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 (test_opt(sb, I_VERSION))
1129                 seq_puts(seq, ",i_version");
1130         if (!test_opt(sb, DELALLOC) &&
1131             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1132                 seq_puts(seq, ",nodelalloc");
1133
1134         if (!test_opt(sb, MBLK_IO_SUBMIT))
1135                 seq_puts(seq, ",nomblk_io_submit");
1136         if (sbi->s_stripe)
1137                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1138         /*
1139          * journal mode get enabled in different ways
1140          * So just print the value even if we didn't specify it
1141          */
1142         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1143                 seq_puts(seq, ",data=journal");
1144         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1145                 seq_puts(seq, ",data=ordered");
1146         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1147                 seq_puts(seq, ",data=writeback");
1148
1149         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1150                 seq_printf(seq, ",inode_readahead_blks=%u",
1151                            sbi->s_inode_readahead_blks);
1152
1153         if (test_opt(sb, DATA_ERR_ABORT))
1154                 seq_puts(seq, ",data_err=abort");
1155
1156         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1157                 seq_puts(seq, ",noauto_da_alloc");
1158
1159         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1160                 seq_puts(seq, ",discard");
1161
1162         if (test_opt(sb, NOLOAD))
1163                 seq_puts(seq, ",norecovery");
1164
1165         if (test_opt(sb, DIOREAD_NOLOCK))
1166                 seq_puts(seq, ",dioread_nolock");
1167
1168         if (test_opt(sb, BLOCK_VALIDITY) &&
1169             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1170                 seq_puts(seq, ",block_validity");
1171
1172         if (!test_opt(sb, INIT_INODE_TABLE))
1173                 seq_puts(seq, ",noinit_itable");
1174         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1175                 seq_printf(seq, ",init_itable=%u",
1176                            (unsigned) sbi->s_li_wait_mult);
1177
1178         ext4_show_quota_options(seq, sb);
1179
1180         return 0;
1181 }
1182
1183 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1184                                         u64 ino, u32 generation)
1185 {
1186         struct inode *inode;
1187
1188         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1189                 return ERR_PTR(-ESTALE);
1190         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1191                 return ERR_PTR(-ESTALE);
1192
1193         /* iget isn't really right if the inode is currently unallocated!!
1194          *
1195          * ext4_read_inode will return a bad_inode if the inode had been
1196          * deleted, so we should be safe.
1197          *
1198          * Currently we don't know the generation for parent directory, so
1199          * a generation of 0 means "accept any"
1200          */
1201         inode = ext4_iget(sb, ino);
1202         if (IS_ERR(inode))
1203                 return ERR_CAST(inode);
1204         if (generation && inode->i_generation != generation) {
1205                 iput(inode);
1206                 return ERR_PTR(-ESTALE);
1207         }
1208
1209         return inode;
1210 }
1211
1212 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1213                                         int fh_len, int fh_type)
1214 {
1215         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1216                                     ext4_nfs_get_inode);
1217 }
1218
1219 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1220                                         int fh_len, int fh_type)
1221 {
1222         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1223                                     ext4_nfs_get_inode);
1224 }
1225
1226 /*
1227  * Try to release metadata pages (indirect blocks, directories) which are
1228  * mapped via the block device.  Since these pages could have journal heads
1229  * which would prevent try_to_free_buffers() from freeing them, we must use
1230  * jbd2 layer's try_to_free_buffers() function to release them.
1231  */
1232 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1233                                  gfp_t wait)
1234 {
1235         journal_t *journal = EXT4_SB(sb)->s_journal;
1236
1237         WARN_ON(PageChecked(page));
1238         if (!page_has_buffers(page))
1239                 return 0;
1240         if (journal)
1241                 return jbd2_journal_try_to_free_buffers(journal, page,
1242                                                         wait & ~__GFP_WAIT);
1243         return try_to_free_buffers(page);
1244 }
1245
1246 #ifdef CONFIG_QUOTA
1247 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1248 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1249
1250 static int ext4_write_dquot(struct dquot *dquot);
1251 static int ext4_acquire_dquot(struct dquot *dquot);
1252 static int ext4_release_dquot(struct dquot *dquot);
1253 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1254 static int ext4_write_info(struct super_block *sb, int type);
1255 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1256                          struct path *path);
1257 static int ext4_quota_off(struct super_block *sb, int type);
1258 static int ext4_quota_on_mount(struct super_block *sb, int type);
1259 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1260                                size_t len, loff_t off);
1261 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1262                                 const char *data, size_t len, loff_t off);
1263
1264 static const struct dquot_operations ext4_quota_operations = {
1265         .get_reserved_space = ext4_get_reserved_space,
1266         .write_dquot    = ext4_write_dquot,
1267         .acquire_dquot  = ext4_acquire_dquot,
1268         .release_dquot  = ext4_release_dquot,
1269         .mark_dirty     = ext4_mark_dquot_dirty,
1270         .write_info     = ext4_write_info,
1271         .alloc_dquot    = dquot_alloc,
1272         .destroy_dquot  = dquot_destroy,
1273 };
1274
1275 static const struct quotactl_ops ext4_qctl_operations = {
1276         .quota_on       = ext4_quota_on,
1277         .quota_off      = ext4_quota_off,
1278         .quota_sync     = dquot_quota_sync,
1279         .get_info       = dquot_get_dqinfo,
1280         .set_info       = dquot_set_dqinfo,
1281         .get_dqblk      = dquot_get_dqblk,
1282         .set_dqblk      = dquot_set_dqblk
1283 };
1284 #endif
1285
1286 static const struct super_operations ext4_sops = {
1287         .alloc_inode    = ext4_alloc_inode,
1288         .destroy_inode  = ext4_destroy_inode,
1289         .write_inode    = ext4_write_inode,
1290         .dirty_inode    = ext4_dirty_inode,
1291         .drop_inode     = ext4_drop_inode,
1292         .evict_inode    = ext4_evict_inode,
1293         .put_super      = ext4_put_super,
1294         .sync_fs        = ext4_sync_fs,
1295         .freeze_fs      = ext4_freeze,
1296         .unfreeze_fs    = ext4_unfreeze,
1297         .statfs         = ext4_statfs,
1298         .remount_fs     = ext4_remount,
1299         .show_options   = ext4_show_options,
1300 #ifdef CONFIG_QUOTA
1301         .quota_read     = ext4_quota_read,
1302         .quota_write    = ext4_quota_write,
1303 #endif
1304         .bdev_try_to_free_page = bdev_try_to_free_page,
1305 };
1306
1307 static const struct super_operations ext4_nojournal_sops = {
1308         .alloc_inode    = ext4_alloc_inode,
1309         .destroy_inode  = ext4_destroy_inode,
1310         .write_inode    = ext4_write_inode,
1311         .dirty_inode    = ext4_dirty_inode,
1312         .drop_inode     = ext4_drop_inode,
1313         .evict_inode    = ext4_evict_inode,
1314         .write_super    = ext4_write_super,
1315         .put_super      = ext4_put_super,
1316         .statfs         = ext4_statfs,
1317         .remount_fs     = ext4_remount,
1318         .show_options   = ext4_show_options,
1319 #ifdef CONFIG_QUOTA
1320         .quota_read     = ext4_quota_read,
1321         .quota_write    = ext4_quota_write,
1322 #endif
1323         .bdev_try_to_free_page = bdev_try_to_free_page,
1324 };
1325
1326 static const struct export_operations ext4_export_ops = {
1327         .fh_to_dentry = ext4_fh_to_dentry,
1328         .fh_to_parent = ext4_fh_to_parent,
1329         .get_parent = ext4_get_parent,
1330 };
1331
1332 enum {
1333         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1334         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1335         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1336         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1337         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1338         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1339         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1340         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1341         Opt_data_err_abort, Opt_data_err_ignore,
1342         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1343         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1344         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1345         Opt_usrquota, Opt_grpquota, Opt_i_version,
1346         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1347         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1348         Opt_inode_readahead_blks, Opt_journal_ioprio,
1349         Opt_dioread_nolock, Opt_dioread_lock,
1350         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1351 };
1352
1353 static const match_table_t tokens = {
1354         {Opt_bsd_df, "bsddf"},
1355         {Opt_minix_df, "minixdf"},
1356         {Opt_grpid, "grpid"},
1357         {Opt_grpid, "bsdgroups"},
1358         {Opt_nogrpid, "nogrpid"},
1359         {Opt_nogrpid, "sysvgroups"},
1360         {Opt_resgid, "resgid=%u"},
1361         {Opt_resuid, "resuid=%u"},
1362         {Opt_sb, "sb=%u"},
1363         {Opt_err_cont, "errors=continue"},
1364         {Opt_err_panic, "errors=panic"},
1365         {Opt_err_ro, "errors=remount-ro"},
1366         {Opt_nouid32, "nouid32"},
1367         {Opt_debug, "debug"},
1368         {Opt_oldalloc, "oldalloc"},
1369         {Opt_orlov, "orlov"},
1370         {Opt_user_xattr, "user_xattr"},
1371         {Opt_nouser_xattr, "nouser_xattr"},
1372         {Opt_acl, "acl"},
1373         {Opt_noacl, "noacl"},
1374         {Opt_noload, "noload"},
1375         {Opt_noload, "norecovery"},
1376         {Opt_nobh, "nobh"},
1377         {Opt_bh, "bh"},
1378         {Opt_commit, "commit=%u"},
1379         {Opt_min_batch_time, "min_batch_time=%u"},
1380         {Opt_max_batch_time, "max_batch_time=%u"},
1381         {Opt_journal_dev, "journal_dev=%u"},
1382         {Opt_journal_checksum, "journal_checksum"},
1383         {Opt_journal_async_commit, "journal_async_commit"},
1384         {Opt_abort, "abort"},
1385         {Opt_data_journal, "data=journal"},
1386         {Opt_data_ordered, "data=ordered"},
1387         {Opt_data_writeback, "data=writeback"},
1388         {Opt_data_err_abort, "data_err=abort"},
1389         {Opt_data_err_ignore, "data_err=ignore"},
1390         {Opt_offusrjquota, "usrjquota="},
1391         {Opt_usrjquota, "usrjquota=%s"},
1392         {Opt_offgrpjquota, "grpjquota="},
1393         {Opt_grpjquota, "grpjquota=%s"},
1394         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1395         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1396         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1397         {Opt_grpquota, "grpquota"},
1398         {Opt_noquota, "noquota"},
1399         {Opt_quota, "quota"},
1400         {Opt_usrquota, "usrquota"},
1401         {Opt_barrier, "barrier=%u"},
1402         {Opt_barrier, "barrier"},
1403         {Opt_nobarrier, "nobarrier"},
1404         {Opt_i_version, "i_version"},
1405         {Opt_stripe, "stripe=%u"},
1406         {Opt_delalloc, "delalloc"},
1407         {Opt_nodelalloc, "nodelalloc"},
1408         {Opt_mblk_io_submit, "mblk_io_submit"},
1409         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1410         {Opt_block_validity, "block_validity"},
1411         {Opt_noblock_validity, "noblock_validity"},
1412         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1413         {Opt_journal_ioprio, "journal_ioprio=%u"},
1414         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1415         {Opt_auto_da_alloc, "auto_da_alloc"},
1416         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1417         {Opt_dioread_nolock, "dioread_nolock"},
1418         {Opt_dioread_lock, "dioread_lock"},
1419         {Opt_discard, "discard"},
1420         {Opt_nodiscard, "nodiscard"},
1421         {Opt_init_itable, "init_itable=%u"},
1422         {Opt_init_itable, "init_itable"},
1423         {Opt_noinit_itable, "noinit_itable"},
1424         {Opt_err, NULL},
1425 };
1426
1427 static ext4_fsblk_t get_sb_block(void **data)
1428 {
1429         ext4_fsblk_t    sb_block;
1430         char            *options = (char *) *data;
1431
1432         if (!options || strncmp(options, "sb=", 3) != 0)
1433                 return 1;       /* Default location */
1434
1435         options += 3;
1436         /* TODO: use simple_strtoll with >32bit ext4 */
1437         sb_block = simple_strtoul(options, &options, 0);
1438         if (*options && *options != ',') {
1439                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1440                        (char *) *data);
1441                 return 1;
1442         }
1443         if (*options == ',')
1444                 options++;
1445         *data = (void *) options;
1446
1447         return sb_block;
1448 }
1449
1450 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1451 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1452         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1453
1454 #ifdef CONFIG_QUOTA
1455 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1456 {
1457         struct ext4_sb_info *sbi = EXT4_SB(sb);
1458         char *qname;
1459
1460         if (sb_any_quota_loaded(sb) &&
1461                 !sbi->s_qf_names[qtype]) {
1462                 ext4_msg(sb, KERN_ERR,
1463                         "Cannot change journaled "
1464                         "quota options when quota turned on");
1465                 return 0;
1466         }
1467         qname = match_strdup(args);
1468         if (!qname) {
1469                 ext4_msg(sb, KERN_ERR,
1470                         "Not enough memory for storing quotafile name");
1471                 return 0;
1472         }
1473         if (sbi->s_qf_names[qtype] &&
1474                 strcmp(sbi->s_qf_names[qtype], qname)) {
1475                 ext4_msg(sb, KERN_ERR,
1476                         "%s quota file already specified", QTYPE2NAME(qtype));
1477                 kfree(qname);
1478                 return 0;
1479         }
1480         sbi->s_qf_names[qtype] = qname;
1481         if (strchr(sbi->s_qf_names[qtype], '/')) {
1482                 ext4_msg(sb, KERN_ERR,
1483                         "quotafile must be on filesystem root");
1484                 kfree(sbi->s_qf_names[qtype]);
1485                 sbi->s_qf_names[qtype] = NULL;
1486                 return 0;
1487         }
1488         set_opt(sb, QUOTA);
1489         return 1;
1490 }
1491
1492 static int clear_qf_name(struct super_block *sb, int qtype)
1493 {
1494
1495         struct ext4_sb_info *sbi = EXT4_SB(sb);
1496
1497         if (sb_any_quota_loaded(sb) &&
1498                 sbi->s_qf_names[qtype]) {
1499                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1500                         " when quota turned on");
1501                 return 0;
1502         }
1503         /*
1504          * The space will be released later when all options are confirmed
1505          * to be correct
1506          */
1507         sbi->s_qf_names[qtype] = NULL;
1508         return 1;
1509 }
1510 #endif
1511
1512 static int parse_options(char *options, struct super_block *sb,
1513                          unsigned long *journal_devnum,
1514                          unsigned int *journal_ioprio,
1515                          int is_remount)
1516 {
1517         struct ext4_sb_info *sbi = EXT4_SB(sb);
1518         char *p;
1519         substring_t args[MAX_OPT_ARGS];
1520         int data_opt = 0;
1521         int option;
1522 #ifdef CONFIG_QUOTA
1523         int qfmt;
1524 #endif
1525
1526         if (!options)
1527                 return 1;
1528
1529         while ((p = strsep(&options, ",")) != NULL) {
1530                 int token;
1531                 if (!*p)
1532                         continue;
1533
1534                 /*
1535                  * Initialize args struct so we know whether arg was
1536                  * found; some options take optional arguments.
