fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static const char *ext4_decode_error(struct super_block *sb, int errno,
  73                                      char nbuf[16]);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static void ext4_write_super(struct super_block *sb);
  78 static int ext4_freeze(struct super_block *sb);
  79 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  80                        const char *dev_name, void *data);
  81 static inline int ext2_feature_set_ok(struct super_block *sb);
  82 static inline int ext3_feature_set_ok(struct super_block *sb);
  83 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  84 static void ext4_destroy_lazyinit_thread(void);
  85 static void ext4_unregister_li_request(struct super_block *sb);
  86 static void ext4_clear_request_list(void);
  87
  88 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  89 static struct file_system_type ext2_fs_type = {
  90         .owner          = THIS_MODULE,
  91         .name           = "ext2",
  92         .mount          = ext4_mount,
  93         .kill_sb        = kill_block_super,
  94         .fs_flags       = FS_REQUIRES_DEV,
  95 };
  96 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  97 #else
  98 #define IS_EXT2_SB(sb) (0)
  99 #endif
 100
 101
 102 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 103 static struct file_system_type ext3_fs_type = {
 104         .owner          = THIS_MODULE,
 105         .name           = "ext3",
 106         .mount          = ext4_mount,
 107         .kill_sb        = kill_block_super,
 108         .fs_flags       = FS_REQUIRES_DEV,
 109 };
 110 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 111 #else
 112 #define IS_EXT3_SB(sb) (0)
 113 #endif
 114
 115 void *ext4_kvmalloc(size_t size, gfp_t flags)
 116 {
 117         void *ret;
 118
 119         ret = kmalloc(size, flags);
 120         if (!ret)
 121                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 122         return ret;
 123 }
 124
 125 void *ext4_kvzalloc(size_t size, gfp_t flags)
 126 {
 127         void *ret;
 128
 129         ret = kzalloc(size, flags);
 130         if (!ret)
 131                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 132         return ret;
 133 }
 134
 135 void ext4_kvfree(void *ptr)
 136 {
 137         if (is_vmalloc_addr(ptr))
 138                 vfree(ptr);
 139         else
 140                 kfree(ptr);
 141
 142 }
 143
 144 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 145                                struct ext4_group_desc *bg)
 146 {
 147         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 148                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 149                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 150 }
 151
 152 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 153                                struct ext4_group_desc *bg)
 154 {
 155         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 156                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 157                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 158 }
 159
 160 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 161                               struct ext4_group_desc *bg)
 162 {
 163         return le32_to_cpu(bg->bg_inode_table_lo) |
 164                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 165                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 166 }
 167
 168 __u32 ext4_free_group_clusters(struct super_block *sb,
 169                                struct ext4_group_desc *bg)
 170 {
 171         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 172                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 173                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 174 }
 175
 176 __u32 ext4_free_inodes_count(struct super_block *sb,
 177                               struct ext4_group_desc *bg)
 178 {
 179         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 180                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 181                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 182 }
 183
 184 __u32 ext4_used_dirs_count(struct super_block *sb,
 185                               struct ext4_group_desc *bg)
 186 {
 187         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 188                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 189                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 190 }
 191
 192 __u32 ext4_itable_unused_count(struct super_block *sb,
 193                               struct ext4_group_desc *bg)
 194 {
 195         return le16_to_cpu(bg->bg_itable_unused_lo) |
 196                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 197                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 198 }
 199
 200 void ext4_block_bitmap_set(struct super_block *sb,
 201                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 202 {
 203         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 204         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 205                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 206 }
 207
 208 void ext4_inode_bitmap_set(struct super_block *sb,
 209                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 210 {
 211         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 212         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 213                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 214 }
 215
 216 void ext4_inode_table_set(struct super_block *sb,
 217                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 218 {
 219         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 220         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 221                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 222 }
 223
 224 void ext4_free_group_clusters_set(struct super_block *sb,
 225                                   struct ext4_group_desc *bg, __u32 count)
 226 {
 227         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 228         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 229                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 230 }
 231
 232 void ext4_free_inodes_set(struct super_block *sb,
 233                           struct ext4_group_desc *bg, __u32 count)
 234 {
 235         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 236         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 237                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 238 }
 239
 240 void ext4_used_dirs_set(struct super_block *sb,
 241                           struct ext4_group_desc *bg, __u32 count)
 242 {
 243         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 244         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 245                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 246 }
 247
 248 void ext4_itable_unused_set(struct super_block *sb,
 249                           struct ext4_group_desc *bg, __u32 count)
 250 {
 251         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 252         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 253                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 254 }
 255
 256
 257 /* Just increment the non-pointer handle value */
 258 static handle_t *ext4_get_nojournal(void)
 259 {
 260         handle_t *handle = current->journal_info;
 261         unsigned long ref_cnt = (unsigned long)handle;
 262
 263         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 264
 265         ref_cnt++;
 266         handle = (handle_t *)ref_cnt;
 267
 268         current->journal_info = handle;
 269         return handle;
 270 }
 271
 272
 273 /* Decrement the non-pointer handle value */
 274 static void ext4_put_nojournal(handle_t *handle)
 275 {
 276         unsigned long ref_cnt = (unsigned long)handle;
 277
 278         BUG_ON(ref_cnt == 0);
 279
 280         ref_cnt--;
 281         handle = (handle_t *)ref_cnt;
 282
 283         current->journal_info = handle;
 284 }
 285
 286 /*
 287  * Wrappers for jbd2_journal_start/end.
 288  *
 289  * The only special thing we need to do here is to make sure that all
 290  * journal_end calls result in the superblock being marked dirty, so
 291  * that sync() will call the filesystem's write_super callback if
 292  * appropriate.
 293  *
 294  * To avoid j_barrier hold in userspace when a user calls freeze(),
 295  * ext4 prevents a new handle from being started by s_frozen, which
 296  * is in an upper layer.
 297  */
 298 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 299 {
 300         journal_t *journal;
 301         handle_t  *handle;
 302
 303         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 304         if (sb->s_flags & MS_RDONLY)
 305                 return ERR_PTR(-EROFS);
 306
 307         journal = EXT4_SB(sb)->s_journal;
 308         handle = ext4_journal_current_handle();
 309
 310         /*
 311          * If a handle has been started, it should be allowed to
 312          * finish, otherwise deadlock could happen between freeze
 313          * and others(e.g. truncate) due to the restart of the
 314          * journal handle if the filesystem is forzen and active
 315          * handles are not stopped.
 316          */
 317         if (!handle)
 318                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 319
 320         if (!journal)
 321                 return ext4_get_nojournal();
 322         /*
 323          * Special case here: if the journal has aborted behind our
 324          * backs (eg. EIO in the commit thread), then we still need to
 325          * take the FS itself readonly cleanly.
 326          */
 327         if (is_journal_aborted(journal)) {
 328                 ext4_abort(sb, "Detected aborted journal");
 329                 return ERR_PTR(-EROFS);
 330         }
 331         return jbd2_journal_start(journal, nblocks);
 332 }
 333
 334 /*
 335  * The only special thing we need to do here is to make sure that all
 336  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 337  * that sync() will call the filesystem's write_super callback if
 338  * appropriate.
 339  */
 340 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 341 {
 342         struct super_block *sb;
 343         int err;
 344         int rc;
 345
 346         if (!ext4_handle_valid(handle)) {
 347                 ext4_put_nojournal(handle);
 348                 return 0;
 349         }
 350         sb = handle->h_transaction->t_journal->j_private;
 351         err = handle->h_err;
 352         rc = jbd2_journal_stop(handle);
 353
 354         if (!err)
 355                 err = rc;
 356         if (err)
 357                 __ext4_std_error(sb, where, line, err);
 358         return err;
 359 }
 360
 361 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 362                                const char *err_fn, struct buffer_head *bh,
 363                                handle_t *handle, int err)
 364 {
 365         char nbuf[16];
 366         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 367
 368         BUG_ON(!ext4_handle_valid(handle));
 369
 370         if (bh)
 371                 BUFFER_TRACE(bh, "abort");
 372
 373         if (!handle->h_err)
 374                 handle->h_err = err;
 375
 376         if (is_handle_aborted(handle))
 377                 return;
 378
 379         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 380                caller, line, errstr, err_fn);
 381
 382         jbd2_journal_abort_handle(handle);
 383 }
 384
 385 static void __save_error_info(struct super_block *sb, const char *func,
 386                             unsigned int line)
 387 {
 388         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 389
 390         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 391         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 392         es->s_last_error_time = cpu_to_le32(get_seconds());
 393         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 394         es->s_last_error_line = cpu_to_le32(line);
 395         if (!es->s_first_error_time) {
 396                 es->s_first_error_time = es->s_last_error_time;
 397                 strncpy(es->s_first_error_func, func,
 398                         sizeof(es->s_first_error_func));
 399                 es->s_first_error_line = cpu_to_le32(line);
 400                 es->s_first_error_ino = es->s_last_error_ino;
 401                 es->s_first_error_block = es->s_last_error_block;
 402         }
 403         /*
 404          * Start the daily error reporting function if it hasn't been
 405          * started already
 406          */
 407         if (!es->s_error_count)
 408                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 409         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 410 }
 411
 412 static void save_error_info(struct super_block *sb, const char *func,
 413                             unsigned int line)
 414 {
 415         __save_error_info(sb, func, line);
 416         ext4_commit_super(sb, 1);
 417 }
 418
 419 /*
 420  * The del_gendisk() function uninitializes the disk-specific data
 421  * structures, including the bdi structure, without telling anyone
 422  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 423  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 424  * This is a kludge to prevent these oops until we can put in a proper
 425  * hook in del_gendisk() to inform the VFS and file system layers.
 426  */
 427 static int block_device_ejected(struct super_block *sb)
 428 {
 429         struct inode *bd_inode = sb->s_bdev->bd_inode;
 430         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 431
 432         return bdi->dev == NULL;
 433 }
 434
 435 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 436 {
 437         struct super_block              *sb = journal->j_private;
 438         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 439         int                             error = is_journal_aborted(journal);
 440         struct ext4_journal_cb_entry    *jce, *tmp;
 441
 442         spin_lock(&sbi->s_md_lock);
 443         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 444                 list_del_init(&jce->jce_list);
 445                 spin_unlock(&sbi->s_md_lock);
 446                 jce->jce_func(sb, jce, error);
 447                 spin_lock(&sbi->s_md_lock);
 448         }
 449         spin_unlock(&sbi->s_md_lock);
 450 }
 451
 452 /* Deal with the reporting of failure conditions on a filesystem such as
 453  * inconsistencies detected or read IO failures.
 454  *
 455  * On ext2, we can store the error state of the filesystem in the
 456  * superblock.  That is not possible on ext4, because we may have other
 457  * write ordering constraints on the superblock which prevent us from
 458  * writing it out straight away; and given that the journal is about to
 459  * be aborted, we can't rely on the current, or future, transactions to
 460  * write out the superblock safely.
 461  *
 462  * We'll just use the jbd2_journal_abort() error code to record an error in
 463  * the journal instead.  On recovery, the journal will complain about
 464  * that error until we've noted it down and cleared it.
 465  */
 466
 467 static void ext4_handle_error(struct super_block *sb)
 468 {
 469         if (sb->s_flags & MS_RDONLY)
 470                 return;
 471
 472         if (!test_opt(sb, ERRORS_CONT)) {
 473                 journal_t *journal = EXT4_SB(sb)->s_journal;
 474
 475                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 476                 if (journal)
 477                         jbd2_journal_abort(journal, -EIO);
 478         }
 479         if (test_opt(sb, ERRORS_RO)) {
 480                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 481                 sb->s_flags |= MS_RDONLY;
 482         }
 483         if (test_opt(sb, ERRORS_PANIC))
 484                 panic("EXT4-fs (device %s): panic forced after error\n",
 485                         sb->s_id);
 486 }
 487
 488 void __ext4_error(struct super_block *sb, const char *function,
 489                   unsigned int line, const char *fmt, ...)
 490 {
 491         struct va_format vaf;
 492         va_list args;
 493
 494         va_start(args, fmt);
 495         vaf.fmt = fmt;
 496         vaf.va = &args;
 497         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 498                sb->s_id, function, line, current->comm, &vaf);
 499         va_end(args);
 500
 501         ext4_handle_error(sb);
 502 }
 503
 504 void ext4_error_inode(struct inode *inode, const char *function,
 505                       unsigned int line, ext4_fsblk_t block,
 506                       const char *fmt, ...)
 507 {
 508         va_list args;
 509         struct va_format vaf;
 510         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 511
 512         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 513         es->s_last_error_block = cpu_to_le64(block);
 514         save_error_info(inode->i_sb, function, line);
 515         va_start(args, fmt);
 516         vaf.fmt = fmt;
 517         vaf.va = &args;
 518         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 519                inode->i_sb->s_id, function, line, inode->i_ino);
 520         if (block)
 521                 printk(KERN_CONT "block %llu: ", block);
 522         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 523         va_end(args);
 524
 525         ext4_handle_error(inode->i_sb);
 526 }
 527
 528 void ext4_error_file(struct file *file, const char *function,
 529                      unsigned int line, ext4_fsblk_t block,
 530                      const char *fmt, ...)
 531 {
 532         va_list args;
 533         struct va_format vaf;
 534         struct ext4_super_block *es;
 535         struct inode *inode = file->f_dentry->d_inode;
 536         char pathname[80], *path;
 537
 538         es = EXT4_SB(inode->i_sb)->s_es;
 539         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 540         save_error_info(inode->i_sb, function, line);
 541         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 542         if (IS_ERR(path))
 543                 path = "(unknown)";
 544         printk(KERN_CRIT
 545                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 546                inode->i_sb->s_id, function, line, inode->i_ino);
 547         if (block)
 548                 printk(KERN_CONT "block %llu: ", block);
 549         va_start(args, fmt);
 550         vaf.fmt = fmt;
 551         vaf.va = &args;
 552         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 553         va_end(args);
 554
 555         ext4_handle_error(inode->i_sb);
 556 }
 557
 558 static const char *ext4_decode_error(struct super_block *sb, int errno,
 559                                      char nbuf[16])
 560 {
 561         char *errstr = NULL;
 562
 563         switch (errno) {
 564         case -EIO:
 565                 errstr = "IO failure";
 566                 break;
 567         case -ENOMEM:
 568                 errstr = "Out of memory";
 569                 break;
 570         case -EROFS:
 571                 if (!sb || (EXT4_SB(sb)->s_journal &&
 572                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 573                         errstr = "Journal has aborted";
 574                 else
 575                         errstr = "Readonly filesystem";
 576                 break;
 577         default:
 578                 /* If the caller passed in an extra buffer for unknown
 579                  * errors, textualise them now.  Else we just return
 580                  * NULL. */
 581                 if (nbuf) {
 582                         /* Check for truncated error codes... */
 583                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 584                                 errstr = nbuf;
 585                 }
 586                 break;
 587         }
 588
 589         return errstr;
 590 }
 591
 592 /* __ext4_std_error decodes expected errors from journaling functions
 593  * automatically and invokes the appropriate error response.  */
 594
 595 void __ext4_std_error(struct super_block *sb, const char *function,
 596                       unsigned int line, int errno)
 597 {
 598         char nbuf[16];
 599         const char *errstr;
 600
 601         /* Special case: if the error is EROFS, and we're not already
 602          * inside a transaction, then there's really no point in logging
 603          * an error. */
 604         if (errno == -EROFS && journal_current_handle() == NULL &&
 605             (sb->s_flags & MS_RDONLY))
 606                 return;
 607
 608         errstr = ext4_decode_error(sb, errno, nbuf);
 609         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 610                sb->s_id, function, line, errstr);
 611         save_error_info(sb, function, line);
 612
 613         ext4_handle_error(sb);
 614 }
 615
 616 /*
 617  * ext4_abort is a much stronger failure handler than ext4_error.  The
 618  * abort function may be used to deal with unrecoverable failures such
 619  * as journal IO errors or ENOMEM at a critical moment in log management.
 620  *
 621  * We unconditionally force the filesystem into an ABORT|READONLY state,
 622  * unless the error response on the fs has been set to panic in which
 623  * case we take the easy way out and panic immediately.
 624  */
 625
 626 void __ext4_abort(struct super_block *sb, const char *function,
 627                 unsigned int line, const char *fmt, ...)
 628 {
 629         va_list args;
 630
 631         save_error_info(sb, function, line);
 632         va_start(args, fmt);
 633         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 634                function, line);
 635         vprintk(fmt, args);
 636         printk("\n");
 637         va_end(args);
 638
 639         if ((sb->s_flags & MS_RDONLY) == 0) {
 640                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 641                 sb->s_flags |= MS_RDONLY;
 642                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 643                 if (EXT4_SB(sb)->s_journal)
 644                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 645                 save_error_info(sb, function, line);
 646         }
 647         if (test_opt(sb, ERRORS_PANIC))
 648                 panic("EXT4-fs panic from previous error\n");
 649 }
 650
 651 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 652 {
 653         struct va_format vaf;
 654         va_list args;
 655
 656         va_start(args, fmt);
 657         vaf.fmt = fmt;
 658         vaf.va = &args;
 659         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 660         va_end(args);
 661 }
 662
 663 void __ext4_warning(struct super_block *sb, const char *function,
 664                     unsigned int line, const char *fmt, ...)
 665 {
 666         struct va_format vaf;
 667         va_list args;
 668
 669         va_start(args, fmt);
 670         vaf.fmt = fmt;
 671         vaf.va = &args;
 672         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 673                sb->s_id, function, line, &vaf);
 674         va_end(args);
 675 }
 676
 677 void __ext4_grp_locked_error(const char *function, unsigned int line,
 678                              struct super_block *sb, ext4_group_t grp,
 679                              unsigned long ino, ext4_fsblk_t block,
 680                              const char *fmt, ...)
 681 __releases(bitlock)
 682 __acquires(bitlock)
 683 {
 684         struct va_format vaf;
 685         va_list args;
 686         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 687
 688         es->s_last_error_ino = cpu_to_le32(ino);
 689         es->s_last_error_block = cpu_to_le64(block);
 690         __save_error_info(sb, function, line);
 691
 692         va_start(args, fmt);
 693
 694         vaf.fmt = fmt;
 695         vaf.va = &args;
 696         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 697                sb->s_id, function, line, grp);
 698         if (ino)
 699                 printk(KERN_CONT "inode %lu: ", ino);
 700         if (block)
 701                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 702         printk(KERN_CONT "%pV\n", &vaf);
 703         va_end(args);
 704
 705         if (test_opt(sb, ERRORS_CONT)) {
 706                 ext4_commit_super(sb, 0);
 707                 return;
 708         }
 709
 710         ext4_unlock_group(sb, grp);
 711         ext4_handle_error(sb);
 712         /*
 713          * We only get here in the ERRORS_RO case; relocking the group
 714          * may be dangerous, but nothing bad will happen since the
 715          * filesystem will have already been marked read/only and the
 716          * journal has been aborted.  We return 1 as a hint to callers
 717          * who might what to use the return value from
 718          * ext4_grp_locked_error() to distinguish between the
 719          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 720          * aggressively from the ext4 function in question, with a
 721          * more appropriate error code.
