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_err, NULL},
1253 };
1254
1255 static ext4_fsblk_t get_sb_block(void **data)
1256 {
1257         ext4_fsblk_t    sb_block;
1258         char            *options = (char *) *data;
1259
1260         if (!options || strncmp(options, "sb=", 3) != 0)
1261                 return 1;       /* Default location */
1262
1263         options += 3;
1264         /* TODO: use simple_strtoll with >32bit ext4 */
1265         sb_block = simple_strtoul(options, &options, 0);
1266         if (*options && *options != ',') {
1267                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1268                        (char *) *data);
1269                 return 1;
1270         }
1271         if (*options == ',')
1272                 options++;
1273         *data = (void *) options;
1274
1275         return sb_block;
1276 }
1277
1278 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1279 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1280         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1281
1282 #ifdef CONFIG_QUOTA
1283 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1284 {
1285         struct ext4_sb_info *sbi = EXT4_SB(sb);
1286         char *qname;
1287
1288         if (sb_any_quota_loaded(sb) &&
1289                 !sbi->s_qf_names[qtype]) {
1290                 ext4_msg(sb, KERN_ERR,
1291                         "Cannot change journaled "
1292                         "quota options when quota turned on");
1293                 return 0;
1294         }
1295         qname = match_strdup(args);
1296         if (!qname) {
1297                 ext4_msg(sb, KERN_ERR,
1298                         "Not enough memory for storing quotafile name");
1299                 return 0;
1300         }
1301         if (sbi->s_qf_names[qtype] &&
1302                 strcmp(sbi->s_qf_names[qtype], qname)) {
1303                 ext4_msg(sb, KERN_ERR,
1304                         "%s quota file already specified", QTYPE2NAME(qtype));
1305                 kfree(qname);
1306                 return 0;
1307         }
1308         sbi->s_qf_names[qtype] = qname;
1309         if (strchr(sbi->s_qf_names[qtype], '/')) {
1310                 ext4_msg(sb, KERN_ERR,
1311                         "quotafile must be on filesystem root");
1312                 kfree(sbi->s_qf_names[qtype]);
1313                 sbi->s_qf_names[qtype] = NULL;
1314                 return 0;
1315         }
1316         set_opt(sb, QUOTA);
1317         return 1;
1318 }
1319
1320 static int clear_qf_name(struct super_block *sb, int qtype)
1321 {
1322
1323         struct ext4_sb_info *sbi = EXT4_SB(sb);
1324
1325         if (sb_any_quota_loaded(sb) &&
1326                 sbi->s_qf_names[qtype]) {
1327                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1328                         " when quota turned on");
1329                 return 0;
1330         }
1331         /*
1332          * The space will be released later when all options are confirmed
1333          * to be correct
1334          */
1335         sbi->s_qf_names[qtype] = NULL;
1336         return 1;
1337 }
1338 #endif
1339
1340 #define MOPT_SET        0x0001
1341 #define MOPT_CLEAR      0x0002
1342 #define MOPT_NOSUPPORT  0x0004
1343 #define MOPT_EXPLICIT   0x0008
1344 #define MOPT_CLEAR_ERR  0x0010
1345 #define MOPT_GTE0       0x0020
1346 #ifdef CONFIG_QUOTA
1347 #define MOPT_Q          0
1348 #define MOPT_QFMT       0x0040
1349 #else
1350 #define MOPT_Q          MOPT_NOSUPPORT
1351 #define MOPT_QFMT       MOPT_NOSUPPORT
1352 #endif
1353 #define MOPT_DATAJ      0x0080
1354
1355 static const struct mount_opts {
1356         int     token;
1357         int     mount_opt;
1358         int     flags;
1359 } ext4_mount_opts[] = {
1360         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1361         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1362         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1363         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1364         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1365         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1366         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1367         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1368         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1369         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1370         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1371         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1372         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1373         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1374         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1375         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1376                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1377         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1378         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1379         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1380         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1381         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1382         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1383         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1384         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1385         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1386         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1387         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1388         {Opt_commit, 0, MOPT_GTE0},
1389         {Opt_max_batch_time, 0, MOPT_GTE0},
1390         {Opt_min_batch_time, 0, MOPT_GTE0},
1391         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1392         {Opt_init_itable, 0, MOPT_GTE0},
1393         {Opt_stripe, 0, MOPT_GTE0},
1394         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1395         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1396         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1397 #ifdef CONFIG_EXT4_FS_XATTR
1398         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1399         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1400 #else
1401         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1402         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1403 #endif
1404 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1405         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1406         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1407 #else
1408         {Opt_acl, 0, MOPT_NOSUPPORT},
1409         {Opt_noacl, 0, MOPT_NOSUPPORT},
1410 #endif
1411         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1412         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1413         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1414         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1415                                                         MOPT_SET | MOPT_Q},
1416         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1417                                                         MOPT_SET | MOPT_Q},
1418         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1419                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1420         {Opt_usrjquota, 0, MOPT_Q},
1421         {Opt_grpjquota, 0, MOPT_Q},
1422         {Opt_offusrjquota, 0, MOPT_Q},
1423         {Opt_offgrpjquota, 0, MOPT_Q},
1424         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1425         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1426         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1427         {Opt_err, 0, 0}
1428 };
1429
1430 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1431                             substring_t *args, unsigned long *journal_devnum,
1432                             unsigned int *journal_ioprio, int is_remount)
1433 {
1434         struct ext4_sb_info *sbi = EXT4_SB(sb);
1435         const struct mount_opts *m;
1436         int arg = 0;
1437
1438         if (args->from && match_int(args, &arg))
1439                 return -1;
1440         switch (token) {
1441         case Opt_sb:
1442                 return 1;       /* handled by get_sb_block() */
1443         case Opt_removed:
1444                 ext4_msg(sb, KERN_WARNING,
1445                          "Ignoring removed %s option", opt);
1446                 return 1;
1447         case Opt_resuid:
1448                 sbi->s_resuid = arg;
1449                 return 1;
1450         case Opt_resgid:
1451                 sbi->s_resgid = arg;
1452                 return 1;
1453         case Opt_abort:
1454                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1455                 return 1;
1456         case Opt_i_version:
1457                 sb->s_flags |= MS_I_VERSION;
1458                 return 1;
1459         case Opt_journal_dev:
1460                 if (is_remount) {
1461                         ext4_msg(sb, KERN_ERR,
1462                                  "Cannot specify journal on remount");
1463                         return -1;
1464                 }
1465                 *journal_devnum = arg;
1466                 return 1;
1467         case Opt_journal_ioprio:
1468                 if (arg < 0 || arg > 7)
1469                         return -1;
1470                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1471                 return 1;
1472         }
1473
1474         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1475                 if (token != m->token)
1476                         continue;
1477                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1478                         return -1;
1479                 if (m->flags & MOPT_EXPLICIT)
1480                         set_opt2(sb, EXPLICIT_DELALLOC);
1481                 if (m->flags & MOPT_CLEAR_ERR)
1482                         clear_opt(sb, ERRORS_MASK);
1483                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1484                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1485                                  "options when quota turned on");
1486                         return -1;
1487                 }
1488
1489                 if (m->flags & MOPT_NOSUPPORT) {
1490                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1491                 } else if (token == Opt_commit) {
1492                         if (arg == 0)
1493                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1494                         sbi->s_commit_interval = HZ * arg;
1495                 } else if (token == Opt_max_batch_time) {
1496                         if (arg == 0)
1497                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1498                         sbi->s_max_batch_time = arg;
1499                 } else if (token == Opt_min_batch_time) {
1500                         sbi->s_min_batch_time = arg;
1501                 } else if (token == Opt_inode_readahead_blks) {
1502                         if (arg > (1 << 30))
1503                                 return -1;
1504                         if (arg && !is_power_of_2(arg)) {
1505                                 ext4_msg(sb, KERN_ERR,
1506                                          "EXT4-fs: inode_readahead_blks"
1507                                          " must be a power of 2");
1508                                 return -1;
1509                         }
1510                         sbi->s_inode_readahead_blks = arg;
1511                 } else if (token == Opt_init_itable) {
1512                         set_opt(sb, INIT_INODE_TABLE);
1513                         if (!args->from)
1514                                 arg = EXT4_DEF_LI_WAIT_MULT;
1515                         sbi->s_li_wait_mult = arg;
1516                 } else if (token == Opt_stripe) {
1517                         sbi->s_stripe = arg;
1518                 } else if (m->flags & MOPT_DATAJ) {
1519                         if (is_remount) {
1520                                 if (!sbi->s_journal)
1521                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1522                                 else if (test_opt(sb, DATA_FLAGS) !=
1523                                          m->mount_opt) {
1524                                         ext4_msg(sb, KERN_ERR,
1525                                          "Cannot change data mode on remount");
1526                                         return -1;
1527                                 }
1528                         } else {
1529                                 clear_opt(sb, DATA_FLAGS);
1530                                 sbi->s_mount_opt |= m->mount_opt;
1531                         }
1532 #ifdef CONFIG_QUOTA
1533                 } else if (token == Opt_usrjquota) {
1534                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1535                                 return -1;
1536                 } else if (token == Opt_grpjquota) {
1537                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1538                                 return -1;
1539                 } else if (token == Opt_offusrjquota) {
1540                         if (!clear_qf_name(sb, USRQUOTA))
1541                                 return -1;
1542                 } else if (token == Opt_offgrpjquota) {
1543                         if (!clear_qf_name(sb, GRPQUOTA))
1544                                 return -1;
1545                 } else if (m->flags & MOPT_QFMT) {
1546                         if (sb_any_quota_loaded(sb) &&
1547                             sbi->s_jquota_fmt != m->mount_opt) {
1548                                 ext4_msg(sb, KERN_ERR, "Cannot "
1549                                          "change journaled quota options "
1550                                          "when quota turned on");
1551                                 return -1;
1552                         }
1553                         sbi->s_jquota_fmt = m->mount_opt;
1554 #endif
1555                 } else {
1556                         if (!args->from)
1557                                 arg = 1;
1558                         if (m->flags & MOPT_CLEAR)
1559                                 arg = !arg;
1560                         else if (unlikely(!(m->flags & MOPT_SET))) {
1561                                 ext4_msg(sb, KERN_WARNING,
1562                                          "buggy handling of option %s", opt);
1563                                 WARN_ON(1);
1564                                 return -1;
1565                         }
1566                         if (arg != 0)
1567                                 sbi->s_mount_opt |= m->mount_opt;
1568                         else
1569                                 sbi->s_mount_opt &= ~m->mount_opt;
1570                 }
1571                 return 1;
1572         }
1573         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1574                  "or missing value", opt);
1575         return -1;
1576 }
1577
1578 static int parse_options(char *options, struct super_block *sb,
1579                          unsigned long *journal_devnum,
1580                          unsigned int *journal_ioprio,
1581                          int is_remount)
1582 {
1583         struct ext4_sb_info *sbi = EXT4_SB(sb);
1584         char *p;
1585         substring_t args[MAX_OPT_ARGS];
1586         int token;
1587
1588         if (!options)
1589                 return 1;
1590
1591         while ((p = strsep(&options, ",")) != NULL) {
1592                 if (!*p)
1593                         continue;
1594                 /*
1595                  * Initialize args struct so we know whether arg was
1596                  * found; some options take optional arguments.