1537                  */
1538                 args[0].to = args[0].from = NULL;
1539                 token = match_token(p, tokens, args);
1540                 switch (token) {
1541                 case Opt_bsd_df:
1542                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1543                         clear_opt(sb, MINIX_DF);
1544                         break;
1545                 case Opt_minix_df:
1546                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1547                         set_opt(sb, MINIX_DF);
1548
1549                         break;
1550                 case Opt_grpid:
1551                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1552                         set_opt(sb, GRPID);
1553
1554                         break;
1555                 case Opt_nogrpid:
1556                         ext4_msg(sb, KERN_WARNING, deprecated_msg, p, "2.6.38");
1557                         clear_opt(sb, GRPID);
1558
1559                         break;
1560                 case Opt_resuid:
1561                         if (match_int(&args[0], &option))
1562                                 return 0;
1563                         sbi->s_resuid = option;
1564                         break;
1565                 case Opt_resgid:
1566                         if (match_int(&args[0], &option))
1567                                 return 0;
1568                         sbi->s_resgid = option;
1569                         break;
1570                 case Opt_sb:
1571                         /* handled by get_sb_block() instead of here */
1572                         /* *sb_block = match_int(&args[0]); */
1573                         break;
1574                 case Opt_err_panic:
1575                         clear_opt(sb, ERRORS_CONT);
1576                         clear_opt(sb, ERRORS_RO);
1577                         set_opt(sb, ERRORS_PANIC);
1578                         break;
1579                 case Opt_err_ro:
1580                         clear_opt(sb, ERRORS_CONT);
1581                         clear_opt(sb, ERRORS_PANIC);
1582                         set_opt(sb, ERRORS_RO);
1583                         break;
1584                 case Opt_err_cont:
1585                         clear_opt(sb, ERRORS_RO);
1586                         clear_opt(sb, ERRORS_PANIC);
1587                         set_opt(sb, ERRORS_CONT);
1588                         break;
1589                 case Opt_nouid32:
1590                         set_opt(sb, NO_UID32);
1591                         break;
1592                 case Opt_debug:
1593                         set_opt(sb, DEBUG);
1594                         break;
1595                 case Opt_oldalloc:
1596                         ext4_msg(sb, KERN_WARNING,
1597                                  "Ignoring deprecated oldalloc option");
1598                         break;
1599                 case Opt_orlov:
1600                         ext4_msg(sb, KERN_WARNING,
1601                                  "Ignoring deprecated orlov option");
1602                         break;
1603 #ifdef CONFIG_EXT4_FS_XATTR
1604                 case Opt_user_xattr:
1605                         set_opt(sb, XATTR_USER);
1606                         break;
1607                 case Opt_nouser_xattr:
1608                         clear_opt(sb, XATTR_USER);
1609                         break;
1610 #else
1611                 case Opt_user_xattr:
1612                 case Opt_nouser_xattr:
1613                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1614                         break;
1615 #endif
1616 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1617                 case Opt_acl:
1618                         set_opt(sb, POSIX_ACL);
1619                         break;
1620                 case Opt_noacl:
1621                         clear_opt(sb, POSIX_ACL);
1622                         break;
1623 #else
1624                 case Opt_acl:
1625                 case Opt_noacl:
1626                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1627                         break;
1628 #endif
1629                 case Opt_journal_dev:
1630                         if (is_remount) {
1631                                 ext4_msg(sb, KERN_ERR,
1632                                         "Cannot specify journal on remount");
1633                                 return 0;
1634                         }
1635                         if (match_int(&args[0], &option))
1636                                 return 0;
1637                         *journal_devnum = option;
1638                         break;
1639                 case Opt_journal_checksum:
1640                         set_opt(sb, JOURNAL_CHECKSUM);
1641                         break;
1642                 case Opt_journal_async_commit:
1643                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1644                         set_opt(sb, JOURNAL_CHECKSUM);
1645                         break;
1646                 case Opt_noload:
1647                         set_opt(sb, NOLOAD);
1648                         break;
1649                 case Opt_commit:
1650                         if (match_int(&args[0], &option))
1651                                 return 0;
1652                         if (option < 0)
1653                                 return 0;
1654                         if (option == 0)
1655                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1656                         sbi->s_commit_interval = HZ * option;
1657                         break;
1658                 case Opt_max_batch_time:
1659                         if (match_int(&args[0], &option))
1660                                 return 0;
1661                         if (option < 0)
1662                                 return 0;
1663                         if (option == 0)
1664                                 option = EXT4_DEF_MAX_BATCH_TIME;
1665                         sbi->s_max_batch_time = option;
1666                         break;
1667                 case Opt_min_batch_time:
1668                         if (match_int(&args[0], &option))
1669                                 return 0;
1670                         if (option < 0)
1671                                 return 0;
1672                         sbi->s_min_batch_time = option;
1673                         break;
1674                 case Opt_data_journal:
1675                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1676                         goto datacheck;
1677                 case Opt_data_ordered:
1678                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1679                         goto datacheck;
1680                 case Opt_data_writeback:
1681                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1682                 datacheck:
1683                         if (is_remount) {
1684                                 if (!sbi->s_journal)
1685                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1686                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1687                                         ext4_msg(sb, KERN_ERR,
1688                                                 "Cannot change data mode on remount");
1689                                         return 0;
1690                                 }
1691                         } else {
1692                                 clear_opt(sb, DATA_FLAGS);
1693                                 sbi->s_mount_opt |= data_opt;
1694                         }
1695                         break;
1696                 case Opt_data_err_abort:
1697                         set_opt(sb, DATA_ERR_ABORT);
1698                         break;
1699                 case Opt_data_err_ignore:
1700                         clear_opt(sb, DATA_ERR_ABORT);
1701                         break;
1702 #ifdef CONFIG_QUOTA
1703                 case Opt_usrjquota:
1704                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1705                                 return 0;
1706                         break;
1707                 case Opt_grpjquota:
1708                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1709                                 return 0;
1710                         break;
1711                 case Opt_offusrjquota:
1712                         if (!clear_qf_name(sb, USRQUOTA))
1713                                 return 0;
1714                         break;
1715                 case Opt_offgrpjquota:
1716                         if (!clear_qf_name(sb, GRPQUOTA))
1717                                 return 0;
1718                         break;
1719
1720                 case Opt_jqfmt_vfsold:
1721                         qfmt = QFMT_VFS_OLD;
1722                         goto set_qf_format;
1723                 case Opt_jqfmt_vfsv0:
1724                         qfmt = QFMT_VFS_V0;
1725                         goto set_qf_format;
1726                 case Opt_jqfmt_vfsv1:
1727                         qfmt = QFMT_VFS_V1;
1728 set_qf_format:
1729                         if (sb_any_quota_loaded(sb) &&
1730                             sbi->s_jquota_fmt != qfmt) {
1731                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1732                                         "journaled quota options when "
1733                                         "quota turned on");
1734                                 return 0;
1735                         }
1736                         sbi->s_jquota_fmt = qfmt;
1737                         break;
1738                 case Opt_quota:
1739                 case Opt_usrquota:
1740                         set_opt(sb, QUOTA);
1741                         set_opt(sb, USRQUOTA);
1742                         break;
1743                 case Opt_grpquota:
1744                         set_opt(sb, QUOTA);
1745                         set_opt(sb, GRPQUOTA);
1746                         break;
1747                 case Opt_noquota:
1748                         if (sb_any_quota_loaded(sb)) {
1749                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1750                                         "options when quota turned on");
1751                                 return 0;
1752                         }
1753                         clear_opt(sb, QUOTA);
1754                         clear_opt(sb, USRQUOTA);
1755                         clear_opt(sb, GRPQUOTA);
1756                         break;
1757 #else
1758                 case Opt_quota:
1759                 case Opt_usrquota:
1760                 case Opt_grpquota:
1761                         ext4_msg(sb, KERN_ERR,
1762                                 "quota options not supported");
1763                         break;
1764                 case Opt_usrjquota:
1765                 case Opt_grpjquota:
1766                 case Opt_offusrjquota:
1767                 case Opt_offgrpjquota:
1768                 case Opt_jqfmt_vfsold:
1769                 case Opt_jqfmt_vfsv0:
1770                 case Opt_jqfmt_vfsv1:
1771                         ext4_msg(sb, KERN_ERR,
1772                                 "journaled quota options not supported");
1773                         break;
1774                 case Opt_noquota:
1775                         break;
1776 #endif
1777                 case Opt_abort:
1778                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1779                         break;
1780                 case Opt_nobarrier:
1781                         clear_opt(sb, BARRIER);
1782                         break;
1783                 case Opt_barrier:
1784                         if (args[0].from) {
1785                                 if (match_int(&args[0], &option))
1786                                         return 0;
1787                         } else
1788                                 option = 1;     /* No argument, default to 1 */
1789                         if (option)
1790                                 set_opt(sb, BARRIER);
1791                         else
1792                                 clear_opt(sb, BARRIER);
1793                         break;
1794                 case Opt_ignore:
1795                         break;
1796                 case Opt_nobh:
1797                         ext4_msg(sb, KERN_WARNING,
1798                                  "Ignoring deprecated nobh option");
1799                         break;
1800                 case Opt_bh:
1801                         ext4_msg(sb, KERN_WARNING,
1802                                  "Ignoring deprecated bh option");
1803                         break;
1804                 case Opt_i_version:
1805                         set_opt(sb, I_VERSION);
1806                         sb->s_flags |= MS_I_VERSION;
1807                         break;
1808                 case Opt_nodelalloc:
1809                         clear_opt(sb, DELALLOC);
1810                         clear_opt2(sb, EXPLICIT_DELALLOC);
1811                         break;
1812                 case Opt_mblk_io_submit:
1813                         set_opt(sb, MBLK_IO_SUBMIT);
1814                         break;
1815                 case Opt_nomblk_io_submit:
1816                         clear_opt(sb, MBLK_IO_SUBMIT);
1817                         break;
1818                 case Opt_stripe:
1819                         if (match_int(&args[0], &option))
1820                                 return 0;
1821                         if (option < 0)
1822                                 return 0;
1823                         sbi->s_stripe = option;
1824                         break;
1825                 case Opt_delalloc:
1826                         set_opt(sb, DELALLOC);
1827                         set_opt2(sb, EXPLICIT_DELALLOC);
1828                         break;
1829                 case Opt_block_validity:
1830                         set_opt(sb, BLOCK_VALIDITY);
1831                         break;
1832                 case Opt_noblock_validity:
1833                         clear_opt(sb, BLOCK_VALIDITY);
1834                         break;
1835                 case Opt_inode_readahead_blks:
1836                         if (match_int(&args[0], &option))
1837                                 return 0;
1838                         if (option < 0 || option > (1 << 30))
1839                                 return 0;
1840                         if (option && !is_power_of_2(option)) {
1841                                 ext4_msg(sb, KERN_ERR,
1842                                          "EXT4-fs: inode_readahead_blks"
1843                                          " must be a power of 2");
1844                                 return 0;
1845                         }
1846                         sbi->s_inode_readahead_blks = option;
1847                         break;
1848                 case Opt_journal_ioprio:
1849                         if (match_int(&args[0], &option))
1850                                 return 0;
1851                         if (option < 0 || option > 7)
1852                                 break;
1853                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1854                                                             option);
1855                         break;
1856                 case Opt_noauto_da_alloc:
1857                         set_opt(sb, NO_AUTO_DA_ALLOC);
1858                         break;
1859                 case Opt_auto_da_alloc:
1860                         if (args[0].from) {
1861                                 if (match_int(&args[0], &option))
1862                                         return 0;
1863                         } else
1864                                 option = 1;     /* No argument, default to 1 */
1865                         if (option)
1866                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1867                         else
1868                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1869                         break;
1870                 case Opt_discard:
1871                         set_opt(sb, DISCARD);
1872                         break;
1873                 case Opt_nodiscard:
1874                         clear_opt(sb, DISCARD);
1875                         break;
1876                 case Opt_dioread_nolock:
1877                         set_opt(sb, DIOREAD_NOLOCK);
1878                         break;
1879                 case Opt_dioread_lock:
1880                         clear_opt(sb, DIOREAD_NOLOCK);
1881                         break;
1882                 case Opt_init_itable:
1883                         set_opt(sb, INIT_INODE_TABLE);
1884                         if (args[0].from) {
1885                                 if (match_int(&args[0], &option))
1886                                         return 0;
1887                         } else
1888                                 option = EXT4_DEF_LI_WAIT_MULT;
1889                         if (option < 0)
1890                                 return 0;
1891                         sbi->s_li_wait_mult = option;
1892                         break;
1893                 case Opt_noinit_itable:
1894                         clear_opt(sb, INIT_INODE_TABLE);
1895                         break;
1896                 default:
1897                         ext4_msg(sb, KERN_ERR,
1898                                "Unrecognized mount option \"%s\" "
1899                                "or missing value", p);
1900                         return 0;
1901                 }
1902         }
1903 #ifdef CONFIG_QUOTA
1904         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1905                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1906                         clear_opt(sb, USRQUOTA);
1907
1908                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1909                         clear_opt(sb, GRPQUOTA);
1910
1911                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1912                         ext4_msg(sb, KERN_ERR, "old and new quota "
1913                                         "format mixing");
1914                         return 0;
1915                 }
1916
1917                 if (!sbi->s_jquota_fmt) {
1918                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1919                                         "not specified");
1920                         return 0;
1921                 }
1922         } else {
1923                 if (sbi->s_jquota_fmt) {
1924                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1925                                         "specified with no journaling "
1926                                         "enabled");
1927                         return 0;
1928                 }
1929         }
1930 #endif
1931         return 1;
1932 }
1933
1934 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1935                             int read_only)
1936 {
1937         struct ext4_sb_info *sbi = EXT4_SB(sb);
1938         int res = 0;
1939
1940         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1941                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1942                          "forcing read-only mode");
1943                 res = MS_RDONLY;
1944         }
1945         if (read_only)
1946                 goto done;
1947         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1948                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1949                          "running e2fsck is recommended");
1950         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1951                 ext4_msg(sb, KERN_WARNING,
1952                          "warning: mounting fs with errors, "
1953                          "running e2fsck is recommended");
1954         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1955                  le16_to_cpu(es->s_mnt_count) >=
1956                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1957                 ext4_msg(sb, KERN_WARNING,
1958                          "warning: maximal mount count reached, "
1959                          "running e2fsck is recommended");
1960         else if (le32_to_cpu(es->s_checkinterval) &&
1961                 (le32_to_cpu(es->s_lastcheck) +
1962                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1963                 ext4_msg(sb, KERN_WARNING,
1964                          "warning: checktime reached, "
1965                          "running e2fsck is recommended");
1966         if (!sbi->s_journal)
1967                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1968         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1969                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1970         le16_add_cpu(&es->s_mnt_count, 1);
1971         es->s_mtime = cpu_to_le32(get_seconds());
1972         ext4_update_dynamic_rev(sb);
1973         if (sbi->s_journal)
1974                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1975
1976         ext4_commit_super(sb, 1);
1977 done:
1978         if (test_opt(sb, DEBUG))
1979                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1980                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1981                         sb->s_blocksize,
1982                         sbi->s_groups_count,
1983                         EXT4_BLOCKS_PER_GROUP(sb),
1984                         EXT4_INODES_PER_GROUP(sb),
1985                         sbi->s_mount_opt, sbi->s_mount_opt2);
1986
1987         cleancache_init_fs(sb);
1988         return res;
1989 }
1990
1991 static int ext4_fill_flex_info(struct super_block *sb)
1992 {
1993         struct ext4_sb_info *sbi = EXT4_SB(sb);
1994         struct ext4_group_desc *gdp = NULL;
1995         ext4_group_t flex_group_count;
1996         ext4_group_t flex_group;
1997         unsigned int groups_per_flex = 0;
1998         size_t size;
1999         int i;
2000
2001         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2002         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2003                 sbi->s_log_groups_per_flex = 0;
2004                 return 1;
2005         }
2006         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
2007
2008         /* We allocate both existing and potentially added groups */
2009         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2010                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2011                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2012         size = flex_group_count * sizeof(struct flex_groups);
2013         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2014         if (sbi->s_flex_groups == NULL) {
2015                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2016                          flex_group_count);
2017                 goto failed;
2018         }
2019
2020         for (i = 0; i < sbi->s_groups_count; i++) {
2021                 gdp = ext4_get_group_desc(sb, i, NULL);
2022
2023                 flex_group = ext4_flex_group(sbi, i);
2024                 atomic_add(ext4_free_inodes_count(sb, gdp),
2025                            &sbi->s_flex_groups[flex_group].free_inodes);
2026                 atomic_add(ext4_free_group_clusters(sb, gdp),
2027                            &sbi->s_flex_groups[flex_group].free_clusters);
2028                 atomic_add(ext4_used_dirs_count(sb, gdp),
2029                            &sbi->s_flex_groups[flex_group].used_dirs);
2030         }
2031
2032         return 1;
2033 failed:
2034         return 0;
2035 }
2036
2037 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2038                             struct ext4_group_desc *gdp)
2039 {
2040         __u16 crc = 0;
2041
2042         if (sbi->s_es->s_feature_ro_compat &
2043             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2044                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2045                 __le32 le_group = cpu_to_le32(block_group);
2046
2047                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2048                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2049                 crc = crc16(crc, (__u8 *)gdp, offset);
2050                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2051                 /* for checksum of struct ext4_group_desc do the rest...*/
2052                 if ((sbi->s_es->s_feature_incompat &
2053                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2054                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2055                         crc = crc16(crc, (__u8 *)gdp + offset,
2056                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2057                                         offset);
2058         }
2059
2060         return cpu_to_le16(crc);
2061 }
2062
2063 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2064                                 struct ext4_group_desc *gdp)
2065 {
2066         if ((sbi->s_es->s_feature_ro_compat &
2067              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2068             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2069                 return 0;
2070
2071         return 1;
2072 }
2073
2074 /* Called at mount-time, super-block is locked */
2075 static int ext4_check_descriptors(struct super_block *sb,
2076                                   ext4_group_t *first_not_zeroed)
2077 {
2078         struct ext4_sb_info *sbi = EXT4_SB(sb);
2079         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2080         ext4_fsblk_t last_block;
2081         ext4_fsblk_t block_bitmap;
2082         ext4_fsblk_t inode_bitmap;
2083         ext4_fsblk_t inode_table;
2084         int flexbg_flag = 0;
2085         ext4_group_t i, grp = sbi->s_groups_count;
2086
2087         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2088                 flexbg_flag = 1;
2089
2090         ext4_debug("Checking group descriptors");
2091
2092         for (i = 0; i < sbi->s_groups_count; i++) {
2093                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2094
2095                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2096                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2097                 else
2098                         last_block = first_block +
2099                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2100
2101                 if ((grp == sbi->s_groups_count) &&
2102                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2103                         grp = i;
2104
2105                 block_bitmap = ext4_block_bitmap(sb, gdp);
2106                 if (block_bitmap < first_block || block_bitmap > last_block) {
2107                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2108                                "Block bitmap for group %u not in group "
2109                                "(block %llu)!", i, block_bitmap);
2110                         return 0;
2111                 }
2112                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2113                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2114                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2115                                "Inode bitmap for group %u not in group "
2116                                "(block %llu)!", i, inode_bitmap);
2117                         return 0;
2118                 }
2119                 inode_table = ext4_inode_table(sb, gdp);
2120                 if (inode_table < first_block ||
2121                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2122                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2123                                "Inode table for group %u not in group "
2124                                "(block %llu)!", i, inode_table);
2125                         return 0;
2126                 }
2127                 ext4_lock_group(sb, i);
2128                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2129                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2130                                  "Checksum for group %u failed (%u!=%u)",
2131                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2132                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2133                         if (!(sb->s_flags & MS_RDONLY)) {
2134                                 ext4_unlock_group(sb, i);
2135                                 return 0;
2136                         }
2137                 }
2138                 ext4_unlock_group(sb, i);
2139                 if (!flexbg_flag)
2140                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2141         }
2142         if (NULL != first_not_zeroed)
2143                 *first_not_zeroed = grp;
2144
2145         ext4_free_blocks_count_set(sbi->s_es,
2146                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2147         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2148         return 1;
2149 }
2150
2151 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2152  * the superblock) which were deleted from all directories, but held open by
2153  * a process at the time of a crash.  We walk the list and try to delete these
2154  * inodes at recovery time (only with a read-write filesystem).