 722          */
 723         ext4_lock_group(sb, grp);
 724         return;
 725 }
 726
 727 void ext4_update_dynamic_rev(struct super_block *sb)
 728 {
 729         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 730
 731         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 732                 return;
 733
 734         ext4_warning(sb,
 735                      "updating to rev %d because of new feature flag, "
 736                      "running e2fsck is recommended",
 737                      EXT4_DYNAMIC_REV);
 738
 739         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 740         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 741         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 742         /* leave es->s_feature_*compat flags alone */
 743         /* es->s_uuid will be set by e2fsck if empty */
 744
 745         /*
 746          * The rest of the superblock fields should be zero, and if not it
 747          * means they are likely already in use, so leave them alone.  We
 748          * can leave it up to e2fsck to clean up any inconsistencies there.
 749          */
 750 }
 751
 752 /*
 753  * Open the external journal device
 754  */
 755 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 756 {
 757         struct block_device *bdev;
 758         char b[BDEVNAME_SIZE];
 759
 760         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 761         if (IS_ERR(bdev))
 762                 goto fail;
 763         return bdev;
 764
 765 fail:
 766         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 767                         __bdevname(dev, b), PTR_ERR(bdev));
 768         return NULL;
 769 }
 770
 771 /*
 772  * Release the journal device
 773  */
 774 static int ext4_blkdev_put(struct block_device *bdev)
 775 {
 776         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 777 }
 778
 779 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 780 {
 781         struct block_device *bdev;
 782         int ret = -ENODEV;
 783
 784         bdev = sbi->journal_bdev;
 785         if (bdev) {
 786                 ret = ext4_blkdev_put(bdev);
 787                 sbi->journal_bdev = NULL;
 788         }
 789         return ret;
 790 }
 791
 792 static inline struct inode *orphan_list_entry(struct list_head *l)
 793 {
 794         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 795 }
 796
 797 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 798 {
 799         struct list_head *l;
 800
 801         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 802                  le32_to_cpu(sbi->s_es->s_last_orphan));
 803
 804         printk(KERN_ERR "sb_info orphan list:\n");
 805         list_for_each(l, &sbi->s_orphan) {
 806                 struct inode *inode = orphan_list_entry(l);
 807                 printk(KERN_ERR "  "
 808                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 809                        inode->i_sb->s_id, inode->i_ino, inode,
 810                        inode->i_mode, inode->i_nlink,
 811                        NEXT_ORPHAN(inode));
 812         }
 813 }
 814
 815 static void ext4_put_super(struct super_block *sb)
 816 {
 817         struct ext4_sb_info *sbi = EXT4_SB(sb);
 818         struct ext4_super_block *es = sbi->s_es;
 819         int i, err;
 820
 821         ext4_unregister_li_request(sb);
 822         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 823
 824         flush_workqueue(sbi->dio_unwritten_wq);
 825         destroy_workqueue(sbi->dio_unwritten_wq);
 826
 827         lock_super(sb);
 828         if (sb->s_dirt)
 829                 ext4_commit_super(sb, 1);
 830
 831         if (sbi->s_journal) {
 832                 err = jbd2_journal_destroy(sbi->s_journal);
 833                 sbi->s_journal = NULL;
 834                 if (err < 0)
 835                         ext4_abort(sb, "Couldn't clean up the journal");
 836         }
 837
 838         del_timer(&sbi->s_err_report);
 839         ext4_release_system_zone(sb);
 840         ext4_mb_release(sb);
 841         ext4_ext_release(sb);
 842         ext4_xattr_put_super(sb);
 843
 844         if (!(sb->s_flags & MS_RDONLY)) {
 845                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 846                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 847                 ext4_commit_super(sb, 1);
 848         }
 849         if (sbi->s_proc) {
 850                 remove_proc_entry("options", sbi->s_proc);
 851                 remove_proc_entry(sb->s_id, ext4_proc_root);
 852         }
 853         kobject_del(&sbi->s_kobj);
 854
 855         for (i = 0; i < sbi->s_gdb_count; i++)
 856                 brelse(sbi->s_group_desc[i]);
 857         ext4_kvfree(sbi->s_group_desc);
 858         ext4_kvfree(sbi->s_flex_groups);
 859         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 860         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 861         percpu_counter_destroy(&sbi->s_dirs_counter);
 862         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 863         brelse(sbi->s_sbh);
 864 #ifdef CONFIG_QUOTA
 865         for (i = 0; i < MAXQUOTAS; i++)
 866                 kfree(sbi->s_qf_names[i]);
 867 #endif
 868
 869         /* Debugging code just in case the in-memory inode orphan list
 870          * isn't empty.  The on-disk one can be non-empty if we've
 871          * detected an error and taken the fs readonly, but the
 872          * in-memory list had better be clean by this point. */
 873         if (!list_empty(&sbi->s_orphan))
 874                 dump_orphan_list(sb, sbi);
 875         J_ASSERT(list_empty(&sbi->s_orphan));
 876
 877         invalidate_bdev(sb->s_bdev);
 878         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 879                 /*
 880                  * Invalidate the journal device's buffers.  We don't want them
 881                  * floating about in memory - the physical journal device may
 882                  * hotswapped, and it breaks the `ro-after' testing code.
 883                  */
 884                 sync_blockdev(sbi->journal_bdev);
 885                 invalidate_bdev(sbi->journal_bdev);
 886                 ext4_blkdev_remove(sbi);
 887         }
 888         if (sbi->s_mmp_tsk)
 889                 kthread_stop(sbi->s_mmp_tsk);
 890         sb->s_fs_info = NULL;
 891         /*
 892          * Now that we are completely done shutting down the
 893          * superblock, we need to actually destroy the kobject.
 894          */
 895         unlock_super(sb);
 896         kobject_put(&sbi->s_kobj);
 897         wait_for_completion(&sbi->s_kobj_unregister);
 898         kfree(sbi->s_blockgroup_lock);
 899         kfree(sbi);
 900 }
 901
 902 static struct kmem_cache *ext4_inode_cachep;
 903
 904 /*
 905  * Called inside transaction, so use GFP_NOFS
 906  */
 907 static struct inode *ext4_alloc_inode(struct super_block *sb)
 908 {
 909         struct ext4_inode_info *ei;
 910
 911         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 912         if (!ei)
 913                 return NULL;
 914
 915         ei->vfs_inode.i_version = 1;
 916         ei->vfs_inode.i_data.writeback_index = 0;
 917         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 918         INIT_LIST_HEAD(&ei->i_prealloc_list);
 919         spin_lock_init(&ei->i_prealloc_lock);
 920         ei->i_reserved_data_blocks = 0;
 921         ei->i_reserved_meta_blocks = 0;
 922         ei->i_allocated_meta_blocks = 0;
 923         ei->i_da_metadata_calc_len = 0;
 924         spin_lock_init(&(ei->i_block_reservation_lock));
 925 #ifdef CONFIG_QUOTA
 926         ei->i_reserved_quota = 0;
 927 #endif
 928         ei->jinode = NULL;
 929         INIT_LIST_HEAD(&ei->i_completed_io_list);
 930         spin_lock_init(&ei->i_completed_io_lock);
 931         ei->cur_aio_dio = NULL;
 932         ei->i_sync_tid = 0;
 933         ei->i_datasync_tid = 0;
 934         atomic_set(&ei->i_ioend_count, 0);
 935         atomic_set(&ei->i_aiodio_unwritten, 0);
 936
 937         return &ei->vfs_inode;
 938 }
 939
 940 static int ext4_drop_inode(struct inode *inode)
 941 {
 942         int drop = generic_drop_inode(inode);
 943
 944         trace_ext4_drop_inode(inode, drop);
 945         return drop;
 946 }
 947
 948 static void ext4_i_callback(struct rcu_head *head)
 949 {
 950         struct inode *inode = container_of(head, struct inode, i_rcu);
 951         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 952 }
 953
 954 static void ext4_destroy_inode(struct inode *inode)
 955 {
 956         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 957                 ext4_msg(inode->i_sb, KERN_ERR,
 958                          "Inode %lu (%p): orphan list check failed!",
 959                          inode->i_ino, EXT4_I(inode));
 960                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 961                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 962                                 true);
 963                 dump_stack();
 964         }
 965         call_rcu(&inode->i_rcu, ext4_i_callback);
 966 }
 967
 968 static void init_once(void *foo)
 969 {
 970         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 971
 972         INIT_LIST_HEAD(&ei->i_orphan);
 973 #ifdef CONFIG_EXT4_FS_XATTR
 974         init_rwsem(&ei->xattr_sem);
 975 #endif
 976         init_rwsem(&ei->i_data_sem);
 977         inode_init_once(&ei->vfs_inode);
 978 }
 979
 980 static int init_inodecache(void)
 981 {
 982         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 983                                              sizeof(struct ext4_inode_info),
 984                                              0, (SLAB_RECLAIM_ACCOUNT|
 985                                                 SLAB_MEM_SPREAD),
 986                                              init_once);
 987         if (ext4_inode_cachep == NULL)
 988                 return -ENOMEM;
 989         return 0;
 990 }
 991
 992 static void destroy_inodecache(void)
 993 {
 994         kmem_cache_destroy(ext4_inode_cachep);
 995 }
 996
 997 void ext4_clear_inode(struct inode *inode)
 998 {
 999         invalidate_inode_buffers(inode);
1000         end_writeback(inode);
1001         dquot_drop(inode);
1002         ext4_discard_preallocations(inode);
1003         if (EXT4_I(inode)->jinode) {
1004                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1005                                                EXT4_I(inode)->jinode);
1006                 jbd2_free_inode(EXT4_I(inode)->jinode);
1007                 EXT4_I(inode)->jinode = NULL;
1008         }
1009 }
1010
1011 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1012                                         u64 ino, u32 generation)
1013 {
1014         struct inode *inode;
1015
1016         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1017                 return ERR_PTR(-ESTALE);
1018         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1019                 return ERR_PTR(-ESTALE);
1020
1021         /* iget isn't really right if the inode is currently unallocated!!
1022          *
1023          * ext4_read_inode will return a bad_inode if the inode had been
1024          * deleted, so we should be safe.
1025          *
1026          * Currently we don't know the generation for parent directory, so
1027          * a generation of 0 means "accept any"
1028          */
1029         inode = ext4_iget(sb, ino);
1030         if (IS_ERR(inode))
1031                 return ERR_CAST(inode);
1032         if (generation && inode->i_generation != generation) {
1033                 iput(inode);
1034                 return ERR_PTR(-ESTALE);
1035         }
1036
1037         return inode;
1038 }
1039
1040 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1041                                         int fh_len, int fh_type)
1042 {
1043         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1044                                     ext4_nfs_get_inode);
1045 }
1046
1047 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1048                                         int fh_len, int fh_type)
1049 {
1050         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1051                                     ext4_nfs_get_inode);
1052 }
1053
1054 /*
1055  * Try to release metadata pages (indirect blocks, directories) which are
1056  * mapped via the block device.  Since these pages could have journal heads
1057  * which would prevent try_to_free_buffers() from freeing them, we must use
1058  * jbd2 layer's try_to_free_buffers() function to release them.
1059  */
1060 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1061                                  gfp_t wait)
1062 {
1063         journal_t *journal = EXT4_SB(sb)->s_journal;
1064
1065         WARN_ON(PageChecked(page));
1066         if (!page_has_buffers(page))
1067                 return 0;
1068         if (journal)
1069                 return jbd2_journal_try_to_free_buffers(journal, page,
1070                                                         wait & ~__GFP_WAIT);
1071         return try_to_free_buffers(page);
1072 }
1073
1074 #ifdef CONFIG_QUOTA
1075 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1076 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1077
1078 static int ext4_write_dquot(struct dquot *dquot);
1079 static int ext4_acquire_dquot(struct dquot *dquot);
1080 static int ext4_release_dquot(struct dquot *dquot);
1081 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1082 static int ext4_write_info(struct super_block *sb, int type);
1083 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1084                          struct path *path);
1085 static int ext4_quota_off(struct super_block *sb, int type);
1086 static int ext4_quota_on_mount(struct super_block *sb, int type);
1087 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1088                                size_t len, loff_t off);
1089 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1090                                 const char *data, size_t len, loff_t off);
1091
1092 static const struct dquot_operations ext4_quota_operations = {
1093         .get_reserved_space = ext4_get_reserved_space,
1094         .write_dquot    = ext4_write_dquot,
1095         .acquire_dquot  = ext4_acquire_dquot,
1096         .release_dquot  = ext4_release_dquot,
1097         .mark_dirty     = ext4_mark_dquot_dirty,
1098         .write_info     = ext4_write_info,
1099         .alloc_dquot    = dquot_alloc,
1100         .destroy_dquot  = dquot_destroy,
1101 };
1102
1103 static const struct quotactl_ops ext4_qctl_operations = {
1104         .quota_on       = ext4_quota_on,
1105         .quota_off      = ext4_quota_off,
1106         .quota_sync     = dquot_quota_sync,
1107         .get_info       = dquot_get_dqinfo,
1108         .set_info       = dquot_set_dqinfo,
1109         .get_dqblk      = dquot_get_dqblk,
1110         .set_dqblk      = dquot_set_dqblk
1111 };
1112 #endif
1113
1114 static const struct super_operations ext4_sops = {
1115         .alloc_inode    = ext4_alloc_inode,
1116         .destroy_inode  = ext4_destroy_inode,
1117         .write_inode    = ext4_write_inode,
1118         .dirty_inode    = ext4_dirty_inode,
1119         .drop_inode     = ext4_drop_inode,
1120         .evict_inode    = ext4_evict_inode,
1121         .put_super      = ext4_put_super,
1122         .sync_fs        = ext4_sync_fs,
1123         .freeze_fs      = ext4_freeze,
1124         .unfreeze_fs    = ext4_unfreeze,
1125         .statfs         = ext4_statfs,
1126         .remount_fs     = ext4_remount,
1127         .show_options   = ext4_show_options,
1128 #ifdef CONFIG_QUOTA
1129         .quota_read     = ext4_quota_read,
1130         .quota_write    = ext4_quota_write,
1131 #endif
1132         .bdev_try_to_free_page = bdev_try_to_free_page,
1133 };
1134
1135 static const struct super_operations ext4_nojournal_sops = {
1136         .alloc_inode    = ext4_alloc_inode,
1137         .destroy_inode  = ext4_destroy_inode,
1138         .write_inode    = ext4_write_inode,
1139         .dirty_inode    = ext4_dirty_inode,
1140         .drop_inode     = ext4_drop_inode,
1141         .evict_inode    = ext4_evict_inode,
1142         .write_super    = ext4_write_super,
1143         .put_super      = ext4_put_super,
1144         .statfs         = ext4_statfs,
1145         .remount_fs     = ext4_remount,
1146         .show_options   = ext4_show_options,
1147 #ifdef CONFIG_QUOTA
1148         .quota_read     = ext4_quota_read,
1149         .quota_write    = ext4_quota_write,
1150 #endif
1151         .bdev_try_to_free_page = bdev_try_to_free_page,
1152 };
1153
1154 static const struct export_operations ext4_export_ops = {
1155         .fh_to_dentry = ext4_fh_to_dentry,
1156         .fh_to_parent = ext4_fh_to_parent,
1157         .get_parent = ext4_get_parent,
1158 };
1159
1160 enum {
1161         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1162         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1163         Opt_nouid32, Opt_debug, Opt_removed,
1164         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1165         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1166         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1167         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1168         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1169         Opt_data_err_abort, Opt_data_err_ignore,
1170         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1171         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1172         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1173         Opt_usrquota, Opt_grpquota, Opt_i_version,
1174         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1175         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1176         Opt_inode_readahead_blks, Opt_journal_ioprio,
1177         Opt_dioread_nolock, Opt_dioread_lock,
1178         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1179 };
1180
1181 static const match_table_t tokens = {
1182         {Opt_bsd_df, "bsddf"},
1183         {Opt_minix_df, "minixdf"},
1184         {Opt_grpid, "grpid"},
1185         {Opt_grpid, "bsdgroups"},
1186         {Opt_nogrpid, "nogrpid"},
1187         {Opt_nogrpid, "sysvgroups"},
1188         {Opt_resgid, "resgid=%u"},
1189         {Opt_resuid, "resuid=%u"},
1190         {Opt_sb, "sb=%u"},
1191         {Opt_err_cont, "errors=continue"},
1192         {Opt_err_panic, "errors=panic"},
1193         {Opt_err_ro, "errors=remount-ro"},
1194         {Opt_nouid32, "nouid32"},
1195         {Opt_debug, "debug"},
1196         {Opt_removed, "oldalloc"},
1197         {Opt_removed, "orlov"},
1198         {Opt_user_xattr, "user_xattr"},
1199         {Opt_nouser_xattr, "nouser_xattr"},
1200         {Opt_acl, "acl"},
1201         {Opt_noacl, "noacl"},
1202         {Opt_noload, "norecovery"},
1203         {Opt_noload, "noload"},
1204         {Opt_removed, "nobh"},
1205         {Opt_removed, "bh"},
1206         {Opt_commit, "commit=%u"},
1207         {Opt_min_batch_time, "min_batch_time=%u"},
1208         {Opt_max_batch_time, "max_batch_time=%u"},
1209         {Opt_journal_dev, "journal_dev=%u"},
1210         {Opt_journal_checksum, "journal_checksum"},
1211         {Opt_journal_async_commit, "journal_async_commit"},
1212         {Opt_abort, "abort"},
1213         {Opt_data_journal, "data=journal"},
1214         {Opt_data_ordered, "data=ordered"},
1215         {Opt_data_writeback, "data=writeback"},
1216         {Opt_data_err_abort, "data_err=abort"},
1217         {Opt_data_err_ignore, "data_err=ignore"},
1218         {Opt_offusrjquota, "usrjquota="},
1219         {Opt_usrjquota, "usrjquota=%s"},
1220         {Opt_offgrpjquota, "grpjquota="},
1221         {Opt_grpjquota, "grpjquota=%s"},
1222         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1223         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1224         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1225         {Opt_grpquota, "grpquota"},
1226         {Opt_noquota, "noquota"},
1227         {Opt_quota, "quota"},
1228         {Opt_usrquota, "usrquota"},
1229         {Opt_barrier, "barrier=%u"},
1230         {Opt_barrier, "barrier"},
1231         {Opt_nobarrier, "nobarrier"},
1232         {Opt_i_version, "i_version"},
1233         {Opt_stripe, "stripe=%u"},
1234         {Opt_delalloc, "delalloc"},
1235         {Opt_nodelalloc, "nodelalloc"},
1236         {Opt_mblk_io_submit, "mblk_io_submit"},
1237         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1238         {Opt_block_validity, "block_validity"},
1239         {Opt_noblock_validity, "noblock_validity"},
1240         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1241         {Opt_journal_ioprio, "journal_ioprio=%u"},
1242         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1243         {Opt_auto_da_alloc, "auto_da_alloc"},
1244         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1245         {Opt_dioread_nolock, "dioread_nolock"},
1246         {Opt_dioread_lock, "dioread_lock"},
1247         {Opt_discard, "discard"},
1248         {Opt_nodiscard, "nodiscard"},
1249         {Opt_init_itable, "init_itable=%u"},
1250         {Opt_init_itable, "init_itable"},
1251         {Opt_noinit_itable, "noinit_itable"},
1252         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1253         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1254         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1255         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1256         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1257         {Opt_err, NULL},
1258 };
1259
1260 static ext4_fsblk_t get_sb_block(void **data)
1261 {
1262         ext4_fsblk_t    sb_block;
1263         char            *options = (char *) *data;
1264
1265         if (!options || strncmp(options, "sb=", 3) != 0)
1266                 return 1;       /* Default location */
1267
1268         options += 3;
1269         /* TODO: use simple_strtoll with >32bit ext4 */
1270         sb_block = simple_strtoul(options, &options, 0);
1271         if (*options && *options != ',') {
1272                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1273                        (char *) *data);
1274                 return 1;
1275         }
1276         if (*options == ',')
1277                 options++;
1278         *data = (void *) options;
1279
1280         return sb_block;
1281 }
1282
1283 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1284 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1285         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1286
1287 #ifdef CONFIG_QUOTA
1288 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1289 {
1290         struct ext4_sb_info *sbi = EXT4_SB(sb);
1291         char *qname;
1292
1293         if (sb_any_quota_loaded(sb) &&
1294                 !sbi->s_qf_names[qtype]) {
1295                 ext4_msg(sb, KERN_ERR,
1296                         "Cannot change journaled "
1297                         "quota options when quota turned on");
1298                 return 0;
1299         }
1300         qname = match_strdup(args);
1301         if (!qname) {
1302                 ext4_msg(sb, KERN_ERR,
1303                         "Not enough memory for storing quotafile name");
1304                 return 0;
1305         }
1306         if (sbi->s_qf_names[qtype] &&
1307                 strcmp(sbi->s_qf_names[qtype], qname)) {
1308                 ext4_msg(sb, KERN_ERR,
1309                         "%s quota file already specified", QTYPE2NAME(qtype));
1310                 kfree(qname);
1311                 return 0;
1312         }
1313         sbi->s_qf_names[qtype] = qname;
1314         if (strchr(sbi->s_qf_names[qtype], '/')) {
1315                 ext4_msg(sb, KERN_ERR,
1316                         "quotafile must be on filesystem root");
1317                 kfree(sbi->s_qf_names[qtype]);
1318                 sbi->s_qf_names[qtype] = NULL;
1319                 return 0;
1320         }
1321         set_opt(sb, QUOTA);
1322         return 1;
1323 }
1324
1325 static int clear_qf_name(struct super_block *sb, int qtype)
1326 {
1327
1328         struct ext4_sb_info *sbi = EXT4_SB(sb);
1329
1330         if (sb_any_quota_loaded(sb) &&
1331                 sbi->s_qf_names[qtype]) {
1332                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1333                         " when quota turned on");
1334                 return 0;
1335         }
1336         /*
1337          * The space will be released later when all options are confirmed
1338          * to be correct
1339          */
1340         sbi->s_qf_names[qtype] = NULL;
1341         return 1;
1342 }
1343 #endif
1344
1345 #define MOPT_SET        0x0001
1346 #define MOPT_CLEAR      0x0002
1347 #define MOPT_NOSUPPORT  0x0004
1348 #define MOPT_EXPLICIT   0x0008
1349 #define MOPT_CLEAR_ERR  0x0010
1350 #define MOPT_GTE0       0x0020
1351 #ifdef CONFIG_QUOTA
1352 #define MOPT_Q          0
1353 #define MOPT_QFMT       0x0040
1354 #else
1355 #define MOPT_Q          MOPT_NOSUPPORT
1356 #define MOPT_QFMT       MOPT_NOSUPPORT
1357 #endif
1358 #define MOPT_DATAJ      0x0080
1359
1360 static const struct mount_opts {
1361         int     token;
1362         int     mount_opt;
1363         int     flags;
1364 } ext4_mount_opts[] = {
1365         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1366         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1367         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1368         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1369         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1370         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1371         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1372         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1373         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1374         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1375         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1376         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1377         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1378         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1379         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1380         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1381                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1382         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1383         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1384         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1385         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1386         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1387         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1388         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1389         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1390         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1391         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1392         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1393         {Opt_commit, 0, MOPT_GTE0},
1394         {Opt_max_batch_time, 0, MOPT_GTE0},
1395         {Opt_min_batch_time, 0, MOPT_GTE0},
1396         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1397         {Opt_init_itable, 0, MOPT_GTE0},
1398         {Opt_stripe, 0, MOPT_GTE0},
1399         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1400         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1401         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1402 #ifdef CONFIG_EXT4_FS_XATTR
1403         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1404         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1405 #else
1406         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1407         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1408 #endif
