2426 }
2427
2428 /* sysfs supprt */
2429
2430 struct ext4_attr {
2431         struct attribute attr;
2432         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2433         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2434                          const char *, size_t);
2435         int offset;
2436 };
2437
2438 static int parse_strtoul(const char *buf,
2439                 unsigned long max, unsigned long *value)
2440 {
2441         char *endp;
2442
2443         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2444         endp = skip_spaces(endp);
2445         if (*endp || *value > max)
2446                 return -EINVAL;
2447
2448         return 0;
2449 }
2450
2451 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2452                                               struct ext4_sb_info *sbi,
2453                                               char *buf)
2454 {
2455         return snprintf(buf, PAGE_SIZE, "%llu\n",
2456                 (s64) EXT4_C2B(sbi,
2457                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2458 }
2459
2460 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2461                                          struct ext4_sb_info *sbi, char *buf)
2462 {
2463         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2464
2465         if (!sb->s_bdev->bd_part)
2466                 return snprintf(buf, PAGE_SIZE, "0\n");
2467         return snprintf(buf, PAGE_SIZE, "%lu\n",
2468                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2469                          sbi->s_sectors_written_start) >> 1);
2470 }
2471
2472 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2473                                           struct ext4_sb_info *sbi, char *buf)
2474 {
2475         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2476
2477         if (!sb->s_bdev->bd_part)
2478                 return snprintf(buf, PAGE_SIZE, "0\n");
2479         return snprintf(buf, PAGE_SIZE, "%llu\n",
2480                         (unsigned long long)(sbi->s_kbytes_written +
2481                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2482                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2483 }
2484
2485 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2486                                       struct ext4_sb_info *sbi, char *buf)
2487 {
2488         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2489 }
2490
2491 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2492                                         struct ext4_sb_info *sbi, char *buf)
2493 {
2494         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2495 }
2496
2497 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2498                                           struct ext4_sb_info *sbi,
2499                                           const char *buf, size_t count)
2500 {
2501         unsigned long t;
2502
2503         if (parse_strtoul(buf, 0x40000000, &t))
2504                 return -EINVAL;
2505
2506         if (t && !is_power_of_2(t))
2507                 return -EINVAL;
2508
2509         sbi->s_inode_readahead_blks = t;
2510         return count;
2511 }
2512
2513 static ssize_t sbi_ui_show(struct ext4_attr *a,
2514                            struct ext4_sb_info *sbi, char *buf)
2515 {
2516         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2517
2518         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2519 }
2520
2521 static ssize_t sbi_ui_store(struct ext4_attr *a,
2522                             struct ext4_sb_info *sbi,
2523                             const char *buf, size_t count)
2524 {
2525         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2526         unsigned long t;
2527
2528         if (parse_strtoul(buf, 0xffffffff, &t))
2529                 return -EINVAL;
2530         *ui = t;
2531         return count;
2532 }
2533
2534 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2535 static struct ext4_attr ext4_attr_##_name = {                   \
2536         .attr = {.name = __stringify(_name), .mode = _mode },   \
2537         .show   = _show,                                        \
2538         .store  = _store,                                       \
2539         .offset = offsetof(struct ext4_sb_info, _elname),       \
2540 }
2541 #define EXT4_ATTR(name, mode, show, store) \
2542 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2543
2544 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2545 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2546 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2547 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2548         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2549 #define ATTR_LIST(name) &ext4_attr_##name.attr
2550
2551 EXT4_RO_ATTR(delayed_allocation_blocks);
2552 EXT4_RO_ATTR(session_write_kbytes);
2553 EXT4_RO_ATTR(lifetime_write_kbytes);
2554 EXT4_RO_ATTR(extent_cache_hits);
2555 EXT4_RO_ATTR(extent_cache_misses);
2556 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2557                  inode_readahead_blks_store, s_inode_readahead_blks);
2558 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2559 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2560 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2561 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2562 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2563 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2564 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2565 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2566
2567 static struct attribute *ext4_attrs[] = {
2568         ATTR_LIST(delayed_allocation_blocks),
2569         ATTR_LIST(session_write_kbytes),
2570         ATTR_LIST(lifetime_write_kbytes),
2571         ATTR_LIST(extent_cache_hits),
2572         ATTR_LIST(extent_cache_misses),
2573         ATTR_LIST(inode_readahead_blks),
2574         ATTR_LIST(inode_goal),
2575         ATTR_LIST(mb_stats),
2576         ATTR_LIST(mb_max_to_scan),
2577         ATTR_LIST(mb_min_to_scan),
2578         ATTR_LIST(mb_order2_req),
2579         ATTR_LIST(mb_stream_req),
2580         ATTR_LIST(mb_group_prealloc),
2581         ATTR_LIST(max_writeback_mb_bump),
2582         NULL,
2583 };
2584
2585 /* Features this copy of ext4 supports */
2586 EXT4_INFO_ATTR(lazy_itable_init);
2587 EXT4_INFO_ATTR(batched_discard);
2588
2589 static struct attribute *ext4_feat_attrs[] = {
2590         ATTR_LIST(lazy_itable_init),
2591         ATTR_LIST(batched_discard),
2592         NULL,
2593 };
2594
2595 static ssize_t ext4_attr_show(struct kobject *kobj,
2596                               struct attribute *attr, char *buf)
2597 {
2598         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2599                                                 s_kobj);
2600         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2601
2602         return a->show ? a->show(a, sbi, buf) : 0;
2603 }
2604
2605 static ssize_t ext4_attr_store(struct kobject *kobj,
2606                                struct attribute *attr,
2607                                const char *buf, size_t len)
2608 {
2609         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2610                                                 s_kobj);
2611         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2612
2613         return a->store ? a->store(a, sbi, buf, len) : 0;
2614 }
2615
2616 static void ext4_sb_release(struct kobject *kobj)
2617 {
2618         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2619                                                 s_kobj);
2620         complete(&sbi->s_kobj_unregister);
2621 }
2622
2623 static const struct sysfs_ops ext4_attr_ops = {
2624         .show   = ext4_attr_show,
2625         .store  = ext4_attr_store,
2626 };
2627
2628 static struct kobj_type ext4_ktype = {
2629         .default_attrs  = ext4_attrs,
2630         .sysfs_ops      = &ext4_attr_ops,
2631         .release        = ext4_sb_release,
2632 };
2633
2634 static void ext4_feat_release(struct kobject *kobj)
2635 {
2636         complete(&ext4_feat->f_kobj_unregister);
2637 }
2638
2639 static struct kobj_type ext4_feat_ktype = {
2640         .default_attrs  = ext4_feat_attrs,
2641         .sysfs_ops      = &ext4_attr_ops,
2642         .release        = ext4_feat_release,
2643 };
2644
2645 /*
2646  * Check whether this filesystem can be mounted based on
2647  * the features present and the RDONLY/RDWR mount requested.
2648  * Returns 1 if this filesystem can be mounted as requested,
2649  * 0 if it cannot be.
2650  */
2651 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2652 {
2653         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2654                 ext4_msg(sb, KERN_ERR,
2655                         "Couldn't mount because of "
2656                         "unsupported optional features (%x)",
2657                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2658                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2659                 return 0;
2660         }
2661
2662         if (readonly)
2663                 return 1;
2664
2665         /* Check that feature set is OK for a read-write mount */
2666         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2667                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2668                          "unsupported optional features (%x)",
2669                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2670                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2671                 return 0;
2672         }
2673         /*
2674          * Large file size enabled file system can only be mounted
2675          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2676          */
2677         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2678                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2679                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2680                                  "cannot be mounted RDWR without "
2681                                  "CONFIG_LBDAF");
2682                         return 0;
2683                 }
2684         }
2685         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2686             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2687                 ext4_msg(sb, KERN_ERR,
2688                          "Can't support bigalloc feature without "
2689                          "extents feature\n");
2690                 return 0;
2691         }
2692         return 1;
2693 }
2694
2695 /*
2696  * This function is called once a day if we have errors logged
2697  * on the file system
2698  */
2699 static void print_daily_error_info(unsigned long arg)
2700 {
2701         struct super_block *sb = (struct super_block *) arg;
2702         struct ext4_sb_info *sbi;
2703         struct ext4_super_block *es;
2704
2705         sbi = EXT4_SB(sb);
2706         es = sbi->s_es;
2707
2708         if (es->s_error_count)
2709                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2710                          le32_to_cpu(es->s_error_count));
2711         if (es->s_first_error_time) {
2712                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2713                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2714                        (int) sizeof(es->s_first_error_func),
2715                        es->s_first_error_func,
2716                        le32_to_cpu(es->s_first_error_line));
2717                 if (es->s_first_error_ino)
2718                         printk(": inode %u",
2719                                le32_to_cpu(es->s_first_error_ino));
2720                 if (es->s_first_error_block)
2721                         printk(": block %llu", (unsigned long long)
2722                                le64_to_cpu(es->s_first_error_block));
2723                 printk("\n");
2724         }
2725         if (es->s_last_error_time) {
2726                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2727                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2728                        (int) sizeof(es->s_last_error_func),
2729                        es->s_last_error_func,
2730                        le32_to_cpu(es->s_last_error_line));
2731                 if (es->s_last_error_ino)
2732                         printk(": inode %u",
2733                                le32_to_cpu(es->s_last_error_ino));
2734                 if (es->s_last_error_block)
2735                         printk(": block %llu", (unsigned long long)
2736                                le64_to_cpu(es->s_last_error_block));
2737                 printk("\n");
2738         }
2739         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2740 }
2741
2742 /* Find next suitable group and run ext4_init_inode_table */
2743 static int ext4_run_li_request(struct ext4_li_request *elr)
2744 {
2745         struct ext4_group_desc *gdp = NULL;
2746         ext4_group_t group, ngroups;
2747         struct super_block *sb;
2748         unsigned long timeout = 0;
2749         int ret = 0;
2750
2751         sb = elr->lr_super;
2752         ngroups = EXT4_SB(sb)->s_groups_count;
2753
2754         for (group = elr->lr_next_group; group < ngroups; group++) {
2755                 gdp = ext4_get_group_desc(sb, group, NULL);
2756                 if (!gdp) {
2757                         ret = 1;
2758                         break;
2759                 }
2760
2761                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2762                         break;
2763         }
2764
2765         if (group == ngroups)
2766                 ret = 1;
2767
2768         if (!ret) {
2769                 timeout = jiffies;
2770                 ret = ext4_init_inode_table(sb, group,
2771                                             elr->lr_timeout ? 0 : 1);
2772                 if (elr->lr_timeout == 0) {
2773                         timeout = (jiffies - timeout) *
2774                                   elr->lr_sbi->s_li_wait_mult;
2775                         elr->lr_timeout = timeout;
2776                 }
2777                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2778                 elr->lr_next_group = group + 1;
2779         }
2780
2781         return ret;
2782 }
2783
2784 /*
2785  * Remove lr_request from the list_request and free the
2786  * request structure. Should be called with li_list_mtx held
2787  */
2788 static void ext4_remove_li_request(struct ext4_li_request *elr)
2789 {
2790         struct ext4_sb_info *sbi;
2791
2792         if (!elr)
2793                 return;
2794
2795         sbi = elr->lr_sbi;
2796
2797         list_del(&elr->lr_request);
2798         sbi->s_li_request = NULL;
2799         kfree(elr);
2800 }
2801
2802 static void ext4_unregister_li_request(struct super_block *sb)
2803 {
2804         mutex_lock(&ext4_li_mtx);
2805         if (!ext4_li_info) {
2806                 mutex_unlock(&ext4_li_mtx);
2807                 return;
2808         }
2809
2810         mutex_lock(&ext4_li_info->li_list_mtx);
2811         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2812         mutex_unlock(&ext4_li_info->li_list_mtx);
2813         mutex_unlock(&ext4_li_mtx);
2814 }
2815
2816 static struct task_struct *ext4_lazyinit_task;
2817
2818 /*
2819  * This is the function where ext4lazyinit thread lives. It walks
2820  * through the request list searching for next scheduled filesystem.