2155  *
2156  * In order to keep the orphan inode chain consistent during traversal (in
2157  * case of crash during recovery), we link each inode into the superblock
2158  * orphan list_head and handle it the same way as an inode deletion during
2159  * normal operation (which journals the operations for us).
2160  *
2161  * We only do an iget() and an iput() on each inode, which is very safe if we
2162  * accidentally point at an in-use or already deleted inode.  The worst that
2163  * can happen in this case is that we get a "bit already cleared" message from
2164  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2165  * e2fsck was run on this filesystem, and it must have already done the orphan
2166  * inode cleanup for us, so we can safely abort without any further action.
2167  */
2168 static void ext4_orphan_cleanup(struct super_block *sb,
2169                                 struct ext4_super_block *es)
2170 {
2171         unsigned int s_flags = sb->s_flags;
2172         int nr_orphans = 0, nr_truncates = 0;
2173 #ifdef CONFIG_QUOTA
2174         int i;
2175 #endif
2176         if (!es->s_last_orphan) {
2177                 jbd_debug(4, "no orphan inodes to clean up\n");
2178                 return;
2179         }
2180
2181         if (bdev_read_only(sb->s_bdev)) {
2182                 ext4_msg(sb, KERN_ERR, "write access "
2183                         "unavailable, skipping orphan cleanup");
2184                 return;
2185         }
2186
2187         /* Check if feature set would not allow a r/w mount */
2188         if (!ext4_feature_set_ok(sb, 0)) {
2189                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2190                          "unknown ROCOMPAT features");
2191                 return;
2192         }
2193
2194         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2195                 if (es->s_last_orphan)
2196                         jbd_debug(1, "Errors on filesystem, "
2197                                   "clearing orphan list.\n");
2198                 es->s_last_orphan = 0;
2199                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2200                 return;
2201         }
2202
2203         if (s_flags & MS_RDONLY) {
2204                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2205                 sb->s_flags &= ~MS_RDONLY;
2206         }
2207 #ifdef CONFIG_QUOTA
2208         /* Needed for iput() to work correctly and not trash data */
2209         sb->s_flags |= MS_ACTIVE;
2210         /* Turn on quotas so that they are updated correctly */
2211         for (i = 0; i < MAXQUOTAS; i++) {
2212                 if (EXT4_SB(sb)->s_qf_names[i]) {
2213                         int ret = ext4_quota_on_mount(sb, i);
2214                         if (ret < 0)
2215                                 ext4_msg(sb, KERN_ERR,
2216                                         "Cannot turn on journaled "
2217                                         "quota: error %d", ret);
2218                 }
2219         }
2220 #endif
2221
2222         while (es->s_last_orphan) {
2223                 struct inode *inode;
2224
2225                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2226                 if (IS_ERR(inode)) {
2227                         es->s_last_orphan = 0;
2228                         break;
2229                 }
2230
2231                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2232                 dquot_initialize(inode);
2233                 if (inode->i_nlink) {
2234                         ext4_msg(sb, KERN_DEBUG,
2235                                 "%s: truncating inode %lu to %lld bytes",
2236                                 __func__, inode->i_ino, inode->i_size);
2237                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2238                                   inode->i_ino, inode->i_size);
2239                         ext4_truncate(inode);
2240                         nr_truncates++;
2241                 } else {
2242                         ext4_msg(sb, KERN_DEBUG,
2243                                 "%s: deleting unreferenced inode %lu",
2244                                 __func__, inode->i_ino);
2245                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2246                                   inode->i_ino);
2247                         nr_orphans++;
2248                 }
2249                 iput(inode);  /* The delete magic happens here! */
2250         }
2251
2252 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2253
2254         if (nr_orphans)
2255                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2256                        PLURAL(nr_orphans));
2257         if (nr_truncates)
2258                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2259                        PLURAL(nr_truncates));
2260 #ifdef CONFIG_QUOTA
2261         /* Turn quotas off */
2262         for (i = 0; i < MAXQUOTAS; i++) {
2263                 if (sb_dqopt(sb)->files[i])
2264                         dquot_quota_off(sb, i);
2265         }
2266 #endif
2267         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2268 }
2269
2270 /*
2271  * Maximal extent format file size.
2272  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2273  * extent format containers, within a sector_t, and within i_blocks
2274  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2275  * so that won't be a limiting factor.
2276  *
2277  * However there is other limiting factor. We do store extents in the form
2278  * of starting block and length, hence the resulting length of the extent
2279  * covering maximum file size must fit into on-disk format containers as
2280  * well. Given that length is always by 1 unit bigger than max unit (because
2281  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2282  *
2283  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2284  */
2285 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2286 {
2287         loff_t res;
2288         loff_t upper_limit = MAX_LFS_FILESIZE;
2289
2290         /* small i_blocks in vfs inode? */
2291         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2292                 /*
2293                  * CONFIG_LBDAF is not enabled implies the inode
2294                  * i_block represent total blocks in 512 bytes
2295                  * 32 == size of vfs inode i_blocks * 8
2296                  */
2297                 upper_limit = (1LL << 32) - 1;
2298
2299                 /* total blocks in file system block size */
2300                 upper_limit >>= (blkbits - 9);
2301                 upper_limit <<= blkbits;
2302         }
2303
2304         /*
2305          * 32-bit extent-start container, ee_block. We lower the maxbytes
2306          * by one fs block, so ee_len can cover the extent of maximum file
2307          * size
2308          */
2309         res = (1LL << 32) - 1;
2310         res <<= blkbits;
2311
2312         /* Sanity check against vm- & vfs- imposed limits */
2313         if (res > upper_limit)
2314                 res = upper_limit;
2315
2316         return res;
2317 }
2318
2319 /*
2320  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2321  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2322  * We need to be 1 filesystem block less than the 2^48 sector limit.
2323  */
2324 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2325 {
2326         loff_t res = EXT4_NDIR_BLOCKS;
2327         int meta_blocks;
2328         loff_t upper_limit;
2329         /* This is calculated to be the largest file size for a dense, block
2330          * mapped file such that the file's total number of 512-byte sectors,
2331          * including data and all indirect blocks, does not exceed (2^48 - 1).
2332          *
2333          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2334          * number of 512-byte sectors of the file.
2335          */
2336
2337         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2338                 /*
2339                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2340                  * the inode i_block field represents total file blocks in
2341                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2342                  */
2343                 upper_limit = (1LL << 32) - 1;
2344
2345                 /* total blocks in file system block size */
2346                 upper_limit >>= (bits - 9);
2347
2348         } else {
2349                 /*
2350                  * We use 48 bit ext4_inode i_blocks
2351                  * With EXT4_HUGE_FILE_FL set the i_blocks
2352                  * represent total number of blocks in
2353                  * file system block size
2354                  */
2355                 upper_limit = (1LL << 48) - 1;
2356
2357         }
2358
2359         /* indirect blocks */
2360         meta_blocks = 1;
2361         /* double indirect blocks */
2362         meta_blocks += 1 + (1LL << (bits-2));
2363         /* tripple indirect blocks */
2364         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2365
2366         upper_limit -= meta_blocks;
2367         upper_limit <<= bits;
2368
2369         res += 1LL << (bits-2);
2370         res += 1LL << (2*(bits-2));
2371         res += 1LL << (3*(bits-2));
2372         res <<= bits;
2373         if (res > upper_limit)
2374                 res = upper_limit;
2375
2376         if (res > MAX_LFS_FILESIZE)
2377                 res = MAX_LFS_FILESIZE;
2378
2379         return res;
2380 }
2381
2382 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2383                                    ext4_fsblk_t logical_sb_block, int nr)
2384 {
2385         struct ext4_sb_info *sbi = EXT4_SB(sb);
2386         ext4_group_t bg, first_meta_bg;
2387         int has_super = 0;
2388
2389         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2390
2391         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2392             nr < first_meta_bg)
2393                 return logical_sb_block + nr + 1;
2394         bg = sbi->s_desc_per_block * nr;
2395         if (ext4_bg_has_super(sb, bg))
2396                 has_super = 1;
2397
2398         return (has_super + ext4_group_first_block_no(sb, bg));
2399 }
2400
2401 /**
2402  * ext4_get_stripe_size: Get the stripe size.
2403  * @sbi: In memory super block info
2404  *
2405  * If we have specified it via mount option, then
2406  * use the mount option value. If the value specified at mount time is
2407  * greater than the blocks per group use the super block value.
2408  * If the super block value is greater than blocks per group return 0.
2409  * Allocator needs it be less than blocks per group.
2410  *
2411  */
2412 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2413 {
2414         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2415         unsigned long stripe_width =
2416                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2417         int ret;
2418
2419         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2420                 ret = sbi->s_stripe;
2421         else if (stripe_width <= sbi->s_blocks_per_group)
2422                 ret = stripe_width;
2423         else if (stride <= sbi->s_blocks_per_group)
2424                 ret = stride;
2425         else
2426                 ret = 0;
2427
2428         /*
2429          * If the stripe width is 1, this makes no sense and
2430          * we set it to 0 to turn off stripe handling code.