1597                  */
1598                 args[0].to = args[0].from = 0;
1599                 token = match_token(p, tokens, args);
1600                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1601                                      journal_ioprio, is_remount) < 0)
1602                         return 0;
1603         }
1604 #ifdef CONFIG_QUOTA
1605         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1606                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1607                         clear_opt(sb, USRQUOTA);
1608
1609                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1610                         clear_opt(sb, GRPQUOTA);
1611
1612                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1613                         ext4_msg(sb, KERN_ERR, "old and new quota "
1614                                         "format mixing");
1615                         return 0;
1616                 }
1617
1618                 if (!sbi->s_jquota_fmt) {
1619                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1620                                         "not specified");
1621                         return 0;
1622                 }
1623         } else {
1624                 if (sbi->s_jquota_fmt) {
1625                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1626                                         "specified with no journaling "
1627                                         "enabled");
1628                         return 0;
1629                 }
1630         }
1631 #endif
1632         return 1;
1633 }
1634
1635 static inline void ext4_show_quota_options(struct seq_file *seq,
1636                                            struct super_block *sb)
1637 {
1638 #if defined(CONFIG_QUOTA)
1639         struct ext4_sb_info *sbi = EXT4_SB(sb);
1640
1641         if (sbi->s_jquota_fmt) {
1642                 char *fmtname = "";
1643
1644                 switch (sbi->s_jquota_fmt) {
1645                 case QFMT_VFS_OLD:
1646                         fmtname = "vfsold";
1647                         break;
1648                 case QFMT_VFS_V0:
1649                         fmtname = "vfsv0";
1650                         break;
1651                 case QFMT_VFS_V1:
1652                         fmtname = "vfsv1";
1653                         break;
1654                 }
1655                 seq_printf(seq, ",jqfmt=%s", fmtname);
1656         }
1657
1658         if (sbi->s_qf_names[USRQUOTA])
1659                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1660
1661         if (sbi->s_qf_names[GRPQUOTA])
1662                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1663
1664         if (test_opt(sb, USRQUOTA))
1665                 seq_puts(seq, ",usrquota");
1666
1667         if (test_opt(sb, GRPQUOTA))
1668                 seq_puts(seq, ",grpquota");
1669 #endif
1670 }
1671
1672 static const char *token2str(int token)
1673 {
1674         static const struct match_token *t;
1675
1676         for (t = tokens; t->token != Opt_err; t++)
1677                 if (t->token == token && !strchr(t->pattern, '='))
1678                         break;
1679         return t->pattern;
1680 }
1681
1682 /*
1683  * Show an option if
1684  *  - it's set to a non-default value OR
1685  *  - if the per-sb default is different from the global default
1686  */
1687 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1688                               int nodefs)
1689 {
1690         struct ext4_sb_info *sbi = EXT4_SB(sb);
1691         struct ext4_super_block *es = sbi->s_es;
1692         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1693         const struct mount_opts *m;
1694         char sep = nodefs ? '\n' : ',';
1695
1696 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1697 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1698
1699         if (sbi->s_sb_block != 1)
1700                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1701
1702         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1703                 int want_set = m->flags & MOPT_SET;
1704                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1705                     (m->flags & MOPT_CLEAR_ERR))
1706                         continue;
1707                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1708                         continue; /* skip if same as the default */
1709                 if ((want_set &&
1710                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1711                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1712                         continue; /* select Opt_noFoo vs Opt_Foo */
1713                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1714         }
1715
1716         if (nodefs || sbi->s_resuid != EXT4_DEF_RESUID ||
1717             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1718                 SEQ_OPTS_PRINT("resuid=%u", sbi->s_resuid);
1719         if (nodefs || sbi->s_resgid != EXT4_DEF_RESGID ||
1720             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1721                 SEQ_OPTS_PRINT("resgid=%u", sbi->s_resgid);
1722         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1723         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1724                 SEQ_OPTS_PUTS("errors=remount-ro");
1725         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1726                 SEQ_OPTS_PUTS("errors=continue");
1727         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1728                 SEQ_OPTS_PUTS("errors=panic");
1729         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1730                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1731         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1732                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1733         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1734                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1735         if (sb->s_flags & MS_I_VERSION)
1736                 SEQ_OPTS_PUTS("i_version");
1737         if (nodefs || sbi->s_stripe)
1738                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1739         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1740                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1741                         SEQ_OPTS_PUTS("data=journal");
1742                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1743                         SEQ_OPTS_PUTS("data=ordered");
1744                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1745                         SEQ_OPTS_PUTS("data=writeback");
1746         }
1747         if (nodefs ||
1748             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1749                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1750                                sbi->s_inode_readahead_blks);
1751
1752         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1753                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1754                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1755
1756         ext4_show_quota_options(seq, sb);
1757         return 0;
1758 }
1759
1760 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1761 {
1762         return _ext4_show_options(seq, root->d_sb, 0);
1763 }
1764
1765 static int options_seq_show(struct seq_file *seq, void *offset)
1766 {
1767         struct super_block *sb = seq->private;
1768         int rc;
1769
1770         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1771         rc = _ext4_show_options(seq, sb, 1);
1772         seq_puts(seq, "\n");
1773         return rc;
1774 }
1775
1776 static int options_open_fs(struct inode *inode, struct file *file)
1777 {
1778         return single_open(file, options_seq_show, PDE(inode)->data);
1779 }
1780
1781 static const struct file_operations ext4_seq_options_fops = {
1782         .owner = THIS_MODULE,
1783         .open = options_open_fs,
1784         .read = seq_read,
1785         .llseek = seq_lseek,
1786         .release = single_release,
1787 };
1788
1789 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1790                             int read_only)
1791 {
1792         struct ext4_sb_info *sbi = EXT4_SB(sb);
1793         int res = 0;
1794
1795         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1796                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1797                          "forcing read-only mode");
1798                 res = MS_RDONLY;
1799         }
1800         if (read_only)
1801                 goto done;
1802         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1803                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1804                          "running e2fsck is recommended");
1805         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1806                 ext4_msg(sb, KERN_WARNING,
1807                          "warning: mounting fs with errors, "
1808                          "running e2fsck is recommended");
1809         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1810                  le16_to_cpu(es->s_mnt_count) >=
1811                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1812                 ext4_msg(sb, KERN_WARNING,
1813                          "warning: maximal mount count reached, "
1814                          "running e2fsck is recommended");
1815         else if (le32_to_cpu(es->s_checkinterval) &&
1816                 (le32_to_cpu(es->s_lastcheck) +
1817                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1818                 ext4_msg(sb, KERN_WARNING,
1819                          "warning: checktime reached, "
1820                          "running e2fsck is recommended");
1821         if (!sbi->s_journal)
1822                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1823         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1824                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1825         le16_add_cpu(&es->s_mnt_count, 1);
1826         es->s_mtime = cpu_to_le32(get_seconds());
1827         ext4_update_dynamic_rev(sb);
1828         if (sbi->s_journal)
1829                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1830
1831         ext4_commit_super(sb, 1);
1832 done:
1833         if (test_opt(sb, DEBUG))
1834                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1835                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1836                         sb->s_blocksize,
1837                         sbi->s_groups_count,
1838                         EXT4_BLOCKS_PER_GROUP(sb),
1839                         EXT4_INODES_PER_GROUP(sb),
1840                         sbi->s_mount_opt, sbi->s_mount_opt2);
1841
1842         cleancache_init_fs(sb);
1843         return res;
1844 }
1845
1846 static int ext4_fill_flex_info(struct super_block *sb)
1847 {
1848         struct ext4_sb_info *sbi = EXT4_SB(sb);
1849         struct ext4_group_desc *gdp = NULL;
1850         ext4_group_t flex_group_count;
1851         ext4_group_t flex_group;
1852         unsigned int groups_per_flex = 0;
1853         size_t size;
1854         int i;
1855
1856         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1857         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1858                 sbi->s_log_groups_per_flex = 0;
1859                 return 1;
1860         }
1861         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1862
1863         /* We allocate both existing and potentially added groups */
1864         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1865                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1866                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1867         size = flex_group_count * sizeof(struct flex_groups);
