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_removed,
1336         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1337         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
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_removed, "oldalloc"},
1369         {Opt_removed, "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_removed, "nobh"},
1377         {Opt_removed, "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_MASK);
1569                         set_opt(sb, ERRORS_PANIC);
1570                         break;
1571                 case Opt_err_ro:
1572                         clear_opt(sb, ERRORS_MASK);
1573                         set_opt(sb, ERRORS_RO);
1574                         break;
1575                 case Opt_err_cont:
1576                         clear_opt(sb, ERRORS_MASK);
1577                         set_opt(sb, ERRORS_CONT);
1578                         break;
1579                 case Opt_nouid32:
1580                         set_opt(sb, NO_UID32);
1581                         break;
1582                 case Opt_debug:
1583                         set_opt(sb, DEBUG);
1584                         break;
1585                 case Opt_removed:
1586                         ext4_msg(sb, KERN_WARNING,
1587                                  "Ignoring deprecated %s option", p);
1588                         break;
1589 #ifdef CONFIG_EXT4_FS_XATTR
1590                 case Opt_user_xattr:
1591                         set_opt(sb, XATTR_USER);
1592                         break;
1593                 case Opt_nouser_xattr:
1594                         clear_opt(sb, XATTR_USER);
1595                         break;
1596 #else
1597                 case Opt_user_xattr:
1598                 case Opt_nouser_xattr:
1599                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1600                         break;
1601 #endif
1602 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1603                 case Opt_acl:
1604                         set_opt(sb, POSIX_ACL);
1605                         break;
1606                 case Opt_noacl:
1607                         clear_opt(sb, POSIX_ACL);
1608                         break;
1609 #else
1610                 case Opt_acl:
1611                 case Opt_noacl:
1612                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1613                         break;
1614 #endif
1615                 case Opt_journal_dev:
1616                         if (is_remount) {
1617                                 ext4_msg(sb, KERN_ERR,
1618                                         "Cannot specify journal on remount");
1619                                 return 0;
1620                         }
1621                         if (match_int(&args[0], &option))
1622                                 return 0;
1623                         *journal_devnum = option;
1624                         break;
1625                 case Opt_journal_checksum:
1626                         set_opt(sb, JOURNAL_CHECKSUM);
1627                         break;
1628                 case Opt_journal_async_commit:
1629                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1630                         set_opt(sb, JOURNAL_CHECKSUM);
1631                         break;
1632                 case Opt_noload:
1633                         set_opt(sb, NOLOAD);
1634                         break;
1635                 case Opt_commit:
1636                         if (match_int(&args[0], &option))
1637                                 return 0;
1638                         if (option < 0)
1639                                 return 0;
1640                         if (option == 0)
1641                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1642                         sbi->s_commit_interval = HZ * option;
1643                         break;
1644                 case Opt_max_batch_time:
1645                         if (match_int(&args[0], &option))
1646                                 return 0;
1647                         if (option < 0)
1648                                 return 0;
1649                         if (option == 0)
1650                                 option = EXT4_DEF_MAX_BATCH_TIME;
1651                         sbi->s_max_batch_time = option;
1652                         break;
1653                 case Opt_min_batch_time:
1654                         if (match_int(&args[0], &option))
1655                                 return 0;
1656                         if (option < 0)
1657                                 return 0;
1658                         sbi->s_min_batch_time = option;
1659                         break;
1660                 case Opt_data_journal:
1661                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1662                         goto datacheck;
1663                 case Opt_data_ordered:
1664                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1665                         goto datacheck;
1666                 case Opt_data_writeback:
1667                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1668                 datacheck:
1669                         if (is_remount) {
1670                                 if (!sbi->s_journal)
1671                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1672                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1673                                         ext4_msg(sb, KERN_ERR,
1674                                                 "Cannot change data mode on remount");
1675                                         return 0;
1676                                 }
1677                         } else {
1678                                 clear_opt(sb, DATA_FLAGS);
1679                                 sbi->s_mount_opt |= data_opt;
1680                         }
1681                         break;
1682                 case Opt_data_err_abort:
1683                         set_opt(sb, DATA_ERR_ABORT);
1684                         break;
1685                 case Opt_data_err_ignore:
1686                         clear_opt(sb, DATA_ERR_ABORT);
1687                         break;
1688 #ifdef CONFIG_QUOTA
1689                 case Opt_usrjquota:
1690                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1691                                 return 0;
1692                         break;
1693                 case Opt_grpjquota:
1694                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1695                                 return 0;
1696                         break;
1697                 case Opt_offusrjquota:
1698                         if (!clear_qf_name(sb, USRQUOTA))
1699                                 return 0;
1700                         break;
1701                 case Opt_offgrpjquota:
1702                         if (!clear_qf_name(sb, GRPQUOTA))
1703                                 return 0;
1704                         break;
1705
1706                 case Opt_jqfmt_vfsold:
1707                         qfmt = QFMT_VFS_OLD;
1708                         goto set_qf_format;
1709                 case Opt_jqfmt_vfsv0:
1710                         qfmt = QFMT_VFS_V0;
1711                         goto set_qf_format;
1712                 case Opt_jqfmt_vfsv1:
1713                         qfmt = QFMT_VFS_V1;
1714 set_qf_format:
1715                         if (sb_any_quota_loaded(sb) &&
1716                             sbi->s_jquota_fmt != qfmt) {
1717                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1718                                         "journaled quota options when "
1719                                         "quota turned on");
1720                                 return 0;
1721                         }
1722                         sbi->s_jquota_fmt = qfmt;
1723                         break;
1724                 case Opt_quota:
1725                 case Opt_usrquota:
1726                         set_opt(sb, QUOTA);
1727                         set_opt(sb, USRQUOTA);
1728                         break;
1729                 case Opt_grpquota:
1730                         set_opt(sb, QUOTA);
1731                         set_opt(sb, GRPQUOTA);
1732                         break;
1733                 case Opt_noquota:
1734                         if (sb_any_quota_loaded(sb)) {
1735                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1736                                         "options when quota turned on");
1737                                 return 0;
1738                         }
1739                         clear_opt(sb, QUOTA);
1740                         clear_opt(sb, USRQUOTA);
1741                         clear_opt(sb, GRPQUOTA);
1742                         break;
1743 #else
1744                 case Opt_quota:
1745                 case Opt_usrquota:
1746                 case Opt_grpquota:
1747                         ext4_msg(sb, KERN_ERR,
1748                                 "quota options not supported");
1749                         break;
1750                 case Opt_usrjquota:
1751                 case Opt_grpjquota:
1752                 case Opt_offusrjquota:
1753                 case Opt_offgrpjquota:
1754                 case Opt_jqfmt_vfsold:
1755                 case Opt_jqfmt_vfsv0:
1756                 case Opt_jqfmt_vfsv1:
1757                         ext4_msg(sb, KERN_ERR,
1758                                 "journaled quota options not supported");
1759                         break;
1760                 case Opt_noquota:
1761                         break;
1762 #endif
1763                 case Opt_abort:
1764                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1765                         break;
1766                 case Opt_nobarrier:
1767                         clear_opt(sb, BARRIER);
1768                         break;
1769                 case Opt_barrier:
1770                         if (args[0].from) {
1771                                 if (match_int(&args[0], &option))
1772                                         return 0;
1773                         } else
1774                                 option = 1;     /* No argument, default to 1 */
1775                         if (option)
1776                                 set_opt(sb, BARRIER);
1777                         else
1778                                 clear_opt(sb, BARRIER);
1779                         break;
1780                 case Opt_i_version:
1781                         sb->s_flags |= MS_I_VERSION;
1782                         break;
1783                 case Opt_nodelalloc:
1784                         clear_opt(sb, DELALLOC);
1785                         clear_opt2(sb, EXPLICIT_DELALLOC);
1786                         break;
1787                 case Opt_mblk_io_submit:
1788                         set_opt(sb, MBLK_IO_SUBMIT);
1789                         break;
1790                 case Opt_nomblk_io_submit:
1791                         clear_opt(sb, MBLK_IO_SUBMIT);
1792                         break;
1793                 case Opt_stripe:
1794                         if (match_int(&args[0], &option))
1795                                 return 0;
1796                         if (option < 0)
1797                                 return 0;
1798                         sbi->s_stripe = option;
1799                         break;
1800                 case Opt_delalloc:
1801                         set_opt(sb, DELALLOC);
1802                         set_opt2(sb, EXPLICIT_DELALLOC);
1803                         break;
1804                 case Opt_block_validity:
1805                         set_opt(sb, BLOCK_VALIDITY);
1806                         break;
1807                 case Opt_noblock_validity:
1808                         clear_opt(sb, BLOCK_VALIDITY);
1809                         break;
1810                 case Opt_inode_readahead_blks:
1811                         if (match_int(&args[0], &option))
1812                                 return 0;
1813                         if (option < 0 || option > (1 << 30))
1814                                 return 0;
1815                         if (option && !is_power_of_2(option)) {
1816                                 ext4_msg(sb, KERN_ERR,
1817                                          "EXT4-fs: inode_readahead_blks"
1818                                          " must be a power of 2");
1819                                 return 0;
1820                         }
1821                         sbi->s_inode_readahead_blks = option;
1822                         break;
1823                 case Opt_journal_ioprio:
1824                         if (match_int(&args[0], &option))
1825                                 return 0;
1826                         if (option < 0 || option > 7)
1827                                 break;
1828                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1829                                                             option);
1830                         break;
1831                 case Opt_noauto_da_alloc:
1832                         set_opt(sb, NO_AUTO_DA_ALLOC);
1833                         break;
1834                 case Opt_auto_da_alloc:
1835                         if (args[0].from) {
1836                                 if (match_int(&args[0], &option))
1837                                         return 0;
1838                         } else
1839                                 option = 1;     /* No argument, default to 1 */
1840                         if (option)
1841                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1842                         else
1843                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1844                         break;
1845                 case Opt_discard:
1846                         set_opt(sb, DISCARD);
1847                         break;
1848                 case Opt_nodiscard:
1849                         clear_opt(sb, DISCARD);
1850                         break;
1851                 case Opt_dioread_nolock:
1852                         set_opt(sb, DIOREAD_NOLOCK);
1853                         break;
1854                 case Opt_dioread_lock:
1855                         clear_opt(sb, DIOREAD_NOLOCK);
1856                         break;
1857                 case Opt_init_itable:
1858                         set_opt(sb, INIT_INODE_TABLE);
1859                         if (args[0].from) {
1860                                 if (match_int(&args[0], &option))
1861                                         return 0;
1862                         } else
1863                                 option = EXT4_DEF_LI_WAIT_MULT;
1864                         if (option < 0)
1865                                 return 0;
1866                         sbi->s_li_wait_mult = option;
1867                         break;
1868                 case Opt_noinit_itable:
1869                         clear_opt(sb, INIT_INODE_TABLE);
1870                         break;
1871                 default:
1872                         ext4_msg(sb, KERN_ERR,
1873                                "Unrecognized mount option \"%s\" "
1874                                "or missing value", p);
1875                         return 0;
1876                 }
1877         }
1878 #ifdef CONFIG_QUOTA
1879         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1880                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1881                         clear_opt(sb, USRQUOTA);
1882
1883                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1884                         clear_opt(sb, GRPQUOTA);
1885
1886                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1887                         ext4_msg(sb, KERN_ERR, "old and new quota "
1888                                         "format mixing");
1889                         return 0;
1890                 }
1891
1892                 if (!sbi->s_jquota_fmt) {
1893                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1894                                         "not specified");
1895                         return 0;
1896                 }
1897         } else {
1898                 if (sbi->s_jquota_fmt) {
1899                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1900                                         "specified with no journaling "
1901                                         "enabled");
1902                         return 0;
1903                 }
1904         }
1905 #endif
1906         return 1;
1907 }
1908
1909 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1910                             int read_only)
1911 {
1912         struct ext4_sb_info *sbi = EXT4_SB(sb);
1913         int res = 0;
1914
1915         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1916                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1917                          "forcing read-only mode");
1918                 res = MS_RDONLY;
1919         }
1920         if (read_only)
1921                 goto done;
1922         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1923                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1924                          "running e2fsck is recommended");
1925         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1926                 ext4_msg(sb, KERN_WARNING,
1927                          "warning: mounting fs with errors, "
1928                          "running e2fsck is recommended");
1929         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1930                  le16_to_cpu(es->s_mnt_count) >=
1931                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1932                 ext4_msg(sb, KERN_WARNING,
1933                          "warning: maximal mount count reached, "
1934                          "running e2fsck is recommended");
1935         else if (le32_to_cpu(es->s_checkinterval) &&
1936                 (le32_to_cpu(es->s_lastcheck) +
1937                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1938                 ext4_msg(sb, KERN_WARNING,
1939                          "warning: checktime reached, "
1940                          "running e2fsck is recommended");
1941         if (!sbi->s_journal)
1942                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1943         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1944                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1945         le16_add_cpu(&es->s_mnt_count, 1);
1946         es->s_mtime = cpu_to_le32(get_seconds());
1947         ext4_update_dynamic_rev(sb);
1948         if (sbi->s_journal)
1949                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1950
1951         ext4_commit_super(sb, 1);
1952 done:
1953         if (test_opt(sb, DEBUG))
1954                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1955                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1956                         sb->s_blocksize,
1957                         sbi->s_groups_count,
1958                         EXT4_BLOCKS_PER_GROUP(sb),
1959                         EXT4_INODES_PER_GROUP(sb),
1960                         sbi->s_mount_opt, sbi->s_mount_opt2);