2821  * When such a fs is found, run the lazy initialization request
2822  * (ext4_rn_li_request) and keep track of the time spend in this
2823  * function. Based on that time we compute next schedule time of
2824  * the request. When walking through the list is complete, compute
2825  * next waking time and put itself into sleep.
2826  */
2827 static int ext4_lazyinit_thread(void *arg)
2828 {
2829         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2830         struct list_head *pos, *n;
2831         struct ext4_li_request *elr;
2832         unsigned long next_wakeup, cur;
2833
2834         BUG_ON(NULL == eli);
2835
2836 cont_thread:
2837         while (true) {
2838                 next_wakeup = MAX_JIFFY_OFFSET;
2839
2840                 mutex_lock(&eli->li_list_mtx);
2841                 if (list_empty(&eli->li_request_list)) {
2842                         mutex_unlock(&eli->li_list_mtx);
2843                         goto exit_thread;
2844                 }
2845
2846                 list_for_each_safe(pos, n, &eli->li_request_list) {
2847                         elr = list_entry(pos, struct ext4_li_request,
2848                                          lr_request);
2849
2850                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2851                                 if (ext4_run_li_request(elr) != 0) {
2852                                         /* error, remove the lazy_init job */
2853                                         ext4_remove_li_request(elr);
2854                                         continue;
2855                                 }
2856                         }
2857
2858                         if (time_before(elr->lr_next_sched, next_wakeup))
2859                                 next_wakeup = elr->lr_next_sched;
2860                 }
2861                 mutex_unlock(&eli->li_list_mtx);
2862
2863                 try_to_freeze();
2864
2865                 cur = jiffies;
2866                 if ((time_after_eq(cur, next_wakeup)) ||
2867                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2868                         cond_resched();
2869                         continue;
2870                 }
2871
2872                 schedule_timeout_interruptible(next_wakeup - cur);
2873
2874                 if (kthread_should_stop()) {
2875                         ext4_clear_request_list();
2876                         goto exit_thread;
2877                 }
2878         }
2879
2880 exit_thread:
2881         /*
2882          * It looks like the request list is empty, but we need
2883          * to check it under the li_list_mtx lock, to prevent any
2884          * additions into it, and of course we should lock ext4_li_mtx
2885          * to atomically free the list and ext4_li_info, because at
2886          * this point another ext4 filesystem could be registering
2887          * new one.
2888          */
2889         mutex_lock(&ext4_li_mtx);
2890         mutex_lock(&eli->li_list_mtx);
2891         if (!list_empty(&eli->li_request_list)) {
2892                 mutex_unlock(&eli->li_list_mtx);
2893                 mutex_unlock(&ext4_li_mtx);
2894                 goto cont_thread;
2895         }
2896         mutex_unlock(&eli->li_list_mtx);
2897         kfree(ext4_li_info);
2898         ext4_li_info = NULL;
2899         mutex_unlock(&ext4_li_mtx);
2900
2901         return 0;
2902 }
2903
2904 static void ext4_clear_request_list(void)
2905 {
2906         struct list_head *pos, *n;
2907         struct ext4_li_request *elr;
2908
2909         mutex_lock(&ext4_li_info->li_list_mtx);
2910         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2911                 elr = list_entry(pos, struct ext4_li_request,
2912                                  lr_request);
2913                 ext4_remove_li_request(elr);
2914         }
2915         mutex_unlock(&ext4_li_info->li_list_mtx);
2916 }
2917
2918 static int ext4_run_lazyinit_thread(void)
2919 {
2920         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2921                                          ext4_li_info, "ext4lazyinit");
2922         if (IS_ERR(ext4_lazyinit_task)) {
2923                 int err = PTR_ERR(ext4_lazyinit_task);
2924                 ext4_clear_request_list();
2925                 kfree(ext4_li_info);
2926                 ext4_li_info = NULL;
2927                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2928                                  "initialization thread\n",
2929                                  err);
2930                 return err;
2931         }
2932         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2933         return 0;
2934 }
2935
2936 /*
2937  * Check whether it make sense to run itable init. thread or not.
2938  * If there is at least one uninitialized inode table, return
2939  * corresponding group number, else the loop goes through all
2940  * groups and return total number of groups.
2941  */
2942 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2943 {
2944         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2945         struct ext4_group_desc *gdp = NULL;
2946
2947         for (group = 0; group < ngroups; group++) {
2948                 gdp = ext4_get_group_desc(sb, group, NULL);
2949                 if (!gdp)
2950                         continue;
2951
2952                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2953                         break;
2954         }
2955
2956         return group;
2957 }
2958
2959 static int ext4_li_info_new(void)
2960 {
2961         struct ext4_lazy_init *eli = NULL;
2962
2963         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2964         if (!eli)
2965                 return -ENOMEM;
2966
2967         INIT_LIST_HEAD(&eli->li_request_list);
2968         mutex_init(&eli->li_list_mtx);
2969
2970         eli->li_state |= EXT4_LAZYINIT_QUIT;
2971
2972         ext4_li_info = eli;
2973
2974         return 0;
2975 }
2976
2977 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2978                                             ext4_group_t start)
2979 {
2980         struct ext4_sb_info *sbi = EXT4_SB(sb);
2981         struct ext4_li_request *elr;
2982         unsigned long rnd;
2983
2984         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2985         if (!elr)
2986                 return NULL;
2987
2988         elr->lr_super = sb;
2989         elr->lr_sbi = sbi;
2990         elr->lr_next_group = start;
2991
2992         /*
2993          * Randomize first schedule time of the request to
2994          * spread the inode table initialization requests
2995          * better.
2996          */
2997         get_random_bytes(&rnd, sizeof(rnd));
2998         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2999                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3000
3001         return elr;
3002 }
3003
3004 static int ext4_register_li_request(struct super_block *sb,
3005                                     ext4_group_t first_not_zeroed)
3006 {
3007         struct ext4_sb_info *sbi = EXT4_SB(sb);
3008         struct ext4_li_request *elr;
3009         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3010         int ret = 0;
3011
3012         if (sbi->s_li_request != NULL) {
3013                 /*
3014                  * Reset timeout so it can be computed again, because
3015                  * s_li_wait_mult might have changed.
3016                  */
3017                 sbi->s_li_request->lr_timeout = 0;
3018                 return 0;
3019         }
3020
3021         if (first_not_zeroed == ngroups ||
3022             (sb->s_flags & MS_RDONLY) ||
3023             !test_opt(sb, INIT_INODE_TABLE))
3024                 return 0;
3025
3026         elr = ext4_li_request_new(sb, first_not_zeroed);
3027         if (!elr)
3028                 return -ENOMEM;
3029
3030         mutex_lock(&ext4_li_mtx);
3031
3032         if (NULL == ext4_li_info) {
3033                 ret = ext4_li_info_new();
3034                 if (ret)
3035                         goto out;
3036         }
3037
3038         mutex_lock(&ext4_li_info->li_list_mtx);
3039         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3040         mutex_unlock(&ext4_li_info->li_list_mtx);
3041
3042         sbi->s_li_request = elr;
3043         /*
3044          * set elr to NULL here since it has been inserted to
3045          * the request_list and the removal and free of it is
3046          * handled by ext4_clear_request_list from now on.
3047          */
3048         elr = NULL;
3049
3050         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3051                 ret = ext4_run_lazyinit_thread();
3052                 if (ret)
3053                         goto out;
3054         }
3055 out:
3056         mutex_unlock(&ext4_li_mtx);
3057         if (ret)
3058                 kfree(elr);
3059         return ret;
3060 }
3061
3062 /*
3063  * We do not need to lock anything since this is called on
3064  * module unload.
3065  */
3066 static void ext4_destroy_lazyinit_thread(void)
3067 {
3068         /*
3069          * If thread exited earlier
3070          * there's nothing to be done.