1868         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1869         if (sbi->s_flex_groups == NULL) {
1870                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1871                          flex_group_count);
1872                 goto failed;
1873         }
1874
1875         for (i = 0; i < sbi->s_groups_count; i++) {
1876                 gdp = ext4_get_group_desc(sb, i, NULL);
1877
1878                 flex_group = ext4_flex_group(sbi, i);
1879                 atomic_add(ext4_free_inodes_count(sb, gdp),
1880                            &sbi->s_flex_groups[flex_group].free_inodes);
1881                 atomic_add(ext4_free_group_clusters(sb, gdp),
1882                            &sbi->s_flex_groups[flex_group].free_clusters);
1883                 atomic_add(ext4_used_dirs_count(sb, gdp),
1884                            &sbi->s_flex_groups[flex_group].used_dirs);
1885         }
1886
1887         return 1;
1888 failed:
1889         return 0;
1890 }
1891
1892 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1893                             struct ext4_group_desc *gdp)
1894 {
1895         __u16 crc = 0;
1896
1897         if (sbi->s_es->s_feature_ro_compat &
1898             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1899                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1900                 __le32 le_group = cpu_to_le32(block_group);
1901
1902                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1903                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1904                 crc = crc16(crc, (__u8 *)gdp, offset);
1905                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1906                 /* for checksum of struct ext4_group_desc do the rest...*/
1907                 if ((sbi->s_es->s_feature_incompat &
1908                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1909                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1910                         crc = crc16(crc, (__u8 *)gdp + offset,
1911                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1912                                         offset);
1913         }
1914
1915         return cpu_to_le16(crc);
1916 }
1917
1918 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1919                                 struct ext4_group_desc *gdp)
1920 {
1921         if ((sbi->s_es->s_feature_ro_compat &
1922              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1923             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1924                 return 0;
1925
1926         return 1;
1927 }
1928
1929 /* Called at mount-time, super-block is locked */
1930 static int ext4_check_descriptors(struct super_block *sb,
1931                                   ext4_group_t *first_not_zeroed)
1932 {
1933         struct ext4_sb_info *sbi = EXT4_SB(sb);
1934         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1935         ext4_fsblk_t last_block;
1936         ext4_fsblk_t block_bitmap;
1937         ext4_fsblk_t inode_bitmap;
1938         ext4_fsblk_t inode_table;
1939         int flexbg_flag = 0;
1940         ext4_group_t i, grp = sbi->s_groups_count;
1941
1942         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1943                 flexbg_flag = 1;
1944
1945         ext4_debug("Checking group descriptors");
1946
1947         for (i = 0; i < sbi->s_groups_count; i++) {
1948                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1949
1950                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1951                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1952                 else
1953                         last_block = first_block +
1954                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1955
1956                 if ((grp == sbi->s_groups_count) &&
1957                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
1958                         grp = i;
1959
1960                 block_bitmap = ext4_block_bitmap(sb, gdp);
1961                 if (block_bitmap < first_block || block_bitmap > last_block) {
1962                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1963                                "Block bitmap for group %u not in group "
1964                                "(block %llu)!", i, block_bitmap);
1965                         return 0;
1966                 }
1967                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1968                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1969                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1970                                "Inode bitmap for group %u not in group "
1971                                "(block %llu)!", i, inode_bitmap);
1972                         return 0;
1973                 }
1974                 inode_table = ext4_inode_table(sb, gdp);
1975                 if (inode_table < first_block ||
1976                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1977                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1978                                "Inode table for group %u not in group "
1979                                "(block %llu)!", i, inode_table);
1980                         return 0;
1981                 }
1982                 ext4_lock_group(sb, i);
1983                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
1984                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1985                                  "Checksum for group %u failed (%u!=%u)",
1986                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
1987                                      gdp)), le16_to_cpu(gdp->bg_checksum));
1988                         if (!(sb->s_flags & MS_RDONLY)) {
1989                                 ext4_unlock_group(sb, i);
1990                                 return 0;
1991                         }
1992                 }
1993                 ext4_unlock_group(sb, i);
1994                 if (!flexbg_flag)
1995                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
1996         }
1997         if (NULL != first_not_zeroed)
1998                 *first_not_zeroed = grp;
1999
2000         ext4_free_blocks_count_set(sbi->s_es,
2001                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2002         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2003         return 1;
2004 }
2005
2006 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2007  * the superblock) which were deleted from all directories, but held open by
2008  * a process at the time of a crash.  We walk the list and try to delete these
2009  * inodes at recovery time (only with a read-write filesystem).
2010  *
2011  * In order to keep the orphan inode chain consistent during traversal (in
2012  * case of crash during recovery), we link each inode into the superblock
2013  * orphan list_head and handle it the same way as an inode deletion during
2014  * normal operation (which journals the operations for us).
2015  *
2016  * We only do an iget() and an iput() on each inode, which is very safe if we
2017  * accidentally point at an in-use or already deleted inode.  The worst that
2018  * can happen in this case is that we get a "bit already cleared" message from
2019  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2020  * e2fsck was run on this filesystem, and it must have already done the orphan
2021  * inode cleanup for us, so we can safely abort without any further action.
2022  */
2023 static void ext4_orphan_cleanup(struct super_block *sb,
2024                                 struct ext4_super_block *es)
2025 {
2026         unsigned int s_flags = sb->s_flags;
2027         int nr_orphans = 0, nr_truncates = 0;
2028 #ifdef CONFIG_QUOTA
2029         int i;
2030 #endif
2031         if (!es->s_last_orphan) {
2032                 jbd_debug(4, "no orphan inodes to clean up\n");
2033                 return;
2034         }
2035
2036         if (bdev_read_only(sb->s_bdev)) {
2037                 ext4_msg(sb, KERN_ERR, "write access "
2038                         "unavailable, skipping orphan cleanup");
2039                 return;
2040         }
2041
2042         /* Check if feature set would not allow a r/w mount */
2043         if (!ext4_feature_set_ok(sb, 0)) {
2044                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2045                          "unknown ROCOMPAT features");
2046                 return;
2047         }
2048
2049         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2050                 if (es->s_last_orphan)
2051                         jbd_debug(1, "Errors on filesystem, "
2052                                   "clearing orphan list.\n");
2053                 es->s_last_orphan = 0;
2054                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2055                 return;
2056         }
2057
2058         if (s_flags & MS_RDONLY) {
2059                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2060                 sb->s_flags &= ~MS_RDONLY;
2061         }
2062 #ifdef CONFIG_QUOTA
2063         /* Needed for iput() to work correctly and not trash data */
2064         sb->s_flags |= MS_ACTIVE;
2065         /* Turn on quotas so that they are updated correctly */
2066         for (i = 0; i < MAXQUOTAS; i++) {
2067                 if (EXT4_SB(sb)->s_qf_names[i]) {
2068                         int ret = ext4_quota_on_mount(sb, i);
2069                         if (ret < 0)
2070                                 ext4_msg(sb, KERN_ERR,
2071                                         "Cannot turn on journaled "
2072                                         "quota: error %d", ret);
2073                 }
2074         }
2075 #endif
2076
2077         while (es->s_last_orphan) {
2078                 struct inode *inode;
2079
2080                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2081                 if (IS_ERR(inode)) {
2082                         es->s_last_orphan = 0;
2083                         break;
2084                 }
2085
2086                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2087                 dquot_initialize(inode);
2088                 if (inode->i_nlink) {
2089                         ext4_msg(sb, KERN_DEBUG,
2090                                 "%s: truncating inode %lu to %lld bytes",
2091                                 __func__, inode->i_ino, inode->i_size);
2092                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2093                                   inode->i_ino, inode->i_size);
2094                         ext4_truncate(inode);
2095                         nr_truncates++;
2096                 } else {
2097                         ext4_msg(sb, KERN_DEBUG,
2098                                 "%s: deleting unreferenced inode %lu",
2099                                 __func__, inode->i_ino);
2100                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2101                                   inode->i_ino);
2102                         nr_orphans++;
2103                 }
2104                 iput(inode);  /* The delete magic happens here! */
2105         }
2106
2107 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2108
2109         if (nr_orphans)
2110                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2111                        PLURAL(nr_orphans));
2112         if (nr_truncates)
2113                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2114                        PLURAL(nr_truncates));
2115 #ifdef CONFIG_QUOTA
2116         /* Turn quotas off */
2117         for (i = 0; i < MAXQUOTAS; i++) {
2118                 if (sb_dqopt(sb)->files[i])
2119                         dquot_quota_off(sb, i);
2120         }
2121 #endif
2122         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2123 }
2124
2125 /*
2126  * Maximal extent format file size.