1961
1962         cleancache_init_fs(sb);
1963         return res;
1964 }
1965
1966 static int ext4_fill_flex_info(struct super_block *sb)
1967 {
1968         struct ext4_sb_info *sbi = EXT4_SB(sb);
1969         struct ext4_group_desc *gdp = NULL;
1970         ext4_group_t flex_group_count;
1971         ext4_group_t flex_group;
1972         unsigned int groups_per_flex = 0;
1973         size_t size;
1974         int i;
1975
1976         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1977         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1978                 sbi->s_log_groups_per_flex = 0;
1979                 return 1;
1980         }
1981         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1982
1983         /* We allocate both existing and potentially added groups */
1984         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1985                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1986                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1987         size = flex_group_count * sizeof(struct flex_groups);
1988         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1989         if (sbi->s_flex_groups == NULL) {
1990                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1991                          flex_group_count);
1992                 goto failed;
1993         }
1994
1995         for (i = 0; i < sbi->s_groups_count; i++) {
1996                 gdp = ext4_get_group_desc(sb, i, NULL);
1997
1998                 flex_group = ext4_flex_group(sbi, i);
1999                 atomic_add(ext4_free_inodes_count(sb, gdp),
2000                            &sbi->s_flex_groups[flex_group].free_inodes);
2001                 atomic_add(ext4_free_group_clusters(sb, gdp),
2002                            &sbi->s_flex_groups[flex_group].free_clusters);
2003                 atomic_add(ext4_used_dirs_count(sb, gdp),
2004                            &sbi->s_flex_groups[flex_group].used_dirs);
2005         }
2006
2007         return 1;
2008 failed:
2009         return 0;
2010 }
2011
2012 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2013                             struct ext4_group_desc *gdp)
2014 {
2015         __u16 crc = 0;
2016
2017         if (sbi->s_es->s_feature_ro_compat &
2018             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2019                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2020                 __le32 le_group = cpu_to_le32(block_group);
2021
2022                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2023                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2024                 crc = crc16(crc, (__u8 *)gdp, offset);
2025                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2026                 /* for checksum of struct ext4_group_desc do the rest...*/
2027                 if ((sbi->s_es->s_feature_incompat &
2028                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2029                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2030                         crc = crc16(crc, (__u8 *)gdp + offset,
2031                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2032                                         offset);
2033         }
2034
2035         return cpu_to_le16(crc);
2036 }
2037
2038 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2039                                 struct ext4_group_desc *gdp)
2040 {
2041         if ((sbi->s_es->s_feature_ro_compat &
2042              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2043             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2044                 return 0;
2045
2046         return 1;
2047 }
2048
2049 /* Called at mount-time, super-block is locked */
2050 static int ext4_check_descriptors(struct super_block *sb,
2051                                   ext4_group_t *first_not_zeroed)
2052 {
2053         struct ext4_sb_info *sbi = EXT4_SB(sb);
2054         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2055         ext4_fsblk_t last_block;
2056         ext4_fsblk_t block_bitmap;
2057         ext4_fsblk_t inode_bitmap;
2058         ext4_fsblk_t inode_table;
2059         int flexbg_flag = 0;
2060         ext4_group_t i, grp = sbi->s_groups_count;
2061
2062         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2063                 flexbg_flag = 1;
2064
2065         ext4_debug("Checking group descriptors");
2066
2067         for (i = 0; i < sbi->s_groups_count; i++) {
2068                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2069
2070                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2071                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2072                 else
2073                         last_block = first_block +
2074                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2075
2076                 if ((grp == sbi->s_groups_count) &&
2077                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2078                         grp = i;
2079
2080                 block_bitmap = ext4_block_bitmap(sb, gdp);
2081                 if (block_bitmap < first_block || block_bitmap > last_block) {
2082                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2083                                "Block bitmap for group %u not in group "
2084                                "(block %llu)!", i, block_bitmap);
2085                         return 0;
2086                 }
2087                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2088                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2089                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2090                                "Inode bitmap for group %u not in group "
2091                                "(block %llu)!", i, inode_bitmap);
2092                         return 0;
2093                 }
2094                 inode_table = ext4_inode_table(sb, gdp);
2095                 if (inode_table < first_block ||
2096                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2097                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2098                                "Inode table for group %u not in group "
2099                                "(block %llu)!", i, inode_table);
2100                         return 0;
2101                 }
2102                 ext4_lock_group(sb, i);
2103                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2104                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2105                                  "Checksum for group %u failed (%u!=%u)",
2106                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2107                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2108                         if (!(sb->s_flags & MS_RDONLY)) {
2109                                 ext4_unlock_group(sb, i);
2110                                 return 0;
2111                         }
2112                 }
2113                 ext4_unlock_group(sb, i);
2114                 if (!flexbg_flag)
2115                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2116         }
2117         if (NULL != first_not_zeroed)
2118                 *first_not_zeroed = grp;
2119
2120         ext4_free_blocks_count_set(sbi->s_es,
2121                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2122         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2123         return 1;
2124 }
2125
2126 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2127  * the superblock) which were deleted from all directories, but held open by
2128  * a process at the time of a crash.  We walk the list and try to delete these
2129  * inodes at recovery time (only with a read-write filesystem).
2130  *
2131  * In order to keep the orphan inode chain consistent during traversal (in
2132  * case of crash during recovery), we link each inode into the superblock
2133  * orphan list_head and handle it the same way as an inode deletion during
2134  * normal operation (which journals the operations for us).
2135  *
2136  * We only do an iget() and an iput() on each inode, which is very safe if we
2137  * accidentally point at an in-use or already deleted inode.  The worst that
2138  * can happen in this case is that we get a "bit already cleared" message from
2139  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2140  * e2fsck was run on this filesystem, and it must have already done the orphan
2141  * inode cleanup for us, so we can safely abort without any further action.
2142  */
2143 static void ext4_orphan_cleanup(struct super_block *sb,
2144                                 struct ext4_super_block *es)
2145 {
2146         unsigned int s_flags = sb->s_flags;
2147         int nr_orphans = 0, nr_truncates = 0;
2148 #ifdef CONFIG_QUOTA
2149         int i;
2150 #endif
2151         if (!es->s_last_orphan) {
2152                 jbd_debug(4, "no orphan inodes to clean up\n");
2153                 return;
2154         }
2155
2156         if (bdev_read_only(sb->s_bdev)) {
2157                 ext4_msg(sb, KERN_ERR, "write access "
2158                         "unavailable, skipping orphan cleanup");
2159                 return;
2160         }
2161
2162         /* Check if feature set would not allow a r/w mount */
2163         if (!ext4_feature_set_ok(sb, 0)) {
2164                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2165                          "unknown ROCOMPAT features");
2166                 return;
2167         }
2168
2169         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2170                 if (es->s_last_orphan)
2171                         jbd_debug(1, "Errors on filesystem, "
2172                                   "clearing orphan list.\n");
2173                 es->s_last_orphan = 0;
2174                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2175                 return;
2176         }
2177
2178         if (s_flags & MS_RDONLY) {
2179                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2180                 sb->s_flags &= ~MS_RDONLY;
2181         }
2182 #ifdef CONFIG_QUOTA
2183         /* Needed for iput() to work correctly and not trash data */
2184         sb->s_flags |= MS_ACTIVE;
2185         /* Turn on quotas so that they are updated correctly */
2186         for (i = 0; i < MAXQUOTAS; i++) {
2187                 if (EXT4_SB(sb)->s_qf_names[i]) {
2188                         int ret = ext4_quota_on_mount(sb, i);
2189                         if (ret < 0)
2190                                 ext4_msg(sb, KERN_ERR,
2191                                         "Cannot turn on journaled "
2192                                         "quota: error %d", ret);
2193                 }
2194         }
2195 #endif
2196
2197         while (es->s_last_orphan) {
2198                 struct inode *inode;
2199
2200                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2201                 if (IS_ERR(inode)) {
2202                         es->s_last_orphan = 0;
2203                         break;
2204                 }
2205
2206                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2207                 dquot_initialize(inode);
2208                 if (inode->i_nlink) {
2209                         ext4_msg(sb, KERN_DEBUG,
2210                                 "%s: truncating inode %lu to %lld bytes",
2211                                 __func__, inode->i_ino, inode->i_size);
2212                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2213                                   inode->i_ino, inode->i_size);
2214                         ext4_truncate(inode);
2215                         nr_truncates++;
2216                 } else {
2217                         ext4_msg(sb, KERN_DEBUG,
2218                                 "%s: deleting unreferenced inode %lu",
2219                                 __func__, inode->i_ino);
2220                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2221                                   inode->i_ino);
2222                         nr_orphans++;
2223                 }
2224                 iput(inode);  /* The delete magic happens here! */
2225         }
2226
2227 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2228
2229         if (nr_orphans)
2230                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2231                        PLURAL(nr_orphans));
2232         if (nr_truncates)
2233                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2234                        PLURAL(nr_truncates));
2235 #ifdef CONFIG_QUOTA
2236         /* Turn quotas off */
2237         for (i = 0; i < MAXQUOTAS; i++) {
2238                 if (sb_dqopt(sb)->files[i])
2239                         dquot_quota_off(sb, i);
2240         }
2241 #endif
2242         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2243 }
2244
2245 /*
2246  * Maximal extent format file size.
2247  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2248  * extent format containers, within a sector_t, and within i_blocks
2249  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2250  * so that won't be a limiting factor.
2251  *
2252  * However there is other limiting factor. We do store extents in the form
2253  * of starting block and length, hence the resulting length of the extent
2254  * covering maximum file size must fit into on-disk format containers as
2255  * well. Given that length is always by 1 unit bigger than max unit (because
2256  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2257  *
2258  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2259  */
2260 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2261 {
2262         loff_t res;
2263         loff_t upper_limit = MAX_LFS_FILESIZE;
2264
2265         /* small i_blocks in vfs inode? */
2266         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2267                 /*
2268                  * CONFIG_LBDAF is not enabled implies the inode
2269                  * i_block represent total blocks in 512 bytes
2270                  * 32 == size of vfs inode i_blocks * 8
2271                  */
2272                 upper_limit = (1LL << 32) - 1;
2273
2274                 /* total blocks in file system block size */
2275                 upper_limit >>= (blkbits - 9);
2276                 upper_limit <<= blkbits;
2277         }
2278
2279         /*
2280          * 32-bit extent-start container, ee_block. We lower the maxbytes
2281          * by one fs block, so ee_len can cover the extent of maximum file
2282          * size
2283          */
2284         res = (1LL << 32) - 1;
2285         res <<= blkbits;
2286
2287         /* Sanity check against vm- & vfs- imposed limits */
2288         if (res > upper_limit)
2289                 res = upper_limit;
2290
2291         return res;
2292 }
2293
2294 /*
2295  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2296  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2297  * We need to be 1 filesystem block less than the 2^48 sector limit.
2298  */
2299 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2300 {
2301         loff_t res = EXT4_NDIR_BLOCKS;
2302         int meta_blocks;
2303         loff_t upper_limit;
2304         /* This is calculated to be the largest file size for a dense, block
2305          * mapped file such that the file's total number of 512-byte sectors,
2306          * including data and all indirect blocks, does not exceed (2^48 - 1).
2307          *
2308          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2309          * number of 512-byte sectors of the file.
2310          */
2311
2312         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2313                 /*
2314                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2315                  * the inode i_block field represents total file blocks in
2316                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2317                  */
2318                 upper_limit = (1LL << 32) - 1;
2319
2320                 /* total blocks in file system block size */
2321                 upper_limit >>= (bits - 9);
2322
2323         } else {
2324                 /*
2325                  * We use 48 bit ext4_inode i_blocks
2326                  * With EXT4_HUGE_FILE_FL set the i_blocks
2327                  * represent total number of blocks in
2328                  * file system block size
2329                  */
2330                 upper_limit = (1LL << 48) - 1;
2331
2332         }
2333
2334         /* indirect blocks */
2335         meta_blocks = 1;
2336         /* double indirect blocks */
2337         meta_blocks += 1 + (1LL << (bits-2));
2338         /* tripple indirect blocks */
2339         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2340
2341         upper_limit -= meta_blocks;
2342         upper_limit <<= bits;
2343
2344         res += 1LL << (bits-2);
2345         res += 1LL << (2*(bits-2));
2346         res += 1LL << (3*(bits-2));
2347         res <<= bits;
2348         if (res > upper_limit)
2349                 res = upper_limit;
2350
2351         if (res > MAX_LFS_FILESIZE)
2352                 res = MAX_LFS_FILESIZE;
2353
2354         return res;
2355 }
2356
2357 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2358                                    ext4_fsblk_t logical_sb_block, int nr)
2359 {
2360         struct ext4_sb_info *sbi = EXT4_SB(sb);
2361         ext4_group_t bg, first_meta_bg;
2362         int has_super = 0;
2363
2364         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2365
2366         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2367             nr < first_meta_bg)
2368                 return logical_sb_block + nr + 1;
2369         bg = sbi->s_desc_per_block * nr;
2370         if (ext4_bg_has_super(sb, bg))
2371                 has_super = 1;
2372
2373         return (has_super + ext4_group_first_block_no(sb, bg));
2374 }
2375
2376 /**
2377  * ext4_get_stripe_size: Get the stripe size.
2378  * @sbi: In memory super block info
2379  *
2380  * If we have specified it via mount option, then
2381  * use the mount option value. If the value specified at mount time is
2382  * greater than the blocks per group use the super block value.
2383  * If the super block value is greater than blocks per group return 0.
2384  * Allocator needs it be less than blocks per group.
2385  *
2386  */
2387 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2388 {
2389         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2390         unsigned long stripe_width =
2391                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2392         int ret;
2393
2394         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2395                 ret = sbi->s_stripe;
2396         else if (stripe_width <= sbi->s_blocks_per_group)
2397                 ret = stripe_width;
2398         else if (stride <= sbi->s_blocks_per_group)
2399                 ret = stride;
2400         else
2401                 ret = 0;
2402
2403         /*
2404          * If the stripe width is 1, this makes no sense and
2405          * we set it to 0 to turn off stripe handling code.