3071          */
3072         if (!ext4_li_info || !ext4_lazyinit_task)
3073                 return;
3074
3075         kthread_stop(ext4_lazyinit_task);
3076 }
3077
3078 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3079 {
3080         char *orig_data = kstrdup(data, GFP_KERNEL);
3081         struct buffer_head *bh;
3082         struct ext4_super_block *es = NULL;
3083         struct ext4_sb_info *sbi;
3084         ext4_fsblk_t block;
3085         ext4_fsblk_t sb_block = get_sb_block(&data);
3086         ext4_fsblk_t logical_sb_block;
3087         unsigned long offset = 0;
3088         unsigned long journal_devnum = 0;
3089         unsigned long def_mount_opts;
3090         struct inode *root;
3091         char *cp;
3092         const char *descr;
3093         int ret = -ENOMEM;
3094         int blocksize, clustersize;
3095         unsigned int db_count;
3096         unsigned int i;
3097         int needs_recovery, has_huge_files, has_bigalloc;
3098         __u64 blocks_count;
3099         int err;
3100         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3101         ext4_group_t first_not_zeroed;
3102
3103         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3104         if (!sbi)
3105                 goto out_free_orig;
3106
3107         sbi->s_blockgroup_lock =
3108                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3109         if (!sbi->s_blockgroup_lock) {
3110                 kfree(sbi);
3111                 goto out_free_orig;
3112         }
3113         sb->s_fs_info = sbi;
3114         sbi->s_mount_opt = 0;
3115         sbi->s_resuid = EXT4_DEF_RESUID;
3116         sbi->s_resgid = EXT4_DEF_RESGID;
3117         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3118         sbi->s_sb_block = sb_block;
3119         if (sb->s_bdev->bd_part)
3120                 sbi->s_sectors_written_start =
3121                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3122
3123         /* Cleanup superblock name */
3124         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3125                 *cp = '!';
3126
3127         ret = -EINVAL;
3128         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3129         if (!blocksize) {
3130                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3131                 goto out_fail;
3132         }
3133
3134         /*
3135          * The ext4 superblock will not be buffer aligned for other than 1kB
3136          * block sizes.  We need to calculate the offset from buffer start.
3137          */
3138         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3139                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3140                 offset = do_div(logical_sb_block, blocksize);
3141         } else {
3142                 logical_sb_block = sb_block;
3143         }
3144
3145         if (!(bh = sb_bread(sb, logical_sb_block))) {
3146                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3147                 goto out_fail;
3148         }
3149         /*
3150          * Note: s_es must be initialized as soon as possible because
3151          *       some ext4 macro-instructions depend on its value
3152          */
3153         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3154         sbi->s_es = es;
3155         sb->s_magic = le16_to_cpu(es->s_magic);
3156         if (sb->s_magic != EXT4_SUPER_MAGIC)
3157                 goto cantfind_ext4;
3158         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3159
3160         /* Set defaults before we parse the mount options */
3161         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3162         set_opt(sb, INIT_INODE_TABLE);
3163         if (def_mount_opts & EXT4_DEFM_DEBUG)
3164                 set_opt(sb, DEBUG);
3165         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3166                 set_opt(sb, GRPID);
3167         if (def_mount_opts & EXT4_DEFM_UID16)
3168                 set_opt(sb, NO_UID32);
3169         /* xattr user namespace & acls are now defaulted on */
3170 #ifdef CONFIG_EXT4_FS_XATTR
3171         set_opt(sb, XATTR_USER);
3172 #endif
3173 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3174         set_opt(sb, POSIX_ACL);
3175 #endif
3176         set_opt(sb, MBLK_IO_SUBMIT);
3177         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3178                 set_opt(sb, JOURNAL_DATA);
3179         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3180                 set_opt(sb, ORDERED_DATA);
3181         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3182                 set_opt(sb, WRITEBACK_DATA);
3183
3184         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3185                 set_opt(sb, ERRORS_PANIC);
3186         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3187                 set_opt(sb, ERRORS_CONT);
3188         else
3189                 set_opt(sb, ERRORS_RO);
3190         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3191                 set_opt(sb, BLOCK_VALIDITY);
3192         if (def_mount_opts & EXT4_DEFM_DISCARD)
3193                 set_opt(sb, DISCARD);
3194
3195         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3196         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3197         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3198         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3199         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3200
3201         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3202                 set_opt(sb, BARRIER);
3203
3204         /*
3205          * enable delayed allocation by default
3206          * Use -o nodelalloc to turn it off
3207          */
3208         if (!IS_EXT3_SB(sb) &&
3209             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3210                 set_opt(sb, DELALLOC);
3211
3212         /*
3213          * set default s_li_wait_mult for lazyinit, for the case there is
3214          * no mount option specified.
3215          */
3216         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3217
3218         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3219                            &journal_devnum, &journal_ioprio, 0)) {
3220                 ext4_msg(sb, KERN_WARNING,
3221                          "failed to parse options in superblock: %s",
3222                          sbi->s_es->s_mount_opts);
3223         }
3224         if (!parse_options((char *) data, sb, &journal_devnum,
3225                            &journal_ioprio, 0))
3226                 goto failed_mount;
3227
3228         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3229                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3230                             "with data=journal disables delayed "
3231                             "allocation and O_DIRECT support!\n");
3232                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3233                         ext4_msg(sb, KERN_ERR, "can't mount with "
3234                                  "both data=journal and delalloc");
3235                         goto failed_mount;
3236                 }
3237                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3238                         ext4_msg(sb, KERN_ERR, "can't mount with "
3239                                  "both data=journal and delalloc");
3240                         goto failed_mount;
3241                 }
3242                 if (test_opt(sb, DELALLOC))
3243                         clear_opt(sb, DELALLOC);
3244         }
3245
3246         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3247         if (test_opt(sb, DIOREAD_NOLOCK)) {
3248                 if (blocksize < PAGE_SIZE) {
3249                         ext4_msg(sb, KERN_ERR, "can't mount with "
3250                                  "dioread_nolock if block size != PAGE_SIZE");
3251                         goto failed_mount;
3252                 }
3253         }
3254
3255         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3256                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3257
3258         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3259             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3260              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3261              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3262                 ext4_msg(sb, KERN_WARNING,
3263                        "feature flags set on rev 0 fs, "
3264                        "running e2fsck is recommended");
3265
3266         if (IS_EXT2_SB(sb)) {
3267                 if (ext2_feature_set_ok(sb))
3268                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3269                                  "using the ext4 subsystem");
3270                 else {
3271                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3272                                  "to feature incompatibilities");
3273                         goto failed_mount;
3274                 }
3275         }
3276
3277         if (IS_EXT3_SB(sb)) {
3278                 if (ext3_feature_set_ok(sb))
3279                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3280                                  "using the ext4 subsystem");
3281                 else {
3282                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3283                                  "to feature incompatibilities");
3284                         goto failed_mount;
3285                 }
3286         }
3287
3288         /*
3289          * Check feature flags regardless of the revision level, since we
3290          * previously didn't change the revision level when setting the flags,
3291          * so there is a chance incompat flags are set on a rev 0 filesystem.
3292          */
3293         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3294                 goto failed_mount;
3295
3296         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3297             blocksize > EXT4_MAX_BLOCK_SIZE) {
3298                 ext4_msg(sb, KERN_ERR,
3299                        "Unsupported filesystem blocksize %d", blocksize);
3300                 goto failed_mount;
3301         }
3302
3303         if (sb->s_blocksize != blocksize) {
3304                 /* Validate the filesystem blocksize */
3305                 if (!sb_set_blocksize(sb, blocksize)) {
3306                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3307                                         blocksize);
3308                         goto failed_mount;
3309                 }
3310
3311                 brelse(bh);
3312                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3313                 offset = do_div(logical_sb_block, blocksize);
3314                 bh = sb_bread(sb, logical_sb_block);
3315                 if (!bh) {
3316                         ext4_msg(sb, KERN_ERR,
3317                                "Can't read superblock on 2nd try");
3318                         goto failed_mount;
3319                 }
3320                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3321                 sbi->s_es = es;
3322                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3323                         ext4_msg(sb, KERN_ERR,
3324                                "Magic mismatch, very weird!");
3325                         goto failed_mount;
3326                 }
3327         }
3328
3329         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3330                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3331         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3332                                                       has_huge_files);
3333         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3334
3335         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3336                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3337                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3338         } else {
3339                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3340                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3341                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3342                     (!is_power_of_2(sbi->s_inode_size)) ||
3343                     (sbi->s_inode_size > blocksize)) {
3344                         ext4_msg(sb, KERN_ERR,
3345                                "unsupported inode size: %d",
3346                                sbi->s_inode_size);
3347                         goto failed_mount;
3348                 }
3349                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3350                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3351         }
3352
3353         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3354         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3355                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3356                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3357                     !is_power_of_2(sbi->s_desc_size)) {
3358                         ext4_msg(sb, KERN_ERR,
3359                                "unsupported descriptor size %lu",
3360                                sbi->s_desc_size);
3361                         goto failed_mount;
3362                 }
3363         } else
3364                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3365
3366         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3367         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3368         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3369                 goto cantfind_ext4;
3370
3371         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3372         if (sbi->s_inodes_per_block == 0)
3373                 goto cantfind_ext4;
3374         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3375                                         sbi->s_inodes_per_block;
3376         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3377         sbi->s_sbh = bh;
3378         sbi->s_mount_state = le16_to_cpu(es->s_state);
3379         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3380         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3381
3382         for (i = 0; i < 4; i++)
3383                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3384         sbi->s_def_hash_version = es->s_def_hash_version;
3385         i = le32_to_cpu(es->s_flags);
3386         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3387                 sbi->s_hash_unsigned = 3;
3388         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3389 #ifdef __CHAR_UNSIGNED__
3390                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3391                 sbi->s_hash_unsigned = 3;
3392 #else
3393                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3394 #endif
3395                 sb->s_dirt = 1;
3396         }
3397
3398         /* Handle clustersize */
3399         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3400         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3401                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3402         if (has_bigalloc) {
3403                 if (clustersize < blocksize) {
3404                         ext4_msg(sb, KERN_ERR,
3405                                  "cluster size (%d) smaller than "
3406                                  "block size (%d)", clustersize, blocksize);
3407                         goto failed_mount;
3408                 }
3409                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3410                         le32_to_cpu(es->s_log_block_size);
3411                 sbi->s_clusters_per_group =
3412                         le32_to_cpu(es->s_clusters_per_group);
3413                 if (sbi->s_clusters_per_group > blocksize * 8) {
3414                         ext4_msg(sb, KERN_ERR,
3415                                  "#clusters per group too big: %lu",
3416                                  sbi->s_clusters_per_group);
3417                         goto failed_mount;
3418                 }
3419                 if (sbi->s_blocks_per_group !=
3420                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3421                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3422                                  "clusters per group (%lu) inconsistent",
3423                                  sbi->s_blocks_per_group,
3424                                  sbi->s_clusters_per_group);
3425                         goto failed_mount;
3426                 }
3427         } else {
3428                 if (clustersize != blocksize) {
3429                         ext4_warning(sb, "fragment/cluster size (%d) != "
3430                                      "block size (%d)", clustersize,
3431                                      blocksize);
3432                         clustersize = blocksize;
3433                 }
3434                 if (sbi->s_blocks_per_group > blocksize * 8) {
3435                         ext4_msg(sb, KERN_ERR,
3436                                  "#blocks per group too big: %lu",
3437                                  sbi->s_blocks_per_group);
3438                         goto failed_mount;
3439                 }
3440                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3441                 sbi->s_cluster_bits = 0;
3442         }
3443         sbi->s_cluster_ratio = clustersize / blocksize;
3444
3445         if (sbi->s_inodes_per_group > blocksize * 8) {
3446                 ext4_msg(sb, KERN_ERR,
3447                        "#inodes per group too big: %lu",
3448                        sbi->s_inodes_per_group);
3449                 goto failed_mount;
3450         }
3451
3452         /*
3453          * Test whether we have more sectors than will fit in sector_t,
3454          * and whether the max offset is addressable by the page cache.