1409 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1410         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1411         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1412 #else
1413         {Opt_acl, 0, MOPT_NOSUPPORT},
1414         {Opt_noacl, 0, MOPT_NOSUPPORT},
1415 #endif
1416         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1417         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1418         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1419         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1420                                                         MOPT_SET | MOPT_Q},
1421         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1422                                                         MOPT_SET | MOPT_Q},
1423         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1424                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1425         {Opt_usrjquota, 0, MOPT_Q},
1426         {Opt_grpjquota, 0, MOPT_Q},
1427         {Opt_offusrjquota, 0, MOPT_Q},
1428         {Opt_offgrpjquota, 0, MOPT_Q},
1429         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1430         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1431         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1432         {Opt_err, 0, 0}
1433 };
1434
1435 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1436                             substring_t *args, unsigned long *journal_devnum,
1437                             unsigned int *journal_ioprio, int is_remount)
1438 {
1439         struct ext4_sb_info *sbi = EXT4_SB(sb);
1440         const struct mount_opts *m;
1441         int arg = 0;
1442
1443         if (args->from && match_int(args, &arg))
1444                 return -1;
1445         switch (token) {
1446         case Opt_noacl:
1447         case Opt_nouser_xattr:
1448                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1449                 break;
1450         case Opt_sb:
1451                 return 1;       /* handled by get_sb_block() */
1452         case Opt_removed:
1453                 ext4_msg(sb, KERN_WARNING,
1454                          "Ignoring removed %s option", opt);
1455                 return 1;
1456         case Opt_resuid:
1457                 sbi->s_resuid = arg;
1458                 return 1;
1459         case Opt_resgid:
1460                 sbi->s_resgid = arg;
1461                 return 1;
1462         case Opt_abort:
1463                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1464                 return 1;
1465         case Opt_i_version:
1466                 sb->s_flags |= MS_I_VERSION;
1467                 return 1;
1468         case Opt_journal_dev:
1469                 if (is_remount) {
1470                         ext4_msg(sb, KERN_ERR,
1471                                  "Cannot specify journal on remount");
1472                         return -1;
1473                 }
1474                 *journal_devnum = arg;
1475                 return 1;
1476         case Opt_journal_ioprio:
1477                 if (arg < 0 || arg > 7)
1478                         return -1;
1479                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1480                 return 1;
1481         }
1482
1483         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1484                 if (token != m->token)
1485                         continue;
1486                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1487                         return -1;
1488                 if (m->flags & MOPT_EXPLICIT)
1489                         set_opt2(sb, EXPLICIT_DELALLOC);
1490                 if (m->flags & MOPT_CLEAR_ERR)
1491                         clear_opt(sb, ERRORS_MASK);
1492                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1493                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1494                                  "options when quota turned on");
1495                         return -1;
1496                 }
1497
1498                 if (m->flags & MOPT_NOSUPPORT) {
1499                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1500                 } else if (token == Opt_commit) {
1501                         if (arg == 0)
1502                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1503                         sbi->s_commit_interval = HZ * arg;
1504                 } else if (token == Opt_max_batch_time) {
1505                         if (arg == 0)
1506                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1507                         sbi->s_max_batch_time = arg;
1508                 } else if (token == Opt_min_batch_time) {
1509                         sbi->s_min_batch_time = arg;
1510                 } else if (token == Opt_inode_readahead_blks) {
1511                         if (arg > (1 << 30))
1512                                 return -1;
1513                         if (arg && !is_power_of_2(arg)) {
1514                                 ext4_msg(sb, KERN_ERR,
1515                                          "EXT4-fs: inode_readahead_blks"
1516                                          " must be a power of 2");
1517                                 return -1;
1518                         }
1519                         sbi->s_inode_readahead_blks = arg;
1520                 } else if (token == Opt_init_itable) {
1521                         set_opt(sb, INIT_INODE_TABLE);
1522                         if (!args->from)
1523                                 arg = EXT4_DEF_LI_WAIT_MULT;
1524                         sbi->s_li_wait_mult = arg;
1525                 } else if (token == Opt_stripe) {
1526                         sbi->s_stripe = arg;
1527                 } else if (m->flags & MOPT_DATAJ) {
1528                         if (is_remount) {
1529                                 if (!sbi->s_journal)
1530                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1531                                 else if (test_opt(sb, DATA_FLAGS) !=
1532                                          m->mount_opt) {
1533                                         ext4_msg(sb, KERN_ERR,
1534                                          "Cannot change data mode on remount");
1535                                         return -1;
1536                                 }
1537                         } else {
1538                                 clear_opt(sb, DATA_FLAGS);
1539                                 sbi->s_mount_opt |= m->mount_opt;
1540                         }
1541 #ifdef CONFIG_QUOTA
1542                 } else if (token == Opt_usrjquota) {
1543                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1544                                 return -1;
1545                 } else if (token == Opt_grpjquota) {
1546                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1547                                 return -1;
1548                 } else if (token == Opt_offusrjquota) {
1549                         if (!clear_qf_name(sb, USRQUOTA))
1550                                 return -1;
1551                 } else if (token == Opt_offgrpjquota) {
1552                         if (!clear_qf_name(sb, GRPQUOTA))
1553                                 return -1;
1554                 } else if (m->flags & MOPT_QFMT) {
1555                         if (sb_any_quota_loaded(sb) &&
1556                             sbi->s_jquota_fmt != m->mount_opt) {
1557                                 ext4_msg(sb, KERN_ERR, "Cannot "
1558                                          "change journaled quota options "
1559                                          "when quota turned on");
1560                                 return -1;
1561                         }
1562                         sbi->s_jquota_fmt = m->mount_opt;
1563 #endif
1564                 } else {
1565                         if (!args->from)
1566                                 arg = 1;
1567                         if (m->flags & MOPT_CLEAR)
1568                                 arg = !arg;
1569                         else if (unlikely(!(m->flags & MOPT_SET))) {
1570                                 ext4_msg(sb, KERN_WARNING,
1571                                          "buggy handling of option %s", opt);
1572                                 WARN_ON(1);
1573                                 return -1;
1574                         }
1575                         if (arg != 0)
1576                                 sbi->s_mount_opt |= m->mount_opt;
1577                         else
1578                                 sbi->s_mount_opt &= ~m->mount_opt;
1579                 }
1580                 return 1;
1581         }
1582         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1583                  "or missing value", opt);
1584         return -1;
1585 }
1586
1587 static int parse_options(char *options, struct super_block *sb,
1588                          unsigned long *journal_devnum,
1589                          unsigned int *journal_ioprio,
1590                          int is_remount)
1591 {
1592         struct ext4_sb_info *sbi = EXT4_SB(sb);
1593         char *p;
1594         substring_t args[MAX_OPT_ARGS];
1595         int token;
1596
1597         if (!options)
1598                 return 1;
1599
1600         while ((p = strsep(&options, ",")) != NULL) {
1601                 if (!*p)
1602                         continue;
1603                 /*
1604                  * Initialize args struct so we know whether arg was
1605                  * found; some options take optional arguments.
1606                  */
1607                 args[0].to = args[0].from = 0;
1608                 token = match_token(p, tokens, args);
1609                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1610                                      journal_ioprio, is_remount) < 0)
1611                         return 0;
1612         }
1613 #ifdef CONFIG_QUOTA
1614         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1615                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1616                         clear_opt(sb, USRQUOTA);
1617
1618                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1619                         clear_opt(sb, GRPQUOTA);
1620
1621                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1622                         ext4_msg(sb, KERN_ERR, "old and new quota "
1623                                         "format mixing");
1624                         return 0;
1625                 }
1626
1627                 if (!sbi->s_jquota_fmt) {
1628                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1629                                         "not specified");
1630                         return 0;
1631                 }
1632         } else {
1633                 if (sbi->s_jquota_fmt) {
1634                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1635                                         "specified with no journaling "
1636                                         "enabled");
1637                         return 0;
1638                 }
1639         }
1640 #endif
1641         return 1;
1642 }
1643
1644 static inline void ext4_show_quota_options(struct seq_file *seq,
1645                                            struct super_block *sb)
1646 {
1647 #if defined(CONFIG_QUOTA)
1648         struct ext4_sb_info *sbi = EXT4_SB(sb);
1649
1650         if (sbi->s_jquota_fmt) {
1651                 char *fmtname = "";
1652
1653                 switch (sbi->s_jquota_fmt) {
1654                 case QFMT_VFS_OLD:
1655                         fmtname = "vfsold";
1656                         break;
1657                 case QFMT_VFS_V0:
1658                         fmtname = "vfsv0";
1659                         break;
1660                 case QFMT_VFS_V1:
1661                         fmtname = "vfsv1";
1662                         break;
1663                 }
1664                 seq_printf(seq, ",jqfmt=%s", fmtname);
1665         }
1666
1667         if (sbi->s_qf_names[USRQUOTA])
1668                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1669
1670         if (sbi->s_qf_names[GRPQUOTA])
1671                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1672
1673         if (test_opt(sb, USRQUOTA))
1674                 seq_puts(seq, ",usrquota");
1675
1676         if (test_opt(sb, GRPQUOTA))
1677                 seq_puts(seq, ",grpquota");
1678 #endif
1679 }
1680
1681 static const char *token2str(int token)
1682 {
1683         static const struct match_token *t;
1684
1685         for (t = tokens; t->token != Opt_err; t++)
1686                 if (t->token == token && !strchr(t->pattern, '='))
1687                         break;
1688         return t->pattern;
1689 }
1690
1691 /*
1692  * Show an option if
1693  *  - it's set to a non-default value OR
1694  *  - if the per-sb default is different from the global default
1695  */
1696 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1697                               int nodefs)
1698 {
1699         struct ext4_sb_info *sbi = EXT4_SB(sb);
1700         struct ext4_super_block *es = sbi->s_es;
1701         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1702         const struct mount_opts *m;
1703         char sep = nodefs ? '\n' : ',';
1704
1705 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1706 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1707
1708         if (sbi->s_sb_block != 1)
1709                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1710
1711         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1712                 int want_set = m->flags & MOPT_SET;
1713                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1714                     (m->flags & MOPT_CLEAR_ERR))
1715                         continue;
1716                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1717                         continue; /* skip if same as the default */
1718                 if ((want_set &&
1719                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1720                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1721                         continue; /* select Opt_noFoo vs Opt_Foo */
1722                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1723         }
1724
1725         if (nodefs || sbi->s_resuid != EXT4_DEF_RESUID ||
1726             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1727                 SEQ_OPTS_PRINT("resuid=%u", sbi->s_resuid);
1728         if (nodefs || sbi->s_resgid != EXT4_DEF_RESGID ||
1729             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1730                 SEQ_OPTS_PRINT("resgid=%u", sbi->s_resgid);
1731         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1732         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1733                 SEQ_OPTS_PUTS("errors=remount-ro");
1734         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1735                 SEQ_OPTS_PUTS("errors=continue");
1736         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1737                 SEQ_OPTS_PUTS("errors=panic");
1738         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1739                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1740         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1741                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1742         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1743                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1744         if (sb->s_flags & MS_I_VERSION)
1745                 SEQ_OPTS_PUTS("i_version");
1746         if (nodefs || sbi->s_stripe)
1747                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1748         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1749                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1750                         SEQ_OPTS_PUTS("data=journal");
1751                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1752                         SEQ_OPTS_PUTS("data=ordered");
1753                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1754                         SEQ_OPTS_PUTS("data=writeback");
1755         }
1756         if (nodefs ||
1757             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1758                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1759                                sbi->s_inode_readahead_blks);
1760
1761         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1762                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1763                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1764
1765         ext4_show_quota_options(seq, sb);
1766         return 0;
1767 }
1768
1769 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1770 {
1771         return _ext4_show_options(seq, root->d_sb, 0);
1772 }
1773
1774 static int options_seq_show(struct seq_file *seq, void *offset)
1775 {
1776         struct super_block *sb = seq->private;
1777         int rc;
1778
1779         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1780         rc = _ext4_show_options(seq, sb, 1);
1781         seq_puts(seq, "\n");
1782         return rc;
1783 }
1784
1785 static int options_open_fs(struct inode *inode, struct file *file)
1786 {
1787         return single_open(file, options_seq_show, PDE(inode)->data);
1788 }
1789
1790 static const struct file_operations ext4_seq_options_fops = {
1791         .owner = THIS_MODULE,
1792         .open = options_open_fs,
1793         .read = seq_read,
1794         .llseek = seq_lseek,
1795         .release = single_release,
1796 };
1797
1798 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1799                             int read_only)
1800 {
1801         struct ext4_sb_info *sbi = EXT4_SB(sb);
1802         int res = 0;
1803
1804         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1805                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1806                          "forcing read-only mode");
1807                 res = MS_RDONLY;
1808         }
1809         if (read_only)
1810                 goto done;
1811         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1812                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1813                          "running e2fsck is recommended");
1814         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1815                 ext4_msg(sb, KERN_WARNING,
1816                          "warning: mounting fs with errors, "
1817                          "running e2fsck is recommended");
1818         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1819                  le16_to_cpu(es->s_mnt_count) >=
1820                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1821                 ext4_msg(sb, KERN_WARNING,
1822                          "warning: maximal mount count reached, "
1823                          "running e2fsck is recommended");
1824         else if (le32_to_cpu(es->s_checkinterval) &&
1825                 (le32_to_cpu(es->s_lastcheck) +
1826                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1827                 ext4_msg(sb, KERN_WARNING,
1828                          "warning: checktime reached, "
1829                          "running e2fsck is recommended");
1830         if (!sbi->s_journal)
1831                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1832         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1833                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1834         le16_add_cpu(&es->s_mnt_count, 1);
1835         es->s_mtime = cpu_to_le32(get_seconds());
1836         ext4_update_dynamic_rev(sb);
1837         if (sbi->s_journal)
1838                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1839
1840         ext4_commit_super(sb, 1);
1841 done:
1842         if (test_opt(sb, DEBUG))
1843                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1844                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1845                         sb->s_blocksize,
1846                         sbi->s_groups_count,
1847                         EXT4_BLOCKS_PER_GROUP(sb),
1848                         EXT4_INODES_PER_GROUP(sb),
1849                         sbi->s_mount_opt, sbi->s_mount_opt2);
1850
1851         cleancache_init_fs(sb);
1852         return res;
1853 }
1854
1855 static int ext4_fill_flex_info(struct super_block *sb)
1856 {
1857         struct ext4_sb_info *sbi = EXT4_SB(sb);
1858         struct ext4_group_desc *gdp = NULL;
1859         ext4_group_t flex_group_count;
1860         ext4_group_t flex_group;
1861         unsigned int groups_per_flex = 0;
1862         size_t size;
1863         int i;
1864
1865         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1866         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1867                 sbi->s_log_groups_per_flex = 0;
1868                 return 1;
1869         }
1870         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1871
1872         /* We allocate both existing and potentially added groups */
1873         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1874                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1875                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1876         size = flex_group_count * sizeof(struct flex_groups);
1877         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1878         if (sbi->s_flex_groups == NULL) {
1879                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1880                          flex_group_count);
1881                 goto failed;
1882         }
1883
1884         for (i = 0; i < sbi->s_groups_count; i++) {
1885                 gdp = ext4_get_group_desc(sb, i, NULL);
1886
1887                 flex_group = ext4_flex_group(sbi, i);
1888                 atomic_add(ext4_free_inodes_count(sb, gdp),
1889                            &sbi->s_flex_groups[flex_group].free_inodes);
1890                 atomic_add(ext4_free_group_clusters(sb, gdp),
1891                            &sbi->s_flex_groups[flex_group].free_clusters);
1892                 atomic_add(ext4_used_dirs_count(sb, gdp),
1893                            &sbi->s_flex_groups[flex_group].used_dirs);
1894         }
1895
1896         return 1;
1897 failed:
1898         return 0;
1899 }
1900
1901 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1902                             struct ext4_group_desc *gdp)
1903 {
1904         __u16 crc = 0;
1905
1906         if (sbi->s_es->s_feature_ro_compat &
1907             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1908                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1909                 __le32 le_group = cpu_to_le32(block_group);
1910
1911                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1912                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1913                 crc = crc16(crc, (__u8 *)gdp, offset);
1914                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1915                 /* for checksum of struct ext4_group_desc do the rest...*/
1916                 if ((sbi->s_es->s_feature_incompat &
1917                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1918                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1919                         crc = crc16(crc, (__u8 *)gdp + offset,
1920                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1921                                         offset);
1922         }
1923
1924         return cpu_to_le16(crc);
1925 }
1926
1927 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1928                                 struct ext4_group_desc *gdp)
1929 {
1930         if ((sbi->s_es->s_feature_ro_compat &
1931              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1932             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1933                 return 0;
1934
1935         return 1;
1936 }
1937
1938 /* Called at mount-time, super-block is locked */
1939 static int ext4_check_descriptors(struct super_block *sb,
1940                                   ext4_group_t *first_not_zeroed)
1941 {
1942         struct ext4_sb_info *sbi = EXT4_SB(sb);
1943         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1944         ext4_fsblk_t last_block;
1945         ext4_fsblk_t block_bitmap;
1946         ext4_fsblk_t inode_bitmap;
1947         ext4_fsblk_t inode_table;
1948         int flexbg_flag = 0;
1949         ext4_group_t i, grp = sbi->s_groups_count;
1950
1951         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1952                 flexbg_flag = 1;
1953
1954         ext4_debug("Checking group descriptors");
1955
1956         for (i = 0; i < sbi->s_groups_count; i++) {
1957                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1958
1959                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1960                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1961                 else
1962                         last_block = first_block +
1963                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1964
1965                 if ((grp == sbi->s_groups_count) &&
1966                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
1967                         grp = i;
1968
1969                 block_bitmap = ext4_block_bitmap(sb, gdp);
1970                 if (block_bitmap < first_block || block_bitmap > last_block) {
1971                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1972                                "Block bitmap for group %u not in group "
1973                                "(block %llu)!", i, block_bitmap);
1974                         return 0;
1975                 }
1976                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1977                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1978                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1979                                "Inode bitmap for group %u not in group "
1980                                "(block %llu)!", i, inode_bitmap);
1981                         return 0;
1982                 }
1983                 inode_table = ext4_inode_table(sb, gdp);
1984                 if (inode_table < first_block ||
1985                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1986                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1987                                "Inode table for group %u not in group "
1988                                "(block %llu)!", i, inode_table);
1989                         return 0;
1990                 }
1991                 ext4_lock_group(sb, i);
1992                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1993                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1994                                  "Checksum for group %u failed (%u!=%u)",
1995                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1996                                      gdp)), le16_to_cpu(gdp->bg_checksum));
1997                         if (!(sb->s_flags & MS_RDONLY)) {
1998                                 ext4_unlock_group(sb, i);
1999                                 return 0;
2000                         }
2001                 }
2002                 ext4_unlock_group(sb, i);
2003                 if (!flexbg_flag)
2004                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2005         }
2006         if (NULL != first_not_zeroed)
2007                 *first_not_zeroed = grp;
2008
2009         ext4_free_blocks_count_set(sbi->s_es,
2010                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2011         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2012         return 1;
2013 }
2014
2015 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2016  * the superblock) which were deleted from all directories, but held open by
2017  * a process at the time of a crash.  We walk the list and try to delete these
2018  * inodes at recovery time (only with a read-write filesystem).