2431          */
2432         if (ret <= 1)
2433                 ret = 0;
2434
2435         return ret;
2436 }
2437
2438 /* sysfs supprt */
2439
2440 struct ext4_attr {
2441         struct attribute attr;
2442         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2443         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2444                          const char *, size_t);
2445         int offset;
2446 };
2447
2448 static int parse_strtoul(const char *buf,
2449                 unsigned long max, unsigned long *value)
2450 {
2451         char *endp;
2452
2453         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2454         endp = skip_spaces(endp);
2455         if (*endp || *value > max)
2456                 return -EINVAL;
2457
2458         return 0;
2459 }
2460
2461 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2462                                               struct ext4_sb_info *sbi,
2463                                               char *buf)
2464 {
2465         return snprintf(buf, PAGE_SIZE, "%llu\n",
2466                 (s64) EXT4_C2B(sbi,
2467                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2468 }
2469
2470 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2471                                          struct ext4_sb_info *sbi, char *buf)
2472 {
2473         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2474
2475         if (!sb->s_bdev->bd_part)
2476                 return snprintf(buf, PAGE_SIZE, "0\n");
2477         return snprintf(buf, PAGE_SIZE, "%lu\n",
2478                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2479                          sbi->s_sectors_written_start) >> 1);
2480 }
2481
2482 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2483                                           struct ext4_sb_info *sbi, char *buf)
2484 {
2485         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2486
2487         if (!sb->s_bdev->bd_part)
2488                 return snprintf(buf, PAGE_SIZE, "0\n");
2489         return snprintf(buf, PAGE_SIZE, "%llu\n",
2490                         (unsigned long long)(sbi->s_kbytes_written +
2491                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2492                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2493 }
2494
2495 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2496                                       struct ext4_sb_info *sbi, char *buf)
2497 {
2498         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2499 }
2500
2501 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2502                                         struct ext4_sb_info *sbi, char *buf)
2503 {
2504         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2505 }
2506
2507 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2508                                           struct ext4_sb_info *sbi,
2509                                           const char *buf, size_t count)
2510 {
2511         unsigned long t;
2512
2513         if (parse_strtoul(buf, 0x40000000, &t))
2514                 return -EINVAL;
2515
2516         if (t && !is_power_of_2(t))
2517                 return -EINVAL;
2518
2519         sbi->s_inode_readahead_blks = t;
2520         return count;
2521 }
2522
2523 static ssize_t sbi_ui_show(struct ext4_attr *a,
2524                            struct ext4_sb_info *sbi, char *buf)
2525 {
2526         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2527
2528         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2529 }
2530
2531 static ssize_t sbi_ui_store(struct ext4_attr *a,
2532                             struct ext4_sb_info *sbi,
2533                             const char *buf, size_t count)
2534 {
2535         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2536         unsigned long t;
2537
2538         if (parse_strtoul(buf, 0xffffffff, &t))
2539                 return -EINVAL;
2540         *ui = t;
2541         return count;
2542 }
2543
2544 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2545 static struct ext4_attr ext4_attr_##_name = {                   \
2546         .attr = {.name = __stringify(_name), .mode = _mode },   \
2547         .show   = _show,                                        \
2548         .store  = _store,                                       \
2549         .offset = offsetof(struct ext4_sb_info, _elname),       \
2550 }
2551 #define EXT4_ATTR(name, mode, show, store) \
2552 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2553
2554 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2555 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2556 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2557 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2558         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2559 #define ATTR_LIST(name) &ext4_attr_##name.attr
2560
2561 EXT4_RO_ATTR(delayed_allocation_blocks);
2562 EXT4_RO_ATTR(session_write_kbytes);
2563 EXT4_RO_ATTR(lifetime_write_kbytes);
2564 EXT4_RO_ATTR(extent_cache_hits);
2565 EXT4_RO_ATTR(extent_cache_misses);
2566 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2567                  inode_readahead_blks_store, s_inode_readahead_blks);
2568 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2569 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2570 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2571 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2572 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2573 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2574 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2575 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2576
2577 static struct attribute *ext4_attrs[] = {
2578         ATTR_LIST(delayed_allocation_blocks),
2579         ATTR_LIST(session_write_kbytes),
2580         ATTR_LIST(lifetime_write_kbytes),
2581         ATTR_LIST(extent_cache_hits),
2582         ATTR_LIST(extent_cache_misses),
2583         ATTR_LIST(inode_readahead_blks),
2584         ATTR_LIST(inode_goal),
2585         ATTR_LIST(mb_stats),
2586         ATTR_LIST(mb_max_to_scan),
2587         ATTR_LIST(mb_min_to_scan),
2588         ATTR_LIST(mb_order2_req),
2589         ATTR_LIST(mb_stream_req),
2590         ATTR_LIST(mb_group_prealloc),
2591         ATTR_LIST(max_writeback_mb_bump),
2592         NULL,
2593 };
2594
2595 /* Features this copy of ext4 supports */
2596 EXT4_INFO_ATTR(lazy_itable_init);
2597 EXT4_INFO_ATTR(batched_discard);
2598
2599 static struct attribute *ext4_feat_attrs[] = {
2600         ATTR_LIST(lazy_itable_init),
2601         ATTR_LIST(batched_discard),
2602         NULL,
2603 };
2604
2605 static ssize_t ext4_attr_show(struct kobject *kobj,
2606                               struct attribute *attr, char *buf)
2607 {
2608         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2609                                                 s_kobj);
2610         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2611
2612         return a->show ? a->show(a, sbi, buf) : 0;
2613 }
2614
2615 static ssize_t ext4_attr_store(struct kobject *kobj,
2616                                struct attribute *attr,
2617                                const char *buf, size_t len)
2618 {
2619         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2620                                                 s_kobj);
2621         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2622
2623         return a->store ? a->store(a, sbi, buf, len) : 0;
2624 }
2625
2626 static void ext4_sb_release(struct kobject *kobj)
2627 {
2628         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2629                                                 s_kobj);
2630         complete(&sbi->s_kobj_unregister);
2631 }
2632
2633 static const struct sysfs_ops ext4_attr_ops = {
2634         .show   = ext4_attr_show,
2635         .store  = ext4_attr_store,
2636 };
2637
2638 static struct kobj_type ext4_ktype = {
2639         .default_attrs  = ext4_attrs,
2640         .sysfs_ops      = &ext4_attr_ops,
2641         .release        = ext4_sb_release,
2642 };
2643
2644 static void ext4_feat_release(struct kobject *kobj)
2645 {
2646         complete(&ext4_feat->f_kobj_unregister);
2647 }
2648
2649 static struct kobj_type ext4_feat_ktype = {
2650         .default_attrs  = ext4_feat_attrs,
2651         .sysfs_ops      = &ext4_attr_ops,
2652         .release        = ext4_feat_release,
2653 };
2654
2655 /*
2656  * Check whether this filesystem can be mounted based on
2657  * the features present and the RDONLY/RDWR mount requested.
2658  * Returns 1 if this filesystem can be mounted as requested,
2659  * 0 if it cannot be.
2660  */
2661 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2662 {
2663         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2664                 ext4_msg(sb, KERN_ERR,
2665                         "Couldn't mount because of "
2666                         "unsupported optional features (%x)",
2667                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2668                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2669                 return 0;
2670         }
2671
2672         if (readonly)
2673                 return 1;
2674
2675         /* Check that feature set is OK for a read-write mount */
2676         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2677                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2678                          "unsupported optional features (%x)",
2679                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2680                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2681                 return 0;
2682         }
2683         /*
2684          * Large file size enabled file system can only be mounted
2685          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2686          */
2687         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2688                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2689                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2690                                  "cannot be mounted RDWR without "
2691                                  "CONFIG_LBDAF");
2692                         return 0;
2693                 }
2694         }
2695         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2696             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2697                 ext4_msg(sb, KERN_ERR,
2698                          "Can't support bigalloc feature without "
2699                          "extents feature\n");
2700                 return 0;
2701         }
2702         return 1;
2703 }
2704
2705 /*
2706  * This function is called once a day if we have errors logged
2707  * on the file system
2708  */
2709 static void print_daily_error_info(unsigned long arg)
2710 {
2711         struct super_block *sb = (struct super_block *) arg;
2712         struct ext4_sb_info *sbi;
2713         struct ext4_super_block *es;
2714
2715         sbi = EXT4_SB(sb);
2716         es = sbi->s_es;
2717
2718         if (es->s_error_count)
2719                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2720                          le32_to_cpu(es->s_error_count));
2721         if (es->s_first_error_time) {
2722                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2723                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2724                        (int) sizeof(es->s_first_error_func),
2725                        es->s_first_error_func,
2726                        le32_to_cpu(es->s_first_error_line));
2727                 if (es->s_first_error_ino)
2728                         printk(": inode %u",
2729                                le32_to_cpu(es->s_first_error_ino));
2730                 if (es->s_first_error_block)
2731                         printk(": block %llu", (unsigned long long)
2732                                le64_to_cpu(es->s_first_error_block));
2733                 printk("\n");
2734         }
2735         if (es->s_last_error_time) {
2736                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2737                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2738                        (int) sizeof(es->s_last_error_func),
2739                        es->s_last_error_func,
2740                        le32_to_cpu(es->s_last_error_line));
2741                 if (es->s_last_error_ino)
2742                         printk(": inode %u",
2743                                le32_to_cpu(es->s_last_error_ino));
2744                 if (es->s_last_error_block)
2745                         printk(": block %llu", (unsigned long long)
2746                                le64_to_cpu(es->s_last_error_block));
2747                 printk("\n");
2748         }
2749         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2750 }
2751
2752 /* Find next suitable group and run ext4_init_inode_table */
2753 static int ext4_run_li_request(struct ext4_li_request *elr)
2754 {
2755         struct ext4_group_desc *gdp = NULL;
2756         ext4_group_t group, ngroups;
2757         struct super_block *sb;
2758         unsigned long timeout = 0;
2759         int ret = 0;
2760
2761         sb = elr->lr_super;
2762         ngroups = EXT4_SB(sb)->s_groups_count;
2763
2764         for (group = elr->lr_next_group; group < ngroups; group++) {
2765                 gdp = ext4_get_group_desc(sb, group, NULL);
2766                 if (!gdp) {
2767                         ret = 1;
2768                         break;
2769                 }
2770
2771                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2772                         break;
2773         }
2774
2775         if (group == ngroups)
2776                 ret = 1;
2777
2778         if (!ret) {
2779                 timeout = jiffies;
2780                 ret = ext4_init_inode_table(sb, group,
2781                                             elr->lr_timeout ? 0 : 1);
2782                 if (elr->lr_timeout == 0) {
2783                         timeout = (jiffies - timeout) *
2784                                   elr->lr_sbi->s_li_wait_mult;
2785                         elr->lr_timeout = timeout;
2786                 }
2787                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2788                 elr->lr_next_group = group + 1;
2789         }
2790
2791         return ret;
2792 }
2793
2794 /*
2795  * Remove lr_request from the list_request and free the
2796  * request structure. Should be called with li_list_mtx held
2797  */
2798 static void ext4_remove_li_request(struct ext4_li_request *elr)
2799 {
2800         struct ext4_sb_info *sbi;
2801
2802         if (!elr)
2803                 return;
2804
2805         sbi = elr->lr_sbi;
2806
2807         list_del(&elr->lr_request);
2808         sbi->s_li_request = NULL;
2809         kfree(elr);
2810 }
2811
2812 static void ext4_unregister_li_request(struct super_block *sb)
2813 {
2814         mutex_lock(&ext4_li_mtx);
2815         if (!ext4_li_info) {
2816                 mutex_unlock(&ext4_li_mtx);
2817                 return;
2818         }
2819
2820         mutex_lock(&ext4_li_info->li_list_mtx);
2821         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2822         mutex_unlock(&ext4_li_info->li_list_mtx);
2823         mutex_unlock(&ext4_li_mtx);
2824 }
2825
2826 static struct task_struct *ext4_lazyinit_task;
2827
2828 /*
2829  * This is the function where ext4lazyinit thread lives. It walks
2830  * through the request list searching for next scheduled filesystem.
2831  * When such a fs is found, run the lazy initialization request
2832  * (ext4_rn_li_request) and keep track of the time spend in this
2833  * function. Based on that time we compute next schedule time of
2834  * the request. When walking through the list is complete, compute
2835  * next waking time and put itself into sleep.
2836  */
2837 static int ext4_lazyinit_thread(void *arg)
2838 {
2839         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2840         struct list_head *pos, *n;
2841         struct ext4_li_request *elr;
2842         unsigned long next_wakeup, cur;
2843
2844         BUG_ON(NULL == eli);
2845
2846 cont_thread:
2847         while (true) {
2848                 next_wakeup = MAX_JIFFY_OFFSET;
2849
2850                 mutex_lock(&eli->li_list_mtx);
2851                 if (list_empty(&eli->li_request_list)) {
2852                         mutex_unlock(&eli->li_list_mtx);
2853                         goto exit_thread;
2854                 }
2855
2856                 list_for_each_safe(pos, n, &eli->li_request_list) {
2857                         elr = list_entry(pos, struct ext4_li_request,
2858                                          lr_request);
2859
2860                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2861                                 if (ext4_run_li_request(elr) != 0) {
2862                                         /* error, remove the lazy_init job */
2863                                         ext4_remove_li_request(elr);
2864                                         continue;
2865                                 }
2866                         }
2867
2868                         if (time_before(elr->lr_next_sched, next_wakeup))
2869                                 next_wakeup = elr->lr_next_sched;
2870                 }
2871                 mutex_unlock(&eli->li_list_mtx);
2872
2873                 try_to_freeze();
2874
2875                 cur = jiffies;
2876                 if ((time_after_eq(cur, next_wakeup)) ||
2877                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2878                         cond_resched();
2879                         continue;
2880                 }
2881
2882                 schedule_timeout_interruptible(next_wakeup - cur);
2883
2884                 if (kthread_should_stop()) {
2885                         ext4_clear_request_list();
2886                         goto exit_thread;
2887                 }
2888         }
2889
2890 exit_thread:
2891         /*
2892          * It looks like the request list is empty, but we need
2893          * to check it under the li_list_mtx lock, to prevent any
2894          * additions into it, and of course we should lock ext4_li_mtx
2895          * to atomically free the list and ext4_li_info, because at
2896          * this point another ext4 filesystem could be registering
2897          * new one.
2898          */
2899         mutex_lock(&ext4_li_mtx);
2900         mutex_lock(&eli->li_list_mtx);
2901         if (!list_empty(&eli->li_request_list)) {
2902                 mutex_unlock(&eli->li_list_mtx);
2903                 mutex_unlock(&ext4_li_mtx);
2904                 goto cont_thread;
2905         }
2906         mutex_unlock(&eli->li_list_mtx);
2907         kfree(ext4_li_info);
2908         ext4_li_info = NULL;
2909         mutex_unlock(&ext4_li_mtx);
2910
2911         return 0;
2912 }
2913
2914 static void ext4_clear_request_list(void)
2915 {
2916         struct list_head *pos, *n;
2917         struct ext4_li_request *elr;
2918
2919         mutex_lock(&ext4_li_info->li_list_mtx);
2920         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2921                 elr = list_entry(pos, struct ext4_li_request,
2922                                  lr_request);
2923                 ext4_remove_li_request(elr);
2924         }
2925         mutex_unlock(&ext4_li_info->li_list_mtx);
2926 }
2927
2928 static int ext4_run_lazyinit_thread(void)
2929 {
2930         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2931                                          ext4_li_info, "ext4lazyinit");
2932         if (IS_ERR(ext4_lazyinit_task)) {
2933                 int err = PTR_ERR(ext4_lazyinit_task);
2934                 ext4_clear_request_list();
2935                 kfree(ext4_li_info);
2936                 ext4_li_info = NULL;
2937                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2938                                  "initialization thread\n",
2939                                  err);
2940                 return err;
2941         }
2942         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2943         return 0;
2944 }
2945
2946 /*
2947  * Check whether it make sense to run itable init. thread or not.
2948  * If there is at least one uninitialized inode table, return
2949  * corresponding group number, else the loop goes through all
2950  * groups and return total number of groups.
2951  */
2952 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2953 {
2954         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2955         struct ext4_group_desc *gdp = NULL;
2956
2957         for (group = 0; group < ngroups; group++) {
2958                 gdp = ext4_get_group_desc(sb, group, NULL);
2959                 if (!gdp)
2960                         continue;
2961
2962                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2963                         break;
2964         }
2965
2966         return group;
2967 }
2968
2969 static int ext4_li_info_new(void)
2970 {
2971         struct ext4_lazy_init *eli = NULL;
2972
2973         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2974         if (!eli)
2975                 return -ENOMEM;
2976
2977         INIT_LIST_HEAD(&eli->li_request_list);
2978         mutex_init(&eli->li_list_mtx);
2979
2980         eli->li_state |= EXT4_LAZYINIT_QUIT;
2981
2982         ext4_li_info = eli;
2983
2984         return 0;
2985 }
2986
2987 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2988                                             ext4_group_t start)
2989 {
2990         struct ext4_sb_info *sbi = EXT4_SB(sb);
2991         struct ext4_li_request *elr;
2992         unsigned long rnd;
2993
2994         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2995         if (!elr)
2996                 return NULL;
2997
2998         elr->lr_super = sb;
2999         elr->lr_sbi = sbi;
3000         elr->lr_next_group = start;
3001
3002         /*
3003          * Randomize first schedule time of the request to
3004          * spread the inode table initialization requests
3005          * better.
3006          */
3007         get_random_bytes(&rnd, sizeof(rnd));
3008         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3009                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3010
3011         return elr;
3012 }
3013
3014 static int ext4_register_li_request(struct super_block *sb,
3015                                     ext4_group_t first_not_zeroed)
3016 {
3017         struct ext4_sb_info *sbi = EXT4_SB(sb);
3018         struct ext4_li_request *elr;
3019         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3020         int ret = 0;
3021
3022         if (sbi->s_li_request != NULL) {
3023                 /*
3024                  * Reset timeout so it can be computed again, because
3025                  * s_li_wait_mult might have changed.
3026                  */
3027                 sbi->s_li_request->lr_timeout = 0;
3028                 return 0;
3029         }
3030
3031         if (first_not_zeroed == ngroups ||
3032             (sb->s_flags & MS_RDONLY) ||
3033             !test_opt(sb, INIT_INODE_TABLE))
3034                 return 0;
3035
3036         elr = ext4_li_request_new(sb, first_not_zeroed);
3037         if (!elr)
3038                 return -ENOMEM;
3039
3040         mutex_lock(&ext4_li_mtx);
3041
3042         if (NULL == ext4_li_info) {
3043                 ret = ext4_li_info_new();
3044                 if (ret)
3045                         goto out;
3046         }
3047
3048         mutex_lock(&ext4_li_info->li_list_mtx);
3049         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3050         mutex_unlock(&ext4_li_info->li_list_mtx);
3051
3052         sbi->s_li_request = elr;
3053         /*
3054          * set elr to NULL here since it has been inserted to
3055          * the request_list and the removal and free of it is
3056          * handled by ext4_clear_request_list from now on.