2127  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2128  * extent format containers, within a sector_t, and within i_blocks
2129  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2130  * so that won't be a limiting factor.
2131  *
2132  * However there is other limiting factor. We do store extents in the form
2133  * of starting block and length, hence the resulting length of the extent
2134  * covering maximum file size must fit into on-disk format containers as
2135  * well. Given that length is always by 1 unit bigger than max unit (because
2136  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2137  *
2138  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2139  */
2140 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2141 {
2142         loff_t res;
2143         loff_t upper_limit = MAX_LFS_FILESIZE;
2144
2145         /* small i_blocks in vfs inode? */
2146         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2147                 /*
2148                  * CONFIG_LBDAF is not enabled implies the inode
2149                  * i_block represent total blocks in 512 bytes
2150                  * 32 == size of vfs inode i_blocks * 8
2151                  */
2152                 upper_limit = (1LL << 32) - 1;
2153
2154                 /* total blocks in file system block size */
2155                 upper_limit >>= (blkbits - 9);
2156                 upper_limit <<= blkbits;
2157         }
2158
2159         /*
2160          * 32-bit extent-start container, ee_block. We lower the maxbytes
2161          * by one fs block, so ee_len can cover the extent of maximum file
2162          * size
2163          */
2164         res = (1LL << 32) - 1;
2165         res <<= blkbits;
2166
2167         /* Sanity check against vm- & vfs- imposed limits */
2168         if (res > upper_limit)
2169                 res = upper_limit;
2170
2171         return res;
2172 }
2173
2174 /*
2175  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2176  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2177  * We need to be 1 filesystem block less than the 2^48 sector limit.
2178  */
2179 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2180 {
2181         loff_t res = EXT4_NDIR_BLOCKS;
2182         int meta_blocks;
2183         loff_t upper_limit;
2184         /* This is calculated to be the largest file size for a dense, block
2185          * mapped file such that the file's total number of 512-byte sectors,
2186          * including data and all indirect blocks, does not exceed (2^48 - 1).
2187          *
2188          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2189          * number of 512-byte sectors of the file.
2190          */
2191
2192         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2193                 /*
2194                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2195                  * the inode i_block field represents total file blocks in
2196                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2197                  */
2198                 upper_limit = (1LL << 32) - 1;
2199
2200                 /* total blocks in file system block size */
2201                 upper_limit >>= (bits - 9);
2202
2203         } else {
2204                 /*
2205                  * We use 48 bit ext4_inode i_blocks
2206                  * With EXT4_HUGE_FILE_FL set the i_blocks
2207                  * represent total number of blocks in
2208                  * file system block size
2209                  */
2210                 upper_limit = (1LL << 48) - 1;
2211
2212         }
2213
2214         /* indirect blocks */
2215         meta_blocks = 1;
2216         /* double indirect blocks */
2217         meta_blocks += 1 + (1LL << (bits-2));
2218         /* tripple indirect blocks */
2219         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2220
2221         upper_limit -= meta_blocks;
2222         upper_limit <<= bits;
2223
2224         res += 1LL << (bits-2);
2225         res += 1LL << (2*(bits-2));
2226         res += 1LL << (3*(bits-2));
2227         res <<= bits;
2228         if (res > upper_limit)
2229                 res = upper_limit;
2230
2231         if (res > MAX_LFS_FILESIZE)
2232                 res = MAX_LFS_FILESIZE;
2233
2234         return res;
2235 }
2236
2237 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2238                                    ext4_fsblk_t logical_sb_block, int nr)
2239 {
2240         struct ext4_sb_info *sbi = EXT4_SB(sb);
2241         ext4_group_t bg, first_meta_bg;
2242         int has_super = 0;
2243
2244         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2245
2246         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2247             nr < first_meta_bg)
2248                 return logical_sb_block + nr + 1;
2249         bg = sbi->s_desc_per_block * nr;
2250         if (ext4_bg_has_super(sb, bg))
2251                 has_super = 1;
2252
2253         return (has_super + ext4_group_first_block_no(sb, bg));
2254 }
2255
2256 /**
2257  * ext4_get_stripe_size: Get the stripe size.
2258  * @sbi: In memory super block info
2259  *
2260  * If we have specified it via mount option, then
2261  * use the mount option value. If the value specified at mount time is
2262  * greater than the blocks per group use the super block value.
2263  * If the super block value is greater than blocks per group return 0.
2264  * Allocator needs it be less than blocks per group.
2265  *
2266  */
2267 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2268 {
2269         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2270         unsigned long stripe_width =
2271                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2272         int ret;
2273
2274         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2275                 ret = sbi->s_stripe;
2276         else if (stripe_width <= sbi->s_blocks_per_group)
2277                 ret = stripe_width;
2278         else if (stride <= sbi->s_blocks_per_group)
2279                 ret = stride;
2280         else
2281                 ret = 0;
2282
2283         /*
2284          * If the stripe width is 1, this makes no sense and
2285          * we set it to 0 to turn off stripe handling code.
2286          */
2287         if (ret <= 1)
2288                 ret = 0;
2289
2290         return ret;
2291 }
2292
2293 /* sysfs supprt */
2294
2295 struct ext4_attr {
2296         struct attribute attr;
2297         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2298         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2299                          const char *, size_t);
2300         int offset;
2301 };
2302
2303 static int parse_strtoul(const char *buf,
2304                 unsigned long max, unsigned long *value)
2305 {
2306         char *endp;
2307
2308         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2309         endp = skip_spaces(endp);
2310         if (*endp || *value > max)
2311                 return -EINVAL;
2312
2313         return 0;
2314 }
2315
2316 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2317                                               struct ext4_sb_info *sbi,
2318                                               char *buf)
2319 {
2320         return snprintf(buf, PAGE_SIZE, "%llu\n",
2321                 (s64) EXT4_C2B(sbi,
2322                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2323 }
2324
2325 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2326                                          struct ext4_sb_info *sbi, char *buf)
2327 {
2328         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2329
2330         if (!sb->s_bdev->bd_part)
2331                 return snprintf(buf, PAGE_SIZE, "0\n");
2332         return snprintf(buf, PAGE_SIZE, "%lu\n",
2333                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2334                          sbi->s_sectors_written_start) >> 1);
2335 }
2336
2337 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2338                                           struct ext4_sb_info *sbi, char *buf)
2339 {
2340         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2341
2342         if (!sb->s_bdev->bd_part)
2343                 return snprintf(buf, PAGE_SIZE, "0\n");
2344         return snprintf(buf, PAGE_SIZE, "%llu\n",
2345                         (unsigned long long)(sbi->s_kbytes_written +
2346                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2347                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2348 }
2349
2350 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2351                                       struct ext4_sb_info *sbi, char *buf)
2352 {
2353         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2354 }
2355
2356 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2357                                         struct ext4_sb_info *sbi, char *buf)
2358 {
2359         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2360 }
2361
2362 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2363                                           struct ext4_sb_info *sbi,
2364                                           const char *buf, size_t count)
2365 {
2366         unsigned long t;
2367
2368         if (parse_strtoul(buf, 0x40000000, &t))
2369                 return -EINVAL;
2370
2371         if (t && !is_power_of_2(t))
2372                 return -EINVAL;
2373
2374         sbi->s_inode_readahead_blks = t;
2375         return count;
2376 }
2377
2378 static ssize_t sbi_ui_show(struct ext4_attr *a,
2379                            struct ext4_sb_info *sbi, char *buf)
2380 {
2381         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2382
2383         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2384 }
2385
2386 static ssize_t sbi_ui_store(struct ext4_attr *a,
2387                             struct ext4_sb_info *sbi,
2388                             const char *buf, size_t count)
2389 {
2390         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2391         unsigned long t;
2392
2393         if (parse_strtoul(buf, 0xffffffff, &t))
2394                 return -EINVAL;
2395         *ui = t;
2396         return count;
2397 }
2398
2399 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2400 static struct ext4_attr ext4_attr_##_name = {                   \
2401         .attr = {.name = __stringify(_name), .mode = _mode },   \
2402         .show   = _show,                                        \
2403         .store  = _store,                                       \
2404         .offset = offsetof(struct ext4_sb_info, _elname),       \
2405 }
2406 #define EXT4_ATTR(name, mode, show, store) \
2407 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2408