2019  *
2020  * In order to keep the orphan inode chain consistent during traversal (in
2021  * case of crash during recovery), we link each inode into the superblock
2022  * orphan list_head and handle it the same way as an inode deletion during
2023  * normal operation (which journals the operations for us).
2024  *
2025  * We only do an iget() and an iput() on each inode, which is very safe if we
2026  * accidentally point at an in-use or already deleted inode.  The worst that
2027  * can happen in this case is that we get a "bit already cleared" message from
2028  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2029  * e2fsck was run on this filesystem, and it must have already done the orphan
2030  * inode cleanup for us, so we can safely abort without any further action.
2031  */
2032 static void ext4_orphan_cleanup(struct super_block *sb,
2033                                 struct ext4_super_block *es)
2034 {
2035         unsigned int s_flags = sb->s_flags;
2036         int nr_orphans = 0, nr_truncates = 0;
2037 #ifdef CONFIG_QUOTA
2038         int i;
2039 #endif
2040         if (!es->s_last_orphan) {
2041                 jbd_debug(4, "no orphan inodes to clean up\n");
2042                 return;
2043         }
2044
2045         if (bdev_read_only(sb->s_bdev)) {
2046                 ext4_msg(sb, KERN_ERR, "write access "
2047                         "unavailable, skipping orphan cleanup");
2048                 return;
2049         }
2050
2051         /* Check if feature set would not allow a r/w mount */
2052         if (!ext4_feature_set_ok(sb, 0)) {
2053                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2054                          "unknown ROCOMPAT features");
2055                 return;
2056         }
2057
2058         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2059                 if (es->s_last_orphan)
2060                         jbd_debug(1, "Errors on filesystem, "
2061                                   "clearing orphan list.\n");
2062                 es->s_last_orphan = 0;
2063                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2064                 return;
2065         }
2066
2067         if (s_flags & MS_RDONLY) {
2068                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2069                 sb->s_flags &= ~MS_RDONLY;
2070         }
2071 #ifdef CONFIG_QUOTA
2072         /* Needed for iput() to work correctly and not trash data */
2073         sb->s_flags |= MS_ACTIVE;
2074         /* Turn on quotas so that they are updated correctly */
2075         for (i = 0; i < MAXQUOTAS; i++) {
2076                 if (EXT4_SB(sb)->s_qf_names[i]) {
2077                         int ret = ext4_quota_on_mount(sb, i);
2078                         if (ret < 0)
2079                                 ext4_msg(sb, KERN_ERR,
2080                                         "Cannot turn on journaled "
2081                                         "quota: error %d", ret);
2082                 }
2083         }
2084 #endif
2085
2086         while (es->s_last_orphan) {
2087                 struct inode *inode;
2088
2089                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2090                 if (IS_ERR(inode)) {
2091                         es->s_last_orphan = 0;
2092                         break;
2093                 }
2094
2095                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2096                 dquot_initialize(inode);
2097                 if (inode->i_nlink) {
2098                         ext4_msg(sb, KERN_DEBUG,
2099                                 "%s: truncating inode %lu to %lld bytes",
2100                                 __func__, inode->i_ino, inode->i_size);
2101                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2102                                   inode->i_ino, inode->i_size);
2103                         ext4_truncate(inode);
2104                         nr_truncates++;
2105                 } else {
2106                         ext4_msg(sb, KERN_DEBUG,
2107                                 "%s: deleting unreferenced inode %lu",
2108                                 __func__, inode->i_ino);
2109                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2110                                   inode->i_ino);
2111                         nr_orphans++;
2112                 }
2113                 iput(inode);  /* The delete magic happens here! */
2114         }
2115
2116 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2117
2118         if (nr_orphans)
2119                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2120                        PLURAL(nr_orphans));
2121         if (nr_truncates)
2122                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2123                        PLURAL(nr_truncates));
2124 #ifdef CONFIG_QUOTA
2125         /* Turn quotas off */
2126         for (i = 0; i < MAXQUOTAS; i++) {
2127                 if (sb_dqopt(sb)->files[i])
2128                         dquot_quota_off(sb, i);
2129         }
2130 #endif
2131         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2132 }
2133
2134 /*
2135  * Maximal extent format file size.
2136  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2137  * extent format containers, within a sector_t, and within i_blocks
2138  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2139  * so that won't be a limiting factor.
2140  *
2141  * However there is other limiting factor. We do store extents in the form
2142  * of starting block and length, hence the resulting length of the extent
2143  * covering maximum file size must fit into on-disk format containers as
2144  * well. Given that length is always by 1 unit bigger than max unit (because
2145  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2146  *
2147  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2148  */
2149 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2150 {
2151         loff_t res;
2152         loff_t upper_limit = MAX_LFS_FILESIZE;
2153
2154         /* small i_blocks in vfs inode? */
2155         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2156                 /*
2157                  * CONFIG_LBDAF is not enabled implies the inode
2158                  * i_block represent total blocks in 512 bytes
2159                  * 32 == size of vfs inode i_blocks * 8
2160                  */
2161                 upper_limit = (1LL << 32) - 1;
2162
2163                 /* total blocks in file system block size */
2164                 upper_limit >>= (blkbits - 9);
2165                 upper_limit <<= blkbits;
2166         }
2167
2168         /*
2169          * 32-bit extent-start container, ee_block. We lower the maxbytes
2170          * by one fs block, so ee_len can cover the extent of maximum file
2171          * size
2172          */
2173         res = (1LL << 32) - 1;
2174         res <<= blkbits;
2175
2176         /* Sanity check against vm- & vfs- imposed limits */
2177         if (res > upper_limit)
2178                 res = upper_limit;
2179
2180         return res;
2181 }
2182
2183 /*
2184  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2185  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2186  * We need to be 1 filesystem block less than the 2^48 sector limit.
2187  */
2188 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2189 {
2190         loff_t res = EXT4_NDIR_BLOCKS;
2191         int meta_blocks;
2192         loff_t upper_limit;
2193         /* This is calculated to be the largest file size for a dense, block
2194          * mapped file such that the file's total number of 512-byte sectors,
2195          * including data and all indirect blocks, does not exceed (2^48 - 1).
2196          *
2197          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2198          * number of 512-byte sectors of the file.
2199          */
2200
2201         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2202                 /*
2203                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2204                  * the inode i_block field represents total file blocks in
2205                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2206                  */
2207                 upper_limit = (1LL << 32) - 1;
2208
2209                 /* total blocks in file system block size */
2210                 upper_limit >>= (bits - 9);
2211
2212         } else {
2213                 /*
2214                  * We use 48 bit ext4_inode i_blocks
2215                  * With EXT4_HUGE_FILE_FL set the i_blocks
2216                  * represent total number of blocks in
2217                  * file system block size
2218                  */
2219                 upper_limit = (1LL << 48) - 1;
2220
2221         }
2222
2223         /* indirect blocks */
2224         meta_blocks = 1;
2225         /* double indirect blocks */
2226         meta_blocks += 1 + (1LL << (bits-2));
2227         /* tripple indirect blocks */
2228         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2229
2230         upper_limit -= meta_blocks;
2231         upper_limit <<= bits;
2232
2233         res += 1LL << (bits-2);
2234         res += 1LL << (2*(bits-2));
2235         res += 1LL << (3*(bits-2));
2236         res <<= bits;
2237         if (res > upper_limit)
2238                 res = upper_limit;
2239
2240         if (res > MAX_LFS_FILESIZE)
2241                 res = MAX_LFS_FILESIZE;
2242
2243         return res;
2244 }
2245
2246 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2247                                    ext4_fsblk_t logical_sb_block, int nr)
2248 {
2249         struct ext4_sb_info *sbi = EXT4_SB(sb);
2250         ext4_group_t bg, first_meta_bg;
2251         int has_super = 0;
2252
2253         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2254
2255         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2256             nr < first_meta_bg)
2257                 return logical_sb_block + nr + 1;
2258         bg = sbi->s_desc_per_block * nr;
2259         if (ext4_bg_has_super(sb, bg))
2260                 has_super = 1;
2261
2262         return (has_super + ext4_group_first_block_no(sb, bg));
2263 }
2264
2265 /**
2266  * ext4_get_stripe_size: Get the stripe size.
2267  * @sbi: In memory super block info
2268  *
2269  * If we have specified it via mount option, then
2270  * use the mount option value. If the value specified at mount time is
2271  * greater than the blocks per group use the super block value.
2272  * If the super block value is greater than blocks per group return 0.
2273  * Allocator needs it be less than blocks per group.
2274  *
2275  */
2276 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2277 {
2278         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2279         unsigned long stripe_width =
2280                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2281         int ret;
2282
2283         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2284                 ret = sbi->s_stripe;
2285         else if (stripe_width <= sbi->s_blocks_per_group)
2286                 ret = stripe_width;
2287         else if (stride <= sbi->s_blocks_per_group)
2288                 ret = stride;
2289         else
2290                 ret = 0;
2291
2292         /*
2293          * If the stripe width is 1, this makes no sense and
2294          * we set it to 0 to turn off stripe handling code.
2295          */
2296         if (ret <= 1)
2297                 ret = 0;
2298
2299         return ret;
2300 }
2301
2302 /* sysfs supprt */
2303
2304 struct ext4_attr {
2305         struct attribute attr;
2306         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2307         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2308                          const char *, size_t);
2309         int offset;
2310 };
2311
2312 static int parse_strtoul(const char *buf,
2313                 unsigned long max, unsigned long *value)
2314 {
2315         char *endp;
2316
2317         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2318         endp = skip_spaces(endp);
2319         if (*endp || *value > max)
2320                 return -EINVAL;
2321
2322         return 0;
2323 }
2324
2325 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2326                                               struct ext4_sb_info *sbi,
2327                                               char *buf)
2328 {
2329         return snprintf(buf, PAGE_SIZE, "%llu\n",
2330                 (s64) EXT4_C2B(sbi,
2331                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2332 }
2333
2334 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2335                                          struct ext4_sb_info *sbi, char *buf)
2336 {
2337         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2338
2339         if (!sb->s_bdev->bd_part)
2340                 return snprintf(buf, PAGE_SIZE, "0\n");
2341         return snprintf(buf, PAGE_SIZE, "%lu\n",
2342                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2343                          sbi->s_sectors_written_start) >> 1);
2344 }
2345
2346 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2347                                           struct ext4_sb_info *sbi, char *buf)
2348 {
2349         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2350
2351         if (!sb->s_bdev->bd_part)
2352                 return snprintf(buf, PAGE_SIZE, "0\n");
2353         return snprintf(buf, PAGE_SIZE, "%llu\n",
2354                         (unsigned long long)(sbi->s_kbytes_written +
2355                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2356                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2357 }
2358
2359 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2360                                       struct ext4_sb_info *sbi, char *buf)
2361 {
2362         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2363 }
2364
2365 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2366                                         struct ext4_sb_info *sbi, char *buf)
2367 {
2368         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2369 }
2370
2371 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2372                                           struct ext4_sb_info *sbi,
2373                                           const char *buf, size_t count)
2374 {
2375         unsigned long t;
2376
2377         if (parse_strtoul(buf, 0x40000000, &t))
2378                 return -EINVAL;
2379
2380         if (t && !is_power_of_2(t))
2381                 return -EINVAL;
2382
2383         sbi->s_inode_readahead_blks = t;
2384         return count;
2385 }
2386
2387 static ssize_t sbi_ui_show(struct ext4_attr *a,
2388                            struct ext4_sb_info *sbi, char *buf)
2389 {
2390         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2391
2392         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2393 }
2394
2395 static ssize_t sbi_ui_store(struct ext4_attr *a,
2396                             struct ext4_sb_info *sbi,
2397                             const char *buf, size_t count)
2398 {
2399         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2400         unsigned long t;
2401
2402         if (parse_strtoul(buf, 0xffffffff, &t))
2403                 return -EINVAL;
2404         *ui = t;
2405         return count;
2406 }
2407
2408 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2409 static struct ext4_attr ext4_attr_##_name = {                   \
2410         .attr = {.name = __stringify(_name), .mode = _mode },   \
2411         .show   = _show,                                        \
2412         .store  = _store,                                       \
2413         .offset = offsetof(struct ext4_sb_info, _elname),       \
2414 }
2415 #define EXT4_ATTR(name, mode, show, store) \
2416 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2417
2418 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2419 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2420 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2421 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2422         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2423 #define ATTR_LIST(name) &ext4_attr_##name.attr
2424
2425 EXT4_RO_ATTR(delayed_allocation_blocks);
2426 EXT4_RO_ATTR(session_write_kbytes);
2427 EXT4_RO_ATTR(lifetime_write_kbytes);
2428 EXT4_RO_ATTR(extent_cache_hits);
2429 EXT4_RO_ATTR(extent_cache_misses);
2430 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2431                  inode_readahead_blks_store, s_inode_readahead_blks);
2432 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2433 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2434 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2435 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2436 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2437 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2438 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2439 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2440
2441 static struct attribute *ext4_attrs[] = {
2442         ATTR_LIST(delayed_allocation_blocks),
2443         ATTR_LIST(session_write_kbytes),
2444         ATTR_LIST(lifetime_write_kbytes),
2445         ATTR_LIST(extent_cache_hits),
2446         ATTR_LIST(extent_cache_misses),
2447         ATTR_LIST(inode_readahead_blks),
2448         ATTR_LIST(inode_goal),
2449         ATTR_LIST(mb_stats),
2450         ATTR_LIST(mb_max_to_scan),
2451         ATTR_LIST(mb_min_to_scan),
2452         ATTR_LIST(mb_order2_req),
2453         ATTR_LIST(mb_stream_req),
2454         ATTR_LIST(mb_group_prealloc),
2455         ATTR_LIST(max_writeback_mb_bump),
2456         NULL,
2457 };
2458
2459 /* Features this copy of ext4 supports */
2460 EXT4_INFO_ATTR(lazy_itable_init);
2461 EXT4_INFO_ATTR(batched_discard);
2462
2463 static struct attribute *ext4_feat_attrs[] = {
2464         ATTR_LIST(lazy_itable_init),
2465         ATTR_LIST(batched_discard),
2466         NULL,
2467 };
2468
2469 static ssize_t ext4_attr_show(struct kobject *kobj,
2470                               struct attribute *attr, char *buf)
2471 {
2472         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2473                                                 s_kobj);
2474         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2475
2476         return a->show ? a->show(a, sbi, buf) : 0;
2477 }
2478
2479 static ssize_t ext4_attr_store(struct kobject *kobj,
2480                                struct attribute *attr,
2481                                const char *buf, size_t len)
2482 {
2483         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2484                                                 s_kobj);
2485         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2486
2487         return a->store ? a->store(a, sbi, buf, len) : 0;
2488 }
2489
2490 static void ext4_sb_release(struct kobject *kobj)
2491 {
2492         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2493                                                 s_kobj);
2494         complete(&sbi->s_kobj_unregister);
2495 }
2496
2497 static const struct sysfs_ops ext4_attr_ops = {
2498         .show   = ext4_attr_show,
2499         .store  = ext4_attr_store,
2500 };
2501
2502 static struct kobj_type ext4_ktype = {
2503         .default_attrs  = ext4_attrs,
2504         .sysfs_ops      = &ext4_attr_ops,
2505         .release        = ext4_sb_release,
2506 };
2507
2508 static void ext4_feat_release(struct kobject *kobj)
2509 {
2510         complete(&ext4_feat->f_kobj_unregister);
2511 }
2512
2513 static struct kobj_type ext4_feat_ktype = {
2514         .default_attrs  = ext4_feat_attrs,
2515         .sysfs_ops      = &ext4_attr_ops,
2516         .release        = ext4_feat_release,
2517 };
2518
2519 /*
2520  * Check whether this filesystem can be mounted based on
2521  * the features present and the RDONLY/RDWR mount requested.