3057          */
3058         elr = NULL;
3059
3060         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3061                 ret = ext4_run_lazyinit_thread();
3062                 if (ret)
3063                         goto out;
3064         }
3065 out:
3066         mutex_unlock(&ext4_li_mtx);
3067         if (ret)
3068                 kfree(elr);
3069         return ret;
3070 }
3071
3072 /*
3073  * We do not need to lock anything since this is called on
3074  * module unload.
3075  */
3076 static void ext4_destroy_lazyinit_thread(void)
3077 {
3078         /*
3079          * If thread exited earlier
3080          * there's nothing to be done.
3081          */
3082         if (!ext4_li_info || !ext4_lazyinit_task)
3083                 return;
3084
3085         kthread_stop(ext4_lazyinit_task);
3086 }
3087
3088 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3089 {
3090         char *orig_data = kstrdup(data, GFP_KERNEL);
3091         struct buffer_head *bh;
3092         struct ext4_super_block *es = NULL;
3093         struct ext4_sb_info *sbi;
3094         ext4_fsblk_t block;
3095         ext4_fsblk_t sb_block = get_sb_block(&data);
3096         ext4_fsblk_t logical_sb_block;
3097         unsigned long offset = 0;
3098         unsigned long journal_devnum = 0;
3099         unsigned long def_mount_opts;
3100         struct inode *root;
3101         char *cp;
3102         const char *descr;
3103         int ret = -ENOMEM;
3104         int blocksize, clustersize;
3105         unsigned int db_count;
3106         unsigned int i;
3107         int needs_recovery, has_huge_files, has_bigalloc;
3108         __u64 blocks_count;
3109         int err;
3110         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3111         ext4_group_t first_not_zeroed;
3112
3113         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3114         if (!sbi)
3115                 goto out_free_orig;
3116
3117         sbi->s_blockgroup_lock =
3118                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3119         if (!sbi->s_blockgroup_lock) {
3120                 kfree(sbi);
3121                 goto out_free_orig;
3122         }
3123         sb->s_fs_info = sbi;
3124         sbi->s_mount_opt = 0;
3125         sbi->s_resuid = EXT4_DEF_RESUID;
3126         sbi->s_resgid = EXT4_DEF_RESGID;
3127         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3128         sbi->s_sb_block = sb_block;
3129         if (sb->s_bdev->bd_part)
3130                 sbi->s_sectors_written_start =
3131                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3132
3133         /* Cleanup superblock name */
3134         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3135                 *cp = '!';
3136
3137         ret = -EINVAL;
3138         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3139         if (!blocksize) {
3140                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3141                 goto out_fail;
3142         }
3143
3144         /*
3145          * The ext4 superblock will not be buffer aligned for other than 1kB
3146          * block sizes.  We need to calculate the offset from buffer start.
3147          */
3148         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3149                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3150                 offset = do_div(logical_sb_block, blocksize);
3151         } else {
3152                 logical_sb_block = sb_block;
3153         }
3154
3155         if (!(bh = sb_bread(sb, logical_sb_block))) {
3156                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3157                 goto out_fail;
3158         }
3159         /*
3160          * Note: s_es must be initialized as soon as possible because
3161          *       some ext4 macro-instructions depend on its value
3162          */
3163         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3164         sbi->s_es = es;
3165         sb->s_magic = le16_to_cpu(es->s_magic);
3166         if (sb->s_magic != EXT4_SUPER_MAGIC)
3167                 goto cantfind_ext4;
3168         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3169
3170         /* Set defaults before we parse the mount options */
3171         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3172         set_opt(sb, INIT_INODE_TABLE);
3173         if (def_mount_opts & EXT4_DEFM_DEBUG)
3174                 set_opt(sb, DEBUG);
3175         if (def_mount_opts & EXT4_DEFM_BSDGROUPS) {
3176                 ext4_msg(sb, KERN_WARNING, deprecated_msg, "bsdgroups",
3177                         "2.6.38");
3178                 set_opt(sb, GRPID);
3179         }
3180         if (def_mount_opts & EXT4_DEFM_UID16)
3181                 set_opt(sb, NO_UID32);
3182         /* xattr user namespace & acls are now defaulted on */
3183 #ifdef CONFIG_EXT4_FS_XATTR
3184         set_opt(sb, XATTR_USER);
3185 #endif
3186 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3187         set_opt(sb, POSIX_ACL);
3188 #endif
3189         set_opt(sb, MBLK_IO_SUBMIT);
3190         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3191                 set_opt(sb, JOURNAL_DATA);
3192         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3193                 set_opt(sb, ORDERED_DATA);
3194         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3195                 set_opt(sb, WRITEBACK_DATA);
3196
3197         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3198                 set_opt(sb, ERRORS_PANIC);
3199         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3200                 set_opt(sb, ERRORS_CONT);
3201         else
3202                 set_opt(sb, ERRORS_RO);
3203         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3204                 set_opt(sb, BLOCK_VALIDITY);
3205         if (def_mount_opts & EXT4_DEFM_DISCARD)
3206                 set_opt(sb, DISCARD);
3207
3208         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3209         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3210         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3211         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3212         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3213
3214         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3215                 set_opt(sb, BARRIER);
3216
3217         /*
3218          * enable delayed allocation by default
3219          * Use -o nodelalloc to turn it off
3220          */
3221         if (!IS_EXT3_SB(sb) &&
3222             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3223                 set_opt(sb, DELALLOC);
3224
3225         /*
3226          * set default s_li_wait_mult for lazyinit, for the case there is
3227          * no mount option specified.
3228          */
3229         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3230
3231         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3232                            &journal_devnum, &journal_ioprio, 0)) {
3233                 ext4_msg(sb, KERN_WARNING,
3234                          "failed to parse options in superblock: %s",
3235                          sbi->s_es->s_mount_opts);
3236         }
3237         if (!parse_options((char *) data, sb, &journal_devnum,
3238                            &journal_ioprio, 0))
3239                 goto failed_mount;
3240
3241         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3242                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3243                             "with data=journal disables delayed "
3244                             "allocation and O_DIRECT support!\n");
3245                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3246                         ext4_msg(sb, KERN_ERR, "can't mount with "
3247                                  "both data=journal and delalloc");
3248                         goto failed_mount;
3249                 }
3250                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3251                         ext4_msg(sb, KERN_ERR, "can't mount with "
3252                                  "both data=journal and delalloc");
3253                         goto failed_mount;
3254                 }
3255                 if (test_opt(sb, DELALLOC))
3256                         clear_opt(sb, DELALLOC);
3257         }
3258
3259         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3260         if (test_opt(sb, DIOREAD_NOLOCK)) {
3261                 if (blocksize < PAGE_SIZE) {
3262                         ext4_msg(sb, KERN_ERR, "can't mount with "
3263                                  "dioread_nolock if block size != PAGE_SIZE");
3264                         goto failed_mount;
3265                 }
3266         }
3267
3268         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3269                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3270
3271         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3272             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3273              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3274              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3275                 ext4_msg(sb, KERN_WARNING,
3276                        "feature flags set on rev 0 fs, "
3277                        "running e2fsck is recommended");
3278
3279         if (IS_EXT2_SB(sb)) {
3280                 if (ext2_feature_set_ok(sb))
3281                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3282                                  "using the ext4 subsystem");
3283                 else {
3284                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3285                                  "to feature incompatibilities");
3286                         goto failed_mount;
3287                 }
3288         }
3289
3290         if (IS_EXT3_SB(sb)) {
3291                 if (ext3_feature_set_ok(sb))
3292                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3293                                  "using the ext4 subsystem");
3294                 else {
3295                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3296                                  "to feature incompatibilities");
3297                         goto failed_mount;
3298                 }
3299         }
3300
3301         /*
3302          * Check feature flags regardless of the revision level, since we
3303          * previously didn't change the revision level when setting the flags,
3304          * so there is a chance incompat flags are set on a rev 0 filesystem.
3305          */
3306         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3307                 goto failed_mount;
3308
3309         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3310             blocksize > EXT4_MAX_BLOCK_SIZE) {
3311                 ext4_msg(sb, KERN_ERR,
3312                        "Unsupported filesystem blocksize %d", blocksize);
3313                 goto failed_mount;
3314         }
3315
3316         if (sb->s_blocksize != blocksize) {
3317                 /* Validate the filesystem blocksize */
3318                 if (!sb_set_blocksize(sb, blocksize)) {
3319                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3320                                         blocksize);
3321                         goto failed_mount;
3322                 }
3323
3324                 brelse(bh);
3325                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3326                 offset = do_div(logical_sb_block, blocksize);
3327                 bh = sb_bread(sb, logical_sb_block);
3328                 if (!bh) {
3329                         ext4_msg(sb, KERN_ERR,
3330                                "Can't read superblock on 2nd try");
3331                         goto failed_mount;
3332                 }
3333                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3334                 sbi->s_es = es;
3335                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3336                         ext4_msg(sb, KERN_ERR,
3337                                "Magic mismatch, very weird!");
3338                         goto failed_mount;
3339                 }
3340         }
3341
3342         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3343                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3344         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3345                                                       has_huge_files);
3346         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3347
3348         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3349                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3350                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3351         } else {
3352                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3353                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3354                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3355                     (!is_power_of_2(sbi->s_inode_size)) ||
3356                     (sbi->s_inode_size > blocksize)) {
3357                         ext4_msg(sb, KERN_ERR,
3358                                "unsupported inode size: %d",
3359                                sbi->s_inode_size);
3360                         goto failed_mount;
3361                 }
3362                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3363                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3364         }
3365
3366         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3367         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3368                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3369                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3370                     !is_power_of_2(sbi->s_desc_size)) {
3371                         ext4_msg(sb, KERN_ERR,
3372                                "unsupported descriptor size %lu",
3373                                sbi->s_desc_size);
3374                         goto failed_mount;
3375                 }
3376         } else
3377                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3378
3379         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3380         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3381         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3382                 goto cantfind_ext4;
3383
3384         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3385         if (sbi->s_inodes_per_block == 0)
3386                 goto cantfind_ext4;
3387         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3388                                         sbi->s_inodes_per_block;
3389         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3390         sbi->s_sbh = bh;
3391         sbi->s_mount_state = le16_to_cpu(es->s_state);
3392         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3393         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3394
3395         for (i = 0; i < 4; i++)
3396                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3397         sbi->s_def_hash_version = es->s_def_hash_version;
3398         i = le32_to_cpu(es->s_flags);
3399         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3400                 sbi->s_hash_unsigned = 3;
3401         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3402 #ifdef __CHAR_UNSIGNED__
3403                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3404                 sbi->s_hash_unsigned = 3;
3405 #else
3406                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3407 #endif
3408                 sb->s_dirt = 1;
3409         }
3410
3411         /* Handle clustersize */
3412         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3413         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3414                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3415         if (has_bigalloc) {
3416                 if (clustersize < blocksize) {
3417                         ext4_msg(sb, KERN_ERR,
3418                                  "cluster size (%d) smaller than "
3419                                  "block size (%d)", clustersize, blocksize);
3420                         goto failed_mount;
3421                 }
3422                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3423                         le32_to_cpu(es->s_log_block_size);
3424                 sbi->s_clusters_per_group =
3425                         le32_to_cpu(es->s_clusters_per_group);
3426                 if (sbi->s_clusters_per_group > blocksize * 8) {
3427                         ext4_msg(sb, KERN_ERR,
3428                                  "#clusters per group too big: %lu",
3429                                  sbi->s_clusters_per_group);
3430                         goto failed_mount;
3431                 }
3432                 if (sbi->s_blocks_per_group !=
3433                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3434                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3435                                  "clusters per group (%lu) inconsistent",
3436                                  sbi->s_blocks_per_group,
3437                                  sbi->s_clusters_per_group);
3438                         goto failed_mount;
3439                 }
3440         } else {
3441                 if (clustersize != blocksize) {
3442                         ext4_warning(sb, "fragment/cluster size (%d) != "
3443                                      "block size (%d)", clustersize,
3444                                      blocksize);
3445                         clustersize = blocksize;
3446                 }
3447                 if (sbi->s_blocks_per_group > blocksize * 8) {
3448                         ext4_msg(sb, KERN_ERR,
3449                                  "#blocks per group too big: %lu",
3450                                  sbi->s_blocks_per_group);
3451                         goto failed_mount;
3452                 }
3453                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3454                 sbi->s_cluster_bits = 0;
3455         }
3456         sbi->s_cluster_ratio = clustersize / blocksize;
3457
3458         if (sbi->s_inodes_per_group > blocksize * 8) {
3459                 ext4_msg(sb, KERN_ERR,
3460                        "#inodes per group too big: %lu",
3461                        sbi->s_inodes_per_group);
3462                 goto failed_mount;
3463         }
3464
3465         /*
3466          * Test whether we have more sectors than will fit in sector_t,
3467          * and whether the max offset is addressable by the page cache.
3468          */
3469         err = generic_check_addressable(sb->s_blocksize_bits,
3470                                         ext4_blocks_count(es));
3471         if (err) {
3472                 ext4_msg(sb, KERN_ERR, "filesystem"
3473                          " too large to mount safely on this system");
3474                 if (sizeof(sector_t) < 8)
3475                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3476                 ret = err;
3477                 goto failed_mount;
3478         }
3479
3480         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3481                 goto cantfind_ext4;
3482
3483         /* check blocks count against device size */
3484         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3485         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3486                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3487                        "exceeds size of device (%llu blocks)",
3488                        ext4_blocks_count(es), blocks_count);
3489                 goto failed_mount;
3490         }
3491
3492         /*
3493          * It makes no sense for the first data block to be beyond the end
3494          * of the filesystem.
3495          */
3496         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3497                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3498                          "block %u is beyond end of filesystem (%llu)",
3499                          le32_to_cpu(es->s_first_data_block),
3500                          ext4_blocks_count(es));
3501                 goto failed_mount;
3502         }
3503         blocks_count = (ext4_blocks_count(es) -
3504                         le32_to_cpu(es->s_first_data_block) +
3505                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3506         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3507         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3508                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3509                        "(block count %llu, first data block %u, "
3510                        "blocks per group %lu)", sbi->s_groups_count,
3511                        ext4_blocks_count(es),
3512                        le32_to_cpu(es->s_first_data_block),
3513                        EXT4_BLOCKS_PER_GROUP(sb));
3514                 goto failed_mount;
3515         }
3516         sbi->s_groups_count = blocks_count;
3517         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3518                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3519         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3520                    EXT4_DESC_PER_BLOCK(sb);
3521         sbi->s_group_desc = ext4_kvmalloc(db_count *
3522                                           sizeof(struct buffer_head *),
3523                                           GFP_KERNEL);
3524         if (sbi->s_group_desc == NULL) {
3525                 ext4_msg(sb, KERN_ERR, "not enough memory");
3526                 goto failed_mount;
3527         }
3528
3529         if (ext4_proc_root)
3530                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3531
3532         bgl_lock_init(sbi->s_blockgroup_lock);
3533
3534         for (i = 0; i < db_count; i++) {
3535                 block = descriptor_loc(sb, logical_sb_block, i);
3536                 sbi->s_group_desc[i] = sb_bread(sb, block);
3537                 if (!sbi->s_group_desc[i]) {
3538                         ext4_msg(sb, KERN_ERR,
3539                                "can't read group descriptor %d", i);
3540                         db_count = i;
3541                         goto failed_mount2;
3542                 }
3543         }
3544         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3545                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3546                 goto failed_mount2;
3547         }
3548         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3549                 if (!ext4_fill_flex_info(sb)) {
3550                         ext4_msg(sb, KERN_ERR,
3551                                "unable to initialize "
3552                                "flex_bg meta info!");
3553                         goto failed_mount2;
3554                 }
3555
3556         sbi->s_gdb_count = db_count;
3557         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3558         spin_lock_init(&sbi->s_next_gen_lock);
3559
3560         init_timer(&sbi->s_err_report);
3561         sbi->s_err_report.function = print_daily_error_info;
3562         sbi->s_err_report.data = (unsigned long) sb;
3563
3564         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3565                         ext4_count_free_clusters(sb));
3566         if (!err) {
3567                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3568                                 ext4_count_free_inodes(sb));
3569         }
3570         if (!err) {
3571                 err = percpu_counter_init(&sbi->s_dirs_counter,
3572                                 ext4_count_dirs(sb));
3573         }
3574         if (!err) {
3575                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3576         }
3577         if (err) {
3578                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3579                 goto failed_mount3;
3580         }
3581
3582         sbi->s_stripe = ext4_get_stripe_size(sbi);
3583         sbi->s_max_writeback_mb_bump = 128;
3584
3585         /*
3586          * set up enough so that it can read an inode
3587          */
3588         if (!test_opt(sb, NOLOAD) &&
3589             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3590                 sb->s_op = &ext4_sops;
3591         else
3592                 sb->s_op = &ext4_nojournal_sops;
3593         sb->s_export_op = &ext4_export_ops;
3594         sb->s_xattr = ext4_xattr_handlers;
3595 #ifdef CONFIG_QUOTA
3596         sb->s_qcop = &ext4_qctl_operations;
3597         sb->dq_op = &ext4_quota_operations;
3598 #endif
3599         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3600
3601         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3602         mutex_init(&sbi->s_orphan_lock);
3603         sbi->s_resize_flags = 0;
3604
3605         sb->s_root = NULL;
3606
3607         needs_recovery = (es->s_last_orphan != 0 ||
3608                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3609                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3610
3611         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3612             !(sb->s_flags & MS_RDONLY))
3613                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3614                         goto failed_mount3;
3615
3616         /*
3617          * The first inode we look at is the journal inode.  Don't try
3618          * root first: it may be modified in the journal!