3455          */
3456         err = generic_check_addressable(sb->s_blocksize_bits,
3457                                         ext4_blocks_count(es));
3458         if (err) {
3459                 ext4_msg(sb, KERN_ERR, "filesystem"
3460                          " too large to mount safely on this system");
3461                 if (sizeof(sector_t) < 8)
3462                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3463                 ret = err;
3464                 goto failed_mount;
3465         }
3466
3467         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3468                 goto cantfind_ext4;
3469
3470         /* check blocks count against device size */
3471         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3472         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3473                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3474                        "exceeds size of device (%llu blocks)",
3475                        ext4_blocks_count(es), blocks_count);
3476                 goto failed_mount;
3477         }
3478
3479         /*
3480          * It makes no sense for the first data block to be beyond the end
3481          * of the filesystem.
3482          */
3483         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3484                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3485                          "block %u is beyond end of filesystem (%llu)",
3486                          le32_to_cpu(es->s_first_data_block),
3487                          ext4_blocks_count(es));
3488                 goto failed_mount;
3489         }
3490         blocks_count = (ext4_blocks_count(es) -
3491                         le32_to_cpu(es->s_first_data_block) +
3492                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3493         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3494         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3495                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3496                        "(block count %llu, first data block %u, "
3497                        "blocks per group %lu)", sbi->s_groups_count,
3498                        ext4_blocks_count(es),
3499                        le32_to_cpu(es->s_first_data_block),
3500                        EXT4_BLOCKS_PER_GROUP(sb));
3501                 goto failed_mount;
3502         }
3503         sbi->s_groups_count = blocks_count;
3504         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3505                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3506         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3507                    EXT4_DESC_PER_BLOCK(sb);
3508         sbi->s_group_desc = ext4_kvmalloc(db_count *
3509                                           sizeof(struct buffer_head *),
3510                                           GFP_KERNEL);
3511         if (sbi->s_group_desc == NULL) {
3512                 ext4_msg(sb, KERN_ERR, "not enough memory");
3513                 goto failed_mount;
3514         }
3515
3516         if (ext4_proc_root)
3517                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3518
3519         bgl_lock_init(sbi->s_blockgroup_lock);
3520
3521         for (i = 0; i < db_count; i++) {
3522                 block = descriptor_loc(sb, logical_sb_block, i);
3523                 sbi->s_group_desc[i] = sb_bread(sb, block);
3524                 if (!sbi->s_group_desc[i]) {
3525                         ext4_msg(sb, KERN_ERR,
3526                                "can't read group descriptor %d", i);
3527                         db_count = i;
3528                         goto failed_mount2;
3529                 }
3530         }
3531         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3532                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3533                 goto failed_mount2;
3534         }
3535         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3536                 if (!ext4_fill_flex_info(sb)) {
3537                         ext4_msg(sb, KERN_ERR,
3538                                "unable to initialize "
3539                                "flex_bg meta info!");
3540                         goto failed_mount2;
3541                 }
3542
3543         sbi->s_gdb_count = db_count;
3544         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3545         spin_lock_init(&sbi->s_next_gen_lock);
3546
3547         init_timer(&sbi->s_err_report);
3548         sbi->s_err_report.function = print_daily_error_info;
3549         sbi->s_err_report.data = (unsigned long) sb;
3550
3551         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3552                         ext4_count_free_clusters(sb));
3553         if (!err) {
3554                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3555                                 ext4_count_free_inodes(sb));
3556         }
3557         if (!err) {
3558                 err = percpu_counter_init(&sbi->s_dirs_counter,
3559                                 ext4_count_dirs(sb));
3560         }
3561         if (!err) {
3562                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3563         }
3564         if (err) {
3565                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3566                 goto failed_mount3;
3567         }
3568
3569         sbi->s_stripe = ext4_get_stripe_size(sbi);
3570         sbi->s_max_writeback_mb_bump = 128;
3571
3572         /*
3573          * set up enough so that it can read an inode
3574          */
3575         if (!test_opt(sb, NOLOAD) &&
3576             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3577                 sb->s_op = &ext4_sops;
3578         else
3579                 sb->s_op = &ext4_nojournal_sops;
3580         sb->s_export_op = &ext4_export_ops;
3581         sb->s_xattr = ext4_xattr_handlers;
3582 #ifdef CONFIG_QUOTA
3583         sb->s_qcop = &ext4_qctl_operations;
3584         sb->dq_op = &ext4_quota_operations;
3585 #endif
3586         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3587
3588         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3589         mutex_init(&sbi->s_orphan_lock);
3590         sbi->s_resize_flags = 0;
3591
3592         sb->s_root = NULL;
3593
3594         needs_recovery = (es->s_last_orphan != 0 ||
3595                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3596                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3597
3598         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3599             !(sb->s_flags & MS_RDONLY))
3600                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3601                         goto failed_mount3;
3602
3603         /*
3604          * The first inode we look at is the journal inode.  Don't try
3605          * root first: it may be modified in the journal!
3606          */
3607         if (!test_opt(sb, NOLOAD) &&
3608             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3609                 if (ext4_load_journal(sb, es, journal_devnum))
3610                         goto failed_mount3;
3611         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3612               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3613                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3614                        "suppressed and not mounted read-only");
3615                 goto failed_mount_wq;
3616         } else {
3617                 clear_opt(sb, DATA_FLAGS);
3618                 sbi->s_journal = NULL;
3619                 needs_recovery = 0;
3620                 goto no_journal;
3621         }
3622
3623         if (ext4_blocks_count(es) > 0xffffffffULL &&
3624             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3625                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3626                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3627                 goto failed_mount_wq;
3628         }
3629
3630         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3631                 jbd2_journal_set_features(sbi->s_journal,
3632                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3633                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3634         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3635                 jbd2_journal_set_features(sbi->s_journal,
3636                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3637                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3638                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3639         } else {
3640                 jbd2_journal_clear_features(sbi->s_journal,
3641                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3642                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3643         }
3644
3645         /* We have now updated the journal if required, so we can
3646          * validate the data journaling mode. */
3647         switch (test_opt(sb, DATA_FLAGS)) {
3648         case 0:
3649                 /* No mode set, assume a default based on the journal
3650                  * capabilities: ORDERED_DATA if the journal can
3651                  * cope, else JOURNAL_DATA
3652                  */
3653                 if (jbd2_journal_check_available_features
3654                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3655                         set_opt(sb, ORDERED_DATA);
3656                 else
3657                         set_opt(sb, JOURNAL_DATA);
3658                 break;
3659
3660         case EXT4_MOUNT_ORDERED_DATA:
3661         case EXT4_MOUNT_WRITEBACK_DATA:
3662                 if (!jbd2_journal_check_available_features
3663                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3664                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3665                                "requested data journaling mode");
3666                         goto failed_mount_wq;
3667                 }
3668         default:
3669                 break;
3670         }
3671         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3672
3673         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3674
3675         /*
3676          * The journal may have updated the bg summary counts, so we
3677          * need to update the global counters.
3678          */
3679         percpu_counter_set(&sbi->s_freeclusters_counter,
3680                            ext4_count_free_clusters(sb));
3681         percpu_counter_set(&sbi->s_freeinodes_counter,
3682                            ext4_count_free_inodes(sb));
3683         percpu_counter_set(&sbi->s_dirs_counter,
3684                            ext4_count_dirs(sb));
3685         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3686
3687 no_journal:
3688         /*
3689          * The maximum number of concurrent works can be high and
3690          * concurrency isn't really necessary.  Limit it to 1.
3691          */
3692         EXT4_SB(sb)->dio_unwritten_wq =
3693                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3694         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3695                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3696                 goto failed_mount_wq;
3697         }
3698
3699         /*
3700          * The jbd2_journal_load will have done any necessary log recovery,
3701          * so we can safely mount the rest of the filesystem now.