2409 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2410 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2411 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2412 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2413         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2414 #define ATTR_LIST(name) &ext4_attr_##name.attr
2415
2416 EXT4_RO_ATTR(delayed_allocation_blocks);
2417 EXT4_RO_ATTR(session_write_kbytes);
2418 EXT4_RO_ATTR(lifetime_write_kbytes);
2419 EXT4_RO_ATTR(extent_cache_hits);
2420 EXT4_RO_ATTR(extent_cache_misses);
2421 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2422                  inode_readahead_blks_store, s_inode_readahead_blks);
2423 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2424 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2425 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2426 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2427 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2428 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2429 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2430 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2431
2432 static struct attribute *ext4_attrs[] = {
2433         ATTR_LIST(delayed_allocation_blocks),
2434         ATTR_LIST(session_write_kbytes),
2435         ATTR_LIST(lifetime_write_kbytes),
2436         ATTR_LIST(extent_cache_hits),
2437         ATTR_LIST(extent_cache_misses),
2438         ATTR_LIST(inode_readahead_blks),
2439         ATTR_LIST(inode_goal),
2440         ATTR_LIST(mb_stats),
2441         ATTR_LIST(mb_max_to_scan),
2442         ATTR_LIST(mb_min_to_scan),
2443         ATTR_LIST(mb_order2_req),
2444         ATTR_LIST(mb_stream_req),
2445         ATTR_LIST(mb_group_prealloc),
2446         ATTR_LIST(max_writeback_mb_bump),
2447         NULL,
2448 };
2449
2450 /* Features this copy of ext4 supports */
2451 EXT4_INFO_ATTR(lazy_itable_init);
2452 EXT4_INFO_ATTR(batched_discard);
2453
2454 static struct attribute *ext4_feat_attrs[] = {
2455         ATTR_LIST(lazy_itable_init),
2456         ATTR_LIST(batched_discard),
2457         NULL,
2458 };
2459
2460 static ssize_t ext4_attr_show(struct kobject *kobj,
2461                               struct attribute *attr, char *buf)
2462 {
2463         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2464                                                 s_kobj);
2465         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2466
2467         return a->show ? a->show(a, sbi, buf) : 0;
2468 }
2469
2470 static ssize_t ext4_attr_store(struct kobject *kobj,
2471                                struct attribute *attr,
2472                                const char *buf, size_t len)
2473 {
2474         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2475                                                 s_kobj);
2476         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2477
2478         return a->store ? a->store(a, sbi, buf, len) : 0;
2479 }
2480
2481 static void ext4_sb_release(struct kobject *kobj)
2482 {
2483         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2484                                                 s_kobj);
2485         complete(&sbi->s_kobj_unregister);
2486 }
2487
2488 static const struct sysfs_ops ext4_attr_ops = {
2489         .show   = ext4_attr_show,
2490         .store  = ext4_attr_store,
2491 };
2492
2493 static struct kobj_type ext4_ktype = {
2494         .default_attrs  = ext4_attrs,
2495         .sysfs_ops      = &ext4_attr_ops,
2496         .release        = ext4_sb_release,
2497 };
2498
2499 static void ext4_feat_release(struct kobject *kobj)
2500 {
2501         complete(&ext4_feat->f_kobj_unregister);
2502 }
2503
2504 static struct kobj_type ext4_feat_ktype = {
2505         .default_attrs  = ext4_feat_attrs,
2506         .sysfs_ops      = &ext4_attr_ops,
2507         .release        = ext4_feat_release,
2508 };
2509
2510 /*
2511  * Check whether this filesystem can be mounted based on
2512  * the features present and the RDONLY/RDWR mount requested.
2513  * Returns 1 if this filesystem can be mounted as requested,
2514  * 0 if it cannot be.
2515  */
2516 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2517 {
2518         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2519                 ext4_msg(sb, KERN_ERR,
2520                         "Couldn't mount because of "
2521                         "unsupported optional features (%x)",
2522                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2523                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2524                 return 0;
2525         }
2526
2527         if (readonly)
2528                 return 1;
2529
2530         /* Check that feature set is OK for a read-write mount */
2531         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2532                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2533                          "unsupported optional features (%x)",
2534                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2535                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2536                 return 0;
2537         }
2538         /*
2539          * Large file size enabled file system can only be mounted
2540          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2541          */
2542         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2543                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2544                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2545                                  "cannot be mounted RDWR without "
2546                                  "CONFIG_LBDAF");
2547                         return 0;
2548                 }
2549         }
2550         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2551             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2552                 ext4_msg(sb, KERN_ERR,
2553                          "Can't support bigalloc feature without "
2554                          "extents feature\n");
2555                 return 0;
2556         }
2557         return 1;
2558 }
2559
2560 /*
2561  * This function is called once a day if we have errors logged
2562  * on the file system
2563  */
2564 static void print_daily_error_info(unsigned long arg)
2565 {
2566         struct super_block *sb = (struct super_block *) arg;
2567         struct ext4_sb_info *sbi;
2568         struct ext4_super_block *es;
2569
2570         sbi = EXT4_SB(sb);
2571         es = sbi->s_es;
2572
2573         if (es->s_error_count)
2574                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2575                          le32_to_cpu(es->s_error_count));
2576         if (es->s_first_error_time) {
2577                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2578                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2579                        (int) sizeof(es->s_first_error_func),
2580                        es->s_first_error_func,
2581                        le32_to_cpu(es->s_first_error_line));
2582                 if (es->s_first_error_ino)
2583                         printk(": inode %u",
2584                                le32_to_cpu(es->s_first_error_ino));
2585                 if (es->s_first_error_block)
2586                         printk(": block %llu", (unsigned long long)
2587                                le64_to_cpu(es->s_first_error_block));
2588                 printk("\n");
2589         }
2590         if (es->s_last_error_time) {
2591                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2592                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2593                        (int) sizeof(es->s_last_error_func),
2594                        es->s_last_error_func,
2595                        le32_to_cpu(es->s_last_error_line));
2596                 if (es->s_last_error_ino)
2597                         printk(": inode %u",
2598                                le32_to_cpu(es->s_last_error_ino));
2599                 if (es->s_last_error_block)
2600                         printk(": block %llu", (unsigned long long)
2601                                le64_to_cpu(es->s_last_error_block));
2602                 printk("\n");
2603         }
2604         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2605 }
2606
2607 /* Find next suitable group and run ext4_init_inode_table */
2608 static int ext4_run_li_request(struct ext4_li_request *elr)
2609 {
2610         struct ext4_group_desc *gdp = NULL;
2611         ext4_group_t group, ngroups;
2612         struct super_block *sb;
2613         unsigned long timeout = 0;
2614         int ret = 0;
2615
2616         sb = elr->lr_super;
2617         ngroups = EXT4_SB(sb)->s_groups_count;
2618
2619         for (group = elr->lr_next_group; group < ngroups; group++) {
2620                 gdp = ext4_get_group_desc(sb, group, NULL);
2621                 if (!gdp) {
2622                         ret = 1;
2623                         break;
2624                 }
2625
2626                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2627                         break;
2628         }
2629
2630         if (group == ngroups)
2631                 ret = 1;
2632
2633         if (!ret) {
2634                 timeout = jiffies;
2635                 ret = ext4_init_inode_table(sb, group,
2636                                             elr->lr_timeout ? 0 : 1);
2637                 if (elr->lr_timeout == 0) {
2638                         timeout = (jiffies - timeout) *
2639                                   elr->lr_sbi->s_li_wait_mult;
2640                         elr->lr_timeout = timeout;
2641                 }
2642                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2643                 elr->lr_next_group = group + 1;
2644         }
2645
2646         return ret;
2647 }
2648
2649 /*
2650  * Remove lr_request from the list_request and free the
2651  * request structure. Should be called with li_list_mtx held
2652  */
2653 static void ext4_remove_li_request(struct ext4_li_request *elr)
2654 {
2655         struct ext4_sb_info *sbi;
2656
2657         if (!elr)
2658                 return;
2659
2660         sbi = elr->lr_sbi;
2661
2662         list_del(&elr->lr_request);
2663         sbi->s_li_request = NULL;
2664         kfree(elr);
2665 }
2666
2667 static void ext4_unregister_li_request(struct super_block *sb)
2668 {
2669         mutex_lock(&ext4_li_mtx);
2670         if (!ext4_li_info) {
2671                 mutex_unlock(&ext4_li_mtx);
2672                 return;
2673         }
2674
2675         mutex_lock(&ext4_li_info->li_list_mtx);
2676         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2677         mutex_unlock(&ext4_li_info->li_list_mtx);
2678         mutex_unlock(&ext4_li_mtx);
2679 }
2680
2681 static struct task_struct *ext4_lazyinit_task;
2682
2683 /*
2684  * This is the function where ext4lazyinit thread lives. It walks
2685  * through the request list searching for next scheduled filesystem.