3619          */
3620         if (!test_opt(sb, NOLOAD) &&
3621             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3622                 if (ext4_load_journal(sb, es, journal_devnum))
3623                         goto failed_mount3;
3624         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3625               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3626                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3627                        "suppressed and not mounted read-only");
3628                 goto failed_mount_wq;
3629         } else {
3630                 clear_opt(sb, DATA_FLAGS);
3631                 sbi->s_journal = NULL;
3632                 needs_recovery = 0;
3633                 goto no_journal;
3634         }
3635
3636         if (ext4_blocks_count(es) > 0xffffffffULL &&
3637             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3638                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3639                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3640                 goto failed_mount_wq;
3641         }
3642
3643         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3644                 jbd2_journal_set_features(sbi->s_journal,
3645                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3646                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3647         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3648                 jbd2_journal_set_features(sbi->s_journal,
3649                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3650                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3651                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3652         } else {
3653                 jbd2_journal_clear_features(sbi->s_journal,
3654                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3655                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3656         }
3657
3658         /* We have now updated the journal if required, so we can
3659          * validate the data journaling mode. */
3660         switch (test_opt(sb, DATA_FLAGS)) {
3661         case 0:
3662                 /* No mode set, assume a default based on the journal
3663                  * capabilities: ORDERED_DATA if the journal can
3664                  * cope, else JOURNAL_DATA
3665                  */
3666                 if (jbd2_journal_check_available_features
3667                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3668                         set_opt(sb, ORDERED_DATA);
3669                 else
3670                         set_opt(sb, JOURNAL_DATA);
3671                 break;
3672
3673         case EXT4_MOUNT_ORDERED_DATA:
3674         case EXT4_MOUNT_WRITEBACK_DATA:
3675                 if (!jbd2_journal_check_available_features
3676                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3677                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3678                                "requested data journaling mode");
3679                         goto failed_mount_wq;
3680                 }
3681         default:
3682                 break;
3683         }
3684         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3685
3686         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3687
3688         /*
3689          * The journal may have updated the bg summary counts, so we
3690          * need to update the global counters.
3691          */
3692         percpu_counter_set(&sbi->s_freeclusters_counter,
3693                            ext4_count_free_clusters(sb));
3694         percpu_counter_set(&sbi->s_freeinodes_counter,
3695                            ext4_count_free_inodes(sb));
3696         percpu_counter_set(&sbi->s_dirs_counter,
3697                            ext4_count_dirs(sb));
3698         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3699
3700 no_journal:
3701         /*
3702          * The maximum number of concurrent works can be high and
3703          * concurrency isn't really necessary.  Limit it to 1.
3704          */
3705         EXT4_SB(sb)->dio_unwritten_wq =
3706                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3707         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3708                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3709                 goto failed_mount_wq;
3710         }
3711
3712         /*
3713          * The jbd2_journal_load will have done any necessary log recovery,
3714          * so we can safely mount the rest of the filesystem now.
3715          */
3716
3717         root = ext4_iget(sb, EXT4_ROOT_INO);
3718         if (IS_ERR(root)) {
3719                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3720                 ret = PTR_ERR(root);
3721                 root = NULL;
3722                 goto failed_mount4;
3723         }
3724         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3725                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3726                 iput(root);
3727                 goto failed_mount4;
3728         }
3729         sb->s_root = d_alloc_root(root);
3730         if (!sb->s_root) {
3731                 iput(root);
3732                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3733                 ret = -ENOMEM;
3734                 goto failed_mount4;
3735         }
3736
3737         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3738
3739         /* determine the minimum size of new large inodes, if present */
3740         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3741                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3742                                                      EXT4_GOOD_OLD_INODE_SIZE;
3743                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3744                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3745                         if (sbi->s_want_extra_isize <
3746                             le16_to_cpu(es->s_want_extra_isize))
3747                                 sbi->s_want_extra_isize =
3748                                         le16_to_cpu(es->s_want_extra_isize);
3749                         if (sbi->s_want_extra_isize <
3750                             le16_to_cpu(es->s_min_extra_isize))
3751                                 sbi->s_want_extra_isize =
3752                                         le16_to_cpu(es->s_min_extra_isize);
3753                 }
3754         }
3755         /* Check if enough inode space is available */
3756         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3757                                                         sbi->s_inode_size) {
3758                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3759                                                        EXT4_GOOD_OLD_INODE_SIZE;
3760                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3761                          "available");
3762         }
3763
3764         err = ext4_setup_system_zone(sb);
3765         if (err) {
3766                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3767                          "zone (%d)", err);
3768                 goto failed_mount4a;
3769         }
3770
3771         ext4_ext_init(sb);
3772         err = ext4_mb_init(sb, needs_recovery);
3773         if (err) {
3774                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3775                          err);
3776                 goto failed_mount5;
3777         }
3778
3779         err = ext4_register_li_request(sb, first_not_zeroed);
3780         if (err)
3781                 goto failed_mount6;
3782
3783         sbi->s_kobj.kset = ext4_kset;
3784         init_completion(&sbi->s_kobj_unregister);
3785         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3786                                    "%s", sb->s_id);
3787         if (err)
3788                 goto failed_mount7;
3789
3790         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3791         ext4_orphan_cleanup(sb, es);
3792         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3793         if (needs_recovery) {
3794                 ext4_msg(sb, KERN_INFO, "recovery complete");
3795                 ext4_mark_recovery_complete(sb, es);
3796         }
3797         if (EXT4_SB(sb)->s_journal) {
3798                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3799                         descr = " journalled data mode";
3800                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3801                         descr = " ordered data mode";
3802                 else
3803                         descr = " writeback data mode";
3804         } else
3805                 descr = "out journal";
3806
3807         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3808                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3809                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3810
3811         if (es->s_error_count)
3812                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3813
3814         kfree(orig_data);
3815         return 0;
3816
3817 cantfind_ext4:
3818         if (!silent)
3819                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3820         goto failed_mount;
3821
3822 failed_mount7:
3823         ext4_unregister_li_request(sb);
3824 failed_mount6:
3825         ext4_mb_release(sb);
3826 failed_mount5:
3827         ext4_ext_release(sb);
3828         ext4_release_system_zone(sb);
3829 failed_mount4a:
3830         dput(sb->s_root);
3831         sb->s_root = NULL;
3832 failed_mount4:
3833         ext4_msg(sb, KERN_ERR, "mount failed");
3834         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3835 failed_mount_wq:
3836         if (sbi->s_journal) {
3837                 jbd2_journal_destroy(sbi->s_journal);
3838                 sbi->s_journal = NULL;
3839         }
3840 failed_mount3:
3841         del_timer(&sbi->s_err_report);
3842         if (sbi->s_flex_groups)
3843                 ext4_kvfree(sbi->s_flex_groups);
3844         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3845         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3846         percpu_counter_destroy(&sbi->s_dirs_counter);
3847         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3848         if (sbi->s_mmp_tsk)
3849                 kthread_stop(sbi->s_mmp_tsk);
3850 failed_mount2:
3851         for (i = 0; i < db_count; i++)
3852                 brelse(sbi->s_group_desc[i]);
3853         ext4_kvfree(sbi->s_group_desc);
3854 failed_mount:
3855         if (sbi->s_proc) {
3856                 remove_proc_entry(sb->s_id, ext4_proc_root);
3857         }
3858 #ifdef CONFIG_QUOTA
3859         for (i = 0; i < MAXQUOTAS; i++)
3860                 kfree(sbi->s_qf_names[i]);
3861 #endif
3862         ext4_blkdev_remove(sbi);
3863         brelse(bh);
3864 out_fail:
3865         sb->s_fs_info = NULL;
3866         kfree(sbi->s_blockgroup_lock);
3867         kfree(sbi);
3868 out_free_orig:
3869         kfree(orig_data);
3870         return ret;
3871 }
3872
3873 /*
3874  * Setup any per-fs journal parameters now.  We'll do this both on
3875  * initial mount, once the journal has been initialised but before we've
3876  * done any recovery; and again on any subsequent remount.
3877  */
3878 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3879 {
3880         struct ext4_sb_info *sbi = EXT4_SB(sb);
3881
3882         journal->j_commit_interval = sbi->s_commit_interval;
3883         journal->j_min_batch_time = sbi->s_min_batch_time;
3884         journal->j_max_batch_time = sbi->s_max_batch_time;
3885
3886         write_lock(&journal->j_state_lock);
3887         if (test_opt(sb, BARRIER))
3888                 journal->j_flags |= JBD2_BARRIER;
3889         else
3890                 journal->j_flags &= ~JBD2_BARRIER;
3891         if (test_opt(sb, DATA_ERR_ABORT))
3892                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3893         else
3894                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3895         write_unlock(&journal->j_state_lock);
3896 }
3897
3898 static journal_t *ext4_get_journal(struct super_block *sb,
3899                                    unsigned int journal_inum)
3900 {
3901         struct inode *journal_inode;
3902         journal_t *journal;
3903
3904         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3905
3906         /* First, test for the existence of a valid inode on disk.  Bad
3907          * things happen if we iget() an unused inode, as the subsequent
3908          * iput() will try to delete it. */
3909
3910         journal_inode = ext4_iget(sb, journal_inum);
3911         if (IS_ERR(journal_inode)) {
3912                 ext4_msg(sb, KERN_ERR, "no journal found");
3913                 return NULL;
3914         }
3915         if (!journal_inode->i_nlink) {
3916                 make_bad_inode(journal_inode);
3917                 iput(journal_inode);
3918                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3919                 return NULL;
3920         }
3921
3922         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3923                   journal_inode, journal_inode->i_size);
3924         if (!S_ISREG(journal_inode->i_mode)) {
3925                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3926                 iput(journal_inode);
3927                 return NULL;
3928         }
3929
3930         journal = jbd2_journal_init_inode(journal_inode);
3931         if (!journal) {
3932                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3933                 iput(journal_inode);
3934                 return NULL;
3935         }
3936         journal->j_private = sb;
3937         ext4_init_journal_params(sb, journal);
3938         return journal;
3939 }
3940
3941 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3942                                        dev_t j_dev)
3943 {
3944         struct buffer_head *bh;
3945         journal_t *journal;
3946         ext4_fsblk_t start;
3947         ext4_fsblk_t len;
3948         int hblock, blocksize;
3949         ext4_fsblk_t sb_block;
3950         unsigned long offset;
3951         struct ext4_super_block *es;
3952         struct block_device *bdev;
3953
3954         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3955
3956         bdev = ext4_blkdev_get(j_dev, sb);
3957         if (bdev == NULL)
3958                 return NULL;
3959
3960         blocksize = sb->s_blocksize;
3961         hblock = bdev_logical_block_size(bdev);
3962         if (blocksize < hblock) {
3963                 ext4_msg(sb, KERN_ERR,
3964                         "blocksize too small for journal device");
3965                 goto out_bdev;
3966         }
3967
3968         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3969         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3970         set_blocksize(bdev, blocksize);
3971         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3972                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3973                        "external journal");
3974                 goto out_bdev;
3975         }
3976
3977         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3978         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3979             !(le32_to_cpu(es->s_feature_incompat) &
3980               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3981                 ext4_msg(sb, KERN_ERR, "external journal has "
3982                                         "bad superblock");
3983                 brelse(bh);
3984                 goto out_bdev;
3985         }
3986
3987         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3988                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3989                 brelse(bh);
3990                 goto out_bdev;
3991         }
3992
3993         len = ext4_blocks_count(es);
3994         start = sb_block + 1;
3995         brelse(bh);     /* we're done with the superblock */
3996
3997         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3998                                         start, len, blocksize);
3999         if (!journal) {
4000                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4001                 goto out_bdev;
4002         }
4003         journal->j_private = sb;
4004         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4005         wait_on_buffer(journal->j_sb_buffer);
4006         if (!buffer_uptodate(journal->j_sb_buffer)) {
4007                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4008                 goto out_journal;
4009         }
4010         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4011                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4012                                         "user (unsupported) - %d",
4013                         be32_to_cpu(journal->j_superblock->s_nr_users));
4014                 goto out_journal;
4015         }
4016         EXT4_SB(sb)->journal_bdev = bdev;
4017         ext4_init_journal_params(sb, journal);
4018         return journal;
4019
4020 out_journal:
4021         jbd2_journal_destroy(journal);
4022 out_bdev:
4023         ext4_blkdev_put(bdev);
4024         return NULL;
4025 }
4026
4027 static int ext4_load_journal(struct super_block *sb,
4028                              struct ext4_super_block *es,
4029                              unsigned long journal_devnum)
4030 {
4031         journal_t *journal;
4032         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4033         dev_t journal_dev;
4034         int err = 0;
4035         int really_read_only;
4036
4037         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4038
4039         if (journal_devnum &&
4040             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4041                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4042                         "numbers have changed");
4043                 journal_dev = new_decode_dev(journal_devnum);
4044         } else
4045                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4046
4047         really_read_only = bdev_read_only(sb->s_bdev);
4048
4049         /*
4050          * Are we loading a blank journal or performing recovery after a
4051          * crash?  For recovery, we need to check in advance whether we
4052          * can get read-write access to the device.