3702          */
3703
3704         root = ext4_iget(sb, EXT4_ROOT_INO);
3705         if (IS_ERR(root)) {
3706                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3707                 ret = PTR_ERR(root);
3708                 root = NULL;
3709                 goto failed_mount4;
3710         }
3711         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3712                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3713                 iput(root);
3714                 goto failed_mount4;
3715         }
3716         sb->s_root = d_alloc_root(root);
3717         if (!sb->s_root) {
3718                 iput(root);
3719                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3720                 ret = -ENOMEM;
3721                 goto failed_mount4;
3722         }
3723
3724         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3725
3726         /* determine the minimum size of new large inodes, if present */
3727         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3728                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3729                                                      EXT4_GOOD_OLD_INODE_SIZE;
3730                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3731                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3732                         if (sbi->s_want_extra_isize <
3733                             le16_to_cpu(es->s_want_extra_isize))
3734                                 sbi->s_want_extra_isize =
3735                                         le16_to_cpu(es->s_want_extra_isize);
3736                         if (sbi->s_want_extra_isize <
3737                             le16_to_cpu(es->s_min_extra_isize))
3738                                 sbi->s_want_extra_isize =
3739                                         le16_to_cpu(es->s_min_extra_isize);
3740                 }
3741         }
3742         /* Check if enough inode space is available */
3743         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3744                                                         sbi->s_inode_size) {
3745                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3746                                                        EXT4_GOOD_OLD_INODE_SIZE;
3747                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3748                          "available");
3749         }
3750
3751         err = ext4_setup_system_zone(sb);
3752         if (err) {
3753                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3754                          "zone (%d)", err);
3755                 goto failed_mount4a;
3756         }
3757
3758         ext4_ext_init(sb);
3759         err = ext4_mb_init(sb, needs_recovery);
3760         if (err) {
3761                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3762                          err);
3763                 goto failed_mount5;
3764         }
3765
3766         err = ext4_register_li_request(sb, first_not_zeroed);
3767         if (err)
3768                 goto failed_mount6;
3769
3770         sbi->s_kobj.kset = ext4_kset;
3771         init_completion(&sbi->s_kobj_unregister);
3772         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3773                                    "%s", sb->s_id);
3774         if (err)
3775                 goto failed_mount7;
3776
3777         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3778         ext4_orphan_cleanup(sb, es);
3779         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3780         if (needs_recovery) {
3781                 ext4_msg(sb, KERN_INFO, "recovery complete");
3782                 ext4_mark_recovery_complete(sb, es);
3783         }
3784         if (EXT4_SB(sb)->s_journal) {
3785                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3786                         descr = " journalled data mode";
3787                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3788                         descr = " ordered data mode";
3789                 else
3790                         descr = " writeback data mode";
3791         } else
3792                 descr = "out journal";
3793
3794         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3795                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3796                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3797
3798         if (es->s_error_count)
3799                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3800
3801         kfree(orig_data);
3802         return 0;
3803
3804 cantfind_ext4:
3805         if (!silent)
3806                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3807         goto failed_mount;
3808
3809 failed_mount7:
3810         ext4_unregister_li_request(sb);
3811 failed_mount6:
3812         ext4_mb_release(sb);
3813 failed_mount5:
3814         ext4_ext_release(sb);
3815         ext4_release_system_zone(sb);
3816 failed_mount4a:
3817         dput(sb->s_root);
3818         sb->s_root = NULL;
3819 failed_mount4:
3820         ext4_msg(sb, KERN_ERR, "mount failed");
3821         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3822 failed_mount_wq:
3823         if (sbi->s_journal) {
3824                 jbd2_journal_destroy(sbi->s_journal);
3825                 sbi->s_journal = NULL;
3826         }
3827 failed_mount3:
3828         del_timer(&sbi->s_err_report);
3829         if (sbi->s_flex_groups)
3830                 ext4_kvfree(sbi->s_flex_groups);
3831         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3832         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3833         percpu_counter_destroy(&sbi->s_dirs_counter);
3834         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3835         if (sbi->s_mmp_tsk)
3836                 kthread_stop(sbi->s_mmp_tsk);
3837 failed_mount2:
3838         for (i = 0; i < db_count; i++)
3839                 brelse(sbi->s_group_desc[i]);
3840         ext4_kvfree(sbi->s_group_desc);
3841 failed_mount:
3842         if (sbi->s_proc) {
3843                 remove_proc_entry(sb->s_id, ext4_proc_root);
3844         }
3845 #ifdef CONFIG_QUOTA
3846         for (i = 0; i < MAXQUOTAS; i++)
3847                 kfree(sbi->s_qf_names[i]);
3848 #endif
3849         ext4_blkdev_remove(sbi);
3850         brelse(bh);
3851 out_fail:
3852         sb->s_fs_info = NULL;
3853         kfree(sbi->s_blockgroup_lock);
3854         kfree(sbi);
3855 out_free_orig:
3856         kfree(orig_data);
3857         return ret;
3858 }
3859
3860 /*
3861  * Setup any per-fs journal parameters now.  We'll do this both on
3862  * initial mount, once the journal has been initialised but before we've
3863  * done any recovery; and again on any subsequent remount.
3864  */
3865 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3866 {
3867         struct ext4_sb_info *sbi = EXT4_SB(sb);
3868
3869         journal->j_commit_interval = sbi->s_commit_interval;
3870         journal->j_min_batch_time = sbi->s_min_batch_time;
3871         journal->j_max_batch_time = sbi->s_max_batch_time;
3872
3873         write_lock(&journal->j_state_lock);
3874         if (test_opt(sb, BARRIER))
3875                 journal->j_flags |= JBD2_BARRIER;
3876         else
3877                 journal->j_flags &= ~JBD2_BARRIER;
3878         if (test_opt(sb, DATA_ERR_ABORT))
3879                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3880         else
3881                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3882         write_unlock(&journal->j_state_lock);
3883 }
3884
3885 static journal_t *ext4_get_journal(struct super_block *sb,
3886                                    unsigned int journal_inum)
3887 {
3888         struct inode *journal_inode;
3889         journal_t *journal;
3890
3891         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3892
3893         /* First, test for the existence of a valid inode on disk.  Bad
3894          * things happen if we iget() an unused inode, as the subsequent
3895          * iput() will try to delete it. */
3896
3897         journal_inode = ext4_iget(sb, journal_inum);
3898         if (IS_ERR(journal_inode)) {
3899                 ext4_msg(sb, KERN_ERR, "no journal found");
3900                 return NULL;
3901         }
3902         if (!journal_inode->i_nlink) {
3903                 make_bad_inode(journal_inode);
3904                 iput(journal_inode);
3905                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3906                 return NULL;
3907         }
3908
3909         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3910                   journal_inode, journal_inode->i_size);
3911         if (!S_ISREG(journal_inode->i_mode)) {
3912                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3913                 iput(journal_inode);
3914                 return NULL;
3915         }
3916
3917         journal = jbd2_journal_init_inode(journal_inode);
3918         if (!journal) {
3919                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3920                 iput(journal_inode);
3921                 return NULL;
3922         }
3923         journal->j_private = sb;
3924         ext4_init_journal_params(sb, journal);
3925         return journal;
3926 }
3927
3928 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3929                                        dev_t j_dev)
3930 {
3931         struct buffer_head *bh;
3932         journal_t *journal;
3933         ext4_fsblk_t start;
3934         ext4_fsblk_t len;
3935         int hblock, blocksize;
3936         ext4_fsblk_t sb_block;
3937         unsigned long offset;
3938         struct ext4_super_block *es;
3939         struct block_device *bdev;
3940
3941         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3942
3943         bdev = ext4_blkdev_get(j_dev, sb);
3944         if (bdev == NULL)
3945                 return NULL;
3946
3947         blocksize = sb->s_blocksize;
3948         hblock = bdev_logical_block_size(bdev);
3949         if (blocksize < hblock) {
3950                 ext4_msg(sb, KERN_ERR,
3951                         "blocksize too small for journal device");
3952                 goto out_bdev;
3953         }
3954
3955         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3956         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3957         set_blocksize(bdev, blocksize);
3958         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3959                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3960                        "external journal");
3961                 goto out_bdev;
3962         }
3963
3964         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3965         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3966             !(le32_to_cpu(es->s_feature_incompat) &
3967               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3968                 ext4_msg(sb, KERN_ERR, "external journal has "
3969                                         "bad superblock");
3970                 brelse(bh);
3971                 goto out_bdev;
3972         }
3973
3974         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3975                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3976                 brelse(bh);
3977                 goto out_bdev;
3978         }
3979
3980         len = ext4_blocks_count(es);
3981         start = sb_block + 1;
3982         brelse(bh);     /* we're done with the superblock */
3983
3984         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3985                                         start, len, blocksize);
3986         if (!journal) {
3987                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3988                 goto out_bdev;
3989         }
3990         journal->j_private = sb;
3991         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3992         wait_on_buffer(journal->j_sb_buffer);
3993         if (!buffer_uptodate(journal->j_sb_buffer)) {
3994                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3995                 goto out_journal;
3996         }
3997         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3998                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3999                                         "user (unsupported) - %d",
4000                         be32_to_cpu(journal->j_superblock->s_nr_users));
4001                 goto out_journal;
4002         }
4003         EXT4_SB(sb)->journal_bdev = bdev;
4004         ext4_init_journal_params(sb, journal);
4005         return journal;
4006
4007 out_journal:
4008         jbd2_journal_destroy(journal);
4009 out_bdev:
4010         ext4_blkdev_put(bdev);
4011         return NULL;
4012 }
4013
4014 static int ext4_load_journal(struct super_block *sb,
4015                              struct ext4_super_block *es,
4016                              unsigned long journal_devnum)
4017 {
4018         journal_t *journal;
4019         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4020         dev_t journal_dev;
4021         int err = 0;
4022         int really_read_only;
4023
4024         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4025
4026         if (journal_devnum &&
4027             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4028                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4029                         "numbers have changed");
4030                 journal_dev = new_decode_dev(journal_devnum);
4031         } else
4032                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4033
4034         really_read_only = bdev_read_only(sb->s_bdev);
4035
4036         /*
4037          * Are we loading a blank journal or performing recovery after a
4038          * crash?  For recovery, we need to check in advance whether we
4039          * can get read-write access to the device.
4040          */
4041         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4042                 if (sb->s_flags & MS_RDONLY) {
4043                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4044                                         "required on readonly filesystem");
4045                         if (really_read_only) {
4046                                 ext4_msg(sb, KERN_ERR, "write access "
4047                                         "unavailable, cannot proceed");
4048                                 return -EROFS;
4049                         }
4050                         ext4_msg(sb, KERN_INFO, "write access will "
4051                                "be enabled during recovery");
4052                 }
4053         }
4054
4055         if (journal_inum && journal_dev) {
4056                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4057                        "and inode journals!");
4058                 return -EINVAL;
4059         }
4060
4061         if (journal_inum) {
4062                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4063                         return -EINVAL;
4064         } else {
4065                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4066                         return -EINVAL;
4067         }
4068
4069         if (!(journal->j_flags & JBD2_BARRIER))
4070                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4071
4072         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4073                 err = jbd2_journal_wipe(journal, !really_read_only);
4074         if (!err) {
4075                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4076                 if (save)
4077                         memcpy(save, ((char *) es) +
4078                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4079                 err = jbd2_journal_load(journal);
4080                 if (save)
4081                         memcpy(((char *) es) + EXT4_S_ERR_START,
4082                                save, EXT4_S_ERR_LEN);
4083                 kfree(save);
4084         }
4085
4086         if (err) {
4087                 ext4_msg(sb, KERN_ERR, "error loading journal");
4088                 jbd2_journal_destroy(journal);
4089                 return err;
4090         }
4091
4092         EXT4_SB(sb)->s_journal = journal;
4093         ext4_clear_journal_err(sb, es);
4094
4095         if (!really_read_only && journal_devnum &&
4096             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4097                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4098
4099                 /* Make sure we flush the recovery flag to disk. */
4100                 ext4_commit_super(sb, 1);
4101         }
4102
4103         return 0;
4104 }
4105
4106 static int ext4_commit_super(struct super_block *sb, int sync)
4107 {
4108         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4109         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4110         int error = 0;
4111
4112         if (!sbh || block_device_ejected(sb))
4113                 return error;
4114         if (buffer_write_io_error(sbh)) {
4115                 /*
4116                  * Oh, dear.  A previous attempt to write the
4117                  * superblock failed.  This could happen because the
4118                  * USB device was yanked out.  Or it could happen to
4119                  * be a transient write error and maybe the block will
4120                  * be remapped.  Nothing we can do but to retry the
4121                  * write and hope for the best.