2686  * When such a fs is found, run the lazy initialization request
2687  * (ext4_rn_li_request) and keep track of the time spend in this
2688  * function. Based on that time we compute next schedule time of
2689  * the request. When walking through the list is complete, compute
2690  * next waking time and put itself into sleep.
2691  */
2692 static int ext4_lazyinit_thread(void *arg)
2693 {
2694         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2695         struct list_head *pos, *n;
2696         struct ext4_li_request *elr;
2697         unsigned long next_wakeup, cur;
2698
2699         BUG_ON(NULL == eli);
2700
2701 cont_thread:
2702         while (true) {
2703                 next_wakeup = MAX_JIFFY_OFFSET;
2704
2705                 mutex_lock(&eli->li_list_mtx);
2706                 if (list_empty(&eli->li_request_list)) {
2707                         mutex_unlock(&eli->li_list_mtx);
2708                         goto exit_thread;
2709                 }
2710
2711                 list_for_each_safe(pos, n, &eli->li_request_list) {
2712                         elr = list_entry(pos, struct ext4_li_request,
2713                                          lr_request);
2714
2715                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2716                                 if (ext4_run_li_request(elr) != 0) {
2717                                         /* error, remove the lazy_init job */
2718                                         ext4_remove_li_request(elr);
2719                                         continue;
2720                                 }
2721                         }
2722
2723                         if (time_before(elr->lr_next_sched, next_wakeup))
2724                                 next_wakeup = elr->lr_next_sched;
2725                 }
2726                 mutex_unlock(&eli->li_list_mtx);
2727
2728                 try_to_freeze();
2729
2730                 cur = jiffies;
2731                 if ((time_after_eq(cur, next_wakeup)) ||
2732                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2733                         cond_resched();
2734                         continue;
2735                 }
2736
2737                 schedule_timeout_interruptible(next_wakeup - cur);
2738
2739                 if (kthread_should_stop()) {
2740                         ext4_clear_request_list();
2741                         goto exit_thread;
2742                 }
2743         }
2744
2745 exit_thread:
2746         /*
2747          * It looks like the request list is empty, but we need
2748          * to check it under the li_list_mtx lock, to prevent any
2749          * additions into it, and of course we should lock ext4_li_mtx
2750          * to atomically free the list and ext4_li_info, because at
2751          * this point another ext4 filesystem could be registering
2752          * new one.
2753          */
2754         mutex_lock(&ext4_li_mtx);
2755         mutex_lock(&eli->li_list_mtx);
2756         if (!list_empty(&eli->li_request_list)) {
2757                 mutex_unlock(&eli->li_list_mtx);
2758                 mutex_unlock(&ext4_li_mtx);
2759                 goto cont_thread;
2760         }
2761         mutex_unlock(&eli->li_list_mtx);
2762         kfree(ext4_li_info);
2763         ext4_li_info = NULL;
2764         mutex_unlock(&ext4_li_mtx);
2765
2766         return 0;
2767 }
2768
2769 static void ext4_clear_request_list(void)
2770 {
2771         struct list_head *pos, *n;
2772         struct ext4_li_request *elr;
2773
2774         mutex_lock(&ext4_li_info->li_list_mtx);
2775         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2776                 elr = list_entry(pos, struct ext4_li_request,
2777                                  lr_request);
2778                 ext4_remove_li_request(elr);
2779         }
2780         mutex_unlock(&ext4_li_info->li_list_mtx);
2781 }
2782
2783 static int ext4_run_lazyinit_thread(void)
2784 {
2785         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2786                                          ext4_li_info, "ext4lazyinit");
2787         if (IS_ERR(ext4_lazyinit_task)) {
2788                 int err = PTR_ERR(ext4_lazyinit_task);
2789                 ext4_clear_request_list();
2790                 kfree(ext4_li_info);
2791                 ext4_li_info = NULL;
2792                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2793                                  "initialization thread\n",
2794                                  err);
2795                 return err;
2796         }
2797         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2798         return 0;
2799 }
2800
2801 /*
2802  * Check whether it make sense to run itable init. thread or not.
2803  * If there is at least one uninitialized inode table, return
2804  * corresponding group number, else the loop goes through all
2805  * groups and return total number of groups.
2806  */
2807 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2808 {
2809         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2810         struct ext4_group_desc *gdp = NULL;
2811
2812         for (group = 0; group < ngroups; group++) {
2813                 gdp = ext4_get_group_desc(sb, group, NULL);
2814                 if (!gdp)
2815                         continue;
2816
2817                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2818                         break;
2819         }
2820
2821         return group;
2822 }
2823
2824 static int ext4_li_info_new(void)
2825 {
2826         struct ext4_lazy_init *eli = NULL;
2827
2828         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2829         if (!eli)
2830                 return -ENOMEM;
2831
2832         INIT_LIST_HEAD(&eli->li_request_list);
2833         mutex_init(&eli->li_list_mtx);
2834
2835         eli->li_state |= EXT4_LAZYINIT_QUIT;
2836
2837         ext4_li_info = eli;
2838
2839         return 0;
2840 }
2841
2842 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2843                                             ext4_group_t start)
2844 {
2845         struct ext4_sb_info *sbi = EXT4_SB(sb);
2846         struct ext4_li_request *elr;
2847         unsigned long rnd;
2848
2849         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2850         if (!elr)
2851                 return NULL;
2852
2853         elr->lr_super = sb;
2854         elr->lr_sbi = sbi;
2855         elr->lr_next_group = start;
2856
2857         /*
2858          * Randomize first schedule time of the request to
2859          * spread the inode table initialization requests
2860          * better.
2861          */
2862         get_random_bytes(&rnd, sizeof(rnd));
2863         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2864                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2865
2866         return elr;
2867 }
2868
2869 static int ext4_register_li_request(struct super_block *sb,
2870                                     ext4_group_t first_not_zeroed)
2871 {
2872         struct ext4_sb_info *sbi = EXT4_SB(sb);
2873         struct ext4_li_request *elr;
2874         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2875         int ret = 0;
2876
2877         if (sbi->s_li_request != NULL) {
2878                 /*
2879                  * Reset timeout so it can be computed again, because
2880                  * s_li_wait_mult might have changed.
2881                  */
2882                 sbi->s_li_request->lr_timeout = 0;
2883                 return 0;
2884         }
2885
2886         if (first_not_zeroed == ngroups ||
2887             (sb->s_flags & MS_RDONLY) ||
2888             !test_opt(sb, INIT_INODE_TABLE))
2889                 return 0;
2890
2891         elr = ext4_li_request_new(sb, first_not_zeroed);
2892         if (!elr)
2893                 return -ENOMEM;
2894
2895         mutex_lock(&ext4_li_mtx);
2896
2897         if (NULL == ext4_li_info) {
2898                 ret = ext4_li_info_new();
2899                 if (ret)
2900                         goto out;
2901         }
2902
2903         mutex_lock(&ext4_li_info->li_list_mtx);
2904         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2905         mutex_unlock(&ext4_li_info->li_list_mtx);
2906
2907         sbi->s_li_request = elr;
2908         /*
2909          * set elr to NULL here since it has been inserted to
2910          * the request_list and the removal and free of it is
2911          * handled by ext4_clear_request_list from now on.
2912          */
2913         elr = NULL;
2914
2915         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2916                 ret = ext4_run_lazyinit_thread();
2917                 if (ret)
2918                         goto out;
2919         }
2920 out:
2921         mutex_unlock(&ext4_li_mtx);
2922         if (ret)
2923                 kfree(elr);
2924         return ret;
2925 }
2926
2927 /*
2928  * We do not need to lock anything since this is called on
2929  * module unload.
2930  */
2931 static void ext4_destroy_lazyinit_thread(void)
2932 {
2933         /*
2934          * If thread exited earlier
2935          * there's nothing to be done.