2522  * Returns 1 if this filesystem can be mounted as requested,
2523  * 0 if it cannot be.
2524  */
2525 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2526 {
2527         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2528                 ext4_msg(sb, KERN_ERR,
2529                         "Couldn't mount because of "
2530                         "unsupported optional features (%x)",
2531                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2532                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2533                 return 0;
2534         }
2535
2536         if (readonly)
2537                 return 1;
2538
2539         /* Check that feature set is OK for a read-write mount */
2540         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2541                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2542                          "unsupported optional features (%x)",
2543                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2544                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2545                 return 0;
2546         }
2547         /*
2548          * Large file size enabled file system can only be mounted
2549          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2550          */
2551         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2552                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2553                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2554                                  "cannot be mounted RDWR without "
2555                                  "CONFIG_LBDAF");
2556                         return 0;
2557                 }
2558         }
2559         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2560             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2561                 ext4_msg(sb, KERN_ERR,
2562                          "Can't support bigalloc feature without "
2563                          "extents feature\n");
2564                 return 0;
2565         }
2566         return 1;
2567 }
2568
2569 /*
2570  * This function is called once a day if we have errors logged
2571  * on the file system
2572  */
2573 static void print_daily_error_info(unsigned long arg)
2574 {
2575         struct super_block *sb = (struct super_block *) arg;
2576         struct ext4_sb_info *sbi;
2577         struct ext4_super_block *es;
2578
2579         sbi = EXT4_SB(sb);
2580         es = sbi->s_es;
2581
2582         if (es->s_error_count)
2583                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2584                          le32_to_cpu(es->s_error_count));
2585         if (es->s_first_error_time) {
2586                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2587                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2588                        (int) sizeof(es->s_first_error_func),
2589                        es->s_first_error_func,
2590                        le32_to_cpu(es->s_first_error_line));
2591                 if (es->s_first_error_ino)
2592                         printk(": inode %u",
2593                                le32_to_cpu(es->s_first_error_ino));
2594                 if (es->s_first_error_block)
2595                         printk(": block %llu", (unsigned long long)
2596                                le64_to_cpu(es->s_first_error_block));
2597                 printk("\n");
2598         }
2599         if (es->s_last_error_time) {
2600                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2601                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2602                        (int) sizeof(es->s_last_error_func),
2603                        es->s_last_error_func,
2604                        le32_to_cpu(es->s_last_error_line));
2605                 if (es->s_last_error_ino)
2606                         printk(": inode %u",
2607                                le32_to_cpu(es->s_last_error_ino));
2608                 if (es->s_last_error_block)
2609                         printk(": block %llu", (unsigned long long)
2610                                le64_to_cpu(es->s_last_error_block));
2611                 printk("\n");
2612         }
2613         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2614 }
2615
2616 /* Find next suitable group and run ext4_init_inode_table */
2617 static int ext4_run_li_request(struct ext4_li_request *elr)
2618 {
2619         struct ext4_group_desc *gdp = NULL;
2620         ext4_group_t group, ngroups;
2621         struct super_block *sb;
2622         unsigned long timeout = 0;
2623         int ret = 0;
2624
2625         sb = elr->lr_super;
2626         ngroups = EXT4_SB(sb)->s_groups_count;
2627
2628         for (group = elr->lr_next_group; group < ngroups; group++) {
2629                 gdp = ext4_get_group_desc(sb, group, NULL);
2630                 if (!gdp) {
2631                         ret = 1;
2632                         break;
2633                 }
2634
2635                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2636                         break;
2637         }
2638
2639         if (group == ngroups)
2640                 ret = 1;
2641
2642         if (!ret) {
2643                 timeout = jiffies;
2644                 ret = ext4_init_inode_table(sb, group,
2645                                             elr->lr_timeout ? 0 : 1);
2646                 if (elr->lr_timeout == 0) {
2647                         timeout = (jiffies - timeout) *
2648                                   elr->lr_sbi->s_li_wait_mult;
2649                         elr->lr_timeout = timeout;
2650                 }
2651                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2652                 elr->lr_next_group = group + 1;
2653         }
2654
2655         return ret;
2656 }
2657
2658 /*
2659  * Remove lr_request from the list_request and free the
2660  * request structure. Should be called with li_list_mtx held
2661  */
2662 static void ext4_remove_li_request(struct ext4_li_request *elr)
2663 {
2664         struct ext4_sb_info *sbi;
2665
2666         if (!elr)
2667                 return;
2668
2669         sbi = elr->lr_sbi;
2670
2671         list_del(&elr->lr_request);
2672         sbi->s_li_request = NULL;
2673         kfree(elr);
2674 }
2675
2676 static void ext4_unregister_li_request(struct super_block *sb)
2677 {
2678         mutex_lock(&ext4_li_mtx);
2679         if (!ext4_li_info) {
2680                 mutex_unlock(&ext4_li_mtx);
2681                 return;
2682         }
2683
2684         mutex_lock(&ext4_li_info->li_list_mtx);
2685         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2686         mutex_unlock(&ext4_li_info->li_list_mtx);
2687         mutex_unlock(&ext4_li_mtx);
2688 }
2689
2690 static struct task_struct *ext4_lazyinit_task;
2691
2692 /*
2693  * This is the function where ext4lazyinit thread lives. It walks
2694  * through the request list searching for next scheduled filesystem.
2695  * When such a fs is found, run the lazy initialization request
2696  * (ext4_rn_li_request) and keep track of the time spend in this
2697  * function. Based on that time we compute next schedule time of
2698  * the request. When walking through the list is complete, compute
2699  * next waking time and put itself into sleep.
2700  */
2701 static int ext4_lazyinit_thread(void *arg)
2702 {
2703         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2704         struct list_head *pos, *n;
2705         struct ext4_li_request *elr;
2706         unsigned long next_wakeup, cur;
2707
2708         BUG_ON(NULL == eli);
2709
2710 cont_thread:
2711         while (true) {
2712                 next_wakeup = MAX_JIFFY_OFFSET;
2713
2714                 mutex_lock(&eli->li_list_mtx);
2715                 if (list_empty(&eli->li_request_list)) {
2716                         mutex_unlock(&eli->li_list_mtx);
2717                         goto exit_thread;
2718                 }
2719
2720                 list_for_each_safe(pos, n, &eli->li_request_list) {
2721                         elr = list_entry(pos, struct ext4_li_request,
2722                                          lr_request);
2723
2724                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2725                                 if (ext4_run_li_request(elr) != 0) {
2726                                         /* error, remove the lazy_init job */
2727                                         ext4_remove_li_request(elr);
2728                                         continue;
2729                                 }
2730                         }
2731
2732                         if (time_before(elr->lr_next_sched, next_wakeup))
2733                                 next_wakeup = elr->lr_next_sched;
2734                 }
2735                 mutex_unlock(&eli->li_list_mtx);
2736
2737                 try_to_freeze();
2738
2739                 cur = jiffies;
2740                 if ((time_after_eq(cur, next_wakeup)) ||
2741                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2742                         cond_resched();
2743                         continue;
2744                 }
2745
2746                 schedule_timeout_interruptible(next_wakeup - cur);
2747
2748                 if (kthread_should_stop()) {
2749                         ext4_clear_request_list();
2750                         goto exit_thread;
2751                 }
2752         }
2753
2754 exit_thread:
2755         /*
2756          * It looks like the request list is empty, but we need
2757          * to check it under the li_list_mtx lock, to prevent any
2758          * additions into it, and of course we should lock ext4_li_mtx
2759          * to atomically free the list and ext4_li_info, because at
2760          * this point another ext4 filesystem could be registering
2761          * new one.
2762          */
2763         mutex_lock(&ext4_li_mtx);
2764         mutex_lock(&eli->li_list_mtx);
2765         if (!list_empty(&eli->li_request_list)) {
2766                 mutex_unlock(&eli->li_list_mtx);
2767                 mutex_unlock(&ext4_li_mtx);
2768                 goto cont_thread;
2769         }
2770         mutex_unlock(&eli->li_list_mtx);
2771         kfree(ext4_li_info);
2772         ext4_li_info = NULL;
2773         mutex_unlock(&ext4_li_mtx);
2774
2775         return 0;
2776 }
2777
2778 static void ext4_clear_request_list(void)
2779 {
2780         struct list_head *pos, *n;
2781         struct ext4_li_request *elr;
2782
2783         mutex_lock(&ext4_li_info->li_list_mtx);
2784         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2785                 elr = list_entry(pos, struct ext4_li_request,
2786                                  lr_request);
2787                 ext4_remove_li_request(elr);
2788         }
2789         mutex_unlock(&ext4_li_info->li_list_mtx);
2790 }
2791
2792 static int ext4_run_lazyinit_thread(void)
2793 {
2794         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2795                                          ext4_li_info, "ext4lazyinit");
2796         if (IS_ERR(ext4_lazyinit_task)) {
2797                 int err = PTR_ERR(ext4_lazyinit_task);
2798                 ext4_clear_request_list();
2799                 kfree(ext4_li_info);
2800                 ext4_li_info = NULL;
2801                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2802                                  "initialization thread\n",
2803                                  err);
2804                 return err;
2805         }
2806         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2807         return 0;
2808 }
2809
2810 /*
2811  * Check whether it make sense to run itable init. thread or not.
2812  * If there is at least one uninitialized inode table, return
2813  * corresponding group number, else the loop goes through all
2814  * groups and return total number of groups.
2815  */
2816 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2817 {
2818         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2819         struct ext4_group_desc *gdp = NULL;
2820
2821         for (group = 0; group < ngroups; group++) {
2822                 gdp = ext4_get_group_desc(sb, group, NULL);
2823                 if (!gdp)
2824                         continue;
2825
2826                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2827                         break;
2828         }
2829
2830         return group;
2831 }
2832
2833 static int ext4_li_info_new(void)
2834 {
2835         struct ext4_lazy_init *eli = NULL;
2836
2837         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2838         if (!eli)
2839                 return -ENOMEM;
2840
2841         INIT_LIST_HEAD(&eli->li_request_list);
2842         mutex_init(&eli->li_list_mtx);
2843
2844         eli->li_state |= EXT4_LAZYINIT_QUIT;
2845
2846         ext4_li_info = eli;
2847
2848         return 0;
2849 }
2850
2851 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2852                                             ext4_group_t start)
2853 {
2854         struct ext4_sb_info *sbi = EXT4_SB(sb);
2855         struct ext4_li_request *elr;
2856         unsigned long rnd;
2857
2858         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2859         if (!elr)
2860                 return NULL;
2861
2862         elr->lr_super = sb;
2863         elr->lr_sbi = sbi;
2864         elr->lr_next_group = start;
2865
2866         /*
2867          * Randomize first schedule time of the request to
2868          * spread the inode table initialization requests
2869          * better.
2870          */
2871         get_random_bytes(&rnd, sizeof(rnd));
2872         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2873                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2874
2875         return elr;
2876 }
2877
2878 static int ext4_register_li_request(struct super_block *sb,
2879                                     ext4_group_t first_not_zeroed)
2880 {
2881         struct ext4_sb_info *sbi = EXT4_SB(sb);
2882         struct ext4_li_request *elr;
2883         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2884         int ret = 0;
2885
2886         if (sbi->s_li_request != NULL) {
2887                 /*
2888                  * Reset timeout so it can be computed again, because
2889                  * s_li_wait_mult might have changed.
2890                  */
2891                 sbi->s_li_request->lr_timeout = 0;
2892                 return 0;
2893         }
2894
2895         if (first_not_zeroed == ngroups ||
2896             (sb->s_flags & MS_RDONLY) ||
2897             !test_opt(sb, INIT_INODE_TABLE))
2898                 return 0;
2899
2900         elr = ext4_li_request_new(sb, first_not_zeroed);
2901         if (!elr)
2902                 return -ENOMEM;
2903
2904         mutex_lock(&ext4_li_mtx);
2905
2906         if (NULL == ext4_li_info) {
2907                 ret = ext4_li_info_new();
2908                 if (ret)
2909                         goto out;
2910         }
2911
2912         mutex_lock(&ext4_li_info->li_list_mtx);
2913         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2914         mutex_unlock(&ext4_li_info->li_list_mtx);
2915
2916         sbi->s_li_request = elr;
2917         /*
2918          * set elr to NULL here since it has been inserted to
2919          * the request_list and the removal and free of it is
2920          * handled by ext4_clear_request_list from now on.
2921          */
2922         elr = NULL;
2923
2924         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2925                 ret = ext4_run_lazyinit_thread();
2926                 if (ret)
2927                         goto out;
2928         }
2929 out:
2930         mutex_unlock(&ext4_li_mtx);
2931         if (ret)
2932                 kfree(elr);
2933         return ret;
2934 }
2935
2936 /*
2937  * We do not need to lock anything since this is called on
2938  * module unload.
2939  */
2940 static void ext4_destroy_lazyinit_thread(void)
2941 {
2942         /*
2943          * If thread exited earlier
2944          * there's nothing to be done.
2945          */
2946         if (!ext4_li_info || !ext4_lazyinit_task)
2947                 return;
2948
2949         kthread_stop(ext4_lazyinit_task);
2950 }
2951
2952 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2953 {
2954         char *orig_data = kstrdup(data, GFP_KERNEL);
2955         struct buffer_head *bh;
2956         struct ext4_super_block *es = NULL;
2957         struct ext4_sb_info *sbi;
2958         ext4_fsblk_t block;
2959         ext4_fsblk_t sb_block = get_sb_block(&data);
2960         ext4_fsblk_t logical_sb_block;
2961         unsigned long offset = 0;
2962         unsigned long journal_devnum = 0;
2963         unsigned long def_mount_opts;
2964         struct inode *root;
2965         char *cp;
2966         const char *descr;
2967         int ret = -ENOMEM;
2968         int blocksize, clustersize;
2969         unsigned int db_count;
2970         unsigned int i;
2971         int needs_recovery, has_huge_files, has_bigalloc;
2972         __u64 blocks_count;
2973         int err;
2974         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2975         ext4_group_t first_not_zeroed;
2976
2977         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2978         if (!sbi)
2979                 goto out_free_orig;
2980
2981         sbi->s_blockgroup_lock =
2982                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2983         if (!sbi->s_blockgroup_lock) {
2984                 kfree(sbi);
2985                 goto out_free_orig;
2986         }
2987         sb->s_fs_info = sbi;
2988         sbi->s_mount_opt = 0;
2989         sbi->s_resuid = EXT4_DEF_RESUID;
2990         sbi->s_resgid = EXT4_DEF_RESGID;
2991         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
2992         sbi->s_sb_block = sb_block;
2993         if (sb->s_bdev->bd_part)
2994                 sbi->s_sectors_written_start =
2995                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
2996
2997         /* Cleanup superblock name */
2998         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
2999                 *cp = '!';
3000
3001         ret = -EINVAL;
3002         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3003         if (!blocksize) {
3004                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3005                 goto out_fail;
3006         }
3007
3008         /*
3009          * The ext4 superblock will not be buffer aligned for other than 1kB
3010          * block sizes.  We need to calculate the offset from buffer start.
3011          */
3012         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3013                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3014                 offset = do_div(logical_sb_block, blocksize);
3015         } else {
3016                 logical_sb_block = sb_block;
3017         }
3018
3019         if (!(bh = sb_bread(sb, logical_sb_block))) {
3020                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3021                 goto out_fail;
3022         }
3023         /*
3024          * Note: s_es must be initialized as soon as possible because
3025          *       some ext4 macro-instructions depend on its value
3026          */
3027         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3028         sbi->s_es = es;
3029         sb->s_magic = le16_to_cpu(es->s_magic);
3030         if (sb->s_magic != EXT4_SUPER_MAGIC)
3031                 goto cantfind_ext4;
3032         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3033
3034         /* Set defaults before we parse the mount options */
3035         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3036         set_opt(sb, INIT_INODE_TABLE);
3037         if (def_mount_opts & EXT4_DEFM_DEBUG)
3038                 set_opt(sb, DEBUG);
3039         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3040                 set_opt(sb, GRPID);
3041         if (def_mount_opts & EXT4_DEFM_UID16)
3042                 set_opt(sb, NO_UID32);
3043         /* xattr user namespace & acls are now defaulted on */
3044 #ifdef CONFIG_EXT4_FS_XATTR
3045         set_opt(sb, XATTR_USER);
3046 #endif
3047 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3048         set_opt(sb, POSIX_ACL);
3049 #endif
3050         set_opt(sb, MBLK_IO_SUBMIT);
3051         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3052                 set_opt(sb, JOURNAL_DATA);
3053         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3054                 set_opt(sb, ORDERED_DATA);
3055         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3056                 set_opt(sb, WRITEBACK_DATA);
3057
3058         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3059                 set_opt(sb, ERRORS_PANIC);
3060         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3061                 set_opt(sb, ERRORS_CONT);
3062         else
3063                 set_opt(sb, ERRORS_RO);
3064         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3065                 set_opt(sb, BLOCK_VALIDITY);
3066         if (def_mount_opts & EXT4_DEFM_DISCARD)
3067                 set_opt(sb, DISCARD);
3068
3069         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3070         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3071         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3072         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3073         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3074
3075         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3076                 set_opt(sb, BARRIER);
3077
3078         /*
3079          * enable delayed allocation by default
3080          * Use -o nodelalloc to turn it off
3081          */
3082         if (!IS_EXT3_SB(sb) &&
3083             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3084                 set_opt(sb, DELALLOC);
3085
3086         /*
3087          * set default s_li_wait_mult for lazyinit, for the case there is
3088          * no mount option specified.