4122                  */
4123                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4124                        "superblock detected");
4125                 clear_buffer_write_io_error(sbh);
4126                 set_buffer_uptodate(sbh);
4127         }
4128         /*
4129          * If the file system is mounted read-only, don't update the
4130          * superblock write time.  This avoids updating the superblock
4131          * write time when we are mounting the root file system
4132          * read/only but we need to replay the journal; at that point,
4133          * for people who are east of GMT and who make their clock
4134          * tick in localtime for Windows bug-for-bug compatibility,
4135          * the clock is set in the future, and this will cause e2fsck
4136          * to complain and force a full file system check.
4137          */
4138         if (!(sb->s_flags & MS_RDONLY))
4139                 es->s_wtime = cpu_to_le32(get_seconds());
4140         if (sb->s_bdev->bd_part)
4141                 es->s_kbytes_written =
4142                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4143                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4144                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4145         else
4146                 es->s_kbytes_written =
4147                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4148         ext4_free_blocks_count_set(es,
4149                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4150                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4151         es->s_free_inodes_count =
4152                 cpu_to_le32(percpu_counter_sum_positive(
4153                                 &EXT4_SB(sb)->s_freeinodes_counter));
4154         sb->s_dirt = 0;
4155         BUFFER_TRACE(sbh, "marking dirty");
4156         mark_buffer_dirty(sbh);
4157         if (sync) {
4158                 error = sync_dirty_buffer(sbh);
4159                 if (error)
4160                         return error;
4161
4162                 error = buffer_write_io_error(sbh);
4163                 if (error) {
4164                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4165                                "superblock");
4166                         clear_buffer_write_io_error(sbh);
4167                         set_buffer_uptodate(sbh);
4168                 }
4169         }
4170         return error;
4171 }
4172
4173 /*
4174  * Have we just finished recovery?  If so, and if we are mounting (or
4175  * remounting) the filesystem readonly, then we will end up with a
4176  * consistent fs on disk.  Record that fact.
4177  */
4178 static void ext4_mark_recovery_complete(struct super_block *sb,
4179                                         struct ext4_super_block *es)
4180 {
4181         journal_t *journal = EXT4_SB(sb)->s_journal;
4182
4183         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4184                 BUG_ON(journal != NULL);
4185                 return;
4186         }
4187         jbd2_journal_lock_updates(journal);
4188         if (jbd2_journal_flush(journal) < 0)
4189                 goto out;
4190
4191         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4192             sb->s_flags & MS_RDONLY) {
4193                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4194                 ext4_commit_super(sb, 1);
4195         }
4196
4197 out:
4198         jbd2_journal_unlock_updates(journal);
4199 }
4200
4201 /*
4202  * If we are mounting (or read-write remounting) a filesystem whose journal
4203  * has recorded an error from a previous lifetime, move that error to the
4204  * main filesystem now.
4205  */
4206 static void ext4_clear_journal_err(struct super_block *sb,
4207                                    struct ext4_super_block *es)
4208 {
4209         journal_t *journal;
4210         int j_errno;
4211         const char *errstr;
4212
4213         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4214
4215         journal = EXT4_SB(sb)->s_journal;
4216
4217         /*
4218          * Now check for any error status which may have been recorded in the
4219          * journal by a prior ext4_error() or ext4_abort()
4220          */
4221
4222         j_errno = jbd2_journal_errno(journal);
4223         if (j_errno) {
4224                 char nbuf[16];
4225
4226                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4227                 ext4_warning(sb, "Filesystem error recorded "
4228                              "from previous mount: %s", errstr);
4229                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4230
4231                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4232                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4233                 ext4_commit_super(sb, 1);
4234
4235                 jbd2_journal_clear_err(journal);
4236         }
4237 }
4238
4239 /*
4240  * Force the running and committing transactions to commit,
4241  * and wait on the commit.
4242  */
4243 int ext4_force_commit(struct super_block *sb)
4244 {
4245         journal_t *journal;
4246         int ret = 0;
4247
4248         if (sb->s_flags & MS_RDONLY)
4249                 return 0;
4250
4251         journal = EXT4_SB(sb)->s_journal;
4252         if (journal) {
4253                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4254                 ret = ext4_journal_force_commit(journal);
4255         }
4256
4257         return ret;
4258 }
4259
4260 static void ext4_write_super(struct super_block *sb)
4261 {
4262         lock_super(sb);
4263         ext4_commit_super(sb, 1);
4264         unlock_super(sb);
4265 }
4266
4267 static int ext4_sync_fs(struct super_block *sb, int wait)
4268 {
4269         int ret = 0;
4270         tid_t target;
4271         struct ext4_sb_info *sbi = EXT4_SB(sb);
4272
4273         trace_ext4_sync_fs(sb, wait);
4274         flush_workqueue(sbi->dio_unwritten_wq);
4275         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4276                 if (wait)
4277                         jbd2_log_wait_commit(sbi->s_journal, target);
4278         }
4279         return ret;
4280 }
4281
4282 /*
4283  * LVM calls this function before a (read-only) snapshot is created.  This
4284  * gives us a chance to flush the journal completely and mark the fs clean.
4285  *
4286  * Note that only this function cannot bring a filesystem to be in a clean
4287  * state independently, because ext4 prevents a new handle from being started
4288  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4289  * the upper layer.
4290  */
4291 static int ext4_freeze(struct super_block *sb)
4292 {
4293         int error = 0;
4294         journal_t *journal;
4295
4296         if (sb->s_flags & MS_RDONLY)
4297                 return 0;
4298
4299         journal = EXT4_SB(sb)->s_journal;
4300
4301         /* Now we set up the journal barrier. */
4302         jbd2_journal_lock_updates(journal);
4303
4304         /*
4305          * Don't clear the needs_recovery flag if we failed to flush
4306          * the journal.
4307          */
4308         error = jbd2_journal_flush(journal);
4309         if (error < 0)
4310                 goto out;
4311
4312         /* Journal blocked and flushed, clear needs_recovery flag. */
4313         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4314         error = ext4_commit_super(sb, 1);
4315 out:
4316         /* we rely on s_frozen to stop further updates */
4317         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4318         return error;
4319 }
4320
4321 /*
4322  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4323  * flag here, even though the filesystem is not technically dirty yet.
4324  */
4325 static int ext4_unfreeze(struct super_block *sb)
4326 {
4327         if (sb->s_flags & MS_RDONLY)
4328                 return 0;
4329
4330         lock_super(sb);
4331         /* Reset the needs_recovery flag before the fs is unlocked. */
4332         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4333         ext4_commit_super(sb, 1);
4334         unlock_super(sb);
4335         return 0;
4336 }
4337
4338 /*
4339  * Structure to save mount options for ext4_remount's benefit
4340  */
4341 struct ext4_mount_options {
4342         unsigned long s_mount_opt;
4343         unsigned long s_mount_opt2;
4344         uid_t s_resuid;
4345         gid_t s_resgid;
4346         unsigned long s_commit_interval;
4347         u32 s_min_batch_time, s_max_batch_time;
4348 #ifdef CONFIG_QUOTA
4349         int s_jquota_fmt;
4350         char *s_qf_names[MAXQUOTAS];
4351 #endif
4352 };
4353
4354 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4355 {
4356         struct ext4_super_block *es;
4357         struct ext4_sb_info *sbi = EXT4_SB(sb);
4358         unsigned long old_sb_flags;
4359         struct ext4_mount_options old_opts;
4360         int enable_quota = 0;
4361         ext4_group_t g;
4362         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4363         int err = 0;
4364 #ifdef CONFIG_QUOTA
4365         int i;
4366 #endif
4367         char *orig_data = kstrdup(data, GFP_KERNEL);
4368
4369         /* Store the original options */
4370         lock_super(sb);
4371         old_sb_flags = sb->s_flags;
4372         old_opts.s_mount_opt = sbi->s_mount_opt;
4373         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4374         old_opts.s_resuid = sbi->s_resuid;
4375         old_opts.s_resgid = sbi->s_resgid;
4376         old_opts.s_commit_interval = sbi->s_commit_interval;
4377         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4378         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4379 #ifdef CONFIG_QUOTA
4380         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4381         for (i = 0; i < MAXQUOTAS; i++)
4382                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4383 #endif
4384         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4385                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4386
4387         /*
4388          * Allow the "check" option to be passed as a remount option.
4389          */
4390         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4391                 err = -EINVAL;
4392                 goto restore_opts;
4393         }
4394
4395         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4396                 ext4_abort(sb, "Abort forced by user");
4397
4398         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4399                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4400
4401         es = sbi->s_es;
4402
4403         if (sbi->s_journal) {
4404                 ext4_init_journal_params(sb, sbi->s_journal);
4405                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4406         }
4407
4408         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4409                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4410                         err = -EROFS;
4411                         goto restore_opts;
4412                 }
4413
4414                 if (*flags & MS_RDONLY) {
4415                         err = dquot_suspend(sb, -1);
4416                         if (err < 0)
4417                                 goto restore_opts;
4418
4419                         /*
4420                          * First of all, the unconditional stuff we have to do
4421                          * to disable replay of the journal when we next remount
4422                          */
4423                         sb->s_flags |= MS_RDONLY;
4424
4425                         /*
4426                          * OK, test if we are remounting a valid rw partition
4427                          * readonly, and if so set the rdonly flag and then
4428                          * mark the partition as valid again.
4429                          */
4430                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4431                             (sbi->s_mount_state & EXT4_VALID_FS))
4432                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4433
4434                         if (sbi->s_journal)
4435                                 ext4_mark_recovery_complete(sb, es);
4436                 } else {
4437                         /* Make sure we can mount this feature set readwrite */
4438                         if (!ext4_feature_set_ok(sb, 0)) {
4439                                 err = -EROFS;
4440                                 goto restore_opts;
4441                         }
4442                         /*
4443                          * Make sure the group descriptor checksums
4444                          * are sane.  If they aren't, refuse to remount r/w.