2936          */
2937         if (!ext4_li_info || !ext4_lazyinit_task)
2938                 return;
2939
2940         kthread_stop(ext4_lazyinit_task);
2941 }
2942
2943 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2944 {
2945         char *orig_data = kstrdup(data, GFP_KERNEL);
2946         struct buffer_head *bh;
2947         struct ext4_super_block *es = NULL;
2948         struct ext4_sb_info *sbi;
2949         ext4_fsblk_t block;
2950         ext4_fsblk_t sb_block = get_sb_block(&data);
2951         ext4_fsblk_t logical_sb_block;
2952         unsigned long offset = 0;
2953         unsigned long journal_devnum = 0;
2954         unsigned long def_mount_opts;
2955         struct inode *root;
2956         char *cp;
2957         const char *descr;
2958         int ret = -ENOMEM;
2959         int blocksize, clustersize;
2960         unsigned int db_count;
2961         unsigned int i;
2962         int needs_recovery, has_huge_files, has_bigalloc;
2963         __u64 blocks_count;
2964         int err;
2965         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2966         ext4_group_t first_not_zeroed;
2967
2968         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2969         if (!sbi)
2970                 goto out_free_orig;
2971
2972         sbi->s_blockgroup_lock =
2973                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2974         if (!sbi->s_blockgroup_lock) {
2975                 kfree(sbi);
2976                 goto out_free_orig;
2977         }
2978         sb->s_fs_info = sbi;
2979         sbi->s_mount_opt = 0;
2980         sbi->s_resuid = EXT4_DEF_RESUID;
2981         sbi->s_resgid = EXT4_DEF_RESGID;
2982         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
2983         sbi->s_sb_block = sb_block;
2984         if (sb->s_bdev->bd_part)
2985                 sbi->s_sectors_written_start =
2986                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
2987
2988         /* Cleanup superblock name */
2989         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
2990                 *cp = '!';
2991
2992         ret = -EINVAL;
2993         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
2994         if (!blocksize) {
2995                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
2996                 goto out_fail;
2997         }
2998
2999         /*
3000          * The ext4 superblock will not be buffer aligned for other than 1kB
3001          * block sizes.  We need to calculate the offset from buffer start.
3002          */
3003         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3004                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3005                 offset = do_div(logical_sb_block, blocksize);
3006         } else {
3007                 logical_sb_block = sb_block;
3008         }
3009
3010         if (!(bh = sb_bread(sb, logical_sb_block))) {
3011                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3012                 goto out_fail;
3013         }
3014         /*
3015          * Note: s_es must be initialized as soon as possible because
3016          *       some ext4 macro-instructions depend on its value
3017          */
3018         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3019         sbi->s_es = es;
3020         sb->s_magic = le16_to_cpu(es->s_magic);
3021         if (sb->s_magic != EXT4_SUPER_MAGIC)
3022                 goto cantfind_ext4;
3023         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3024
3025         /* Set defaults before we parse the mount options */
3026         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3027         set_opt(sb, INIT_INODE_TABLE);
3028         if (def_mount_opts & EXT4_DEFM_DEBUG)
3029                 set_opt(sb, DEBUG);
3030         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3031                 set_opt(sb, GRPID);
3032         if (def_mount_opts & EXT4_DEFM_UID16)
3033                 set_opt(sb, NO_UID32);
3034         /* xattr user namespace & acls are now defaulted on */
3035 #ifdef CONFIG_EXT4_FS_XATTR
3036         set_opt(sb, XATTR_USER);
3037 #endif
3038 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3039         set_opt(sb, POSIX_ACL);
3040 #endif
3041         set_opt(sb, MBLK_IO_SUBMIT);
3042         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3043                 set_opt(sb, JOURNAL_DATA);
3044         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3045                 set_opt(sb, ORDERED_DATA);
3046         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3047                 set_opt(sb, WRITEBACK_DATA);
3048
3049         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3050                 set_opt(sb, ERRORS_PANIC);
3051         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3052                 set_opt(sb, ERRORS_CONT);
3053         else
3054                 set_opt(sb, ERRORS_RO);
3055         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3056                 set_opt(sb, BLOCK_VALIDITY);
3057         if (def_mount_opts & EXT4_DEFM_DISCARD)
3058                 set_opt(sb, DISCARD);
3059
3060         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3061         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3062         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3063         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3064         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3065
3066         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3067                 set_opt(sb, BARRIER);
3068
3069         /*
3070          * enable delayed allocation by default
3071          * Use -o nodelalloc to turn it off
3072          */
3073         if (!IS_EXT3_SB(sb) &&
3074             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3075                 set_opt(sb, DELALLOC);
3076
3077         /*
3078          * set default s_li_wait_mult for lazyinit, for the case there is
3079          * no mount option specified.
3080          */
3081         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3082
3083         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3084                            &journal_devnum, &journal_ioprio, 0)) {
3085                 ext4_msg(sb, KERN_WARNING,
3086                          "failed to parse options in superblock: %s",
3087                          sbi->s_es->s_mount_opts);
3088         }
3089         sbi->s_def_mount_opt = sbi->s_mount_opt;
3090         if (!parse_options((char *) data, sb, &journal_devnum,
3091                            &journal_ioprio, 0))
3092                 goto failed_mount;
3093
3094         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3095                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3096                             "with data=journal disables delayed "
3097                             "allocation and O_DIRECT support!\n");
3098                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3099                         ext4_msg(sb, KERN_ERR, "can't mount with "
3100                                  "both data=journal and delalloc");
3101                         goto failed_mount;
3102                 }
3103                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3104                         ext4_msg(sb, KERN_ERR, "can't mount with "
3105                                  "both data=journal and delalloc");
3106                         goto failed_mount;
3107                 }
3108                 if (test_opt(sb, DELALLOC))
3109                         clear_opt(sb, DELALLOC);
3110         }
3111
3112         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3113         if (test_opt(sb, DIOREAD_NOLOCK)) {
3114                 if (blocksize < PAGE_SIZE) {
3115                         ext4_msg(sb, KERN_ERR, "can't mount with "
3116                                  "dioread_nolock if block size != PAGE_SIZE");
3117                         goto failed_mount;
3118                 }
3119         }
3120
3121         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3122                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3123
3124         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3125             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3126              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3127              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3128                 ext4_msg(sb, KERN_WARNING,
3129                        "feature flags set on rev 0 fs, "
3130                        "running e2fsck is recommended");
3131
3132         if (IS_EXT2_SB(sb)) {
3133                 if (ext2_feature_set_ok(sb))
3134                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3135                                  "using the ext4 subsystem");
3136                 else {
3137                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3138                                  "to feature incompatibilities");
3139                         goto failed_mount;
3140                 }
3141         }
3142
3143         if (IS_EXT3_SB(sb)) {
3144                 if (ext3_feature_set_ok(sb))
3145                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3146                                  "using the ext4 subsystem");
3147                 else {
3148                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3149                                  "to feature incompatibilities");
3150                         goto failed_mount;
3151                 }
3152         }
3153
3154         /*
3155          * Check feature flags regardless of the revision level, since we
3156          * previously didn't change the revision level when setting the flags,
3157          * so there is a chance incompat flags are set on a rev 0 filesystem.
3158          */
3159         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3160                 goto failed_mount;
3161
3162         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3163             blocksize > EXT4_MAX_BLOCK_SIZE) {
3164                 ext4_msg(sb, KERN_ERR,
3165                        "Unsupported filesystem blocksize %d", blocksize);
3166                 goto failed_mount;
3167         }
3168
3169         if (sb->s_blocksize != blocksize) {
3170                 /* Validate the filesystem blocksize */
3171                 if (!sb_set_blocksize(sb, blocksize)) {
3172                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3173                                         blocksize);
3174                         goto failed_mount;
3175                 }
3176
3177                 brelse(bh);
3178                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3179                 offset = do_div(logical_sb_block, blocksize);
3180                 bh = sb_bread(sb, logical_sb_block);
3181                 if (!bh) {
3182                         ext4_msg(sb, KERN_ERR,
3183                                "Can't read superblock on 2nd try");
3184                         goto failed_mount;
3185                 }
3186                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3187                 sbi->s_es = es;
3188                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3189                         ext4_msg(sb, KERN_ERR,
3190                                "Magic mismatch, very weird!");
3191                         goto failed_mount;
3192                 }
3193         }
3194
3195         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3196                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3197         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3198                                                       has_huge_files);
3199         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3200
3201         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3202                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3203                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3204         } else {
3205                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3206                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3207                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3208                     (!is_power_of_2(sbi->s_inode_size)) ||
3209                     (sbi->s_inode_size > blocksize)) {
3210                         ext4_msg(sb, KERN_ERR,
3211                                "unsupported inode size: %d",
3212                                sbi->s_inode_size);
3213                         goto failed_mount;
3214                 }
3215                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3216                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3217         }
3218
3219         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3220         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3221                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3222                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3223                     !is_power_of_2(sbi->s_desc_size)) {
3224                         ext4_msg(sb, KERN_ERR,
3225                                "unsupported descriptor size %lu",
3226                                sbi->s_desc_size);
3227                         goto failed_mount;
3228                 }
3229         } else
3230                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3231
3232         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3233         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3234         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3235                 goto cantfind_ext4;
3236
3237         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3238         if (sbi->s_inodes_per_block == 0)
3239                 goto cantfind_ext4;
3240         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3241                                         sbi->s_inodes_per_block;
3242         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3243         sbi->s_sbh = bh;
3244         sbi->s_mount_state = le16_to_cpu(es->s_state);
3245         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3246         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3247
3248         for (i = 0; i < 4; i++)
3249                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3250         sbi->s_def_hash_version = es->s_def_hash_version;
3251         i = le32_to_cpu(es->s_flags);
3252         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3253                 sbi->s_hash_unsigned = 3;
3254         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3255 #ifdef __CHAR_UNSIGNED__
3256                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3257                 sbi->s_hash_unsigned = 3;
3258 #else
3259                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3260 #endif
3261                 sb->s_dirt = 1;
3262         }
3263
3264         /* Handle clustersize */
3265         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3266         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3267                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3268         if (has_bigalloc) {
3269                 if (clustersize < blocksize) {
3270                         ext4_msg(sb, KERN_ERR,
3271                                  "cluster size (%d) smaller than "
3272                                  "block size (%d)", clustersize, blocksize);
3273                         goto failed_mount;
3274                 }
3275                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3276                         le32_to_cpu(es->s_log_block_size);
3277                 sbi->s_clusters_per_group =
3278                         le32_to_cpu(es->s_clusters_per_group);
3279                 if (sbi->s_clusters_per_group > blocksize * 8) {
3280                         ext4_msg(sb, KERN_ERR,
3281                                  "#clusters per group too big: %lu",
3282                                  sbi->s_clusters_per_group);
3283                         goto failed_mount;
3284                 }
3285                 if (sbi->s_blocks_per_group !=
3286                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3287                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3288                                  "clusters per group (%lu) inconsistent",
3289                                  sbi->s_blocks_per_group,
3290                                  sbi->s_clusters_per_group);
3291                         goto failed_mount;
3292                 }
3293         } else {
3294                 if (clustersize != blocksize) {
3295                         ext4_warning(sb, "fragment/cluster size (%d) != "
3296                                      "block size (%d)", clustersize,
3297                                      blocksize);
3298                         clustersize = blocksize;
3299                 }
3300                 if (sbi->s_blocks_per_group > blocksize * 8) {
3301                         ext4_msg(sb, KERN_ERR,
3302                                  "#blocks per group too big: %lu",
3303                                  sbi->s_blocks_per_group);
3304                         goto failed_mount;
3305                 }
3306                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3307                 sbi->s_cluster_bits = 0;
3308         }
3309         sbi->s_cluster_ratio = clustersize / blocksize;
3310
3311         if (sbi->s_inodes_per_group > blocksize * 8) {
3312                 ext4_msg(sb, KERN_ERR,
3313                        "#inodes per group too big: %lu",
3314                        sbi->s_inodes_per_group);
3315                 goto failed_mount;
3316         }
3317
3318         /*
3319          * Test whether we have more sectors than will fit in sector_t,
3320          * and whether the max offset is addressable by the page cache.