3089          */
3090         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3091
3092         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3093                            &journal_devnum, &journal_ioprio, 0)) {
3094                 ext4_msg(sb, KERN_WARNING,
3095                          "failed to parse options in superblock: %s",
3096                          sbi->s_es->s_mount_opts);
3097         }
3098         sbi->s_def_mount_opt = sbi->s_mount_opt;
3099         if (!parse_options((char *) data, sb, &journal_devnum,
3100                            &journal_ioprio, 0))
3101                 goto failed_mount;
3102
3103         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3104                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3105                             "with data=journal disables delayed "
3106                             "allocation and O_DIRECT support!\n");
3107                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3108                         ext4_msg(sb, KERN_ERR, "can't mount with "
3109                                  "both data=journal and delalloc");
3110                         goto failed_mount;
3111                 }
3112                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3113                         ext4_msg(sb, KERN_ERR, "can't mount with "
3114                                  "both data=journal and delalloc");
3115                         goto failed_mount;
3116                 }
3117                 if (test_opt(sb, DELALLOC))
3118                         clear_opt(sb, DELALLOC);
3119         }
3120
3121         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3122         if (test_opt(sb, DIOREAD_NOLOCK)) {
3123                 if (blocksize < PAGE_SIZE) {
3124                         ext4_msg(sb, KERN_ERR, "can't mount with "
3125                                  "dioread_nolock if block size != PAGE_SIZE");
3126                         goto failed_mount;
3127                 }
3128         }
3129
3130         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3131                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3132
3133         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3134             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3135              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3136              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3137                 ext4_msg(sb, KERN_WARNING,
3138                        "feature flags set on rev 0 fs, "
3139                        "running e2fsck is recommended");
3140
3141         if (IS_EXT2_SB(sb)) {
3142                 if (ext2_feature_set_ok(sb))
3143                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3144                                  "using the ext4 subsystem");
3145                 else {
3146                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3147                                  "to feature incompatibilities");
3148                         goto failed_mount;
3149                 }
3150         }
3151
3152         if (IS_EXT3_SB(sb)) {
3153                 if (ext3_feature_set_ok(sb))
3154                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3155                                  "using the ext4 subsystem");
3156                 else {
3157                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3158                                  "to feature incompatibilities");
3159                         goto failed_mount;
3160                 }
3161         }
3162
3163         /*
3164          * Check feature flags regardless of the revision level, since we
3165          * previously didn't change the revision level when setting the flags,
3166          * so there is a chance incompat flags are set on a rev 0 filesystem.
3167          */
3168         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3169                 goto failed_mount;
3170
3171         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3172             blocksize > EXT4_MAX_BLOCK_SIZE) {
3173                 ext4_msg(sb, KERN_ERR,
3174                        "Unsupported filesystem blocksize %d", blocksize);
3175                 goto failed_mount;
3176         }
3177
3178         if (sb->s_blocksize != blocksize) {
3179                 /* Validate the filesystem blocksize */
3180                 if (!sb_set_blocksize(sb, blocksize)) {
3181                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3182                                         blocksize);
3183                         goto failed_mount;
3184                 }
3185
3186                 brelse(bh);
3187                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3188                 offset = do_div(logical_sb_block, blocksize);
3189                 bh = sb_bread(sb, logical_sb_block);
3190                 if (!bh) {
3191                         ext4_msg(sb, KERN_ERR,
3192                                "Can't read superblock on 2nd try");
3193                         goto failed_mount;
3194                 }
3195                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3196                 sbi->s_es = es;
3197                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3198                         ext4_msg(sb, KERN_ERR,
3199                                "Magic mismatch, very weird!");
3200                         goto failed_mount;
3201                 }
3202         }
3203
3204         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3205                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3206         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3207                                                       has_huge_files);
3208         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3209
3210         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3211                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3212                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3213         } else {
3214                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3215                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3216                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3217                     (!is_power_of_2(sbi->s_inode_size)) ||
3218                     (sbi->s_inode_size > blocksize)) {
3219                         ext4_msg(sb, KERN_ERR,
3220                                "unsupported inode size: %d",
3221                                sbi->s_inode_size);
3222                         goto failed_mount;
3223                 }
3224                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3225                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3226         }
3227
3228         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3229         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3230                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3231                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3232                     !is_power_of_2(sbi->s_desc_size)) {
3233                         ext4_msg(sb, KERN_ERR,
3234                                "unsupported descriptor size %lu",
3235                                sbi->s_desc_size);
3236                         goto failed_mount;
3237                 }
3238         } else
3239                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3240
3241         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3242         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3243         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3244                 goto cantfind_ext4;
3245
3246         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3247         if (sbi->s_inodes_per_block == 0)
3248                 goto cantfind_ext4;
3249         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3250                                         sbi->s_inodes_per_block;
3251         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3252         sbi->s_sbh = bh;
3253         sbi->s_mount_state = le16_to_cpu(es->s_state);
3254         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3255         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3256
3257         for (i = 0; i < 4; i++)
3258                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3259         sbi->s_def_hash_version = es->s_def_hash_version;
3260         i = le32_to_cpu(es->s_flags);
3261         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3262                 sbi->s_hash_unsigned = 3;
3263         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3264 #ifdef __CHAR_UNSIGNED__
3265                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3266                 sbi->s_hash_unsigned = 3;
3267 #else
3268                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3269 #endif
3270                 sb->s_dirt = 1;
3271         }
3272
3273         /* Handle clustersize */
3274         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3275         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3276                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3277         if (has_bigalloc) {
3278                 if (clustersize < blocksize) {
3279                         ext4_msg(sb, KERN_ERR,
3280                                  "cluster size (%d) smaller than "
3281                                  "block size (%d)", clustersize, blocksize);
3282                         goto failed_mount;
3283                 }
3284                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3285                         le32_to_cpu(es->s_log_block_size);
3286                 sbi->s_clusters_per_group =
3287                         le32_to_cpu(es->s_clusters_per_group);
3288                 if (sbi->s_clusters_per_group > blocksize * 8) {
3289                         ext4_msg(sb, KERN_ERR,
3290                                  "#clusters per group too big: %lu",
3291                                  sbi->s_clusters_per_group);
3292                         goto failed_mount;
3293                 }
3294                 if (sbi->s_blocks_per_group !=
3295                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3296                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3297                                  "clusters per group (%lu) inconsistent",
3298                                  sbi->s_blocks_per_group,
3299                                  sbi->s_clusters_per_group);
3300                         goto failed_mount;
3301                 }
3302         } else {
3303                 if (clustersize != blocksize) {
3304                         ext4_warning(sb, "fragment/cluster size (%d) != "
3305                                      "block size (%d)", clustersize,
3306                                      blocksize);
3307                         clustersize = blocksize;
3308                 }
3309                 if (sbi->s_blocks_per_group > blocksize * 8) {
3310                         ext4_msg(sb, KERN_ERR,
3311                                  "#blocks per group too big: %lu",
3312                                  sbi->s_blocks_per_group);
3313                         goto failed_mount;
3314                 }
3315                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3316                 sbi->s_cluster_bits = 0;
3317         }
3318         sbi->s_cluster_ratio = clustersize / blocksize;
3319
3320         if (sbi->s_inodes_per_group > blocksize * 8) {
3321                 ext4_msg(sb, KERN_ERR,
3322                        "#inodes per group too big: %lu",
3323                        sbi->s_inodes_per_group);
3324                 goto failed_mount;
3325         }
3326
3327         /*
3328          * Test whether we have more sectors than will fit in sector_t,
3329          * and whether the max offset is addressable by the page cache.
3330          */
3331         err = generic_check_addressable(sb->s_blocksize_bits,
3332                                         ext4_blocks_count(es));
3333         if (err) {
3334                 ext4_msg(sb, KERN_ERR, "filesystem"
3335                          " too large to mount safely on this system");
3336                 if (sizeof(sector_t) < 8)
3337                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3338                 ret = err;
3339                 goto failed_mount;
3340         }
3341
3342         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3343                 goto cantfind_ext4;
3344
3345         /* check blocks count against device size */
3346         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3347         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3348                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3349                        "exceeds size of device (%llu blocks)",
3350                        ext4_blocks_count(es), blocks_count);
3351                 goto failed_mount;
3352         }
3353
3354         /*
3355          * It makes no sense for the first data block to be beyond the end
3356          * of the filesystem.
3357          */
3358         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3359                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3360                          "block %u is beyond end of filesystem (%llu)",
3361                          le32_to_cpu(es->s_first_data_block),
3362                          ext4_blocks_count(es));
3363                 goto failed_mount;
3364         }
3365         blocks_count = (ext4_blocks_count(es) -
3366                         le32_to_cpu(es->s_first_data_block) +
3367                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3368         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3369         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3370                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3371                        "(block count %llu, first data block %u, "
3372                        "blocks per group %lu)", sbi->s_groups_count,
3373                        ext4_blocks_count(es),
3374                        le32_to_cpu(es->s_first_data_block),
3375                        EXT4_BLOCKS_PER_GROUP(sb));
3376                 goto failed_mount;
3377         }
3378         sbi->s_groups_count = blocks_count;
3379         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3380                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3381         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3382                    EXT4_DESC_PER_BLOCK(sb);
3383         sbi->s_group_desc = ext4_kvmalloc(db_count *
3384                                           sizeof(struct buffer_head *),
3385                                           GFP_KERNEL);
3386         if (sbi->s_group_desc == NULL) {
3387                 ext4_msg(sb, KERN_ERR, "not enough memory");
3388                 goto failed_mount;
3389         }
3390
3391         if (ext4_proc_root)
3392                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3393
3394         if (sbi->s_proc)
3395                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3396                                  &ext4_seq_options_fops, sb);
3397
3398         bgl_lock_init(sbi->s_blockgroup_lock);
3399
3400         for (i = 0; i < db_count; i++) {
3401                 block = descriptor_loc(sb, logical_sb_block, i);
3402                 sbi->s_group_desc[i] = sb_bread(sb, block);
3403                 if (!sbi->s_group_desc[i]) {
3404                         ext4_msg(sb, KERN_ERR,
3405                                "can't read group descriptor %d", i);
3406                         db_count = i;
3407                         goto failed_mount2;
3408                 }
3409         }
3410         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3411                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3412                 goto failed_mount2;
3413         }
3414         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3415                 if (!ext4_fill_flex_info(sb)) {
3416                         ext4_msg(sb, KERN_ERR,
3417                                "unable to initialize "
3418                                "flex_bg meta info!");
3419                         goto failed_mount2;
3420                 }
3421
3422         sbi->s_gdb_count = db_count;
3423         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3424         spin_lock_init(&sbi->s_next_gen_lock);
3425
3426         init_timer(&sbi->s_err_report);
3427         sbi->s_err_report.function = print_daily_error_info;
3428         sbi->s_err_report.data = (unsigned long) sb;
3429
3430         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3431                         ext4_count_free_clusters(sb));
3432         if (!err) {
3433                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3434                                 ext4_count_free_inodes(sb));
3435         }
3436         if (!err) {
3437                 err = percpu_counter_init(&sbi->s_dirs_counter,
3438                                 ext4_count_dirs(sb));
3439         }
3440         if (!err) {
3441                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3442         }
3443         if (err) {
3444                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3445                 goto failed_mount3;
3446         }
3447
3448         sbi->s_stripe = ext4_get_stripe_size(sbi);
3449         sbi->s_max_writeback_mb_bump = 128;
3450
3451         /*
3452          * set up enough so that it can read an inode
3453          */
3454         if (!test_opt(sb, NOLOAD) &&
3455             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3456                 sb->s_op = &ext4_sops;
3457         else
3458                 sb->s_op = &ext4_nojournal_sops;
3459         sb->s_export_op = &ext4_export_ops;
3460         sb->s_xattr = ext4_xattr_handlers;
3461 #ifdef CONFIG_QUOTA
3462         sb->s_qcop = &ext4_qctl_operations;
3463         sb->dq_op = &ext4_quota_operations;
3464 #endif
3465         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3466
3467         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3468         mutex_init(&sbi->s_orphan_lock);
3469         sbi->s_resize_flags = 0;
3470
3471         sb->s_root = NULL;
3472
3473         needs_recovery = (es->s_last_orphan != 0 ||
3474                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3475                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3476
3477         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3478             !(sb->s_flags & MS_RDONLY))
3479                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3480                         goto failed_mount3;
3481
3482         /*
3483          * The first inode we look at is the journal inode.  Don't try
3484          * root first: it may be modified in the journal!
3485          */
3486         if (!test_opt(sb, NOLOAD) &&
3487             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3488                 if (ext4_load_journal(sb, es, journal_devnum))
3489                         goto failed_mount3;
3490         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3491               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3492                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3493                        "suppressed and not mounted read-only");
3494                 goto failed_mount_wq;
3495         } else {
3496                 clear_opt(sb, DATA_FLAGS);
3497                 sbi->s_journal = NULL;
3498                 needs_recovery = 0;
3499                 goto no_journal;
3500         }
3501
3502         if (ext4_blocks_count(es) > 0xffffffffULL &&
3503             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3504                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3505                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3506                 goto failed_mount_wq;
3507         }
3508
3509         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3510                 jbd2_journal_set_features(sbi->s_journal,
3511                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3512                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3513         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3514                 jbd2_journal_set_features(sbi->s_journal,
3515                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3516                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3517                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3518         } else {
3519                 jbd2_journal_clear_features(sbi->s_journal,
3520                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3521                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3522         }
3523
3524         /* We have now updated the journal if required, so we can
3525          * validate the data journaling mode. */
3526         switch (test_opt(sb, DATA_FLAGS)) {
3527         case 0:
3528                 /* No mode set, assume a default based on the journal
3529                  * capabilities: ORDERED_DATA if the journal can
3530                  * cope, else JOURNAL_DATA
3531                  */
3532                 if (jbd2_journal_check_available_features
3533                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3534                         set_opt(sb, ORDERED_DATA);
3535                 else
3536                         set_opt(sb, JOURNAL_DATA);
3537                 break;
3538
3539         case EXT4_MOUNT_ORDERED_DATA:
3540         case EXT4_MOUNT_WRITEBACK_DATA:
3541                 if (!jbd2_journal_check_available_features
3542                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3543                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3544                                "requested data journaling mode");
3545                         goto failed_mount_wq;
3546                 }
3547         default:
3548                 break;
3549         }
3550         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3551
3552         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3553
3554         /*
3555          * The journal may have updated the bg summary counts, so we
3556          * need to update the global counters.
3557          */
3558         percpu_counter_set(&sbi->s_freeclusters_counter,
3559                            ext4_count_free_clusters(sb));
3560         percpu_counter_set(&sbi->s_freeinodes_counter,
3561                            ext4_count_free_inodes(sb));
3562         percpu_counter_set(&sbi->s_dirs_counter,
3563                            ext4_count_dirs(sb));
3564         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3565
3566 no_journal:
3567         /*
3568          * The maximum number of concurrent works can be high and
3569          * concurrency isn't really necessary.  Limit it to 1.
3570          */
3571         EXT4_SB(sb)->dio_unwritten_wq =
3572                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3573         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3574                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3575                 goto failed_mount_wq;
3576         }
3577
3578         /*
3579          * The jbd2_journal_load will have done any necessary log recovery,
3580          * so we can safely mount the rest of the filesystem now.
3581          */
3582
3583         root = ext4_iget(sb, EXT4_ROOT_INO);
3584         if (IS_ERR(root)) {
3585                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3586                 ret = PTR_ERR(root);
3587                 root = NULL;
3588                 goto failed_mount4;
3589         }
3590         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3591                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3592                 iput(root);
3593                 goto failed_mount4;
3594         }
3595         sb->s_root = d_alloc_root(root);
3596         if (!sb->s_root) {
3597                 iput(root);
3598                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3599                 ret = -ENOMEM;
3600                 goto failed_mount4;
3601         }
3602
3603         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3604
3605         /* determine the minimum size of new large inodes, if present */
3606         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3607                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3608                                                      EXT4_GOOD_OLD_INODE_SIZE;
3609                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3610                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3611                         if (sbi->s_want_extra_isize <
3612                             le16_to_cpu(es->s_want_extra_isize))
3613                                 sbi->s_want_extra_isize =
3614                                         le16_to_cpu(es->s_want_extra_isize);
3615                         if (sbi->s_want_extra_isize <
3616                             le16_to_cpu(es->s_min_extra_isize))
3617                                 sbi->s_want_extra_isize =
3618                                         le16_to_cpu(es->s_min_extra_isize);
3619                 }
3620         }
3621         /* Check if enough inode space is available */
3622         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3623                                                         sbi->s_inode_size) {
3624                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3625                                                        EXT4_GOOD_OLD_INODE_SIZE;
3626                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3627                          "available");
3628         }
3629
3630         err = ext4_setup_system_zone(sb);
3631         if (err) {
3632                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3633                          "zone (%d)", err);
3634                 goto failed_mount4a;
3635         }
3636
3637         ext4_ext_init(sb);
3638         err = ext4_mb_init(sb, needs_recovery);
3639         if (err) {
3640                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3641                          err);
3642                 goto failed_mount5;
3643         }
3644
3645         err = ext4_register_li_request(sb, first_not_zeroed);
3646         if (err)
3647                 goto failed_mount6;
3648
3649         sbi->s_kobj.kset = ext4_kset;
3650         init_completion(&sbi->s_kobj_unregister);
3651         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3652                                    "%s", sb->s_id);
3653         if (err)
3654                 goto failed_mount7;
3655
3656         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3657         ext4_orphan_cleanup(sb, es);
3658         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3659         if (needs_recovery) {
3660                 ext4_msg(sb, KERN_INFO, "recovery complete");
3661                 ext4_mark_recovery_complete(sb, es);
3662         }
3663         if (EXT4_SB(sb)->s_journal) {
3664                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3665                         descr = " journalled data mode";
3666                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3667                         descr = " ordered data mode";
3668                 else
3669                         descr = " writeback data mode";
3670         } else
3671                 descr = "out journal";
3672
3673         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3674                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3675                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3676
3677         if (es->s_error_count)
3678                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3679
3680         kfree(orig_data);
3681         return 0;
3682
3683 cantfind_ext4:
3684         if (!silent)
3685                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3686         goto failed_mount;
3687
3688 failed_mount7:
3689         ext4_unregister_li_request(sb);
3690 failed_mount6:
3691         ext4_mb_release(sb);
3692 failed_mount5:
3693         ext4_ext_release(sb);
3694         ext4_release_system_zone(sb);
3695 failed_mount4a:
3696         dput(sb->s_root);
3697         sb->s_root = NULL;
3698 failed_mount4:
3699         ext4_msg(sb, KERN_ERR, "mount failed");
3700         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3701 failed_mount_wq:
3702         if (sbi->s_journal) {
3703                 jbd2_journal_destroy(sbi->s_journal);
3704                 sbi->s_journal = NULL;
3705         }
3706 failed_mount3:
3707         del_timer(&sbi->s_err_report);
3708         if (sbi->s_flex_groups)
3709                 ext4_kvfree(sbi->s_flex_groups);
3710         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3711         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3712         percpu_counter_destroy(&sbi->s_dirs_counter);
3713         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3714         if (sbi->s_mmp_tsk)
3715                 kthread_stop(sbi->s_mmp_tsk);
3716 failed_mount2:
3717         for (i = 0; i < db_count; i++)
3718                 brelse(sbi->s_group_desc[i]);
3719         ext4_kvfree(sbi->s_group_desc);
3720 failed_mount:
3721         if (sbi->s_proc) {
3722                 remove_proc_entry("options", sbi->s_proc);
3723                 remove_proc_entry(sb->s_id, ext4_proc_root);
3724         }
3725 #ifdef CONFIG_QUOTA
3726         for (i = 0; i < MAXQUOTAS; i++)
3727                 kfree(sbi->s_qf_names[i]);
3728 #endif
3729         ext4_blkdev_remove(sbi);
3730         brelse(bh);
3731 out_fail:
3732         sb->s_fs_info = NULL;
3733         kfree(sbi->s_blockgroup_lock);
3734         kfree(sbi);
3735 out_free_orig:
3736         kfree(orig_data);
3737         return ret;
3738 }
3739
3740 /*
3741  * Setup any per-fs journal parameters now.  We'll do this both on
3742  * initial mount, once the journal has been initialised but before we've
3743  * done any recovery; and again on any subsequent remount.