4445                          */
4446                         for (g = 0; g < sbi->s_groups_count; g++) {
4447                                 struct ext4_group_desc *gdp =
4448                                         ext4_get_group_desc(sb, g, NULL);
4449
4450                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4451                                         ext4_msg(sb, KERN_ERR,
4452                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4453                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4454                                                le16_to_cpu(gdp->bg_checksum));
4455                                         err = -EINVAL;
4456                                         goto restore_opts;
4457                                 }
4458                         }
4459
4460                         /*
4461                          * If we have an unprocessed orphan list hanging
4462                          * around from a previously readonly bdev mount,
4463                          * require a full umount/remount for now.
4464                          */
4465                         if (es->s_last_orphan) {
4466                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4467                                        "remount RDWR because of unprocessed "
4468                                        "orphan inode list.  Please "
4469                                        "umount/remount instead");
4470                                 err = -EINVAL;
4471                                 goto restore_opts;
4472                         }
4473
4474                         /*
4475                          * Mounting a RDONLY partition read-write, so reread
4476                          * and store the current valid flag.  (It may have
4477                          * been changed by e2fsck since we originally mounted
4478                          * the partition.)
4479                          */
4480                         if (sbi->s_journal)
4481                                 ext4_clear_journal_err(sb, es);
4482                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4483                         if (!ext4_setup_super(sb, es, 0))
4484                                 sb->s_flags &= ~MS_RDONLY;
4485                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4486                                                      EXT4_FEATURE_INCOMPAT_MMP))
4487                                 if (ext4_multi_mount_protect(sb,
4488                                                 le64_to_cpu(es->s_mmp_block))) {
4489                                         err = -EROFS;
4490                                         goto restore_opts;
4491                                 }
4492                         enable_quota = 1;
4493                 }
4494         }
4495
4496         /*
4497          * Reinitialize lazy itable initialization thread based on
4498          * current settings
4499          */
4500         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4501                 ext4_unregister_li_request(sb);
4502         else {
4503                 ext4_group_t first_not_zeroed;
4504                 first_not_zeroed = ext4_has_uninit_itable(sb);
4505                 ext4_register_li_request(sb, first_not_zeroed);
4506         }
4507
4508         ext4_setup_system_zone(sb);
4509         if (sbi->s_journal == NULL)
4510                 ext4_commit_super(sb, 1);
4511
4512 #ifdef CONFIG_QUOTA
4513         /* Release old quota file names */
4514         for (i = 0; i < MAXQUOTAS; i++)
4515                 if (old_opts.s_qf_names[i] &&
4516                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4517                         kfree(old_opts.s_qf_names[i]);
4518 #endif
4519         unlock_super(sb);
4520         if (enable_quota)
4521                 dquot_resume(sb, -1);
4522
4523         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4524         kfree(orig_data);
4525         return 0;
4526
4527 restore_opts:
4528         sb->s_flags = old_sb_flags;
4529         sbi->s_mount_opt = old_opts.s_mount_opt;
4530         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4531         sbi->s_resuid = old_opts.s_resuid;
4532         sbi->s_resgid = old_opts.s_resgid;
4533         sbi->s_commit_interval = old_opts.s_commit_interval;
4534         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4535         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4536 #ifdef CONFIG_QUOTA
4537         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4538         for (i = 0; i < MAXQUOTAS; i++) {
4539                 if (sbi->s_qf_names[i] &&
4540                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4541                         kfree(sbi->s_qf_names[i]);
4542                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4543         }
4544 #endif
4545         unlock_super(sb);
4546         kfree(orig_data);
4547         return err;
4548 }
4549
4550 /*
4551  * Note: calculating the overhead so we can be compatible with
4552  * historical BSD practice is quite difficult in the face of
4553  * clusters/bigalloc.  This is because multiple metadata blocks from
4554  * different block group can end up in the same allocation cluster.
4555  * Calculating the exact overhead in the face of clustered allocation
4556  * requires either O(all block bitmaps) in memory or O(number of block
4557  * groups**2) in time.  We will still calculate the superblock for
4558  * older file systems --- and if we come across with a bigalloc file
4559  * system with zero in s_overhead_clusters the estimate will be close to
4560  * correct especially for very large cluster sizes --- but for newer
4561  * file systems, it's better to calculate this figure once at mkfs
4562  * time, and store it in the superblock.  If the superblock value is
4563  * present (even for non-bigalloc file systems), we will use it.
4564  */
4565 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4566 {
4567         struct super_block *sb = dentry->d_sb;
4568         struct ext4_sb_info *sbi = EXT4_SB(sb);
4569         struct ext4_super_block *es = sbi->s_es;
4570         struct ext4_group_desc *gdp;
4571         u64 fsid;
4572         s64 bfree;
4573
4574         if (test_opt(sb, MINIX_DF)) {
4575                 sbi->s_overhead_last = 0;
4576         } else if (es->s_overhead_clusters) {
4577                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4578         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4579                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4580                 ext4_fsblk_t overhead = 0;
4581
4582                 /*
4583                  * Compute the overhead (FS structures).  This is constant
4584                  * for a given filesystem unless the number of block groups
4585                  * changes so we cache the previous value until it does.
4586                  */
4587
4588                 /*
4589                  * All of the blocks before first_data_block are
4590                  * overhead
4591                  */
4592                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4593
4594                 /*
4595                  * Add the overhead found in each block group
4596                  */
4597                 for (i = 0; i < ngroups; i++) {
4598                         gdp = ext4_get_group_desc(sb, i, NULL);
4599                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4600                         cond_resched();
4601                 }
4602                 sbi->s_overhead_last = overhead;
4603                 smp_wmb();
4604                 sbi->s_blocks_last = ext4_blocks_count(es);
4605         }
4606
4607         buf->f_type = EXT4_SUPER_MAGIC;
4608         buf->f_bsize = sb->s_blocksize;
4609         buf->f_blocks = (ext4_blocks_count(es) -
4610                          EXT4_C2B(sbi, sbi->s_overhead_last));
4611         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4612                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4613         /* prevent underflow in case that few free space is available */
4614         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4615         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4616         if (buf->f_bfree < ext4_r_blocks_count(es))
4617                 buf->f_bavail = 0;
4618         buf->f_files = le32_to_cpu(es->s_inodes_count);
4619         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4620         buf->f_namelen = EXT4_NAME_LEN;
4621         fsid = le64_to_cpup((void *)es->s_uuid) ^
4622                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4623         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4624         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4625
4626         return 0;
4627 }
4628
4629 /* Helper function for writing quotas on sync - we need to start transaction
4630  * before quota file is locked for write. Otherwise the are possible deadlocks:
4631  * Process 1                         Process 2
4632  * ext4_create()                     quota_sync()
4633  *   jbd2_journal_start()                  write_dquot()
4634  *   dquot_initialize()                         down(dqio_mutex)
4635  *     down(dqio_mutex)                    jbd2_journal_start()
4636  *
4637  */
4638
4639 #ifdef CONFIG_QUOTA
4640
4641 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4642 {
4643         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4644 }
4645
4646 static int ext4_write_dquot(struct dquot *dquot)
4647 {
4648         int ret, err;
4649         handle_t *handle;
4650         struct inode *inode;
4651
4652         inode = dquot_to_inode(dquot);
4653         handle = ext4_journal_start(inode,
4654                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4655         if (IS_ERR(handle))
4656                 return PTR_ERR(handle);
4657         ret = dquot_commit(dquot);
4658         err = ext4_journal_stop(handle);
4659         if (!ret)
4660                 ret = err;
4661         return ret;
4662 }
4663
4664 static int ext4_acquire_dquot(struct dquot *dquot)
4665 {
4666         int ret, err;
4667         handle_t *handle;
4668
4669         handle = ext4_journal_start(dquot_to_inode(dquot),
4670                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4671         if (IS_ERR(handle))
4672                 return PTR_ERR(handle);
4673         ret = dquot_acquire(dquot);
4674         err = ext4_journal_stop(handle);
4675         if (!ret)
4676                 ret = err;
4677         return ret;
4678 }
4679
4680 static int ext4_release_dquot(struct dquot *dquot)
4681 {
4682         int ret, err;
4683         handle_t *handle;
4684
4685         handle = ext4_journal_start(dquot_to_inode(dquot),
4686                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4687         if (IS_ERR(handle)) {
4688                 /* Release dquot anyway to avoid endless cycle in dqput() */
4689                 dquot_release(dquot);
4690                 return PTR_ERR(handle);
4691         }
4692         ret = dquot_release(dquot);
4693         err = ext4_journal_stop(handle);
4694         if (!ret)
4695                 ret = err;
4696         return ret;
4697 }
4698
4699 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4700 {
4701         /* Are we journaling quotas? */
4702         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4703             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4704                 dquot_mark_dquot_dirty(dquot);
4705                 return ext4_write_dquot(dquot);
4706         } else {
4707                 return dquot_mark_dquot_dirty(dquot);
4708         }
4709 }
4710
4711 static int ext4_write_info(struct super_block *sb, int type)
4712 {
4713         int ret, err;
4714         handle_t *handle;
4715
4716         /* Data block + inode block */
4717         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4718         if (IS_ERR(handle))
4719                 return PTR_ERR(handle);
4720         ret = dquot_commit_info(sb, type);
4721         err = ext4_journal_stop(handle);
4722         if (!ret)
4723                 ret = err;
4724         return ret;
4725 }
4726
4727 /*
4728  * Turn on quotas during mount time - we need to find
4729  * the quota file and such...
4730  */
4731 static int ext4_quota_on_mount(struct super_block *sb, int type)
4732 {
4733         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4734                                         EXT4_SB(sb)->s_jquota_fmt, type);
4735 }
4736
4737 /*
4738  * Standard function to be called on quota_on
4739  */
4740 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4741                          struct path *path)
4742 {
4743         int err;
4744
4745         if (!test_opt(sb, QUOTA))
4746                 return -EINVAL;
4747
4748         /* Quotafile not on the same filesystem? */
4749         if (path->dentry->d_sb != sb)
4750                 return -EXDEV;
4751         /* Journaling quota? */
4752         if (EXT4_SB(sb)->s_qf_names[type]) {
4753                 /* Quotafile not in fs root? */
4754                 if (path->dentry->d_parent != sb->s_root)
4755                         ext4_msg(sb, KERN_WARNING,
4756                                 "Quota file not on filesystem root. "
4757                                 "Journaled quota will not work");
4758         }
4759
4760         /*
4761          * When we journal data on quota file, we have to flush journal to see
4762          * all updates to the file when we bypass pagecache...