2406          */
2407         if (ret <= 1)
2408                 ret = 0;
2409
2410         return ret;
2411 }
2412
2413 /* sysfs supprt */
2414
2415 struct ext4_attr {
2416         struct attribute attr;
2417         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2418         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2419                          const char *, size_t);
2420         int offset;
2421 };
2422
2423 static int parse_strtoul(const char *buf,
2424                 unsigned long max, unsigned long *value)
2425 {
2426         char *endp;
2427
2428         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2429         endp = skip_spaces(endp);
2430         if (*endp || *value > max)
2431                 return -EINVAL;
2432
2433         return 0;
2434 }
2435
2436 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2437                                               struct ext4_sb_info *sbi,
2438                                               char *buf)
2439 {
2440         return snprintf(buf, PAGE_SIZE, "%llu\n",
2441                 (s64) EXT4_C2B(sbi,
2442                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2443 }
2444
2445 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2446                                          struct ext4_sb_info *sbi, char *buf)
2447 {
2448         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2449
2450         if (!sb->s_bdev->bd_part)
2451                 return snprintf(buf, PAGE_SIZE, "0\n");
2452         return snprintf(buf, PAGE_SIZE, "%lu\n",
2453                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2454                          sbi->s_sectors_written_start) >> 1);
2455 }
2456
2457 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2458                                           struct ext4_sb_info *sbi, char *buf)
2459 {
2460         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2461
2462         if (!sb->s_bdev->bd_part)
2463                 return snprintf(buf, PAGE_SIZE, "0\n");
2464         return snprintf(buf, PAGE_SIZE, "%llu\n",
2465                         (unsigned long long)(sbi->s_kbytes_written +
2466                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2467                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2468 }
2469
2470 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2471                                       struct ext4_sb_info *sbi, char *buf)
2472 {
2473         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2474 }
2475
2476 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2477                                         struct ext4_sb_info *sbi, char *buf)
2478 {
2479         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2480 }
2481
2482 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2483                                           struct ext4_sb_info *sbi,
2484                                           const char *buf, size_t count)
2485 {
2486         unsigned long t;
2487
2488         if (parse_strtoul(buf, 0x40000000, &t))
2489                 return -EINVAL;
2490
2491         if (t && !is_power_of_2(t))
2492                 return -EINVAL;
2493
2494         sbi->s_inode_readahead_blks = t;
2495         return count;
2496 }
2497
2498 static ssize_t sbi_ui_show(struct ext4_attr *a,
2499                            struct ext4_sb_info *sbi, char *buf)
2500 {
2501         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2502
2503         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2504 }
2505
2506 static ssize_t sbi_ui_store(struct ext4_attr *a,
2507                             struct ext4_sb_info *sbi,
2508                             const char *buf, size_t count)
2509 {
2510         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2511         unsigned long t;
2512
2513         if (parse_strtoul(buf, 0xffffffff, &t))
2514                 return -EINVAL;
2515         *ui = t;
2516         return count;
2517 }
2518
2519 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2520 static struct ext4_attr ext4_attr_##_name = {                   \
2521         .attr = {.name = __stringify(_name), .mode = _mode },   \
2522         .show   = _show,                                        \
2523         .store  = _store,                                       \
2524         .offset = offsetof(struct ext4_sb_info, _elname),       \
2525 }
2526 #define EXT4_ATTR(name, mode, show, store) \
2527 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2528
2529 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2530 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2531 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2532 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2533         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2534 #define ATTR_LIST(name) &ext4_attr_##name.attr
2535
2536 EXT4_RO_ATTR(delayed_allocation_blocks);
2537 EXT4_RO_ATTR(session_write_kbytes);
2538 EXT4_RO_ATTR(lifetime_write_kbytes);
2539 EXT4_RO_ATTR(extent_cache_hits);
2540 EXT4_RO_ATTR(extent_cache_misses);
2541 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2542                  inode_readahead_blks_store, s_inode_readahead_blks);
2543 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2544 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2545 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2546 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2547 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2548 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2549 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2550 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2551
2552 static struct attribute *ext4_attrs[] = {
2553         ATTR_LIST(delayed_allocation_blocks),
2554         ATTR_LIST(session_write_kbytes),
2555         ATTR_LIST(lifetime_write_kbytes),
2556         ATTR_LIST(extent_cache_hits),
2557         ATTR_LIST(extent_cache_misses),
2558         ATTR_LIST(inode_readahead_blks),
2559         ATTR_LIST(inode_goal),
2560         ATTR_LIST(mb_stats),
2561         ATTR_LIST(mb_max_to_scan),
2562         ATTR_LIST(mb_min_to_scan),
2563         ATTR_LIST(mb_order2_req),
2564         ATTR_LIST(mb_stream_req),
2565         ATTR_LIST(mb_group_prealloc),
2566         ATTR_LIST(max_writeback_mb_bump),
2567         NULL,
2568 };
2569
2570 /* Features this copy of ext4 supports */
2571 EXT4_INFO_ATTR(lazy_itable_init);
2572 EXT4_INFO_ATTR(batched_discard);
2573
2574 static struct attribute *ext4_feat_attrs[] = {
2575         ATTR_LIST(lazy_itable_init),
2576         ATTR_LIST(batched_discard),
2577         NULL,
2578 };
2579
2580 static ssize_t ext4_attr_show(struct kobject *kobj,
2581                               struct attribute *attr, char *buf)
2582 {
2583         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2584                                                 s_kobj);
2585         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2586
2587         return a->show ? a->show(a, sbi, buf) : 0;
2588 }
2589
2590 static ssize_t ext4_attr_store(struct kobject *kobj,
2591                                struct attribute *attr,
2592                                const char *buf, size_t len)
2593 {
2594         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2595                                                 s_kobj);
2596         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2597
2598         return a->store ? a->store(a, sbi, buf, len) : 0;
2599 }
2600
2601 static void ext4_sb_release(struct kobject *kobj)
2602 {
2603         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2604                                                 s_kobj);
2605         complete(&sbi->s_kobj_unregister);
2606 }
2607
2608 static const struct sysfs_ops ext4_attr_ops = {
2609         .show   = ext4_attr_show,
2610         .store  = ext4_attr_store,
2611 };
2612
2613 static struct kobj_type ext4_ktype = {
2614         .default_attrs  = ext4_attrs,
2615         .sysfs_ops      = &ext4_attr_ops,
2616         .release        = ext4_sb_release,
2617 };
2618
2619 static void ext4_feat_release(struct kobject *kobj)
2620 {
2621         complete(&ext4_feat->f_kobj_unregister);
2622 }
2623
2624 static struct kobj_type ext4_feat_ktype = {
2625         .default_attrs  = ext4_feat_attrs,
2626         .sysfs_ops      = &ext4_attr_ops,
2627         .release        = ext4_feat_release,
2628 };
2629
2630 /*
2631  * Check whether this filesystem can be mounted based on
2632  * the features present and the RDONLY/RDWR mount requested.
2633  * Returns 1 if this filesystem can be mounted as requested,
2634  * 0 if it cannot be.
2635  */
2636 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2637 {
2638         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2639                 ext4_msg(sb, KERN_ERR,
2640                         "Couldn't mount because of "
2641                         "unsupported optional features (%x)",
2642                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2643                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2644                 return 0;
2645         }
2646
2647         if (readonly)
2648                 return 1;
2649
2650         /* Check that feature set is OK for a read-write mount */
2651         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2652                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2653                          "unsupported optional features (%x)",
2654                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2655                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2656                 return 0;
2657         }
2658         /*
2659          * Large file size enabled file system can only be mounted
2660          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2661          */
2662         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2663                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2664                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2665                                  "cannot be mounted RDWR without "
2666                                  "CONFIG_LBDAF");
2667                         return 0;
2668                 }
2669         }
2670         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2671             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2672                 ext4_msg(sb, KERN_ERR,
2673                          "Can't support bigalloc feature without "
2674                          "extents feature\n");
2675                 return 0;
2676         }
2677         return 1;
2678 }
2679
2680 /*
2681  * This function is called once a day if we have errors logged
2682  * on the file system
2683  */
2684 static void print_daily_error_info(unsigned long arg)
2685 {
2686         struct super_block *sb = (struct super_block *) arg;
2687         struct ext4_sb_info *sbi;
2688         struct ext4_super_block *es;
2689
2690         sbi = EXT4_SB(sb);
2691         es = sbi->s_es;
2692
2693         if (es->s_error_count)
2694                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2695                          le32_to_cpu(es->s_error_count));
2696         if (es->s_first_error_time) {
2697                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2698                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2699                        (int) sizeof(es->s_first_error_func),
2700                        es->s_first_error_func,
2701                        le32_to_cpu(es->s_first_error_line));
2702                 if (es->s_first_error_ino)
2703                         printk(": inode %u",
2704                                le32_to_cpu(es->s_first_error_ino));
2705                 if (es->s_first_error_block)
2706                         printk(": block %llu", (unsigned long long)
2707                                le64_to_cpu(es->s_first_error_block));
2708                 printk("\n");
2709         }
2710         if (es->s_last_error_time) {
2711                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2712                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2713                        (int) sizeof(es->s_last_error_func),
2714                        es->s_last_error_func,
2715                        le32_to_cpu(es->s_last_error_line));
2716                 if (es->s_last_error_ino)
2717                         printk(": inode %u",
2718                                le32_to_cpu(es->s_last_error_ino));
2719                 if (es->s_last_error_block)
2720                         printk(": block %llu", (unsigned long long)
2721                                le64_to_cpu(es->s_last_error_block));
2722                 printk("\n");
2723         }
2724         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2725 }
2726
2727 /* Find next suitable group and run ext4_init_inode_table */
2728 static int ext4_run_li_request(struct ext4_li_request *elr)
2729 {
2730         struct ext4_group_desc *gdp = NULL;
2731         ext4_group_t group, ngroups;
2732         struct super_block *sb;
2733         unsigned long timeout = 0;
2734         int ret = 0;
2735
2736         sb = elr->lr_super;
2737         ngroups = EXT4_SB(sb)->s_groups_count;
2738
2739         for (group = elr->lr_next_group; group < ngroups; group++) {
2740                 gdp = ext4_get_group_desc(sb, group, NULL);
2741                 if (!gdp) {
2742                         ret = 1;
2743                         break;
2744                 }
2745
2746                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2747                         break;
2748         }
2749
2750         if (group == ngroups)
2751                 ret = 1;
2752
2753         if (!ret) {
2754                 timeout = jiffies;
2755                 ret = ext4_init_inode_table(sb, group,
2756                                             elr->lr_timeout ? 0 : 1);
2757                 if (elr->lr_timeout == 0) {
2758                         timeout = (jiffies - timeout) *
2759                                   elr->lr_sbi->s_li_wait_mult;
2760                         elr->lr_timeout = timeout;
2761                 }
2762                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2763                 elr->lr_next_group = group + 1;
2764         }
2765
2766         return ret;
2767 }
2768
2769 /*
2770  * Remove lr_request from the list_request and free the
2771  * request structure. Should be called with li_list_mtx held
2772  */
2773 static void ext4_remove_li_request(struct ext4_li_request *elr)
2774 {
2775         struct ext4_sb_info *sbi;
2776
2777         if (!elr)
2778                 return;
2779
2780         sbi = elr->lr_sbi;
2781
2782         list_del(&elr->lr_request);
2783         sbi->s_li_request = NULL;
2784         kfree(elr);
2785 }
2786
2787 static void ext4_unregister_li_request(struct super_block *sb)
2788 {
2789         mutex_lock(&ext4_li_mtx);
2790         if (!ext4_li_info) {
2791                 mutex_unlock(&ext4_li_mtx);
2792                 return;
2793         }
2794
2795         mutex_lock(&ext4_li_info->li_list_mtx);
2796         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2797         mutex_unlock(&ext4_li_info->li_list_mtx);
2798         mutex_unlock(&ext4_li_mtx);
2799 }
2800
2801 static struct task_struct *ext4_lazyinit_task;
2802
2803 /*
2804  * This is the function where ext4lazyinit thread lives. It walks
2805  * through the request list searching for next scheduled filesystem.
2806  * When such a fs is found, run the lazy initialization request
2807  * (ext4_rn_li_request) and keep track of the time spend in this
2808  * function. Based on that time we compute next schedule time of
2809  * the request. When walking through the list is complete, compute
2810  * next waking time and put itself into sleep.
2811  */
2812 static int ext4_lazyinit_thread(void *arg)
2813 {
2814         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2815         struct list_head *pos, *n;
2816         struct ext4_li_request *elr;
2817         unsigned long next_wakeup, cur;
2818
2819         BUG_ON(NULL == eli);
2820
2821 cont_thread:
2822         while (true) {
2823                 next_wakeup = MAX_JIFFY_OFFSET;
2824
2825                 mutex_lock(&eli->li_list_mtx);
2826                 if (list_empty(&eli->li_request_list)) {
2827                         mutex_unlock(&eli->li_list_mtx);
2828                         goto exit_thread;
2829                 }
2830
2831                 list_for_each_safe(pos, n, &eli->li_request_list) {
2832                         elr = list_entry(pos, struct ext4_li_request,
2833                                          lr_request);
2834
2835                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2836                                 if (ext4_run_li_request(elr) != 0) {
2837                                         /* error, remove the lazy_init job */
2838                                         ext4_remove_li_request(elr);
2839                                         continue;
2840                                 }
2841                         }
2842
2843                         if (time_before(elr->lr_next_sched, next_wakeup))
2844                                 next_wakeup = elr->lr_next_sched;
2845                 }
2846                 mutex_unlock(&eli->li_list_mtx);
2847
2848                 try_to_freeze();
2849
2850                 cur = jiffies;
2851                 if ((time_after_eq(cur, next_wakeup)) ||
2852                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2853                         cond_resched();
2854                         continue;
2855                 }
2856
2857                 schedule_timeout_interruptible(next_wakeup - cur);
2858
2859                 if (kthread_should_stop()) {
2860                         ext4_clear_request_list();
2861                         goto exit_thread;
2862                 }
2863         }
2864
2865 exit_thread:
2866         /*
2867          * It looks like the request list is empty, but we need
2868          * to check it under the li_list_mtx lock, to prevent any
2869          * additions into it, and of course we should lock ext4_li_mtx
2870          * to atomically free the list and ext4_li_info, because at
2871          * this point another ext4 filesystem could be registering
2872          * new one.