3744  */
3745 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3746 {
3747         struct ext4_sb_info *sbi = EXT4_SB(sb);
3748
3749         journal->j_commit_interval = sbi->s_commit_interval;
3750         journal->j_min_batch_time = sbi->s_min_batch_time;
3751         journal->j_max_batch_time = sbi->s_max_batch_time;
3752
3753         write_lock(&journal->j_state_lock);
3754         if (test_opt(sb, BARRIER))
3755                 journal->j_flags |= JBD2_BARRIER;
3756         else
3757                 journal->j_flags &= ~JBD2_BARRIER;
3758         if (test_opt(sb, DATA_ERR_ABORT))
3759                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3760         else
3761                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3762         write_unlock(&journal->j_state_lock);
3763 }
3764
3765 static journal_t *ext4_get_journal(struct super_block *sb,
3766                                    unsigned int journal_inum)
3767 {
3768         struct inode *journal_inode;
3769         journal_t *journal;
3770
3771         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3772
3773         /* First, test for the existence of a valid inode on disk.  Bad
3774          * things happen if we iget() an unused inode, as the subsequent
3775          * iput() will try to delete it. */
3776
3777         journal_inode = ext4_iget(sb, journal_inum);
3778         if (IS_ERR(journal_inode)) {
3779                 ext4_msg(sb, KERN_ERR, "no journal found");
3780                 return NULL;
3781         }
3782         if (!journal_inode->i_nlink) {
3783                 make_bad_inode(journal_inode);
3784                 iput(journal_inode);
3785                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3786                 return NULL;
3787         }
3788
3789         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3790                   journal_inode, journal_inode->i_size);
3791         if (!S_ISREG(journal_inode->i_mode)) {
3792                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3793                 iput(journal_inode);
3794                 return NULL;
3795         }
3796
3797         journal = jbd2_journal_init_inode(journal_inode);
3798         if (!journal) {
3799                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3800                 iput(journal_inode);
3801                 return NULL;
3802         }
3803         journal->j_private = sb;
3804         ext4_init_journal_params(sb, journal);
3805         return journal;
3806 }
3807
3808 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3809                                        dev_t j_dev)
3810 {
3811         struct buffer_head *bh;
3812         journal_t *journal;
3813         ext4_fsblk_t start;
3814         ext4_fsblk_t len;
3815         int hblock, blocksize;
3816         ext4_fsblk_t sb_block;
3817         unsigned long offset;
3818         struct ext4_super_block *es;
3819         struct block_device *bdev;
3820
3821         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3822
3823         bdev = ext4_blkdev_get(j_dev, sb);
3824         if (bdev == NULL)
3825                 return NULL;
3826
3827         blocksize = sb->s_blocksize;
3828         hblock = bdev_logical_block_size(bdev);
3829         if (blocksize < hblock) {
3830                 ext4_msg(sb, KERN_ERR,
3831                         "blocksize too small for journal device");
3832                 goto out_bdev;
3833         }
3834
3835         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3836         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3837         set_blocksize(bdev, blocksize);
3838         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3839                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3840                        "external journal");
3841                 goto out_bdev;
3842         }
3843
3844         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3845         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3846             !(le32_to_cpu(es->s_feature_incompat) &
3847               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3848                 ext4_msg(sb, KERN_ERR, "external journal has "
3849                                         "bad superblock");
3850                 brelse(bh);
3851                 goto out_bdev;
3852         }
3853
3854         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3855                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3856                 brelse(bh);
3857                 goto out_bdev;
3858         }
3859
3860         len = ext4_blocks_count(es);
3861         start = sb_block + 1;
3862         brelse(bh);     /* we're done with the superblock */
3863
3864         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3865                                         start, len, blocksize);
3866         if (!journal) {
3867                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3868                 goto out_bdev;
3869         }
3870         journal->j_private = sb;
3871         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3872         wait_on_buffer(journal->j_sb_buffer);
3873         if (!buffer_uptodate(journal->j_sb_buffer)) {
3874                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3875                 goto out_journal;
3876         }
3877         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3878                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3879                                         "user (unsupported) - %d",
3880                         be32_to_cpu(journal->j_superblock->s_nr_users));
3881                 goto out_journal;
3882         }
3883         EXT4_SB(sb)->journal_bdev = bdev;
3884         ext4_init_journal_params(sb, journal);
3885         return journal;
3886
3887 out_journal:
3888         jbd2_journal_destroy(journal);
3889 out_bdev:
3890         ext4_blkdev_put(bdev);
3891         return NULL;
3892 }
3893
3894 static int ext4_load_journal(struct super_block *sb,
3895                              struct ext4_super_block *es,
3896                              unsigned long journal_devnum)
3897 {
3898         journal_t *journal;
3899         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3900         dev_t journal_dev;
3901         int err = 0;
3902         int really_read_only;
3903
3904         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3905
3906         if (journal_devnum &&
3907             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3908                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3909                         "numbers have changed");
3910                 journal_dev = new_decode_dev(journal_devnum);
3911         } else
3912                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3913
3914         really_read_only = bdev_read_only(sb->s_bdev);
3915
3916         /*
3917          * Are we loading a blank journal or performing recovery after a
3918          * crash?  For recovery, we need to check in advance whether we
3919          * can get read-write access to the device.
3920          */
3921         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3922                 if (sb->s_flags & MS_RDONLY) {
3923                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3924                                         "required on readonly filesystem");
3925                         if (really_read_only) {
3926                                 ext4_msg(sb, KERN_ERR, "write access "
3927                                         "unavailable, cannot proceed");
3928                                 return -EROFS;
3929                         }
3930                         ext4_msg(sb, KERN_INFO, "write access will "
3931                                "be enabled during recovery");
3932                 }
3933         }
3934
3935         if (journal_inum && journal_dev) {
3936                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3937                        "and inode journals!");
3938                 return -EINVAL;
3939         }
3940
3941         if (journal_inum) {
3942                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3943                         return -EINVAL;
3944         } else {
3945                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3946                         return -EINVAL;
3947         }
3948
3949         if (!(journal->j_flags & JBD2_BARRIER))
3950                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3951
3952         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3953                 err = jbd2_journal_wipe(journal, !really_read_only);
3954         if (!err) {
3955                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3956                 if (save)
3957                         memcpy(save, ((char *) es) +
3958                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3959                 err = jbd2_journal_load(journal);
3960                 if (save)
3961                         memcpy(((char *) es) + EXT4_S_ERR_START,
3962                                save, EXT4_S_ERR_LEN);
3963                 kfree(save);
3964         }
3965
3966         if (err) {
3967                 ext4_msg(sb, KERN_ERR, "error loading journal");
3968                 jbd2_journal_destroy(journal);
3969                 return err;
3970         }
3971
3972         EXT4_SB(sb)->s_journal = journal;
3973         ext4_clear_journal_err(sb, es);
3974
3975         if (!really_read_only && journal_devnum &&
3976             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3977                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3978
3979                 /* Make sure we flush the recovery flag to disk. */
3980                 ext4_commit_super(sb, 1);
3981         }
3982
3983         return 0;
3984 }
3985
3986 static int ext4_commit_super(struct super_block *sb, int sync)
3987 {
3988         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3989         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3990         int error = 0;
3991
3992         if (!sbh || block_device_ejected(sb))
3993                 return error;
3994         if (buffer_write_io_error(sbh)) {
3995                 /*
3996                  * Oh, dear.  A previous attempt to write the
3997                  * superblock failed.  This could happen because the
3998                  * USB device was yanked out.  Or it could happen to
3999                  * be a transient write error and maybe the block will
4000                  * be remapped.  Nothing we can do but to retry the
4001                  * write and hope for the best.
4002                  */
4003                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4004                        "superblock detected");
4005                 clear_buffer_write_io_error(sbh);
4006                 set_buffer_uptodate(sbh);
4007         }
4008         /*
4009          * If the file system is mounted read-only, don't update the
4010          * superblock write time.  This avoids updating the superblock
4011          * write time when we are mounting the root file system
4012          * read/only but we need to replay the journal; at that point,
4013          * for people who are east of GMT and who make their clock
4014          * tick in localtime for Windows bug-for-bug compatibility,
4015          * the clock is set in the future, and this will cause e2fsck
4016          * to complain and force a full file system check.
4017          */
4018         if (!(sb->s_flags & MS_RDONLY))
4019                 es->s_wtime = cpu_to_le32(get_seconds());
4020         if (sb->s_bdev->bd_part)
4021                 es->s_kbytes_written =
4022                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4023                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4024                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4025         else
4026                 es->s_kbytes_written =
4027                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4028         ext4_free_blocks_count_set(es,
4029                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4030                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4031         es->s_free_inodes_count =
4032                 cpu_to_le32(percpu_counter_sum_positive(
4033                                 &EXT4_SB(sb)->s_freeinodes_counter));
4034         sb->s_dirt = 0;
4035         BUFFER_TRACE(sbh, "marking dirty");
4036         mark_buffer_dirty(sbh);
4037         if (sync) {
4038                 error = sync_dirty_buffer(sbh);
4039                 if (error)
4040                         return error;
4041
4042                 error = buffer_write_io_error(sbh);
4043                 if (error) {
4044                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4045                                "superblock");
4046                         clear_buffer_write_io_error(sbh);
4047                         set_buffer_uptodate(sbh);
4048                 }
4049         }
4050         return error;
4051 }
4052
4053 /*
4054  * Have we just finished recovery?  If so, and if we are mounting (or
4055  * remounting) the filesystem readonly, then we will end up with a
4056  * consistent fs on disk.  Record that fact.
4057  */
4058 static void ext4_mark_recovery_complete(struct super_block *sb,
4059                                         struct ext4_super_block *es)
4060 {
4061         journal_t *journal = EXT4_SB(sb)->s_journal;
4062
4063         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4064                 BUG_ON(journal != NULL);
4065                 return;
4066         }
4067         jbd2_journal_lock_updates(journal);
4068         if (jbd2_journal_flush(journal) < 0)
4069                 goto out;
4070
4071         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4072             sb->s_flags & MS_RDONLY) {
4073                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4074                 ext4_commit_super(sb, 1);
4075         }
4076
4077 out:
4078         jbd2_journal_unlock_updates(journal);
4079 }
4080
4081 /*
4082  * If we are mounting (or read-write remounting) a filesystem whose journal
4083  * has recorded an error from a previous lifetime, move that error to the
4084  * main filesystem now.
4085  */
4086 static void ext4_clear_journal_err(struct super_block *sb,
4087                                    struct ext4_super_block *es)
4088 {
4089         journal_t *journal;
4090         int j_errno;
4091         const char *errstr;
4092
4093         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4094
4095         journal = EXT4_SB(sb)->s_journal;
4096
4097         /*
4098          * Now check for any error status which may have been recorded in the
4099          * journal by a prior ext4_error() or ext4_abort()
4100          */
4101
4102         j_errno = jbd2_journal_errno(journal);
4103         if (j_errno) {
4104                 char nbuf[16];
4105
4106                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4107                 ext4_warning(sb, "Filesystem error recorded "
4108                              "from previous mount: %s", errstr);
4109                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4110
4111                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4112                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4113                 ext4_commit_super(sb, 1);
4114
4115                 jbd2_journal_clear_err(journal);
4116         }
4117 }
4118
4119 /*
4120  * Force the running and committing transactions to commit,
4121  * and wait on the commit.
4122  */
4123 int ext4_force_commit(struct super_block *sb)
4124 {
4125         journal_t *journal;
4126         int ret = 0;
4127
4128         if (sb->s_flags & MS_RDONLY)
4129                 return 0;
4130
4131         journal = EXT4_SB(sb)->s_journal;
4132         if (journal) {
4133                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4134                 ret = ext4_journal_force_commit(journal);
4135         }
4136
4137         return ret;
4138 }
4139
4140 static void ext4_write_super(struct super_block *sb)
4141 {
4142         lock_super(sb);
4143         ext4_commit_super(sb, 1);
4144         unlock_super(sb);
4145 }
4146
4147 static int ext4_sync_fs(struct super_block *sb, int wait)
4148 {
4149         int ret = 0;
4150         tid_t target;
4151         struct ext4_sb_info *sbi = EXT4_SB(sb);
4152
4153         trace_ext4_sync_fs(sb, wait);
4154         flush_workqueue(sbi->dio_unwritten_wq);
4155         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4156                 if (wait)
4157                         jbd2_log_wait_commit(sbi->s_journal, target);
4158         }
4159         return ret;
4160 }
4161
4162 /*
4163  * LVM calls this function before a (read-only) snapshot is created.  This
4164  * gives us a chance to flush the journal completely and mark the fs clean.
4165  *
4166  * Note that only this function cannot bring a filesystem to be in a clean
4167  * state independently, because ext4 prevents a new handle from being started
4168  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4169  * the upper layer.
4170  */
4171 static int ext4_freeze(struct super_block *sb)
4172 {
4173         int error = 0;
4174         journal_t *journal;
4175
4176         if (sb->s_flags & MS_RDONLY)
4177                 return 0;
4178
4179         journal = EXT4_SB(sb)->s_journal;
4180
4181         /* Now we set up the journal barrier. */
4182         jbd2_journal_lock_updates(journal);
4183
4184         /*
4185          * Don't clear the needs_recovery flag if we failed to flush
4186          * the journal.
4187          */
4188         error = jbd2_journal_flush(journal);
4189         if (error < 0)
4190                 goto out;
4191
4192         /* Journal blocked and flushed, clear needs_recovery flag. */
4193         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4194         error = ext4_commit_super(sb, 1);
4195 out:
4196         /* we rely on s_frozen to stop further updates */
4197         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4198         return error;
4199 }
4200
4201 /*
4202  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4203  * flag here, even though the filesystem is not technically dirty yet.
4204  */
4205 static int ext4_unfreeze(struct super_block *sb)
4206 {
4207         if (sb->s_flags & MS_RDONLY)
4208                 return 0;
4209
4210         lock_super(sb);
4211         /* Reset the needs_recovery flag before the fs is unlocked. */
4212         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4213         ext4_commit_super(sb, 1);
4214         unlock_super(sb);
4215         return 0;
4216 }
4217
4218 /*
4219  * Structure to save mount options for ext4_remount's benefit
4220  */
4221 struct ext4_mount_options {
4222         unsigned long s_mount_opt;
4223         unsigned long s_mount_opt2;
4224         uid_t s_resuid;
4225         gid_t s_resgid;
4226         unsigned long s_commit_interval;
4227         u32 s_min_batch_time, s_max_batch_time;
4228 #ifdef CONFIG_QUOTA
4229         int s_jquota_fmt;
4230         char *s_qf_names[MAXQUOTAS];
4231 #endif
4232 };
4233
4234 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4235 {
4236         struct ext4_super_block *es;
4237         struct ext4_sb_info *sbi = EXT4_SB(sb);
4238         unsigned long old_sb_flags;
4239         struct ext4_mount_options old_opts;
4240         int enable_quota = 0;
4241         ext4_group_t g;
4242         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4243         int err = 0;
4244 #ifdef CONFIG_QUOTA
4245         int i;
4246 #endif
4247         char *orig_data = kstrdup(data, GFP_KERNEL);
4248
4249         /* Store the original options */
4250         lock_super(sb);
4251         old_sb_flags = sb->s_flags;
4252         old_opts.s_mount_opt = sbi->s_mount_opt;
4253         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4254         old_opts.s_resuid = sbi->s_resuid;
4255         old_opts.s_resgid = sbi->s_resgid;
4256         old_opts.s_commit_interval = sbi->s_commit_interval;
4257         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4258         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4259 #ifdef CONFIG_QUOTA
4260         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4261         for (i = 0; i < MAXQUOTAS; i++)
4262                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4263 #endif
4264         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4265                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4266
4267         /*
4268          * Allow the "check" option to be passed as a remount option.
4269          */
4270         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4271                 err = -EINVAL;
4272                 goto restore_opts;
4273         }
4274
4275         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4276                 ext4_abort(sb, "Abort forced by user");
4277
4278         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4279                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4280
4281         es = sbi->s_es;
4282
4283         if (sbi->s_journal) {
4284                 ext4_init_journal_params(sb, sbi->s_journal);
4285                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4286         }
4287
4288         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4289                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4290                         err = -EROFS;
4291                         goto restore_opts;
4292                 }
4293
4294                 if (*flags & MS_RDONLY) {
4295                         err = dquot_suspend(sb, -1);
4296                         if (err < 0)
4297                                 goto restore_opts;
4298
4299                         /*
4300                          * First of all, the unconditional stuff we have to do
4301                          * to disable replay of the journal when we next remount
4302                          */
4303                         sb->s_flags |= MS_RDONLY;
4304
4305                         /*
4306                          * OK, test if we are remounting a valid rw partition
4307                          * readonly, and if so set the rdonly flag and then
4308                          * mark the partition as valid again.
4309                          */
4310                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4311                             (sbi->s_mount_state & EXT4_VALID_FS))
4312                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4313
4314                         if (sbi->s_journal)
4315                                 ext4_mark_recovery_complete(sb, es);
4316                 } else {
4317                         /* Make sure we can mount this feature set readwrite */
4318                         if (!ext4_feature_set_ok(sb, 0)) {
4319                                 err = -EROFS;
4320                                 goto restore_opts;
4321                         }
4322                         /*
4323                          * Make sure the group descriptor checksums
4324                          * are sane.  If they aren't, refuse to remount r/w.
4325                          */
4326                         for (g = 0; g < sbi->s_groups_count; g++) {
4327                                 struct ext4_group_desc *gdp =
4328                                         ext4_get_group_desc(sb, g, NULL);
4329
4330                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4331                                         ext4_msg(sb, KERN_ERR,
4332                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4333                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4334                                                le16_to_cpu(gdp->bg_checksum));
4335                                         err = -EINVAL;
4336                                         goto restore_opts;
4337                                 }
4338                         }
4339
4340                         /*
4341                          * If we have an unprocessed orphan list hanging
4342                          * around from a previously readonly bdev mount,
4343                          * require a full umount/remount for now.
4344                          */
4345                         if (es->s_last_orphan) {
4346                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4347                                        "remount RDWR because of unprocessed "
4348                                        "orphan inode list.  Please "
4349                                        "umount/remount instead");
4350                                 err = -EINVAL;
4351                                 goto restore_opts;
4352                         }
4353
4354                         /*
4355                          * Mounting a RDONLY partition read-write, so reread
4356                          * and store the current valid flag.  (It may have
4357                          * been changed by e2fsck since we originally mounted
4358                          * the partition.)