4053          */
4054         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4055                 if (sb->s_flags & MS_RDONLY) {
4056                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4057                                         "required on readonly filesystem");
4058                         if (really_read_only) {
4059                                 ext4_msg(sb, KERN_ERR, "write access "
4060                                         "unavailable, cannot proceed");
4061                                 return -EROFS;
4062                         }
4063                         ext4_msg(sb, KERN_INFO, "write access will "
4064                                "be enabled during recovery");
4065                 }
4066         }
4067
4068         if (journal_inum && journal_dev) {
4069                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4070                        "and inode journals!");
4071                 return -EINVAL;
4072         }
4073
4074         if (journal_inum) {
4075                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4076                         return -EINVAL;
4077         } else {
4078                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4079                         return -EINVAL;
4080         }
4081
4082         if (!(journal->j_flags & JBD2_BARRIER))
4083                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4084
4085         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4086                 err = jbd2_journal_wipe(journal, !really_read_only);
4087         if (!err) {
4088                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4089                 if (save)
4090                         memcpy(save, ((char *) es) +
4091                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4092                 err = jbd2_journal_load(journal);
4093                 if (save)
4094                         memcpy(((char *) es) + EXT4_S_ERR_START,
4095                                save, EXT4_S_ERR_LEN);
4096                 kfree(save);
4097         }
4098
4099         if (err) {
4100                 ext4_msg(sb, KERN_ERR, "error loading journal");
4101                 jbd2_journal_destroy(journal);
4102                 return err;
4103         }
4104
4105         EXT4_SB(sb)->s_journal = journal;
4106         ext4_clear_journal_err(sb, es);
4107
4108         if (!really_read_only && journal_devnum &&
4109             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4110                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4111
4112                 /* Make sure we flush the recovery flag to disk. */
4113                 ext4_commit_super(sb, 1);
4114         }
4115
4116         return 0;
4117 }
4118
4119 static int ext4_commit_super(struct super_block *sb, int sync)
4120 {
4121         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4122         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4123         int error = 0;
4124
4125         if (!sbh || block_device_ejected(sb))
4126                 return error;
4127         if (buffer_write_io_error(sbh)) {
4128                 /*
4129                  * Oh, dear.  A previous attempt to write the
4130                  * superblock failed.  This could happen because the
4131                  * USB device was yanked out.  Or it could happen to
4132                  * be a transient write error and maybe the block will
4133                  * be remapped.  Nothing we can do but to retry the
4134                  * write and hope for the best.
4135                  */
4136                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4137                        "superblock detected");
4138                 clear_buffer_write_io_error(sbh);
4139                 set_buffer_uptodate(sbh);
4140         }
4141         /*
4142          * If the file system is mounted read-only, don't update the
4143          * superblock write time.  This avoids updating the superblock
4144          * write time when we are mounting the root file system
4145          * read/only but we need to replay the journal; at that point,
4146          * for people who are east of GMT and who make their clock
4147          * tick in localtime for Windows bug-for-bug compatibility,
4148          * the clock is set in the future, and this will cause e2fsck
4149          * to complain and force a full file system check.
4150          */
4151         if (!(sb->s_flags & MS_RDONLY))
4152                 es->s_wtime = cpu_to_le32(get_seconds());
4153         if (sb->s_bdev->bd_part)
4154                 es->s_kbytes_written =
4155                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4156                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4157                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4158         else
4159                 es->s_kbytes_written =
4160                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4161         ext4_free_blocks_count_set(es,
4162                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4163                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4164         es->s_free_inodes_count =
4165                 cpu_to_le32(percpu_counter_sum_positive(
4166                                 &EXT4_SB(sb)->s_freeinodes_counter));
4167         sb->s_dirt = 0;
4168         BUFFER_TRACE(sbh, "marking dirty");
4169         mark_buffer_dirty(sbh);
4170         if (sync) {
4171                 error = sync_dirty_buffer(sbh);
4172                 if (error)
4173                         return error;
4174
4175                 error = buffer_write_io_error(sbh);
4176                 if (error) {
4177                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4178                                "superblock");
4179                         clear_buffer_write_io_error(sbh);
4180                         set_buffer_uptodate(sbh);
4181                 }
4182         }
4183         return error;
4184 }
4185
4186 /*
4187  * Have we just finished recovery?  If so, and if we are mounting (or
4188  * remounting) the filesystem readonly, then we will end up with a
4189  * consistent fs on disk.  Record that fact.
4190  */
4191 static void ext4_mark_recovery_complete(struct super_block *sb,
4192                                         struct ext4_super_block *es)
4193 {
4194         journal_t *journal = EXT4_SB(sb)->s_journal;
4195
4196         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4197                 BUG_ON(journal != NULL);
4198                 return;
4199         }
4200         jbd2_journal_lock_updates(journal);
4201         if (jbd2_journal_flush(journal) < 0)
4202                 goto out;
4203
4204         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4205             sb->s_flags & MS_RDONLY) {
4206                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4207                 ext4_commit_super(sb, 1);
4208         }
4209
4210 out:
4211         jbd2_journal_unlock_updates(journal);
4212 }
4213
4214 /*
4215  * If we are mounting (or read-write remounting) a filesystem whose journal
4216  * has recorded an error from a previous lifetime, move that error to the
4217  * main filesystem now.
4218  */
4219 static void ext4_clear_journal_err(struct super_block *sb,
4220                                    struct ext4_super_block *es)
4221 {
4222         journal_t *journal;
4223         int j_errno;
4224         const char *errstr;
4225
4226         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4227
4228         journal = EXT4_SB(sb)->s_journal;
4229
4230         /*
4231          * Now check for any error status which may have been recorded in the
4232          * journal by a prior ext4_error() or ext4_abort()
4233          */
4234
4235         j_errno = jbd2_journal_errno(journal);
4236         if (j_errno) {
4237                 char nbuf[16];
4238
4239                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4240                 ext4_warning(sb, "Filesystem error recorded "
4241                              "from previous mount: %s", errstr);
4242                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4243
4244                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4245                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4246                 ext4_commit_super(sb, 1);
4247
4248                 jbd2_journal_clear_err(journal);
4249         }
4250 }
4251
4252 /*
4253  * Force the running and committing transactions to commit,
4254  * and wait on the commit.
4255  */
4256 int ext4_force_commit(struct super_block *sb)
4257 {
4258         journal_t *journal;
4259         int ret = 0;
4260
4261         if (sb->s_flags & MS_RDONLY)
4262                 return 0;
4263
4264         journal = EXT4_SB(sb)->s_journal;
4265         if (journal) {
4266                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4267                 ret = ext4_journal_force_commit(journal);
4268         }
4269
4270         return ret;
4271 }
4272
4273 static void ext4_write_super(struct super_block *sb)
4274 {
4275         lock_super(sb);
4276         ext4_commit_super(sb, 1);
4277         unlock_super(sb);
4278 }
4279
4280 static int ext4_sync_fs(struct super_block *sb, int wait)
4281 {
4282         int ret = 0;
4283         tid_t target;
4284         struct ext4_sb_info *sbi = EXT4_SB(sb);
4285
4286         trace_ext4_sync_fs(sb, wait);
4287         flush_workqueue(sbi->dio_unwritten_wq);
4288         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4289                 if (wait)
4290                         jbd2_log_wait_commit(sbi->s_journal, target);
4291         }
4292         return ret;
4293 }
4294
4295 /*
4296  * LVM calls this function before a (read-only) snapshot is created.  This
4297  * gives us a chance to flush the journal completely and mark the fs clean.
4298  *
4299  * Note that only this function cannot bring a filesystem to be in a clean
4300  * state independently, because ext4 prevents a new handle from being started
4301  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4302  * the upper layer.
4303  */
4304 static int ext4_freeze(struct super_block *sb)
4305 {
4306         int error = 0;
4307         journal_t *journal;
4308
4309         if (sb->s_flags & MS_RDONLY)
4310                 return 0;
4311
4312         journal = EXT4_SB(sb)->s_journal;
4313
4314         /* Now we set up the journal barrier. */
4315         jbd2_journal_lock_updates(journal);
4316
4317         /*
4318          * Don't clear the needs_recovery flag if we failed to flush
4319          * the journal.
4320          */
4321         error = jbd2_journal_flush(journal);
4322         if (error < 0)
4323                 goto out;
4324
4325         /* Journal blocked and flushed, clear needs_recovery flag. */
4326         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4327         error = ext4_commit_super(sb, 1);
4328 out:
4329         /* we rely on s_frozen to stop further updates */
4330         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4331         return error;
4332 }
4333
4334 /*
4335  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4336  * flag here, even though the filesystem is not technically dirty yet.
4337  */
4338 static int ext4_unfreeze(struct super_block *sb)
4339 {
4340         if (sb->s_flags & MS_RDONLY)
4341                 return 0;
4342
4343         lock_super(sb);
4344         /* Reset the needs_recovery flag before the fs is unlocked. */
4345         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4346         ext4_commit_super(sb, 1);
4347         unlock_super(sb);
4348         return 0;
4349 }
4350
4351 /*
4352  * Structure to save mount options for ext4_remount's benefit
4353  */
4354 struct ext4_mount_options {
4355         unsigned long s_mount_opt;
4356         unsigned long s_mount_opt2;
4357         uid_t s_resuid;
4358         gid_t s_resgid;
4359         unsigned long s_commit_interval;
4360         u32 s_min_batch_time, s_max_batch_time;
4361 #ifdef CONFIG_QUOTA
4362         int s_jquota_fmt;
4363         char *s_qf_names[MAXQUOTAS];
4364 #endif
4365 };
4366
4367 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4368 {
4369         struct ext4_super_block *es;
4370         struct ext4_sb_info *sbi = EXT4_SB(sb);
4371         unsigned long old_sb_flags;
4372         struct ext4_mount_options old_opts;
4373         int enable_quota = 0;
4374         ext4_group_t g;
4375         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4376         int err = 0;
4377 #ifdef CONFIG_QUOTA
4378         int i;
4379 #endif
4380         char *orig_data = kstrdup(data, GFP_KERNEL);
4381
4382         /* Store the original options */
4383         lock_super(sb);
4384         old_sb_flags = sb->s_flags;
4385         old_opts.s_mount_opt = sbi->s_mount_opt;
4386         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4387         old_opts.s_resuid = sbi->s_resuid;
4388         old_opts.s_resgid = sbi->s_resgid;
4389         old_opts.s_commit_interval = sbi->s_commit_interval;
4390         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4391         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4392 #ifdef CONFIG_QUOTA
4393         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4394         for (i = 0; i < MAXQUOTAS; i++)
4395                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4396 #endif
4397         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4398                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4399
4400         /*
4401          * Allow the "check" option to be passed as a remount option.
4402          */
4403         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4404                 err = -EINVAL;
4405                 goto restore_opts;
4406         }
4407
4408         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4409                 ext4_abort(sb, "Abort forced by user");
4410
4411         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4412                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4413
4414         es = sbi->s_es;
4415
4416         if (sbi->s_journal) {
4417                 ext4_init_journal_params(sb, sbi->s_journal);
4418                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4419         }
4420
4421         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4422                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4423                         err = -EROFS;
4424                         goto restore_opts;
4425                 }
4426
4427                 if (*flags & MS_RDONLY) {
4428                         err = dquot_suspend(sb, -1);
4429                         if (err < 0)
4430                                 goto restore_opts;
4431
4432                         /*
4433                          * First of all, the unconditional stuff we have to do
4434                          * to disable replay of the journal when we next remount
4435                          */
4436                         sb->s_flags |= MS_RDONLY;
4437
4438                         /*
4439                          * OK, test if we are remounting a valid rw partition
4440                          * readonly, and if so set the rdonly flag and then
4441                          * mark the partition as valid again.
4442                          */
4443                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4444                             (sbi->s_mount_state & EXT4_VALID_FS))
4445                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4446
4447                         if (sbi->s_journal)
4448                                 ext4_mark_recovery_complete(sb, es);
4449                 } else {
4450                         /* Make sure we can mount this feature set readwrite */
4451                         if (!ext4_feature_set_ok(sb, 0)) {
4452                                 err = -EROFS;
4453                                 goto restore_opts;
4454                         }
4455                         /*
4456                          * Make sure the group descriptor checksums
4457                          * are sane.  If they aren't, refuse to remount r/w.
4458                          */
4459                         for (g = 0; g < sbi->s_groups_count; g++) {
4460                                 struct ext4_group_desc *gdp =
4461                                         ext4_get_group_desc(sb, g, NULL);
4462
4463                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4464                                         ext4_msg(sb, KERN_ERR,
4465                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4466                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4467                                                le16_to_cpu(gdp->bg_checksum));
4468                                         err = -EINVAL;
4469                                         goto restore_opts;
4470                                 }
4471                         }
4472
4473                         /*
4474                          * If we have an unprocessed orphan list hanging
4475                          * around from a previously readonly bdev mount,
4476                          * require a full umount/remount for now.
4477                          */
4478                         if (es->s_last_orphan) {
4479                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4480                                        "remount RDWR because of unprocessed "
4481                                        "orphan inode list.  Please "
4482                                        "umount/remount instead");
4483                                 err = -EINVAL;
4484                                 goto restore_opts;
4485                         }
4486
4487                         /*
4488                          * Mounting a RDONLY partition read-write, so reread
4489                          * and store the current valid flag.  (It may have
4490                          * been changed by e2fsck since we originally mounted
4491                          * the partition.)
4492                          */
4493                         if (sbi->s_journal)
4494                                 ext4_clear_journal_err(sb, es);
4495                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4496                         if (!ext4_setup_super(sb, es, 0))
4497                                 sb->s_flags &= ~MS_RDONLY;
4498                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4499                                                      EXT4_FEATURE_INCOMPAT_MMP))
4500                                 if (ext4_multi_mount_protect(sb,
4501                                                 le64_to_cpu(es->s_mmp_block))) {
4502                                         err = -EROFS;
4503                                         goto restore_opts;
4504                                 }
4505                         enable_quota = 1;
4506                 }
4507         }
4508
4509         /*
4510          * Reinitialize lazy itable initialization thread based on
4511          * current settings
4512          */
4513         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4514                 ext4_unregister_li_request(sb);
4515         else {
4516                 ext4_group_t first_not_zeroed;
4517                 first_not_zeroed = ext4_has_uninit_itable(sb);
4518                 ext4_register_li_request(sb, first_not_zeroed);
4519         }
4520
4521         ext4_setup_system_zone(sb);
4522         if (sbi->s_journal == NULL)
4523                 ext4_commit_super(sb, 1);
4524
4525 #ifdef CONFIG_QUOTA
4526         /* Release old quota file names */
4527         for (i = 0; i < MAXQUOTAS; i++)
4528                 if (old_opts.s_qf_names[i] &&
4529                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4530                         kfree(old_opts.s_qf_names[i]);
4531 #endif
4532         unlock_super(sb);
4533         if (enable_quota)
4534                 dquot_resume(sb, -1);
4535
4536         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4537         kfree(orig_data);
4538         return 0;
4539
4540 restore_opts:
4541         sb->s_flags = old_sb_flags;
4542         sbi->s_mount_opt = old_opts.s_mount_opt;
4543         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4544         sbi->s_resuid = old_opts.s_resuid;
4545         sbi->s_resgid = old_opts.s_resgid;
4546         sbi->s_commit_interval = old_opts.s_commit_interval;
4547         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4548         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4549 #ifdef CONFIG_QUOTA
4550         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4551         for (i = 0; i < MAXQUOTAS; i++) {
4552                 if (sbi->s_qf_names[i] &&
4553                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4554                         kfree(sbi->s_qf_names[i]);
4555                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4556         }
4557 #endif
4558         unlock_super(sb);
4559         kfree(orig_data);
4560         return err;
4561 }
4562
4563 /*
4564  * Note: calculating the overhead so we can be compatible with
4565  * historical BSD practice is quite difficult in the face of
4566  * clusters/bigalloc.  This is because multiple metadata blocks from
4567  * different block group can end up in the same allocation cluster.
4568  * Calculating the exact overhead in the face of clustered allocation
4569  * requires either O(all block bitmaps) in memory or O(number of block
4570  * groups**2) in time.  We will still calculate the superblock for
4571  * older file systems --- and if we come across with a bigalloc file
4572  * system with zero in s_overhead_clusters the estimate will be close to
4573  * correct especially for very large cluster sizes --- but for newer
4574  * file systems, it's better to calculate this figure once at mkfs
4575  * time, and store it in the superblock.  If the superblock value is
4576  * present (even for non-bigalloc file systems), we will use it.