2873          */
2874         mutex_lock(&ext4_li_mtx);
2875         mutex_lock(&eli->li_list_mtx);
2876         if (!list_empty(&eli->li_request_list)) {
2877                 mutex_unlock(&eli->li_list_mtx);
2878                 mutex_unlock(&ext4_li_mtx);
2879                 goto cont_thread;
2880         }
2881         mutex_unlock(&eli->li_list_mtx);
2882         kfree(ext4_li_info);
2883         ext4_li_info = NULL;
2884         mutex_unlock(&ext4_li_mtx);
2885
2886         return 0;
2887 }
2888
2889 static void ext4_clear_request_list(void)
2890 {
2891         struct list_head *pos, *n;
2892         struct ext4_li_request *elr;
2893
2894         mutex_lock(&ext4_li_info->li_list_mtx);
2895         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2896                 elr = list_entry(pos, struct ext4_li_request,
2897                                  lr_request);
2898                 ext4_remove_li_request(elr);
2899         }
2900         mutex_unlock(&ext4_li_info->li_list_mtx);
2901 }
2902
2903 static int ext4_run_lazyinit_thread(void)
2904 {
2905         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2906                                          ext4_li_info, "ext4lazyinit");
2907         if (IS_ERR(ext4_lazyinit_task)) {
2908                 int err = PTR_ERR(ext4_lazyinit_task);
2909                 ext4_clear_request_list();
2910                 kfree(ext4_li_info);
2911                 ext4_li_info = NULL;
2912                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2913                                  "initialization thread\n",
2914                                  err);
2915                 return err;
2916         }
2917         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2918         return 0;
2919 }
2920
2921 /*
2922  * Check whether it make sense to run itable init. thread or not.
2923  * If there is at least one uninitialized inode table, return
2924  * corresponding group number, else the loop goes through all
2925  * groups and return total number of groups.
2926  */
2927 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2928 {
2929         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2930         struct ext4_group_desc *gdp = NULL;
2931
2932         for (group = 0; group < ngroups; group++) {
2933                 gdp = ext4_get_group_desc(sb, group, NULL);
2934                 if (!gdp)
2935                         continue;
2936
2937                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2938                         break;
2939         }
2940
2941         return group;
2942 }
2943
2944 static int ext4_li_info_new(void)
2945 {
2946         struct ext4_lazy_init *eli = NULL;
2947
2948         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2949         if (!eli)
2950                 return -ENOMEM;
2951
2952         INIT_LIST_HEAD(&eli->li_request_list);
2953         mutex_init(&eli->li_list_mtx);
2954
2955         eli->li_state |= EXT4_LAZYINIT_QUIT;
2956
2957         ext4_li_info = eli;
2958
2959         return 0;
2960 }
2961
2962 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2963                                             ext4_group_t start)
2964 {
2965         struct ext4_sb_info *sbi = EXT4_SB(sb);
2966         struct ext4_li_request *elr;
2967         unsigned long rnd;
2968
2969         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2970         if (!elr)
2971                 return NULL;
2972
2973         elr->lr_super = sb;
2974         elr->lr_sbi = sbi;
2975         elr->lr_next_group = start;
2976
2977         /*
2978          * Randomize first schedule time of the request to
2979          * spread the inode table initialization requests
2980          * better.
2981          */
2982         get_random_bytes(&rnd, sizeof(rnd));
2983         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2984                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2985
2986         return elr;
2987 }
2988
2989 static int ext4_register_li_request(struct super_block *sb,
2990                                     ext4_group_t first_not_zeroed)
2991 {
2992         struct ext4_sb_info *sbi = EXT4_SB(sb);
2993         struct ext4_li_request *elr;
2994         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2995         int ret = 0;
2996
2997         if (sbi->s_li_request != NULL) {
2998                 /*
2999                  * Reset timeout so it can be computed again, because
3000                  * s_li_wait_mult might have changed.
3001                  */
3002                 sbi->s_li_request->lr_timeout = 0;
3003                 return 0;
3004         }
3005
3006         if (first_not_zeroed == ngroups ||
3007             (sb->s_flags & MS_RDONLY) ||
3008             !test_opt(sb, INIT_INODE_TABLE))
3009                 return 0;
3010
3011         elr = ext4_li_request_new(sb, first_not_zeroed);
3012         if (!elr)
3013                 return -ENOMEM;
3014
3015         mutex_lock(&ext4_li_mtx);
3016
3017         if (NULL == ext4_li_info) {
3018                 ret = ext4_li_info_new();
3019                 if (ret)
3020                         goto out;
3021         }
3022
3023         mutex_lock(&ext4_li_info->li_list_mtx);
3024         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3025         mutex_unlock(&ext4_li_info->li_list_mtx);
3026
3027         sbi->s_li_request = elr;
3028         /*
3029          * set elr to NULL here since it has been inserted to
3030          * the request_list and the removal and free of it is
3031          * handled by ext4_clear_request_list from now on.
3032          */
3033         elr = NULL;
3034
3035         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3036                 ret = ext4_run_lazyinit_thread();
3037                 if (ret)
3038                         goto out;
3039         }
3040 out:
3041         mutex_unlock(&ext4_li_mtx);
3042         if (ret)
3043                 kfree(elr);
3044         return ret;
3045 }
3046
3047 /*
3048  * We do not need to lock anything since this is called on
3049  * module unload.
3050  */
3051 static void ext4_destroy_lazyinit_thread(void)
3052 {
3053         /*
3054          * If thread exited earlier
3055          * there's nothing to be done.
3056          */
3057         if (!ext4_li_info || !ext4_lazyinit_task)
3058                 return;
3059
3060         kthread_stop(ext4_lazyinit_task);
3061 }
3062
3063 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3064 {
3065         char *orig_data = kstrdup(data, GFP_KERNEL);
3066         struct buffer_head *bh;
3067         struct ext4_super_block *es = NULL;
3068         struct ext4_sb_info *sbi;
3069         ext4_fsblk_t block;
3070         ext4_fsblk_t sb_block = get_sb_block(&data);
3071         ext4_fsblk_t logical_sb_block;
3072         unsigned long offset = 0;
3073         unsigned long journal_devnum = 0;
3074         unsigned long def_mount_opts;
3075         struct inode *root;
3076         char *cp;
3077         const char *descr;
3078         int ret = -ENOMEM;
3079         int blocksize, clustersize;
3080         unsigned int db_count;
3081         unsigned int i;
3082         int needs_recovery, has_huge_files, has_bigalloc;
3083         __u64 blocks_count;
3084         int err;
3085         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3086         ext4_group_t first_not_zeroed;
3087
3088         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3089         if (!sbi)
3090                 goto out_free_orig;
3091
3092         sbi->s_blockgroup_lock =
3093                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3094         if (!sbi->s_blockgroup_lock) {
3095                 kfree(sbi);
3096                 goto out_free_orig;
3097         }
3098         sb->s_fs_info = sbi;
3099         sbi->s_mount_opt = 0;
3100         sbi->s_resuid = EXT4_DEF_RESUID;
3101         sbi->s_resgid = EXT4_DEF_RESGID;
3102         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3103         sbi->s_sb_block = sb_block;
3104         if (sb->s_bdev->bd_part)
3105                 sbi->s_sectors_written_start =
3106                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3107
3108         /* Cleanup superblock name */
3109         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3110                 *cp = '!';
3111
3112         ret = -EINVAL;
3113         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3114         if (!blocksize) {
3115                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3116                 goto out_fail;
3117         }
3118
3119         /*
3120          * The ext4 superblock will not be buffer aligned for other than 1kB
3121          * block sizes.  We need to calculate the offset from buffer start.
3122          */
3123         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3124                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3125                 offset = do_div(logical_sb_block, blocksize);
3126         } else {
3127                 logical_sb_block = sb_block;
3128         }
3129
3130         if (!(bh = sb_bread(sb, logical_sb_block))) {
3131                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3132                 goto out_fail;
3133         }
3134         /*
3135          * Note: s_es must be initialized as soon as possible because
3136          *       some ext4 macro-instructions depend on its value
3137          */
3138         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3139         sbi->s_es = es;
3140         sb->s_magic = le16_to_cpu(es->s_magic);
3141         if (sb->s_magic != EXT4_SUPER_MAGIC)
3142                 goto cantfind_ext4;
3143         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3144
3145         /* Set defaults before we parse the mount options */
3146         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3147         set_opt(sb, INIT_INODE_TABLE);
3148         if (def_mount_opts & EXT4_DEFM_DEBUG)
3149                 set_opt(sb, DEBUG);
3150         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3151                 set_opt(sb, GRPID);
3152         if (def_mount_opts & EXT4_DEFM_UID16)
3153                 set_opt(sb, NO_UID32);
3154         /* xattr user namespace & acls are now defaulted on */
3155 #ifdef CONFIG_EXT4_FS_XATTR
3156         set_opt(sb, XATTR_USER);
3157 #endif
3158 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3159         set_opt(sb, POSIX_ACL);
3160 #endif
3161         set_opt(sb, MBLK_IO_SUBMIT);
3162         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3163                 set_opt(sb, JOURNAL_DATA);
3164         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3165                 set_opt(sb, ORDERED_DATA);
3166         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3167                 set_opt(sb, WRITEBACK_DATA);
3168
3169         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3170                 set_opt(sb, ERRORS_PANIC);
3171         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3172                 set_opt(sb, ERRORS_CONT);
3173         else
3174                 set_opt(sb, ERRORS_RO);
3175         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3176                 set_opt(sb, BLOCK_VALIDITY);
3177         if (def_mount_opts & EXT4_DEFM_DISCARD)
3178                 set_opt(sb, DISCARD);
3179
3180         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3181         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3182         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3183         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3184         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3185
3186         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3187                 set_opt(sb, BARRIER);
3188
3189         /*
3190          * enable delayed allocation by default
3191          * Use -o nodelalloc to turn it off
3192          */
3193         if (!IS_EXT3_SB(sb) &&
3194             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3195                 set_opt(sb, DELALLOC);
3196
3197         /*
3198          * set default s_li_wait_mult for lazyinit, for the case there is
3199          * no mount option specified.
3200          */
3201         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3202
3203         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3204                            &journal_devnum, &journal_ioprio, 0)) {
3205                 ext4_msg(sb, KERN_WARNING,
3206                          "failed to parse options in superblock: %s",
3207                          sbi->s_es->s_mount_opts);
3208         }
3209         if (!parse_options((char *) data, sb, &journal_devnum,
3210                            &journal_ioprio, 0))
3211                 goto failed_mount;
3212
3213         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3214                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3215                             "with data=journal disables delayed "
3216                             "allocation and O_DIRECT support!\n");
3217                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3218                         ext4_msg(sb, KERN_ERR, "can't mount with "
3219                                  "both data=journal and delalloc");
3220                         goto failed_mount;
3221                 }
3222                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3223                         ext4_msg(sb, KERN_ERR, "can't mount with "
3224                                  "both data=journal and delalloc");
3225                         goto failed_mount;
3226                 }
3227                 if (test_opt(sb, DELALLOC))
3228                         clear_opt(sb, DELALLOC);
3229         }
3230
3231         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3232         if (test_opt(sb, DIOREAD_NOLOCK)) {
3233                 if (blocksize < PAGE_SIZE) {
3234                         ext4_msg(sb, KERN_ERR, "can't mount with "
3235                                  "dioread_nolock if block size != PAGE_SIZE");
3236                         goto failed_mount;
3237                 }
3238         }
3239
3240         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3241                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3242
3243         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3244             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3245              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3246              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3247                 ext4_msg(sb, KERN_WARNING,
3248                        "feature flags set on rev 0 fs, "
3249                        "running e2fsck is recommended");
3250
3251         if (IS_EXT2_SB(sb)) {
3252                 if (ext2_feature_set_ok(sb))
3253                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3254                                  "using the ext4 subsystem");
3255                 else {
3256                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3257                                  "to feature incompatibilities");
3258                         goto failed_mount;
3259                 }
3260         }
3261
3262         if (IS_EXT3_SB(sb)) {
3263                 if (ext3_feature_set_ok(sb))
3264                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3265                                  "using the ext4 subsystem");
3266                 else {
3267                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3268                                  "to feature incompatibilities");
3269                         goto failed_mount;
3270                 }
3271         }
3272
3273         /*
3274          * Check feature flags regardless of the revision level, since we
3275          * previously didn't change the revision level when setting the flags,
3276          * so there is a chance incompat flags are set on a rev 0 filesystem.