4359                          */
4360                         if (sbi->s_journal)
4361                                 ext4_clear_journal_err(sb, es);
4362                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4363                         if (!ext4_setup_super(sb, es, 0))
4364                                 sb->s_flags &= ~MS_RDONLY;
4365                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4366                                                      EXT4_FEATURE_INCOMPAT_MMP))
4367                                 if (ext4_multi_mount_protect(sb,
4368                                                 le64_to_cpu(es->s_mmp_block))) {
4369                                         err = -EROFS;
4370                                         goto restore_opts;
4371                                 }
4372                         enable_quota = 1;
4373                 }
4374         }
4375
4376         /*
4377          * Reinitialize lazy itable initialization thread based on
4378          * current settings
4379          */
4380         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4381                 ext4_unregister_li_request(sb);
4382         else {
4383                 ext4_group_t first_not_zeroed;
4384                 first_not_zeroed = ext4_has_uninit_itable(sb);
4385                 ext4_register_li_request(sb, first_not_zeroed);
4386         }
4387
4388         ext4_setup_system_zone(sb);
4389         if (sbi->s_journal == NULL)
4390                 ext4_commit_super(sb, 1);
4391
4392 #ifdef CONFIG_QUOTA
4393         /* Release old quota file names */
4394         for (i = 0; i < MAXQUOTAS; i++)
4395                 if (old_opts.s_qf_names[i] &&
4396                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4397                         kfree(old_opts.s_qf_names[i]);
4398 #endif
4399         unlock_super(sb);
4400         if (enable_quota)
4401                 dquot_resume(sb, -1);
4402
4403         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4404         kfree(orig_data);
4405         return 0;
4406
4407 restore_opts:
4408         sb->s_flags = old_sb_flags;
4409         sbi->s_mount_opt = old_opts.s_mount_opt;
4410         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4411         sbi->s_resuid = old_opts.s_resuid;
4412         sbi->s_resgid = old_opts.s_resgid;
4413         sbi->s_commit_interval = old_opts.s_commit_interval;
4414         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4415         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4416 #ifdef CONFIG_QUOTA
4417         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4418         for (i = 0; i < MAXQUOTAS; i++) {
4419                 if (sbi->s_qf_names[i] &&
4420                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4421                         kfree(sbi->s_qf_names[i]);
4422                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4423         }
4424 #endif
4425         unlock_super(sb);
4426         kfree(orig_data);
4427         return err;
4428 }
4429
4430 /*
4431  * Note: calculating the overhead so we can be compatible with
4432  * historical BSD practice is quite difficult in the face of
4433  * clusters/bigalloc.  This is because multiple metadata blocks from
4434  * different block group can end up in the same allocation cluster.
4435  * Calculating the exact overhead in the face of clustered allocation
4436  * requires either O(all block bitmaps) in memory or O(number of block
4437  * groups**2) in time.  We will still calculate the superblock for
4438  * older file systems --- and if we come across with a bigalloc file
4439  * system with zero in s_overhead_clusters the estimate will be close to
4440  * correct especially for very large cluster sizes --- but for newer
4441  * file systems, it's better to calculate this figure once at mkfs
4442  * time, and store it in the superblock.  If the superblock value is
4443  * present (even for non-bigalloc file systems), we will use it.
4444  */
4445 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4446 {
4447         struct super_block *sb = dentry->d_sb;
4448         struct ext4_sb_info *sbi = EXT4_SB(sb);
4449         struct ext4_super_block *es = sbi->s_es;
4450         struct ext4_group_desc *gdp;
4451         u64 fsid;
4452         s64 bfree;
4453
4454         if (test_opt(sb, MINIX_DF)) {
4455                 sbi->s_overhead_last = 0;
4456         } else if (es->s_overhead_clusters) {
4457                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4458         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4459                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4460                 ext4_fsblk_t overhead = 0;
4461
4462                 /*
4463                  * Compute the overhead (FS structures).  This is constant
4464                  * for a given filesystem unless the number of block groups
4465                  * changes so we cache the previous value until it does.
4466                  */
4467
4468                 /*
4469                  * All of the blocks before first_data_block are
4470                  * overhead
4471                  */
4472                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4473
4474                 /*
4475                  * Add the overhead found in each block group
4476                  */
4477                 for (i = 0; i < ngroups; i++) {
4478                         gdp = ext4_get_group_desc(sb, i, NULL);
4479                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4480                         cond_resched();
4481                 }
4482                 sbi->s_overhead_last = overhead;
4483                 smp_wmb();
4484                 sbi->s_blocks_last = ext4_blocks_count(es);
4485         }
4486
4487         buf->f_type = EXT4_SUPER_MAGIC;
4488         buf->f_bsize = sb->s_blocksize;
4489         buf->f_blocks = (ext4_blocks_count(es) -
4490                          EXT4_C2B(sbi, sbi->s_overhead_last));
4491         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4492                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4493         /* prevent underflow in case that few free space is available */
4494         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4495         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4496         if (buf->f_bfree < ext4_r_blocks_count(es))
4497                 buf->f_bavail = 0;
4498         buf->f_files = le32_to_cpu(es->s_inodes_count);
4499         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4500         buf->f_namelen = EXT4_NAME_LEN;
4501         fsid = le64_to_cpup((void *)es->s_uuid) ^
4502                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4503         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4504         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4505
4506         return 0;
4507 }
4508
4509 /* Helper function for writing quotas on sync - we need to start transaction
4510  * before quota file is locked for write. Otherwise the are possible deadlocks:
4511  * Process 1                         Process 2
4512  * ext4_create()                     quota_sync()
4513  *   jbd2_journal_start()                  write_dquot()
4514  *   dquot_initialize()                         down(dqio_mutex)
4515  *     down(dqio_mutex)                    jbd2_journal_start()
4516  *
4517  */
4518
4519 #ifdef CONFIG_QUOTA
4520
4521 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4522 {
4523         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4524 }
4525
4526 static int ext4_write_dquot(struct dquot *dquot)
4527 {
4528         int ret, err;
4529         handle_t *handle;
4530         struct inode *inode;
4531
4532         inode = dquot_to_inode(dquot);
4533         handle = ext4_journal_start(inode,
4534                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4535         if (IS_ERR(handle))
4536                 return PTR_ERR(handle);
4537         ret = dquot_commit(dquot);
4538         err = ext4_journal_stop(handle);
4539         if (!ret)
4540                 ret = err;
4541         return ret;
4542 }
4543
4544 static int ext4_acquire_dquot(struct dquot *dquot)
4545 {
4546         int ret, err;
4547         handle_t *handle;
4548
4549         handle = ext4_journal_start(dquot_to_inode(dquot),
4550                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4551         if (IS_ERR(handle))
4552                 return PTR_ERR(handle);
4553         ret = dquot_acquire(dquot);
4554         err = ext4_journal_stop(handle);
4555         if (!ret)
4556                 ret = err;
4557         return ret;
4558 }
4559
4560 static int ext4_release_dquot(struct dquot *dquot)
4561 {
4562         int ret, err;
4563         handle_t *handle;
4564
4565         handle = ext4_journal_start(dquot_to_inode(dquot),
4566                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4567         if (IS_ERR(handle)) {
4568                 /* Release dquot anyway to avoid endless cycle in dqput() */
4569                 dquot_release(dquot);
4570                 return PTR_ERR(handle);
4571         }
4572         ret = dquot_release(dquot);
4573         err = ext4_journal_stop(handle);
4574         if (!ret)
4575                 ret = err;
4576         return ret;
4577 }
4578
4579 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4580 {
4581         /* Are we journaling quotas? */
4582         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4583             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4584                 dquot_mark_dquot_dirty(dquot);
4585                 return ext4_write_dquot(dquot);
4586         } else {
4587                 return dquot_mark_dquot_dirty(dquot);
4588         }
4589 }
4590
4591 static int ext4_write_info(struct super_block *sb, int type)
4592 {
4593         int ret, err;
4594         handle_t *handle;
4595
4596         /* Data block + inode block */
4597         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4598         if (IS_ERR(handle))
4599                 return PTR_ERR(handle);
4600         ret = dquot_commit_info(sb, type);
4601         err = ext4_journal_stop(handle);
4602         if (!ret)
4603                 ret = err;
4604         return ret;
4605 }
4606
4607 /*
4608  * Turn on quotas during mount time - we need to find
4609  * the quota file and such...
4610  */
4611 static int ext4_quota_on_mount(struct super_block *sb, int type)
4612 {
4613         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4614                                         EXT4_SB(sb)->s_jquota_fmt, type);
4615 }
4616
4617 /*
4618  * Standard function to be called on quota_on
4619  */
4620 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4621                          struct path *path)
4622 {
4623         int err;
4624
4625         if (!test_opt(sb, QUOTA))
4626                 return -EINVAL;
4627
4628         /* Quotafile not on the same filesystem? */
4629         if (path->dentry->d_sb != sb)
4630                 return -EXDEV;
4631         /* Journaling quota? */
4632         if (EXT4_SB(sb)->s_qf_names[type]) {
4633                 /* Quotafile not in fs root? */
4634                 if (path->dentry->d_parent != sb->s_root)
4635                         ext4_msg(sb, KERN_WARNING,
4636                                 "Quota file not on filesystem root. "
4637                                 "Journaled quota will not work");
4638         }
4639
4640         /*
4641          * When we journal data on quota file, we have to flush journal to see
4642          * all updates to the file when we bypass pagecache...
4643          */
4644         if (EXT4_SB(sb)->s_journal &&
4645             ext4_should_journal_data(path->dentry->d_inode)) {
4646                 /*
4647                  * We don't need to lock updates but journal_flush() could
4648                  * otherwise be livelocked...
4649                  */
4650                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4651                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4652                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4653                 if (err)
4654                         return err;
4655         }
4656
4657         return dquot_quota_on(sb, type, format_id, path);
4658 }
4659
4660 static int ext4_quota_off(struct super_block *sb, int type)
4661 {
4662         struct inode *inode = sb_dqopt(sb)->files[type];
4663         handle_t *handle;
4664
4665         /* Force all delayed allocation blocks to be allocated.
4666          * Caller already holds s_umount sem */
4667         if (test_opt(sb, DELALLOC))
4668                 sync_filesystem(sb);
4669
4670         if (!inode)
4671                 goto out;
4672
4673         /* Update modification times of quota files when userspace can
4674          * start looking at them */
4675         handle = ext4_journal_start(inode, 1);
4676         if (IS_ERR(handle))
4677                 goto out;
4678         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4679         ext4_mark_inode_dirty(handle, inode);
4680         ext4_journal_stop(handle);
4681
4682 out:
4683         return dquot_quota_off(sb, type);
4684 }
4685
4686 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4687  * acquiring the locks... As quota files are never truncated and quota code
4688  * itself serializes the operations (and no one else should touch the files)
4689  * we don't have to be afraid of races */
4690 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4691                                size_t len, loff_t off)
4692 {
4693         struct inode *inode = sb_dqopt(sb)->files[type];
4694         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4695         int err = 0;
4696         int offset = off & (sb->s_blocksize - 1);
4697         int tocopy;
4698         size_t toread;
4699         struct buffer_head *bh;
4700         loff_t i_size = i_size_read(inode);
4701
4702         if (off > i_size)
4703                 return 0;
4704         if (off+len > i_size)
4705                 len = i_size-off;
4706         toread = len;
4707         while (toread > 0) {
4708                 tocopy = sb->s_blocksize - offset < toread ?
4709                                 sb->s_blocksize - offset : toread;
4710                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4711                 if (err)
4712                         return err;
4713                 if (!bh)        /* A hole? */
4714                         memset(data, 0, tocopy);
4715                 else
4716                         memcpy(data, bh->b_data+offset, tocopy);
4717                 brelse(bh);
4718                 offset = 0;
4719                 toread -= tocopy;
4720                 data += tocopy;
4721                 blk++;
4722         }
4723         return len;
4724 }
4725
4726 /* Write to quotafile (we know the transaction is already started and has
4727  * enough credits) */
4728 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4729                                 const char *data, size_t len, loff_t off)
4730 {
4731         struct inode *inode = sb_dqopt(sb)->files[type];
4732         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4733         int err = 0;
4734         int offset = off & (sb->s_blocksize - 1);
4735         struct buffer_head *bh;
4736         handle_t *handle = journal_current_handle();
4737
4738         if (EXT4_SB(sb)->s_journal && !handle) {
4739                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4740                         " cancelled because transaction is not started",
4741                         (unsigned long long)off, (unsigned long long)len);
4742                 return -EIO;
4743         }
4744         /*
4745          * Since we account only one data block in transaction credits,
4746          * then it is impossible to cross a block boundary.
4747          */
4748         if (sb->s_blocksize - offset < len) {
4749                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4750                         " cancelled because not block aligned",
4751                         (unsigned long long)off, (unsigned long long)len);
4752                 return -EIO;
4753         }
4754
4755         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4756         bh = ext4_bread(handle, inode, blk, 1, &err);
4757         if (!bh)
4758                 goto out;
4759         err = ext4_journal_get_write_access(handle, bh);
4760         if (err) {
4761                 brelse(bh);
4762                 goto out;
4763         }
4764         lock_buffer(bh);
4765         memcpy(bh->b_data+offset, data, len);
4766         flush_dcache_page(bh->b_page);
4767         unlock_buffer(bh);
4768         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4769         brelse(bh);
4770 out:
4771         if (err) {
4772                 mutex_unlock(&inode->i_mutex);
4773                 return err;
4774         }
4775         if (inode->i_size < off + len) {
4776                 i_size_write(inode, off + len);
4777                 EXT4_I(inode)->i_disksize = inode->i_size;
4778                 ext4_mark_inode_dirty(handle, inode);
4779         }
4780         mutex_unlock(&inode->i_mutex);
4781         return len;
4782 }
4783
4784 #endif
4785
4786 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4787                        const char *dev_name, void *data)
4788 {
4789         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4790 }
4791
4792 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4793 static inline void register_as_ext2(void)
4794 {
4795         int err = register_filesystem(&ext2_fs_type);
4796         if (err)
4797                 printk(KERN_WARNING
4798                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4799 }
4800
4801 static inline void unregister_as_ext2(void)
4802 {
4803         unregister_filesystem(&ext2_fs_type);
4804 }
4805
4806 static inline int ext2_feature_set_ok(struct super_block *sb)
4807 {
4808         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4809                 return 0;
4810         if (sb->s_flags & MS_RDONLY)
4811                 return 1;
4812         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4813                 return 0;
4814         return 1;
4815 }
4816 MODULE_ALIAS("ext2");
4817 #else
4818 static inline void register_as_ext2(void) { }
4819 static inline void unregister_as_ext2(void) { }
4820 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4821 #endif
4822
4823 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4824 static inline void register_as_ext3(void)
4825 {
4826         int err = register_filesystem(&ext3_fs_type);
4827         if (err)
4828                 printk(KERN_WARNING
4829                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4830 }
4831
4832 static inline void unregister_as_ext3(void)
4833 {
4834         unregister_filesystem(&ext3_fs_type);
4835 }
4836
4837 static inline int ext3_feature_set_ok(struct super_block *sb)
4838 {
4839         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4840                 return 0;
4841         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4842                 return 0;
4843         if (sb->s_flags & MS_RDONLY)
4844                 return 1;
4845         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4846                 return 0;
4847         return 1;
4848 }
4849 MODULE_ALIAS("ext3");
4850 #else
4851 static inline void register_as_ext3(void) { }
4852 static inline void unregister_as_ext3(void) { }
4853 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4854 #endif
4855
4856 static struct file_system_type ext4_fs_type = {
4857         .owner          = THIS_MODULE,
4858         .name           = "ext4",
4859         .mount          = ext4_mount,
4860         .kill_sb        = kill_block_super,
4861         .fs_flags       = FS_REQUIRES_DEV,
4862 };
4863
4864 static int __init ext4_init_feat_adverts(void)
4865 {
4866         struct ext4_features *ef;
4867         int ret = -ENOMEM;
4868
4869         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4870         if (!ef)
4871                 goto out;
4872
4873         ef->f_kobj.kset = ext4_kset;
4874         init_completion(&ef->f_kobj_unregister);
4875         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4876                                    "features");
4877         if (ret) {
4878                 kfree(ef);
4879                 goto out;
4880         }
4881
4882         ext4_feat = ef;
4883         ret = 0;
4884 out:
4885         return ret;
4886 }
4887
4888 static void ext4_exit_feat_adverts(void)
4889 {
4890         kobject_put(&ext4_feat->f_kobj);
4891         wait_for_completion(&ext4_feat->f_kobj_unregister);
4892         kfree(ext4_feat);
4893 }
4894
4895 /* Shared across all ext4 file systems */
4896 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4897 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4898
4899 static int __init ext4_init_fs(void)
4900 {
4901         int i, err;
4902
4903         ext4_check_flag_values();
4904
4905         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4906                 mutex_init(&ext4__aio_mutex[i]);
4907                 init_waitqueue_head(&ext4__ioend_wq[i]);
4908         }
4909
4910         err = ext4_init_pageio();
4911         if (err)
4912                 return err;
4913         err = ext4_init_system_zone();
4914         if (err)
4915                 goto out6;
4916         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4917         if (!ext4_kset)
4918                 goto out5;
4919         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4920
4921         err = ext4_init_feat_adverts();
4922         if (err)
4923                 goto out4;
4924
4925         err = ext4_init_mballoc();
4926         if (err)
4927                 goto out3;
4928
4929         err = ext4_init_xattr();
4930         if (err)
4931                 goto out2;
4932         err = init_inodecache();
4933         if (err)
4934                 goto out1;
4935         register_as_ext3();
4936         register_as_ext2();
4937         err = register_filesystem(&ext4_fs_type);
4938         if (err)
4939                 goto out;
4940
4941         ext4_li_info = NULL;
4942         mutex_init(&ext4_li_mtx);
4943         return 0;
4944 out:
4945         unregister_as_ext2();
4946         unregister_as_ext3();
4947         destroy_inodecache();
4948 out1:
4949         ext4_exit_xattr();
4950 out2:
4951         ext4_exit_mballoc();
4952 out3:
4953         ext4_exit_feat_adverts();
4954 out4:
4955         if (ext4_proc_root)
4956                 remove_proc_entry("fs/ext4", NULL);
4957         kset_unregister(ext4_kset);
4958 out5:
4959         ext4_exit_system_zone();
4960 out6:
4961         ext4_exit_pageio();
4962         return err;
4963 }
4964
4965 static void __exit ext4_exit_fs(void)
4966 {
4967         ext4_destroy_lazyinit_thread();
4968         unregister_as_ext2();
4969         unregister_as_ext3();
4970         unregister_filesystem(&ext4_fs_type);
4971         destroy_inodecache();
4972         ext4_exit_xattr();
4973         ext4_exit_mballoc();
4974         ext4_exit_feat_adverts();
4975         remove_proc_entry("fs/ext4", NULL);
4976         kset_unregister(ext4_kset);
4977         ext4_exit_system_zone();
4978         ext4_exit_pageio();
4979 }
4980
4981 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4982 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4983 MODULE_LICENSE("GPL");
4984 module_init(ext4_init_fs)
4985 module_exit(ext4_exit_fs)