4763          */
4764         if (EXT4_SB(sb)->s_journal &&
4765             ext4_should_journal_data(path->dentry->d_inode)) {
4766                 /*
4767                  * We don't need to lock updates but journal_flush() could
4768                  * otherwise be livelocked...
4769                  */
4770                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4771                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4772                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4773                 if (err)
4774                         return err;
4775         }
4776
4777         return dquot_quota_on(sb, type, format_id, path);
4778 }
4779
4780 static int ext4_quota_off(struct super_block *sb, int type)
4781 {
4782         struct inode *inode = sb_dqopt(sb)->files[type];
4783         handle_t *handle;
4784
4785         /* Force all delayed allocation blocks to be allocated.
4786          * Caller already holds s_umount sem */
4787         if (test_opt(sb, DELALLOC))
4788                 sync_filesystem(sb);
4789
4790         if (!inode)
4791                 goto out;
4792
4793         /* Update modification times of quota files when userspace can
4794          * start looking at them */
4795         handle = ext4_journal_start(inode, 1);
4796         if (IS_ERR(handle))
4797                 goto out;
4798         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4799         ext4_mark_inode_dirty(handle, inode);
4800         ext4_journal_stop(handle);
4801
4802 out:
4803         return dquot_quota_off(sb, type);
4804 }
4805
4806 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4807  * acquiring the locks... As quota files are never truncated and quota code
4808  * itself serializes the operations (and no one else should touch the files)
4809  * we don't have to be afraid of races */
4810 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4811                                size_t len, loff_t off)
4812 {
4813         struct inode *inode = sb_dqopt(sb)->files[type];
4814         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4815         int err = 0;
4816         int offset = off & (sb->s_blocksize - 1);
4817         int tocopy;
4818         size_t toread;
4819         struct buffer_head *bh;
4820         loff_t i_size = i_size_read(inode);
4821
4822         if (off > i_size)
4823                 return 0;
4824         if (off+len > i_size)
4825                 len = i_size-off;
4826         toread = len;
4827         while (toread > 0) {
4828                 tocopy = sb->s_blocksize - offset < toread ?
4829                                 sb->s_blocksize - offset : toread;
4830                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4831                 if (err)
4832                         return err;
4833                 if (!bh)        /* A hole? */
4834                         memset(data, 0, tocopy);
4835                 else
4836                         memcpy(data, bh->b_data+offset, tocopy);
4837                 brelse(bh);
4838                 offset = 0;
4839                 toread -= tocopy;
4840                 data += tocopy;
4841                 blk++;
4842         }
4843         return len;
4844 }
4845
4846 /* Write to quotafile (we know the transaction is already started and has
4847  * enough credits) */
4848 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4849                                 const char *data, size_t len, loff_t off)
4850 {
4851         struct inode *inode = sb_dqopt(sb)->files[type];
4852         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4853         int err = 0;
4854         int offset = off & (sb->s_blocksize - 1);
4855         struct buffer_head *bh;
4856         handle_t *handle = journal_current_handle();
4857
4858         if (EXT4_SB(sb)->s_journal && !handle) {
4859                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4860                         " cancelled because transaction is not started",
4861                         (unsigned long long)off, (unsigned long long)len);
4862                 return -EIO;
4863         }
4864         /*
4865          * Since we account only one data block in transaction credits,
4866          * then it is impossible to cross a block boundary.
4867          */
4868         if (sb->s_blocksize - offset < len) {
4869                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4870                         " cancelled because not block aligned",
4871                         (unsigned long long)off, (unsigned long long)len);
4872                 return -EIO;
4873         }
4874
4875         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4876         bh = ext4_bread(handle, inode, blk, 1, &err);
4877         if (!bh)
4878                 goto out;
4879         err = ext4_journal_get_write_access(handle, bh);
4880         if (err) {
4881                 brelse(bh);
4882                 goto out;
4883         }
4884         lock_buffer(bh);
4885         memcpy(bh->b_data+offset, data, len);
4886         flush_dcache_page(bh->b_page);
4887         unlock_buffer(bh);
4888         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4889         brelse(bh);
4890 out:
4891         if (err) {
4892                 mutex_unlock(&inode->i_mutex);
4893                 return err;
4894         }
4895         if (inode->i_size < off + len) {
4896                 i_size_write(inode, off + len);
4897                 EXT4_I(inode)->i_disksize = inode->i_size;
4898                 ext4_mark_inode_dirty(handle, inode);
4899         }
4900         mutex_unlock(&inode->i_mutex);
4901         return len;
4902 }
4903
4904 #endif
4905
4906 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4907                        const char *dev_name, void *data)
4908 {
4909         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4910 }
4911
4912 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4913 static inline void register_as_ext2(void)
4914 {
4915         int err = register_filesystem(&ext2_fs_type);
4916         if (err)
4917                 printk(KERN_WARNING
4918                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4919 }
4920
4921 static inline void unregister_as_ext2(void)
4922 {
4923         unregister_filesystem(&ext2_fs_type);
4924 }
4925
4926 static inline int ext2_feature_set_ok(struct super_block *sb)
4927 {
4928         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4929                 return 0;
4930         if (sb->s_flags & MS_RDONLY)
4931                 return 1;
4932         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4933                 return 0;
4934         return 1;
4935 }
4936 MODULE_ALIAS("ext2");
4937 #else
4938 static inline void register_as_ext2(void) { }
4939 static inline void unregister_as_ext2(void) { }
4940 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4941 #endif
4942
4943 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4944 static inline void register_as_ext3(void)
4945 {
4946         int err = register_filesystem(&ext3_fs_type);
4947         if (err)
4948                 printk(KERN_WARNING
4949                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4950 }
4951
4952 static inline void unregister_as_ext3(void)
4953 {
4954         unregister_filesystem(&ext3_fs_type);
4955 }
4956
4957 static inline int ext3_feature_set_ok(struct super_block *sb)
4958 {
4959         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4960                 return 0;
4961         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4962                 return 0;
4963         if (sb->s_flags & MS_RDONLY)
4964                 return 1;
4965         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4966                 return 0;
4967         return 1;
4968 }
4969 MODULE_ALIAS("ext3");
4970 #else
4971 static inline void register_as_ext3(void) { }
4972 static inline void unregister_as_ext3(void) { }
4973 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4974 #endif
4975
4976 static struct file_system_type ext4_fs_type = {
4977         .owner          = THIS_MODULE,
4978         .name           = "ext4",
4979         .mount          = ext4_mount,
4980         .kill_sb        = kill_block_super,
4981         .fs_flags       = FS_REQUIRES_DEV,
4982 };
4983
4984 static int __init ext4_init_feat_adverts(void)
4985 {
4986         struct ext4_features *ef;
4987         int ret = -ENOMEM;
4988
4989         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4990         if (!ef)
4991                 goto out;
4992
4993         ef->f_kobj.kset = ext4_kset;
4994         init_completion(&ef->f_kobj_unregister);
4995         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4996                                    "features");
4997         if (ret) {
4998                 kfree(ef);
4999                 goto out;
5000         }
5001
5002         ext4_feat = ef;
5003         ret = 0;
5004 out:
5005         return ret;
5006 }
5007
5008 static void ext4_exit_feat_adverts(void)
5009 {
5010         kobject_put(&ext4_feat->f_kobj);
5011         wait_for_completion(&ext4_feat->f_kobj_unregister);
5012         kfree(ext4_feat);
5013 }
5014
5015 /* Shared across all ext4 file systems */
5016 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5017 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5018
5019 static int __init ext4_init_fs(void)
5020 {
5021         int i, err;
5022
5023         ext4_check_flag_values();
5024
5025         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5026                 mutex_init(&ext4__aio_mutex[i]);
5027                 init_waitqueue_head(&ext4__ioend_wq[i]);
5028         }
5029
5030         err = ext4_init_pageio();
5031         if (err)
5032                 return err;
5033         err = ext4_init_system_zone();
5034         if (err)
5035                 goto out6;
5036         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5037         if (!ext4_kset)
5038                 goto out5;
5039         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5040
5041         err = ext4_init_feat_adverts();
5042         if (err)
5043                 goto out4;
5044
5045         err = ext4_init_mballoc();
5046         if (err)
5047                 goto out3;
5048
5049         err = ext4_init_xattr();
5050         if (err)
5051                 goto out2;
5052         err = init_inodecache();
5053         if (err)
5054                 goto out1;
5055         register_as_ext3();
5056         register_as_ext2();
5057         err = register_filesystem(&ext4_fs_type);
5058         if (err)
5059                 goto out;
5060
5061         ext4_li_info = NULL;
5062         mutex_init(&ext4_li_mtx);
5063         return 0;
5064 out:
5065         unregister_as_ext2();
5066         unregister_as_ext3();
5067         destroy_inodecache();
5068 out1:
5069         ext4_exit_xattr();
5070 out2:
5071         ext4_exit_mballoc();
5072 out3:
5073         ext4_exit_feat_adverts();
5074 out4:
5075         if (ext4_proc_root)
5076                 remove_proc_entry("fs/ext4", NULL);
5077         kset_unregister(ext4_kset);
5078 out5:
5079         ext4_exit_system_zone();
5080 out6:
5081         ext4_exit_pageio();
5082         return err;
5083 }
5084
5085 static void __exit ext4_exit_fs(void)
5086 {
5087         ext4_destroy_lazyinit_thread();
5088         unregister_as_ext2();
5089         unregister_as_ext3();
5090         unregister_filesystem(&ext4_fs_type);
5091         destroy_inodecache();
5092         ext4_exit_xattr();
5093         ext4_exit_mballoc();
5094         ext4_exit_feat_adverts();
5095         remove_proc_entry("fs/ext4", NULL);
5096         kset_unregister(ext4_kset);
5097         ext4_exit_system_zone();
5098         ext4_exit_pageio();
5099 }
5100
5101 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5102 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5103 MODULE_LICENSE("GPL");
5104 module_init(ext4_init_fs)
5105 module_exit(ext4_exit_fs)