3277          */
3278         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3279                 goto failed_mount;
3280
3281         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3282             blocksize > EXT4_MAX_BLOCK_SIZE) {
3283                 ext4_msg(sb, KERN_ERR,
3284                        "Unsupported filesystem blocksize %d", blocksize);
3285                 goto failed_mount;
3286         }
3287
3288         if (sb->s_blocksize != blocksize) {
3289                 /* Validate the filesystem blocksize */
3290                 if (!sb_set_blocksize(sb, blocksize)) {
3291                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3292                                         blocksize);
3293                         goto failed_mount;
3294                 }
3295
3296                 brelse(bh);
3297                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3298                 offset = do_div(logical_sb_block, blocksize);
3299                 bh = sb_bread(sb, logical_sb_block);
3300                 if (!bh) {
3301                         ext4_msg(sb, KERN_ERR,
3302                                "Can't read superblock on 2nd try");
3303                         goto failed_mount;
3304                 }
3305                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3306                 sbi->s_es = es;
3307                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3308                         ext4_msg(sb, KERN_ERR,
3309                                "Magic mismatch, very weird!");
3310                         goto failed_mount;
3311                 }
3312         }
3313
3314         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3315                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3316         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3317                                                       has_huge_files);
3318         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3319
3320         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3321                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3322                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3323         } else {
3324                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3325                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3326                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3327                     (!is_power_of_2(sbi->s_inode_size)) ||
3328                     (sbi->s_inode_size > blocksize)) {
3329                         ext4_msg(sb, KERN_ERR,
3330                                "unsupported inode size: %d",
3331                                sbi->s_inode_size);
3332                         goto failed_mount;
3333                 }
3334                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3335                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3336         }
3337
3338         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3339         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3340                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3341                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3342                     !is_power_of_2(sbi->s_desc_size)) {
3343                         ext4_msg(sb, KERN_ERR,
3344                                "unsupported descriptor size %lu",
3345                                sbi->s_desc_size);
3346                         goto failed_mount;
3347                 }
3348         } else
3349                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3350
3351         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3352         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3353         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3354                 goto cantfind_ext4;
3355
3356         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3357         if (sbi->s_inodes_per_block == 0)
3358                 goto cantfind_ext4;
3359         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3360                                         sbi->s_inodes_per_block;
3361         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3362         sbi->s_sbh = bh;
3363         sbi->s_mount_state = le16_to_cpu(es->s_state);
3364         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3365         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3366
3367         for (i = 0; i < 4; i++)
3368                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3369         sbi->s_def_hash_version = es->s_def_hash_version;
3370         i = le32_to_cpu(es->s_flags);
3371         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3372                 sbi->s_hash_unsigned = 3;
3373         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3374 #ifdef __CHAR_UNSIGNED__
3375                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3376                 sbi->s_hash_unsigned = 3;
3377 #else
3378                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3379 #endif
3380                 sb->s_dirt = 1;
3381         }
3382
3383         /* Handle clustersize */
3384         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3385         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3386                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3387         if (has_bigalloc) {
3388                 if (clustersize < blocksize) {
3389                         ext4_msg(sb, KERN_ERR,
3390                                  "cluster size (%d) smaller than "
3391                                  "block size (%d)", clustersize, blocksize);
3392                         goto failed_mount;
3393                 }
3394                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3395                         le32_to_cpu(es->s_log_block_size);
3396                 sbi->s_clusters_per_group =
3397                         le32_to_cpu(es->s_clusters_per_group);
3398                 if (sbi->s_clusters_per_group > blocksize * 8) {
3399                         ext4_msg(sb, KERN_ERR,
3400                                  "#clusters per group too big: %lu",
3401                                  sbi->s_clusters_per_group);
3402                         goto failed_mount;
3403                 }
3404                 if (sbi->s_blocks_per_group !=
3405                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3406                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3407                                  "clusters per group (%lu) inconsistent",
3408                                  sbi->s_blocks_per_group,
3409                                  sbi->s_clusters_per_group);
3410                         goto failed_mount;
3411                 }
3412         } else {
3413                 if (clustersize != blocksize) {
3414                         ext4_warning(sb, "fragment/cluster size (%d) != "
3415                                      "block size (%d)", clustersize,
3416                                      blocksize);
3417                         clustersize = blocksize;
3418                 }
3419                 if (sbi->s_blocks_per_group > blocksize * 8) {
3420                         ext4_msg(sb, KERN_ERR,
3421                                  "#blocks per group too big: %lu",
3422                                  sbi->s_blocks_per_group);
3423                         goto failed_mount;
3424                 }
3425                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3426                 sbi->s_cluster_bits = 0;
3427         }
3428         sbi->s_cluster_ratio = clustersize / blocksize;
3429
3430         if (sbi->s_inodes_per_group > blocksize * 8) {
3431                 ext4_msg(sb, KERN_ERR,
3432                        "#inodes per group too big: %lu",
3433                        sbi->s_inodes_per_group);
3434                 goto failed_mount;
3435         }
3436
3437         /*
3438          * Test whether we have more sectors than will fit in sector_t,
3439          * and whether the max offset is addressable by the page cache.
3440          */
3441         err = generic_check_addressable(sb->s_blocksize_bits,
3442                                         ext4_blocks_count(es));
3443         if (err) {
3444                 ext4_msg(sb, KERN_ERR, "filesystem"
3445                          " too large to mount safely on this system");
3446                 if (sizeof(sector_t) < 8)
3447                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3448                 ret = err;
3449                 goto failed_mount;
3450         }
3451
3452         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3453                 goto cantfind_ext4;
3454
3455         /* check blocks count against device size */
3456         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3457         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3458                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3459                        "exceeds size of device (%llu blocks)",
3460                        ext4_blocks_count(es), blocks_count);
3461                 goto failed_mount;
3462         }
3463
3464         /*
3465          * It makes no sense for the first data block to be beyond the end
3466          * of the filesystem.
3467          */
3468         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3469                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3470                          "block %u is beyond end of filesystem (%llu)",
3471                          le32_to_cpu(es->s_first_data_block),
3472                          ext4_blocks_count(es));
3473                 goto failed_mount;
3474         }
3475         blocks_count = (ext4_blocks_count(es) -
3476                         le32_to_cpu(es->s_first_data_block) +
3477                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3478         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3479         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3480                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3481                        "(block count %llu, first data block %u, "
3482                        "blocks per group %lu)", sbi->s_groups_count,
3483                        ext4_blocks_count(es),
3484                        le32_to_cpu(es->s_first_data_block),
3485                        EXT4_BLOCKS_PER_GROUP(sb));
3486                 goto failed_mount;
3487         }
3488         sbi->s_groups_count = blocks_count;
3489         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3490                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3491         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3492                    EXT4_DESC_PER_BLOCK(sb);
3493         sbi->s_group_desc = ext4_kvmalloc(db_count *
3494                                           sizeof(struct buffer_head *),
3495                                           GFP_KERNEL);
3496         if (sbi->s_group_desc == NULL) {
3497                 ext4_msg(sb, KERN_ERR, "not enough memory");
3498                 goto failed_mount;
3499         }
3500
3501         if (ext4_proc_root)
3502                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3503
3504         bgl_lock_init(sbi->s_blockgroup_lock);
3505
3506         for (i = 0; i < db_count; i++) {
3507                 block = descriptor_loc(sb, logical_sb_block, i);
3508                 sbi->s_group_desc[i] = sb_bread(sb, block);
3509                 if (!sbi->s_group_desc[i]) {
3510                         ext4_msg(sb, KERN_ERR,
3511                                "can't read group descriptor %d", i);
3512                         db_count = i;
3513                         goto failed_mount2;
3514                 }
3515         }
3516         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3517                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3518                 goto failed_mount2;
3519         }
3520         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3521                 if (!ext4_fill_flex_info(sb)) {
3522                         ext4_msg(sb, KERN_ERR,
3523                                "unable to initialize "
3524                                "flex_bg meta info!");
3525                         goto failed_mount2;
3526                 }
3527
3528         sbi->s_gdb_count = db_count;
3529         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3530         spin_lock_init(&sbi->s_next_gen_lock);
3531
3532         init_timer(&sbi->s_err_report);
3533         sbi->s_err_report.function = print_daily_error_info;
3534         sbi->s_err_report.data = (unsigned long) sb;
3535
3536         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3537                         ext4_count_free_clusters(sb));
3538         if (!err) {
3539                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3540                                 ext4_count_free_inodes(sb));
3541         }
3542         if (!err) {
3543                 err = percpu_counter_init(&sbi->s_dirs_counter,
3544                                 ext4_count_dirs(sb));
3545         }
3546         if (!err) {
3547                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3548         }
3549         if (err) {
3550                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3551                 goto failed_mount3;
3552         }
3553
3554         sbi->s_stripe = ext4_get_stripe_size(sbi);
3555         sbi->s_max_writeback_mb_bump = 128;
3556
3557         /*
3558          * set up enough so that it can read an inode
3559          */
3560         if (!test_opt(sb, NOLOAD) &&
3561             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3562                 sb->s_op = &ext4_sops;
3563         else
3564                 sb->s_op = &ext4_nojournal_sops;
3565         sb->s_export_op = &ext4_export_ops;
3566         sb->s_xattr = ext4_xattr_handlers;
3567 #ifdef CONFIG_QUOTA
3568         sb->s_qcop = &ext4_qctl_operations;
3569         sb->dq_op = &ext4_quota_operations;
3570 #endif
3571         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3572
3573         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3574         mutex_init(&sbi->s_orphan_lock);
3575         sbi->s_resize_flags = 0;
3576
3577         sb->s_root = NULL;
3578
3579         needs_recovery = (es->s_last_orphan != 0 ||
3580                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3581                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3582
3583         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3584             !(sb->s_flags & MS_RDONLY))
3585                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3586                         goto failed_mount3;
3587
3588         /*
3589          * The first inode we look at is the journal inode.  Don't try
3590          * root first: it may be modified in the journal!
3591          */
3592         if (!test_opt(sb, NOLOAD) &&
3593             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3594                 if (ext4_load_journal(sb, es, journal_devnum))
3595                         goto failed_mount3;
3596         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3597               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3598                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3599                        "suppressed and not mounted read-only");
3600                 goto failed_mount_wq;
3601         } else {
3602                 clear_opt(sb, DATA_FLAGS);
3603                 sbi->s_journal = NULL;
3604                 needs_recovery = 0;
3605                 goto no_journal;
3606         }
3607
3608         if (ext4_blocks_count(es) > 0xffffffffULL &&
3609             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3610                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3611                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3612                 goto failed_mount_wq;
3613         }
3614
3615         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3616                 jbd2_journal_set_features(sbi->s_journal,
3617                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3618                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3619         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3620                 jbd2_journal_set_features(sbi->s_journal,
3621                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3622                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3623                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3624         } else {
3625                 jbd2_journal_clear_features(sbi->s_journal,
3626                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3627                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3628         }
3629
3630         /* We have now updated the journal if required, so we can
3631          * validate the data journaling mode. */
3632         switch (test_opt(sb, DATA_FLAGS)) {
3633         case 0:
3634                 /* No mode set, assume a default based on the journal
3635                  * capabilities: ORDERED_DATA if the journal can
3636                  * cope, else JOURNAL_DATA
3637                  */
3638                 if (jbd2_journal_check_available_features
3639                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3640                         set_opt(sb, ORDERED_DATA);
3641                 else
3642                         set_opt(sb, JOURNAL_DATA);
3643                 break;
3644
3645         case EXT4_MOUNT_ORDERED_DATA:
3646         case EXT4_MOUNT_WRITEBACK_DATA:
3647                 if (!jbd2_journal_check_available_features
3648                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3649                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3650                                "requested data journaling mode");
3651                         goto failed_mount_wq;
3652                 }
3653         default:
3654                 break;
3655         }
3656         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3657
3658         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3659
3660         /*
3661          * The journal may have updated the bg summary counts, so we
3662          * need to update the global counters.
3663          */
3664         percpu_counter_set(&sbi->s_freeclusters_counter,
3665                            ext4_count_free_clusters(sb));
3666         percpu_counter_set(&sbi->s_freeinodes_counter,
3667                            ext4_count_free_inodes(sb));
3668         percpu_counter_set(&sbi->s_dirs_counter,
3669                            ext4_count_dirs(sb));
3670         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3671
3672 no_journal:
3673         /*
3674          * The maximum number of concurrent works can be high and
3675          * concurrency isn't really necessary.  Limit it to 1.
3676          */
3677         EXT4_SB(sb)->dio_unwritten_wq =
3678                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3679         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3680                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3681                 goto failed_mount_wq;
3682         }
3683
3684         /*
3685          * The jbd2_journal_load will have done any necessary log recovery,
3686          * so we can safely mount the rest of the filesystem now.