4577  */
4578 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4579 {
4580         struct super_block *sb = dentry->d_sb;
4581         struct ext4_sb_info *sbi = EXT4_SB(sb);
4582         struct ext4_super_block *es = sbi->s_es;
4583         struct ext4_group_desc *gdp;
4584         u64 fsid;
4585         s64 bfree;
4586
4587         if (test_opt(sb, MINIX_DF)) {
4588                 sbi->s_overhead_last = 0;
4589         } else if (es->s_overhead_clusters) {
4590                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4591         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4592                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4593                 ext4_fsblk_t overhead = 0;
4594
4595                 /*
4596                  * Compute the overhead (FS structures).  This is constant
4597                  * for a given filesystem unless the number of block groups
4598                  * changes so we cache the previous value until it does.
4599                  */
4600
4601                 /*
4602                  * All of the blocks before first_data_block are
4603                  * overhead
4604                  */
4605                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4606
4607                 /*
4608                  * Add the overhead found in each block group
4609                  */
4610                 for (i = 0; i < ngroups; i++) {
4611                         gdp = ext4_get_group_desc(sb, i, NULL);
4612                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4613                         cond_resched();
4614                 }
4615                 sbi->s_overhead_last = overhead;
4616                 smp_wmb();
4617                 sbi->s_blocks_last = ext4_blocks_count(es);
4618         }
4619
4620         buf->f_type = EXT4_SUPER_MAGIC;
4621         buf->f_bsize = sb->s_blocksize;
4622         buf->f_blocks = (ext4_blocks_count(es) -
4623                          EXT4_C2B(sbi, sbi->s_overhead_last));
4624         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4625                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4626         /* prevent underflow in case that few free space is available */
4627         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4628         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4629         if (buf->f_bfree < ext4_r_blocks_count(es))
4630                 buf->f_bavail = 0;
4631         buf->f_files = le32_to_cpu(es->s_inodes_count);
4632         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4633         buf->f_namelen = EXT4_NAME_LEN;
4634         fsid = le64_to_cpup((void *)es->s_uuid) ^
4635                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4636         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4637         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4638
4639         return 0;
4640 }
4641
4642 /* Helper function for writing quotas on sync - we need to start transaction
4643  * before quota file is locked for write. Otherwise the are possible deadlocks:
4644  * Process 1                         Process 2
4645  * ext4_create()                     quota_sync()
4646  *   jbd2_journal_start()                  write_dquot()
4647  *   dquot_initialize()                         down(dqio_mutex)
4648  *     down(dqio_mutex)                    jbd2_journal_start()
4649  *
4650  */
4651
4652 #ifdef CONFIG_QUOTA
4653
4654 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4655 {
4656         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4657 }
4658
4659 static int ext4_write_dquot(struct dquot *dquot)
4660 {
4661         int ret, err;
4662         handle_t *handle;
4663         struct inode *inode;
4664
4665         inode = dquot_to_inode(dquot);
4666         handle = ext4_journal_start(inode,
4667                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4668         if (IS_ERR(handle))
4669                 return PTR_ERR(handle);
4670         ret = dquot_commit(dquot);
4671         err = ext4_journal_stop(handle);
4672         if (!ret)
4673                 ret = err;
4674         return ret;
4675 }
4676
4677 static int ext4_acquire_dquot(struct dquot *dquot)
4678 {
4679         int ret, err;
4680         handle_t *handle;
4681
4682         handle = ext4_journal_start(dquot_to_inode(dquot),
4683                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4684         if (IS_ERR(handle))
4685                 return PTR_ERR(handle);
4686         ret = dquot_acquire(dquot);
4687         err = ext4_journal_stop(handle);
4688         if (!ret)
4689                 ret = err;
4690         return ret;
4691 }
4692
4693 static int ext4_release_dquot(struct dquot *dquot)
4694 {
4695         int ret, err;
4696         handle_t *handle;
4697
4698         handle = ext4_journal_start(dquot_to_inode(dquot),
4699                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4700         if (IS_ERR(handle)) {
4701                 /* Release dquot anyway to avoid endless cycle in dqput() */
4702                 dquot_release(dquot);
4703                 return PTR_ERR(handle);
4704         }
4705         ret = dquot_release(dquot);
4706         err = ext4_journal_stop(handle);
4707         if (!ret)
4708                 ret = err;
4709         return ret;
4710 }
4711
4712 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4713 {
4714         /* Are we journaling quotas? */
4715         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4716             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4717                 dquot_mark_dquot_dirty(dquot);
4718                 return ext4_write_dquot(dquot);
4719         } else {
4720                 return dquot_mark_dquot_dirty(dquot);
4721         }
4722 }
4723
4724 static int ext4_write_info(struct super_block *sb, int type)
4725 {
4726         int ret, err;
4727         handle_t *handle;
4728
4729         /* Data block + inode block */
4730         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4731         if (IS_ERR(handle))
4732                 return PTR_ERR(handle);
4733         ret = dquot_commit_info(sb, type);
4734         err = ext4_journal_stop(handle);
4735         if (!ret)
4736                 ret = err;
4737         return ret;
4738 }
4739
4740 /*
4741  * Turn on quotas during mount time - we need to find
4742  * the quota file and such...
4743  */
4744 static int ext4_quota_on_mount(struct super_block *sb, int type)
4745 {
4746         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4747                                         EXT4_SB(sb)->s_jquota_fmt, type);
4748 }
4749
4750 /*
4751  * Standard function to be called on quota_on
4752  */
4753 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4754                          struct path *path)
4755 {
4756         int err;
4757
4758         if (!test_opt(sb, QUOTA))
4759                 return -EINVAL;
4760
4761         /* Quotafile not on the same filesystem? */
4762         if (path->dentry->d_sb != sb)
4763                 return -EXDEV;
4764         /* Journaling quota? */
4765         if (EXT4_SB(sb)->s_qf_names[type]) {
4766                 /* Quotafile not in fs root? */
4767                 if (path->dentry->d_parent != sb->s_root)
4768                         ext4_msg(sb, KERN_WARNING,
4769                                 "Quota file not on filesystem root. "
4770                                 "Journaled quota will not work");
4771         }
4772
4773         /*
4774          * When we journal data on quota file, we have to flush journal to see
4775          * all updates to the file when we bypass pagecache...
4776          */
4777         if (EXT4_SB(sb)->s_journal &&
4778             ext4_should_journal_data(path->dentry->d_inode)) {
4779                 /*
4780                  * We don't need to lock updates but journal_flush() could
4781                  * otherwise be livelocked...
4782                  */
4783                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4784                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4785                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4786                 if (err)
4787                         return err;
4788         }
4789
4790         return dquot_quota_on(sb, type, format_id, path);
4791 }
4792
4793 static int ext4_quota_off(struct super_block *sb, int type)
4794 {
4795         struct inode *inode = sb_dqopt(sb)->files[type];
4796         handle_t *handle;
4797
4798         /* Force all delayed allocation blocks to be allocated.
4799          * Caller already holds s_umount sem */
4800         if (test_opt(sb, DELALLOC))
4801                 sync_filesystem(sb);
4802
4803         if (!inode)
4804                 goto out;
4805
4806         /* Update modification times of quota files when userspace can
4807          * start looking at them */
4808         handle = ext4_journal_start(inode, 1);
4809         if (IS_ERR(handle))
4810                 goto out;
4811         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4812         ext4_mark_inode_dirty(handle, inode);
4813         ext4_journal_stop(handle);
4814
4815 out:
4816         return dquot_quota_off(sb, type);
4817 }
4818
4819 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4820  * acquiring the locks... As quota files are never truncated and quota code
4821  * itself serializes the operations (and no one else should touch the files)
4822  * we don't have to be afraid of races */
4823 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4824                                size_t len, loff_t off)
4825 {
4826         struct inode *inode = sb_dqopt(sb)->files[type];
4827         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4828         int err = 0;
4829         int offset = off & (sb->s_blocksize - 1);
4830         int tocopy;
4831         size_t toread;
4832         struct buffer_head *bh;
4833         loff_t i_size = i_size_read(inode);
4834
4835         if (off > i_size)
4836                 return 0;
4837         if (off+len > i_size)
4838                 len = i_size-off;
4839         toread = len;
4840         while (toread > 0) {
4841                 tocopy = sb->s_blocksize - offset < toread ?
4842                                 sb->s_blocksize - offset : toread;
4843                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4844                 if (err)
4845                         return err;
4846                 if (!bh)        /* A hole? */
4847                         memset(data, 0, tocopy);
4848                 else
4849                         memcpy(data, bh->b_data+offset, tocopy);
4850                 brelse(bh);
4851                 offset = 0;
4852                 toread -= tocopy;
4853                 data += tocopy;
4854                 blk++;
4855         }
4856         return len;
4857 }
4858
4859 /* Write to quotafile (we know the transaction is already started and has
4860  * enough credits) */
4861 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4862                                 const char *data, size_t len, loff_t off)
4863 {
4864         struct inode *inode = sb_dqopt(sb)->files[type];
4865         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4866         int err = 0;
4867         int offset = off & (sb->s_blocksize - 1);
4868         struct buffer_head *bh;
4869         handle_t *handle = journal_current_handle();
4870
4871         if (EXT4_SB(sb)->s_journal && !handle) {
4872                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4873                         " cancelled because transaction is not started",
4874                         (unsigned long long)off, (unsigned long long)len);
4875                 return -EIO;
4876         }
4877         /*
4878          * Since we account only one data block in transaction credits,
4879          * then it is impossible to cross a block boundary.
4880          */
4881         if (sb->s_blocksize - offset < len) {
4882                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4883                         " cancelled because not block aligned",
4884                         (unsigned long long)off, (unsigned long long)len);
4885                 return -EIO;
4886         }
4887
4888         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4889         bh = ext4_bread(handle, inode, blk, 1, &err);
4890         if (!bh)
4891                 goto out;
4892         err = ext4_journal_get_write_access(handle, bh);
4893         if (err) {
4894                 brelse(bh);
4895                 goto out;
4896         }
4897         lock_buffer(bh);
4898         memcpy(bh->b_data+offset, data, len);
4899         flush_dcache_page(bh->b_page);
4900         unlock_buffer(bh);
4901         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4902         brelse(bh);
4903 out:
4904         if (err) {
4905                 mutex_unlock(&inode->i_mutex);
4906                 return err;
4907         }
4908         if (inode->i_size < off + len) {
4909                 i_size_write(inode, off + len);
4910                 EXT4_I(inode)->i_disksize = inode->i_size;
4911                 ext4_mark_inode_dirty(handle, inode);
4912         }
4913         mutex_unlock(&inode->i_mutex);
4914         return len;
4915 }
4916
4917 #endif
4918
4919 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4920                        const char *dev_name, void *data)
4921 {
4922         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4923 }
4924
4925 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4926 static inline void register_as_ext2(void)
4927 {
4928         int err = register_filesystem(&ext2_fs_type);
4929         if (err)
4930                 printk(KERN_WARNING
4931                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4932 }
4933
4934 static inline void unregister_as_ext2(void)
4935 {
4936         unregister_filesystem(&ext2_fs_type);
4937 }
4938
4939 static inline int ext2_feature_set_ok(struct super_block *sb)
4940 {
4941         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4942                 return 0;
4943         if (sb->s_flags & MS_RDONLY)
4944                 return 1;
4945         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4946                 return 0;
4947         return 1;
4948 }
4949 MODULE_ALIAS("ext2");
4950 #else
4951 static inline void register_as_ext2(void) { }
4952 static inline void unregister_as_ext2(void) { }
4953 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4954 #endif
4955
4956 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4957 static inline void register_as_ext3(void)
4958 {
4959         int err = register_filesystem(&ext3_fs_type);
4960         if (err)
4961                 printk(KERN_WARNING
4962                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4963 }
4964
4965 static inline void unregister_as_ext3(void)
4966 {
4967         unregister_filesystem(&ext3_fs_type);
4968 }
4969
4970 static inline int ext3_feature_set_ok(struct super_block *sb)
4971 {
4972         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4973                 return 0;
4974         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4975                 return 0;
4976         if (sb->s_flags & MS_RDONLY)
4977                 return 1;
4978         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4979                 return 0;
4980         return 1;
4981 }
4982 MODULE_ALIAS("ext3");
4983 #else
4984 static inline void register_as_ext3(void) { }
4985 static inline void unregister_as_ext3(void) { }
4986 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4987 #endif
4988
4989 static struct file_system_type ext4_fs_type = {
4990         .owner          = THIS_MODULE,
4991         .name           = "ext4",
4992         .mount          = ext4_mount,
4993         .kill_sb        = kill_block_super,
4994         .fs_flags       = FS_REQUIRES_DEV,
4995 };
4996
4997 static int __init ext4_init_feat_adverts(void)
4998 {
4999         struct ext4_features *ef;
5000         int ret = -ENOMEM;
5001
5002         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5003         if (!ef)
5004                 goto out;
5005
5006         ef->f_kobj.kset = ext4_kset;
5007         init_completion(&ef->f_kobj_unregister);
5008         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5009                                    "features");
5010         if (ret) {
5011                 kfree(ef);
5012                 goto out;
5013         }
5014
5015         ext4_feat = ef;
5016         ret = 0;
5017 out:
5018         return ret;
5019 }
5020
5021 static void ext4_exit_feat_adverts(void)
5022 {
5023         kobject_put(&ext4_feat->f_kobj);
5024         wait_for_completion(&ext4_feat->f_kobj_unregister);
5025         kfree(ext4_feat);
5026 }
5027
5028 /* Shared across all ext4 file systems */
5029 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5030 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5031
5032 static int __init ext4_init_fs(void)
5033 {
5034         int i, err;
5035
5036         ext4_check_flag_values();
5037
5038         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5039                 mutex_init(&ext4__aio_mutex[i]);
5040                 init_waitqueue_head(&ext4__ioend_wq[i]);
5041         }
5042
5043         err = ext4_init_pageio();
5044         if (err)
5045                 return err;
5046         err = ext4_init_system_zone();
5047         if (err)
5048                 goto out6;
5049         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5050         if (!ext4_kset)
5051                 goto out5;
5052         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5053
5054         err = ext4_init_feat_adverts();
5055         if (err)
5056                 goto out4;
5057
5058         err = ext4_init_mballoc();
5059         if (err)
5060                 goto out3;
5061
5062         err = ext4_init_xattr();
5063         if (err)
5064                 goto out2;
5065         err = init_inodecache();
5066         if (err)
5067                 goto out1;
5068         register_as_ext3();
5069         register_as_ext2();
5070         err = register_filesystem(&ext4_fs_type);
5071         if (err)
5072                 goto out;
5073
5074         ext4_li_info = NULL;
5075         mutex_init(&ext4_li_mtx);
5076         return 0;
5077 out:
5078         unregister_as_ext2();
5079         unregister_as_ext3();
5080         destroy_inodecache();
5081 out1:
5082         ext4_exit_xattr();
5083 out2:
5084         ext4_exit_mballoc();
5085 out3:
5086         ext4_exit_feat_adverts();
5087 out4:
5088         if (ext4_proc_root)
5089                 remove_proc_entry("fs/ext4", NULL);
5090         kset_unregister(ext4_kset);
5091 out5:
5092         ext4_exit_system_zone();
5093 out6:
5094         ext4_exit_pageio();
5095         return err;
5096 }
5097
5098 static void __exit ext4_exit_fs(void)
5099 {
5100         ext4_destroy_lazyinit_thread();
5101         unregister_as_ext2();
5102         unregister_as_ext3();
5103         unregister_filesystem(&ext4_fs_type);
5104         destroy_inodecache();
5105         ext4_exit_xattr();
5106         ext4_exit_mballoc();
5107         ext4_exit_feat_adverts();
5108         remove_proc_entry("fs/ext4", NULL);
5109         kset_unregister(ext4_kset);
5110         ext4_exit_system_zone();
5111         ext4_exit_pageio();
5112 }
5113
5114 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5115 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5116 MODULE_LICENSE("GPL");
5117 module_init(ext4_init_fs)
5118 module_exit(ext4_exit_fs)