3687          */
3688
3689         root = ext4_iget(sb, EXT4_ROOT_INO);
3690         if (IS_ERR(root)) {
3691                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3692                 ret = PTR_ERR(root);
3693                 root = NULL;
3694                 goto failed_mount4;
3695         }
3696         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3697                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3698                 iput(root);
3699                 goto failed_mount4;
3700         }
3701         sb->s_root = d_alloc_root(root);
3702         if (!sb->s_root) {
3703                 iput(root);
3704                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3705                 ret = -ENOMEM;
3706                 goto failed_mount4;
3707         }
3708
3709         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3710
3711         /* determine the minimum size of new large inodes, if present */
3712         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3713                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3714                                                      EXT4_GOOD_OLD_INODE_SIZE;
3715                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3716                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3717                         if (sbi->s_want_extra_isize <
3718                             le16_to_cpu(es->s_want_extra_isize))
3719                                 sbi->s_want_extra_isize =
3720                                         le16_to_cpu(es->s_want_extra_isize);
3721                         if (sbi->s_want_extra_isize <
3722                             le16_to_cpu(es->s_min_extra_isize))
3723                                 sbi->s_want_extra_isize =
3724                                         le16_to_cpu(es->s_min_extra_isize);
3725                 }
3726         }
3727         /* Check if enough inode space is available */
3728         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3729                                                         sbi->s_inode_size) {
3730                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3731                                                        EXT4_GOOD_OLD_INODE_SIZE;
3732                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3733                          "available");
3734         }
3735
3736         err = ext4_setup_system_zone(sb);
3737         if (err) {
3738                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3739                          "zone (%d)", err);
3740                 goto failed_mount4a;
3741         }
3742
3743         ext4_ext_init(sb);
3744         err = ext4_mb_init(sb, needs_recovery);
3745         if (err) {
3746                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3747                          err);
3748                 goto failed_mount5;
3749         }
3750
3751         err = ext4_register_li_request(sb, first_not_zeroed);
3752         if (err)
3753                 goto failed_mount6;
3754
3755         sbi->s_kobj.kset = ext4_kset;
3756         init_completion(&sbi->s_kobj_unregister);
3757         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3758                                    "%s", sb->s_id);
3759         if (err)
3760                 goto failed_mount7;
3761
3762         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3763         ext4_orphan_cleanup(sb, es);
3764         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3765         if (needs_recovery) {
3766                 ext4_msg(sb, KERN_INFO, "recovery complete");
3767                 ext4_mark_recovery_complete(sb, es);
3768         }
3769         if (EXT4_SB(sb)->s_journal) {
3770                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3771                         descr = " journalled data mode";
3772                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3773                         descr = " ordered data mode";
3774                 else
3775                         descr = " writeback data mode";
3776         } else
3777                 descr = "out journal";
3778
3779         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3780                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3781                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3782
3783         if (es->s_error_count)
3784                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3785
3786         kfree(orig_data);
3787         return 0;
3788
3789 cantfind_ext4:
3790         if (!silent)
3791                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3792         goto failed_mount;
3793
3794 failed_mount7:
3795         ext4_unregister_li_request(sb);
3796 failed_mount6:
3797         ext4_mb_release(sb);
3798 failed_mount5:
3799         ext4_ext_release(sb);
3800         ext4_release_system_zone(sb);
3801 failed_mount4a:
3802         dput(sb->s_root);
3803         sb->s_root = NULL;
3804 failed_mount4:
3805         ext4_msg(sb, KERN_ERR, "mount failed");
3806         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3807 failed_mount_wq:
3808         if (sbi->s_journal) {
3809                 jbd2_journal_destroy(sbi->s_journal);
3810                 sbi->s_journal = NULL;
3811         }
3812 failed_mount3:
3813         del_timer(&sbi->s_err_report);
3814         if (sbi->s_flex_groups)
3815                 ext4_kvfree(sbi->s_flex_groups);
3816         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3817         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3818         percpu_counter_destroy(&sbi->s_dirs_counter);
3819         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3820         if (sbi->s_mmp_tsk)
3821                 kthread_stop(sbi->s_mmp_tsk);
3822 failed_mount2:
3823         for (i = 0; i < db_count; i++)
3824                 brelse(sbi->s_group_desc[i]);
3825         ext4_kvfree(sbi->s_group_desc);
3826 failed_mount:
3827         if (sbi->s_proc) {
3828                 remove_proc_entry(sb->s_id, ext4_proc_root);
3829         }
3830 #ifdef CONFIG_QUOTA
3831         for (i = 0; i < MAXQUOTAS; i++)
3832                 kfree(sbi->s_qf_names[i]);
3833 #endif
3834         ext4_blkdev_remove(sbi);
3835         brelse(bh);
3836 out_fail:
3837         sb->s_fs_info = NULL;
3838         kfree(sbi->s_blockgroup_lock);
3839         kfree(sbi);
3840 out_free_orig:
3841         kfree(orig_data);
3842         return ret;
3843 }
3844
3845 /*
3846  * Setup any per-fs journal parameters now.  We'll do this both on
3847  * initial mount, once the journal has been initialised but before we've
3848  * done any recovery; and again on any subsequent remount.
3849  */
3850 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3851 {
3852         struct ext4_sb_info *sbi = EXT4_SB(sb);
3853
3854         journal->j_commit_interval = sbi->s_commit_interval;
3855         journal->j_min_batch_time = sbi->s_min_batch_time;
3856         journal->j_max_batch_time = sbi->s_max_batch_time;
3857
3858         write_lock(&journal->j_state_lock);
3859         if (test_opt(sb, BARRIER))
3860                 journal->j_flags |= JBD2_BARRIER;
3861         else
3862                 journal->j_flags &= ~JBD2_BARRIER;
3863         if (test_opt(sb, DATA_ERR_ABORT))
3864                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3865         else
3866                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3867         write_unlock(&journal->j_state_lock);
3868 }
3869
3870 static journal_t *ext4_get_journal(struct super_block *sb,
3871                                    unsigned int journal_inum)
3872 {
3873         struct inode *journal_inode;
3874         journal_t *journal;
3875
3876         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3877
3878         /* First, test for the existence of a valid inode on disk.  Bad
3879          * things happen if we iget() an unused inode, as the subsequent
3880          * iput() will try to delete it. */
3881
3882         journal_inode = ext4_iget(sb, journal_inum);
3883         if (IS_ERR(journal_inode)) {
3884                 ext4_msg(sb, KERN_ERR, "no journal found");
3885                 return NULL;
3886         }
3887         if (!journal_inode->i_nlink) {
3888                 make_bad_inode(journal_inode);
3889                 iput(journal_inode);
3890                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3891                 return NULL;
3892         }
3893
3894         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3895                   journal_inode, journal_inode->i_size);
3896         if (!S_ISREG(journal_inode->i_mode)) {
3897                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3898                 iput(journal_inode);
3899                 return NULL;
3900         }
3901
3902         journal = jbd2_journal_init_inode(journal_inode);
3903         if (!journal) {
3904                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3905                 iput(journal_inode);
3906                 return NULL;
3907         }
3908         journal->j_private = sb;
3909         ext4_init_journal_params(sb, journal);
3910         return journal;
3911 }
3912
3913 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3914                                        dev_t j_dev)
3915 {
3916         struct buffer_head *bh;
3917         journal_t *journal;
3918         ext4_fsblk_t start;
3919         ext4_fsblk_t len;
3920         int hblock, blocksize;
3921         ext4_fsblk_t sb_block;
3922         unsigned long offset;
3923         struct ext4_super_block *es;
3924         struct block_device *bdev;
3925
3926         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3927
3928         bdev = ext4_blkdev_get(j_dev, sb);
3929         if (bdev == NULL)
3930                 return NULL;
3931
3932         blocksize = sb->s_blocksize;
3933         hblock = bdev_logical_block_size(bdev);
3934         if (blocksize < hblock) {
3935                 ext4_msg(sb, KERN_ERR,
3936                         "blocksize too small for journal device");
3937                 goto out_bdev;
3938         }
3939
3940         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3941         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3942         set_blocksize(bdev, blocksize);
3943         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3944                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3945                        "external journal");
3946                 goto out_bdev;
3947         }
3948
3949         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3950         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3951             !(le32_to_cpu(es->s_feature_incompat) &
3952               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3953                 ext4_msg(sb, KERN_ERR, "external journal has "
3954                                         "bad superblock");
3955                 brelse(bh);
3956                 goto out_bdev;
3957         }
3958
3959         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3960                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3961                 brelse(bh);
3962                 goto out_bdev;
3963         }
3964
3965         len = ext4_blocks_count(es);
3966         start = sb_block + 1;
3967         brelse(bh);     /* we're done with the superblock */
3968
3969         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3970                                         start, len, blocksize);
3971         if (!journal) {
3972                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3973                 goto out_bdev;
3974         }
3975         journal->j_private = sb;
3976         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3977         wait_on_buffer(journal->j_sb_buffer);
3978         if (!buffer_uptodate(journal->j_sb_buffer)) {
3979                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3980                 goto out_journal;
3981         }
3982         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3983                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3984                                         "user (unsupported) - %d",
3985                         be32_to_cpu(journal->j_superblock->s_nr_users));
3986                 goto out_journal;
3987         }
3988         EXT4_SB(sb)->journal_bdev = bdev;
3989         ext4_init_journal_params(sb, journal);
3990         return journal;
3991
3992 out_journal:
3993         jbd2_journal_destroy(journal);
3994 out_bdev:
3995         ext4_blkdev_put(bdev);
3996         return NULL;
3997 }
3998
3999 static int ext4_load_journal(struct super_block *sb,
4000                              struct ext4_super_block *es,
4001                              unsigned long journal_devnum)
4002 {
4003         journal_t *journal;
4004         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4005         dev_t journal_dev;
4006         int err = 0;
4007         int really_read_only;
4008
4009         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4010
4011         if (journal_devnum &&
4012             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4013                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4014                         "numbers have changed");
4015                 journal_dev = new_decode_dev(journal_devnum);
4016         } else
4017                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4018
4019         really_read_only = bdev_read_only(sb->s_bdev);
4020
4021         /*
4022          * Are we loading a blank journal or performing recovery after a
4023          * crash?  For recovery, we need to check in advance whether we
4024          * can get read-write access to the device.
4025          */
4026         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4027                 if (sb->s_flags & MS_RDONLY) {
4028                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4029                                         "required on readonly filesystem");
4030                         if (really_read_only) {
4031                                 ext4_msg(sb, KERN_ERR, "write access "
4032                                         "unavailable, cannot proceed");
4033                                 return -EROFS;
4034                         }
4035                         ext4_msg(sb, KERN_INFO, "write access will "
4036                                "be enabled during recovery");
4037                 }
4038         }
4039
4040         if (journal_inum && journal_dev) {
4041                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4042                        "and inode journals!");
4043                 return -EINVAL;
4044         }
4045
4046         if (journal_inum) {
4047                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4048                         return -EINVAL;
4049         } else {
4050                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4051                         return -EINVAL;
4052         }
4053
4054         if (!(journal->j_flags & JBD2_BARRIER))
4055                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4056
4057         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4058                 err = jbd2_journal_wipe(journal, !really_read_only);
4059         if (!err) {
4060                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4061                 if (save)
4062                         memcpy(save, ((char *) es) +
4063                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4064                 err = jbd2_journal_load(journal);
4065                 if (save)
4066                         memcpy(((char *) es) + EXT4_S_ERR_START,
4067                                save, EXT4_S_ERR_LEN);
4068                 kfree(save);
4069         }
4070
4071         if (err) {
4072                 ext4_msg(sb, KERN_ERR, "error loading journal");
4073                 jbd2_journal_destroy(journal);
4074                 return err;
4075         }
4076
4077         EXT4_SB(sb)->s_journal = journal;
4078         ext4_clear_journal_err(sb, es);
4079
4080         if (!really_read_only && journal_devnum &&
4081             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4082                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4083
4084                 /* Make sure we flush the recovery flag to disk. */
4085                 ext4_commit_super(sb, 1);
4086         }
4087
4088         return 0;
4089 }
4090
4091 static int ext4_commit_super(struct super_block *sb, int sync)
4092 {
4093         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4094         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4095         int error = 0;
4096
4097         if (!sbh || block_device_ejected(sb))
4098                 return error;
4099         if (buffer_write_io_error(sbh)) {
4100                 /*
4101                  * Oh, dear.  A previous attempt to write the
4102                  * superblock failed.  This could happen because the
4103                  * USB device was yanked out.  Or it could happen to
4104                  * be a transient write error and maybe the block will
4105                  * be remapped.  Nothing we can do but to retry the
4106                  * write and hope for the best.
4107                  */
4108                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4109                        "superblock detected");
4110                 clear_buffer_write_io_error(sbh);
4111                 set_buffer_uptodate(sbh);
4112         }
4113         /*
4114          * If the file system is mounted read-only, don't update the
4115          * superblock write time.  This avoids updating the superblock
4116          * write time when we are mounting the root file system
4117          * read/only but we need to replay the journal; at that point,
4118          * for people who are east of GMT and who make their clock
4119          * tick in localtime for Windows bug-for-bug compatibility,
4120          * the clock is set in the future, and this will cause e2fsck
4121          * to complain and force a full file system check.
4122          */
4123         if (!(sb->s_flags & MS_RDONLY))
4124                 es->s_wtime = cpu_to_le32(get_seconds());
4125         if (sb->s_bdev->bd_part)
4126                 es->s_kbytes_written =
4127                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4128                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4129                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4130         else
4131                 es->s_kbytes_written =
4132                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4133         ext4_free_blocks_count_set(es,
4134                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4135                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4136         es->s_free_inodes_count =
4137                 cpu_to_le32(percpu_counter_sum_positive(
4138                                 &EXT4_SB(sb)->s_freeinodes_counter));
4139         sb->s_dirt = 0;
4140         BUFFER_TRACE(sbh, "marking dirty");
4141         mark_buffer_dirty(sbh);
4142         if (sync) {
4143                 error = sync_dirty_buffer(sbh);
4144                 if (error)
4145                         return error;
4146
4147                 error = buffer_write_io_error(sbh);
4148                 if (error) {
4149                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4150                                "superblock");
4151                         clear_buffer_write_io_error(sbh);
4152                         set_buffer_uptodate(sbh);
4153                 }
4154         }
4155         return error;
4156 }
4157
4158 /*
4159  * Have we just finished recovery?  If so, and if we are mounting (or
4160  * remounting) the filesystem readonly, then we will end up with a
4161  * consistent fs on disk.  Record that fact.
4162  */
4163 static void ext4_mark_recovery_complete(struct super_block *sb,
4164                                         struct ext4_super_block *es)
4165 {
4166         journal_t *journal = EXT4_SB(sb)->s_journal;
4167
4168         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4169                 BUG_ON(journal != NULL);
4170                 return;
4171         }
4172         jbd2_journal_lock_updates(journal);
4173         if (jbd2_journal_flush(journal) < 0)
4174                 goto out;
4175
4176         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4177             sb->s_flags & MS_RDONLY) {
4178                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4179                 ext4_commit_super(sb, 1);
4180         }
4181
4182 out:
4183         jbd2_journal_unlock_updates(journal);
4184 }
4185
4186 /*
4187  * If we are mounting (or read-write remounting) a filesystem whose journal
4188  * has recorded an error from a previous lifetime, move that error to the
4189  * main filesystem now.
4190  */
4191 static void ext4_clear_journal_err(struct super_block *sb,
4192                                    struct ext4_super_block *es)
4193 {
4194         journal_t *journal;
4195         int j_errno;
4196         const char *errstr;
4197
4198         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4199
4200         journal = EXT4_SB(sb)->s_journal;
4201
4202         /*
4203          * Now check for any error status which may have been recorded in the
4204          * journal by a prior ext4_error() or ext4_abort()
4205          */
4206
4207         j_errno = jbd2_journal_errno(journal);
4208         if (j_errno) {
4209                 char nbuf[16];
4210
4211                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4212                 ext4_warning(sb, "Filesystem error recorded "
4213                              "from previous mount: %s", errstr);
4214                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4215
4216                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4217                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4218                 ext4_commit_super(sb, 1);
4219
4220                 jbd2_journal_clear_err(journal);
4221         }
4222 }
4223
4224 /*
4225  * Force the running and committing transactions to commit,
4226  * and wait on the commit.
4227  */
4228 int ext4_force_commit(struct super_block *sb)
4229 {
4230         journal_t *journal;
4231         int ret = 0;
4232
4233         if (sb->s_flags & MS_RDONLY)
4234                 return 0;
4235
4236         journal = EXT4_SB(sb)->s_journal;
4237         if (journal) {
4238                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4239                 ret = ext4_journal_force_commit(journal);
4240         }
4241
4242         return ret;
4243 }
4244
4245 static void ext4_write_super(struct super_block *sb)
4246 {
4247         lock_super(sb);
4248         ext4_commit_super(sb, 1);
4249         unlock_super(sb);
4250 }
4251
4252 static int ext4_sync_fs(struct super_block *sb, int wait)
4253 {
4254         int ret = 0;
4255         tid_t target;
4256         struct ext4_sb_info *sbi = EXT4_SB(sb);
4257
4258         trace_ext4_sync_fs(sb, wait);
4259         flush_workqueue(sbi->dio_unwritten_wq);
4260         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4261                 if (wait)
4262                         jbd2_log_wait_commit(sbi->s_journal, target);
4263         }
4264         return ret;
4265 }
4266
4267 /*
4268  * LVM calls this function before a (read-only) snapshot is created.  This
4269  * gives us a chance to flush the journal completely and mark the fs clean.
4270  *
4271  * Note that only this function cannot bring a filesystem to be in a clean
4272  * state independently, because ext4 prevents a new handle from being started
4273  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4274  * the upper layer.
4275  */
4276 static int ext4_freeze(struct super_block *sb)
4277 {
4278         int error = 0;
4279         journal_t *journal;
4280
4281         if (sb->s_flags & MS_RDONLY)
4282                 return 0;
4283
4284         journal = EXT4_SB(sb)->s_journal;
4285
4286         /* Now we set up the journal barrier. */
4287         jbd2_journal_lock_updates(journal);
4288
4289         /*
4290          * Don't clear the needs_recovery flag if we failed to flush
4291          * the journal.
4292          */
4293         error = jbd2_journal_flush(journal);
4294         if (error < 0)
4295                 goto out;
4296
4297         /* Journal blocked and flushed, clear needs_recovery flag. */
4298         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4299         error = ext4_commit_super(sb, 1);
4300 out:
4301         /* we rely on s_frozen to stop further updates */
4302         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4303         return error;
4304 }
4305
4306 /*
4307  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4308  * flag here, even though the filesystem is not technically dirty yet.
4309  */
4310 static int ext4_unfreeze(struct super_block *sb)
4311 {
4312         if (sb->s_flags & MS_RDONLY)
4313                 return 0;
4314
4315         lock_super(sb);
4316         /* Reset the needs_recovery flag before the fs is unlocked. */
4317         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4318         ext4_commit_super(sb, 1);
4319         unlock_super(sb);
4320         return 0;
4321 }
4322
4323 /*
4324  * Structure to save mount options for ext4_remount's benefit
4325  */
4326 struct ext4_mount_options {
4327         unsigned long s_mount_opt;
4328         unsigned long s_mount_opt2;
4329         uid_t s_resuid;
4330         gid_t s_resgid;
4331         unsigned long s_commit_interval;
4332         u32 s_min_batch_time, s_max_batch_time;
4333 #ifdef CONFIG_QUOTA
4334         int s_jquota_fmt;
4335         char *s_qf_names[MAXQUOTAS];
4336 #endif
4337 };
4338
4339 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4340 {
4341         struct ext4_super_block *es;
4342         struct ext4_sb_info *sbi = EXT4_SB(sb);
4343         unsigned long old_sb_flags;
4344         struct ext4_mount_options old_opts;
4345         int enable_quota = 0;
4346         ext4_group_t g;
4347         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4348         int err = 0;
4349 #ifdef CONFIG_QUOTA
4350         int i;
4351 #endif
4352         char *orig_data = kstrdup(data, GFP_KERNEL);
4353
4354         /* Store the original options */
4355         lock_super(sb);
4356         old_sb_flags = sb->s_flags;
4357         old_opts.s_mount_opt = sbi->s_mount_opt;
4358         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4359         old_opts.s_resuid = sbi->s_resuid;
4360         old_opts.s_resgid = sbi->s_resgid;
4361         old_opts.s_commit_interval = sbi->s_commit_interval;
4362         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4363         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4364 #ifdef CONFIG_QUOTA
4365         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4366         for (i = 0; i < MAXQUOTAS; i++)
4367                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4368 #endif
4369         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4370                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4371
4372         /*
4373          * Allow the "check" option to be passed as a remount option.
4374          */
4375         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4376                 err = -EINVAL;
4377                 goto restore_opts;
4378         }
4379
4380         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4381                 ext4_abort(sb, "Abort forced by user");
4382
4383         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4384                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4385
4386         es = sbi->s_es;
4387
4388         if (sbi->s_journal) {
4389                 ext4_init_journal_params(sb, sbi->s_journal);
4390                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4391         }
4392
4393         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4394                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4395                         err = -EROFS;
4396                         goto restore_opts;
4397                 }
4398
4399                 if (*flags & MS_RDONLY) {
4400                         err = dquot_suspend(sb, -1);
4401                         if (err < 0)
4402                                 goto restore_opts;
4403
4404                         /*
4405                          * First of all, the unconditional stuff we have to do
4406                          * to disable replay of the journal when we next remount
4407                          */
4408                         sb->s_flags |= MS_RDONLY;
4409
4410                         /*
4411                          * OK, test if we are remounting a valid rw partition
4412                          * readonly, and if so set the rdonly flag and then
4413                          * mark the partition as valid again.
4414                          */
4415                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4416                             (sbi->s_mount_state & EXT4_VALID_FS))
4417                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4418
4419                         if (sbi->s_journal)
4420                                 ext4_mark_recovery_complete(sb, es);
4421                 } else {
4422                         /* Make sure we can mount this feature set readwrite */
4423                         if (!ext4_feature_set_ok(sb, 0)) {
4424                                 err = -EROFS;
4425                                 goto restore_opts;
4426                         }
4427                         /*
4428                          * Make sure the group descriptor checksums
4429                          * are sane.  If they aren't, refuse to remount r/w.
4430                          */
4431                         for (g = 0; g < sbi->s_groups_count; g++) {
4432                                 struct ext4_group_desc *gdp =
4433                                         ext4_get_group_desc(sb, g, NULL);
4434
4435                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4436                                         ext4_msg(sb, KERN_ERR,
4437                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4438                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4439                                                le16_to_cpu(gdp->bg_checksum));
4440                                         err = -EINVAL;
4441                                         goto restore_opts;
4442                                 }
4443                         }
4444
4445                         /*
4446                          * If we have an unprocessed orphan list hanging
4447                          * around from a previously readonly bdev mount,
4448                          * require a full umount/remount for now.
4449                          */
4450                         if (es->s_last_orphan) {
4451                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4452                                        "remount RDWR because of unprocessed "
4453                                        "orphan inode list.  Please "
4454                                        "umount/remount instead");
4455                                 err = -EINVAL;
4456                                 goto restore_opts;
4457                         }
4458
4459                         /*
4460                          * Mounting a RDONLY partition read-write, so reread
4461                          * and store the current valid flag.  (It may have
4462                          * been changed by e2fsck since we originally mounted
4463                          * the partition.)
4464                          */
4465                         if (sbi->s_journal)
4466                                 ext4_clear_journal_err(sb, es);
4467                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4468                         if (!ext4_setup_super(sb, es, 0))
4469                                 sb->s_flags &= ~MS_RDONLY;
4470                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4471                                                      EXT4_FEATURE_INCOMPAT_MMP))
4472                                 if (ext4_multi_mount_protect(sb,
4473                                                 le64_to_cpu(es->s_mmp_block))) {
4474                                         err = -EROFS;
4475                                         goto restore_opts;
4476                                 }
4477                         enable_quota = 1;
4478                 }
4479         }
4480
4481         /*
4482          * Reinitialize lazy itable initialization thread based on
4483          * current settings
4484          */
4485         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4486                 ext4_unregister_li_request(sb);
4487         else {
4488                 ext4_group_t first_not_zeroed;
4489                 first_not_zeroed = ext4_has_uninit_itable(sb);
4490                 ext4_register_li_request(sb, first_not_zeroed);
4491         }
4492
4493         ext4_setup_system_zone(sb);
4494         if (sbi->s_journal == NULL)
4495                 ext4_commit_super(sb, 1);
4496
4497 #ifdef CONFIG_QUOTA
4498         /* Release old quota file names */
4499         for (i = 0; i < MAXQUOTAS; i++)
4500                 if (old_opts.s_qf_names[i] &&
4501                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4502                         kfree(old_opts.s_qf_names[i]);
4503 #endif
4504         unlock_super(sb);
4505         if (enable_quota)
4506                 dquot_resume(sb, -1);
4507
4508         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4509         kfree(orig_data);
4510         return 0;
4511
4512 restore_opts:
4513         sb->s_flags = old_sb_flags;
4514         sbi->s_mount_opt = old_opts.s_mount_opt;
4515         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4516         sbi->s_resuid = old_opts.s_resuid;
4517         sbi->s_resgid = old_opts.s_resgid;
4518         sbi->s_commit_interval = old_opts.s_commit_interval;
4519         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4520         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4521 #ifdef CONFIG_QUOTA
4522         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4523         for (i = 0; i < MAXQUOTAS; i++) {
4524                 if (sbi->s_qf_names[i] &&
4525                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4526                         kfree(sbi->s_qf_names[i]);
4527                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4528         }
4529 #endif
4530         unlock_super(sb);
4531         kfree(orig_data);
4532         return err;
4533 }
4534
4535 /*
4536  * Note: calculating the overhead so we can be compatible with
4537  * historical BSD practice is quite difficult in the face of
4538  * clusters/bigalloc.  This is because multiple metadata blocks from
4539  * different block group can end up in the same allocation cluster.
4540  * Calculating the exact overhead in the face of clustered allocation
4541  * requires either O(all block bitmaps) in memory or O(number of block
4542  * groups**2) in time.  We will still calculate the superblock for
4543  * older file systems --- and if we come across with a bigalloc file
4544  * system with zero in s_overhead_clusters the estimate will be close to
4545  * correct especially for very large cluster sizes --- but for newer
4546  * file systems, it's better to calculate this figure once at mkfs
4547  * time, and store it in the superblock.  If the superblock value is
4548  * present (even for non-bigalloc file systems), we will use it.
4549  */
4550 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4551 {
4552         struct super_block *sb = dentry->d_sb;
4553         struct ext4_sb_info *sbi = EXT4_SB(sb);
4554         struct ext4_super_block *es = sbi->s_es;
4555         struct ext4_group_desc *gdp;
4556         u64 fsid;
4557         s64 bfree;
4558
4559         if (test_opt(sb, MINIX_DF)) {
4560                 sbi->s_overhead_last = 0;
4561         } else if (es->s_overhead_clusters) {
4562                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4563         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4564                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4565                 ext4_fsblk_t overhead = 0;
4566
4567                 /*
4568                  * Compute the overhead (FS structures).  This is constant
4569                  * for a given filesystem unless the number of block groups
4570                  * changes so we cache the previous value until it does.
4571                  */
4572
4573                 /*
4574                  * All of the blocks before first_data_block are
4575                  * overhead
4576                  */
4577                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4578
4579                 /*
4580                  * Add the overhead found in each block group
4581                  */
4582                 for (i = 0; i < ngroups; i++) {
4583                         gdp = ext4_get_group_desc(sb, i, NULL);
4584                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4585                         cond_resched();
4586                 }
4587                 sbi->s_overhead_last = overhead;
4588                 smp_wmb();
4589                 sbi->s_blocks_last = ext4_blocks_count(es);
4590         }
4591
4592         buf->f_type = EXT4_SUPER_MAGIC;
4593         buf->f_bsize = sb->s_blocksize;
4594         buf->f_blocks = (ext4_blocks_count(es) -
4595                          EXT4_C2B(sbi, sbi->s_overhead_last));
4596         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4597                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4598         /* prevent underflow in case that few free space is available */
4599         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4600         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4601         if (buf->f_bfree < ext4_r_blocks_count(es))
4602                 buf->f_bavail = 0;
4603         buf->f_files = le32_to_cpu(es->s_inodes_count);
4604         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4605         buf->f_namelen = EXT4_NAME_LEN;
4606         fsid = le64_to_cpup((void *)es->s_uuid) ^
4607                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4608         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4609         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4610
4611         return 0;
4612 }
4613
4614 /* Helper function for writing quotas on sync - we need to start transaction
4615  * before quota file is locked for write. Otherwise the are possible deadlocks:
4616  * Process 1                         Process 2
4617  * ext4_create()                     quota_sync()
4618  *   jbd2_journal_start()                  write_dquot()
4619  *   dquot_initialize()                         down(dqio_mutex)
4620  *     down(dqio_mutex)                    jbd2_journal_start()
4621  *
4622  */
4623
4624 #ifdef CONFIG_QUOTA
4625
4626 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4627 {
4628         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4629 }
4630
4631 static int ext4_write_dquot(struct dquot *dquot)
4632 {
4633         int ret, err;
4634         handle_t *handle;
4635         struct inode *inode;
4636
4637         inode = dquot_to_inode(dquot);
4638         handle = ext4_journal_start(inode,
4639                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4640         if (IS_ERR(handle))
4641                 return PTR_ERR(handle);
4642         ret = dquot_commit(dquot);
4643         err = ext4_journal_stop(handle);
4644         if (!ret)
4645                 ret = err;
4646         return ret;
4647 }
4648
4649 static int ext4_acquire_dquot(struct dquot *dquot)
4650 {
4651         int ret, err;
4652         handle_t *handle;
4653
4654         handle = ext4_journal_start(dquot_to_inode(dquot),
4655                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4656         if (IS_ERR(handle))
4657                 return PTR_ERR(handle);
4658         ret = dquot_acquire(dquot);
4659         err = ext4_journal_stop(handle);
4660         if (!ret)
4661                 ret = err;
4662         return ret;
4663 }
4664
4665 static int ext4_release_dquot(struct dquot *dquot)
4666 {
4667         int ret, err;
4668         handle_t *handle;
4669
4670         handle = ext4_journal_start(dquot_to_inode(dquot),
4671                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4672         if (IS_ERR(handle)) {
4673                 /* Release dquot anyway to avoid endless cycle in dqput() */
4674                 dquot_release(dquot);
4675                 return PTR_ERR(handle);
4676         }
4677         ret = dquot_release(dquot);
4678         err = ext4_journal_stop(handle);
4679         if (!ret)
4680                 ret = err;
4681         return ret;
4682 }
4683
4684 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4685 {
4686         /* Are we journaling quotas? */
4687         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4688             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4689                 dquot_mark_dquot_dirty(dquot);
4690                 return ext4_write_dquot(dquot);
4691         } else {
4692                 return dquot_mark_dquot_dirty(dquot);
4693         }
4694 }
4695
4696 static int ext4_write_info(struct super_block *sb, int type)
4697 {
4698         int ret, err;
4699         handle_t *handle;
4700
4701         /* Data block + inode block */
4702         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4703         if (IS_ERR(handle))
4704                 return PTR_ERR(handle);
4705         ret = dquot_commit_info(sb, type);
4706         err = ext4_journal_stop(handle);
4707         if (!ret)
4708                 ret = err;
4709         return ret;
4710 }
4711
4712 /*
4713  * Turn on quotas during mount time - we need to find
4714  * the quota file and such...
4715  */
4716 static int ext4_quota_on_mount(struct super_block *sb, int type)
4717 {
4718         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4719                                         EXT4_SB(sb)->s_jquota_fmt, type);
4720 }
4721
4722 /*
4723  * Standard function to be called on quota_on
4724  */
4725 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4726                          struct path *path)
4727 {
4728         int err;
4729
4730         if (!test_opt(sb, QUOTA))
4731                 return -EINVAL;
4732
4733         /* Quotafile not on the same filesystem? */
4734         if (path->dentry->d_sb != sb)
4735                 return -EXDEV;
4736         /* Journaling quota? */
4737         if (EXT4_SB(sb)->s_qf_names[type]) {
4738                 /* Quotafile not in fs root? */
4739                 if (path->dentry->d_parent != sb->s_root)
4740                         ext4_msg(sb, KERN_WARNING,
4741                                 "Quota file not on filesystem root. "
4742                                 "Journaled quota will not work");
4743         }
4744
4745         /*
4746          * When we journal data on quota file, we have to flush journal to see
4747          * all updates to the file when we bypass pagecache...
4748          */
4749         if (EXT4_SB(sb)->s_journal &&
4750             ext4_should_journal_data(path->dentry->d_inode)) {
4751                 /*
4752                  * We don't need to lock updates but journal_flush() could
4753                  * otherwise be livelocked...
4754                  */
4755                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4756                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4757                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4758                 if (err)
4759                         return err;
4760         }
4761
4762         return dquot_quota_on(sb, type, format_id, path);
4763 }
4764
4765 static int ext4_quota_off(struct super_block *sb, int type)
4766 {
4767         struct inode *inode = sb_dqopt(sb)->files[type];
4768         handle_t *handle;
4769
4770         /* Force all delayed allocation blocks to be allocated.
4771          * Caller already holds s_umount sem */
4772         if (test_opt(sb, DELALLOC))
4773                 sync_filesystem(sb);
4774
4775         if (!inode)
4776                 goto out;
4777
4778         /* Update modification times of quota files when userspace can
4779          * start looking at them */
4780         handle = ext4_journal_start(inode, 1);
4781         if (IS_ERR(handle))
4782                 goto out;
4783         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4784         ext4_mark_inode_dirty(handle, inode);
4785         ext4_journal_stop(handle);
4786
4787 out:
4788         return dquot_quota_off(sb, type);
4789 }
4790
4791 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4792  * acquiring the locks... As quota files are never truncated and quota code
4793  * itself serializes the operations (and no one else should touch the files)
4794  * we don't have to be afraid of races */
4795 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4796                                size_t len, loff_t off)
4797 {
4798         struct inode *inode = sb_dqopt(sb)->files[type];
4799         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4800         int err = 0;
4801         int offset = off & (sb->s_blocksize - 1);
4802         int tocopy;
4803         size_t toread;
4804         struct buffer_head *bh;
4805         loff_t i_size = i_size_read(inode);
4806
4807         if (off > i_size)
4808                 return 0;
4809         if (off+len > i_size)
4810                 len = i_size-off;
4811         toread = len;
4812         while (toread > 0) {
4813                 tocopy = sb->s_blocksize - offset < toread ?
4814                                 sb->s_blocksize - offset : toread;
4815                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4816                 if (err)
4817                         return err;
4818                 if (!bh)        /* A hole? */
4819                         memset(data, 0, tocopy);
4820                 else
4821                         memcpy(data, bh->b_data+offset, tocopy);
4822                 brelse(bh);
4823                 offset = 0;
4824                 toread -= tocopy;
4825                 data += tocopy;
4826                 blk++;
4827         }
4828         return len;
4829 }
4830
4831 /* Write to quotafile (we know the transaction is already started and has
4832  * enough credits) */
4833 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4834                                 const char *data, size_t len, loff_t off)
4835 {
4836         struct inode *inode = sb_dqopt(sb)->files[type];
4837         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4838         int err = 0;
4839         int offset = off & (sb->s_blocksize - 1);
4840         struct buffer_head *bh;
4841         handle_t *handle = journal_current_handle();
4842
4843         if (EXT4_SB(sb)->s_journal && !handle) {
4844                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4845                         " cancelled because transaction is not started",
4846                         (unsigned long long)off, (unsigned long long)len);
4847                 return -EIO;
4848         }
4849         /*
4850          * Since we account only one data block in transaction credits,
4851          * then it is impossible to cross a block boundary.
4852          */
4853         if (sb->s_blocksize - offset < len) {
4854                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4855                         " cancelled because not block aligned",
4856                         (unsigned long long)off, (unsigned long long)len);
4857                 return -EIO;
4858         }
4859
4860         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4861         bh = ext4_bread(handle, inode, blk, 1, &err);
4862         if (!bh)
4863                 goto out;
4864         err = ext4_journal_get_write_access(handle, bh);
4865         if (err) {
4866                 brelse(bh);
4867                 goto out;
4868         }
4869         lock_buffer(bh);
4870         memcpy(bh->b_data+offset, data, len);
4871         flush_dcache_page(bh->b_page);
4872         unlock_buffer(bh);
4873         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4874         brelse(bh);
4875 out:
4876         if (err) {
4877                 mutex_unlock(&inode->i_mutex);
4878                 return err;
4879         }
4880         if (inode->i_size < off + len) {
4881                 i_size_write(inode, off + len);
4882                 EXT4_I(inode)->i_disksize = inode->i_size;
4883                 ext4_mark_inode_dirty(handle, inode);
4884         }
4885         mutex_unlock(&inode->i_mutex);
4886         return len;
4887 }
4888
4889 #endif
4890
4891 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4892                        const char *dev_name, void *data)
4893 {
4894         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4895 }
4896
4897 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4898 static inline void register_as_ext2(void)
4899 {
4900         int err = register_filesystem(&ext2_fs_type);
4901         if (err)
4902                 printk(KERN_WARNING
4903                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4904 }
4905
4906 static inline void unregister_as_ext2(void)
4907 {
4908         unregister_filesystem(&ext2_fs_type);
4909 }
4910
4911 static inline int ext2_feature_set_ok(struct super_block *sb)
4912 {
4913         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4914                 return 0;
4915         if (sb->s_flags & MS_RDONLY)
4916                 return 1;
4917         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4918                 return 0;
4919         return 1;
4920 }
4921 MODULE_ALIAS("ext2");
4922 #else
4923 static inline void register_as_ext2(void) { }
4924 static inline void unregister_as_ext2(void) { }
4925 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4926 #endif
4927
4928 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4929 static inline void register_as_ext3(void)
4930 {
4931         int err = register_filesystem(&ext3_fs_type);
4932         if (err)
4933                 printk(KERN_WARNING
4934                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4935 }
4936
4937 static inline void unregister_as_ext3(void)
4938 {
4939         unregister_filesystem(&ext3_fs_type);
4940 }
4941
4942 static inline int ext3_feature_set_ok(struct super_block *sb)
4943 {
4944         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4945                 return 0;
4946         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4947                 return 0;
4948         if (sb->s_flags & MS_RDONLY)
4949                 return 1;
4950         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4951                 return 0;
4952         return 1;
4953 }
4954 MODULE_ALIAS("ext3");
4955 #else
4956 static inline void register_as_ext3(void) { }
4957 static inline void unregister_as_ext3(void) { }
4958 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4959 #endif
4960
4961 static struct file_system_type ext4_fs_type = {
4962         .owner          = THIS_MODULE,
4963         .name           = "ext4",
4964         .mount          = ext4_mount,
4965         .kill_sb        = kill_block_super,
4966         .fs_flags       = FS_REQUIRES_DEV,
4967 };
4968
4969 static int __init ext4_init_feat_adverts(void)
4970 {
4971         struct ext4_features *ef;
4972         int ret = -ENOMEM;
4973
4974         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4975         if (!ef)
4976                 goto out;
4977
4978         ef->f_kobj.kset = ext4_kset;
4979         init_completion(&ef->f_kobj_unregister);
4980         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4981                                    "features");
4982         if (ret) {
4983                 kfree(ef);
4984                 goto out;
4985         }
4986
4987         ext4_feat = ef;
4988         ret = 0;
4989 out:
4990         return ret;
4991 }
4992
4993 static void ext4_exit_feat_adverts(void)
4994 {
4995         kobject_put(&ext4_feat->f_kobj);
4996         wait_for_completion(&ext4_feat->f_kobj_unregister);
4997         kfree(ext4_feat);
4998 }
4999
5000 /* Shared across all ext4 file systems */
5001 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5002 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5003
5004 static int __init ext4_init_fs(void)
5005 {
5006         int i, err;
5007
5008         ext4_check_flag_values();
5009
5010         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5011                 mutex_init(&ext4__aio_mutex[i]);
5012                 init_waitqueue_head(&ext4__ioend_wq[i]);
5013         }
5014
5015         err = ext4_init_pageio();
5016         if (err)
5017                 return err;
5018         err = ext4_init_system_zone();
5019         if (err)
5020                 goto out6;
5021         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5022         if (!ext4_kset)
5023                 goto out5;
5024         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5025
5026         err = ext4_init_feat_adverts();
5027         if (err)
5028                 goto out4;
5029
5030         err = ext4_init_mballoc();
5031         if (err)
5032                 goto out3;
5033
5034         err = ext4_init_xattr();
5035         if (err)
5036                 goto out2;
5037         err = init_inodecache();
5038         if (err)
5039                 goto out1;
5040         register_as_ext3();
5041         register_as_ext2();
5042         err = register_filesystem(&ext4_fs_type);
5043         if (err)
5044                 goto out;
5045
5046         ext4_li_info = NULL;
5047         mutex_init(&ext4_li_mtx);
5048         return 0;
5049 out:
5050         unregister_as_ext2();
5051         unregister_as_ext3();
5052         destroy_inodecache();
5053 out1:
5054         ext4_exit_xattr();
5055 out2:
5056         ext4_exit_mballoc();
5057 out3:
5058         ext4_exit_feat_adverts();
5059 out4:
5060         if (ext4_proc_root)
5061                 remove_proc_entry("fs/ext4", NULL);
5062         kset_unregister(ext4_kset);
5063 out5:
5064         ext4_exit_system_zone();
5065 out6:
5066         ext4_exit_pageio();
5067         return err;
5068 }
5069
5070 static void __exit ext4_exit_fs(void)
5071 {
5072         ext4_destroy_lazyinit_thread();
5073         unregister_as_ext2();
5074         unregister_as_ext3();
5075         unregister_filesystem(&ext4_fs_type);
5076         destroy_inodecache();
5077         ext4_exit_xattr();
5078         ext4_exit_mballoc();
5079         ext4_exit_feat_adverts();
5080         remove_proc_entry("fs/ext4", NULL);
5081         kset_unregister(ext4_kset);
5082         ext4_exit_system_zone();
5083         ext4_exit_pageio();
5084 }
5085
5086 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5087 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5088 MODULE_LICENSE("GPL");
5089 module_init(ext4_init_fs)
5090 module_exit(ext4_exit_fs)