3321          */
3322         err = generic_check_addressable(sb->s_blocksize_bits,
3323                                         ext4_blocks_count(es));
3324         if (err) {
3325                 ext4_msg(sb, KERN_ERR, "filesystem"
3326                          " too large to mount safely on this system");
3327                 if (sizeof(sector_t) < 8)
3328                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3329                 ret = err;
3330                 goto failed_mount;
3331         }
3332
3333         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3334                 goto cantfind_ext4;
3335
3336         /* check blocks count against device size */
3337         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3338         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3339                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3340                        "exceeds size of device (%llu blocks)",
3341                        ext4_blocks_count(es), blocks_count);
3342                 goto failed_mount;
3343         }
3344
3345         /*
3346          * It makes no sense for the first data block to be beyond the end
3347          * of the filesystem.
3348          */
3349         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3350                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3351                          "block %u is beyond end of filesystem (%llu)",
3352                          le32_to_cpu(es->s_first_data_block),
3353                          ext4_blocks_count(es));
3354                 goto failed_mount;
3355         }
3356         blocks_count = (ext4_blocks_count(es) -
3357                         le32_to_cpu(es->s_first_data_block) +
3358                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3359         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3360         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3361                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3362                        "(block count %llu, first data block %u, "
3363                        "blocks per group %lu)", sbi->s_groups_count,
3364                        ext4_blocks_count(es),
3365                        le32_to_cpu(es->s_first_data_block),
3366                        EXT4_BLOCKS_PER_GROUP(sb));
3367                 goto failed_mount;
3368         }
3369         sbi->s_groups_count = blocks_count;
3370         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3371                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3372         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3373                    EXT4_DESC_PER_BLOCK(sb);
3374         sbi->s_group_desc = ext4_kvmalloc(db_count *
3375                                           sizeof(struct buffer_head *),
3376                                           GFP_KERNEL);
3377         if (sbi->s_group_desc == NULL) {
3378                 ext4_msg(sb, KERN_ERR, "not enough memory");
3379                 goto failed_mount;
3380         }
3381
3382         if (ext4_proc_root)
3383                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3384
3385         if (sbi->s_proc)
3386                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3387                                  &ext4_seq_options_fops, sb);
3388
3389         bgl_lock_init(sbi->s_blockgroup_lock);
3390
3391         for (i = 0; i < db_count; i++) {
3392                 block = descriptor_loc(sb, logical_sb_block, i);
3393                 sbi->s_group_desc[i] = sb_bread(sb, block);
3394                 if (!sbi->s_group_desc[i]) {
3395                         ext4_msg(sb, KERN_ERR,
3396                                "can't read group descriptor %d", i);
3397                         db_count = i;
3398                         goto failed_mount2;
3399                 }
3400         }
3401         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3402                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3403                 goto failed_mount2;
3404         }
3405         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3406                 if (!ext4_fill_flex_info(sb)) {
3407                         ext4_msg(sb, KERN_ERR,
3408                                "unable to initialize "
3409                                "flex_bg meta info!");
3410                         goto failed_mount2;
3411                 }
3412
3413         sbi->s_gdb_count = db_count;
3414         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3415         spin_lock_init(&sbi->s_next_gen_lock);
3416
3417         init_timer(&sbi->s_err_report);
3418         sbi->s_err_report.function = print_daily_error_info;
3419         sbi->s_err_report.data = (unsigned long) sb;
3420
3421         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3422                         ext4_count_free_clusters(sb));
3423         if (!err) {
3424                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3425                                 ext4_count_free_inodes(sb));
3426         }
3427         if (!err) {
3428                 err = percpu_counter_init(&sbi->s_dirs_counter,
3429                                 ext4_count_dirs(sb));
3430         }
3431         if (!err) {
3432                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3433         }
3434         if (err) {
3435                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3436                 goto failed_mount3;
3437         }
3438
3439         sbi->s_stripe = ext4_get_stripe_size(sbi);
3440         sbi->s_max_writeback_mb_bump = 128;
3441
3442         /*
3443          * set up enough so that it can read an inode
3444          */
3445         if (!test_opt(sb, NOLOAD) &&
3446             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3447                 sb->s_op = &ext4_sops;
3448         else
3449                 sb->s_op = &ext4_nojournal_sops;
3450         sb->s_export_op = &ext4_export_ops;
3451         sb->s_xattr = ext4_xattr_handlers;
3452 #ifdef CONFIG_QUOTA
3453         sb->s_qcop = &ext4_qctl_operations;
3454         sb->dq_op = &ext4_quota_operations;
3455 #endif
3456         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3457
3458         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3459         mutex_init(&sbi->s_orphan_lock);
3460         sbi->s_resize_flags = 0;
3461
3462         sb->s_root = NULL;
3463
3464         needs_recovery = (es->s_last_orphan != 0 ||
3465                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3466                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3467
3468         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3469             !(sb->s_flags & MS_RDONLY))
3470                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3471                         goto failed_mount3;
3472
3473         /*
3474          * The first inode we look at is the journal inode.  Don't try
3475          * root first: it may be modified in the journal!
3476          */
3477         if (!test_opt(sb, NOLOAD) &&
3478             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3479                 if (ext4_load_journal(sb, es, journal_devnum))
3480                         goto failed_mount3;
3481         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3482               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3483                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3484                        "suppressed and not mounted read-only");
3485                 goto failed_mount_wq;
3486         } else {
3487                 clear_opt(sb, DATA_FLAGS);
3488                 sbi->s_journal = NULL;
3489                 needs_recovery = 0;
3490                 goto no_journal;
3491         }
3492
3493         if (ext4_blocks_count(es) > 0xffffffffULL &&
3494             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3495                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3496                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3497                 goto failed_mount_wq;
3498         }
3499
3500         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3501                 jbd2_journal_set_features(sbi->s_journal,
3502                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3503                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3504         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3505                 jbd2_journal_set_features(sbi->s_journal,
3506                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3507                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3508                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3509         } else {
3510                 jbd2_journal_clear_features(sbi->s_journal,
3511                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3512                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3513         }
3514
3515         /* We have now updated the journal if required, so we can
3516          * validate the data journaling mode. */
3517         switch (test_opt(sb, DATA_FLAGS)) {
3518         case 0:
3519                 /* No mode set, assume a default based on the journal
3520                  * capabilities: ORDERED_DATA if the journal can
3521                  * cope, else JOURNAL_DATA
3522                  */
3523                 if (jbd2_journal_check_available_features
3524                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3525                         set_opt(sb, ORDERED_DATA);
3526                 else
3527                         set_opt(sb, JOURNAL_DATA);
3528                 break;
3529
3530         case EXT4_MOUNT_ORDERED_DATA:
3531         case EXT4_MOUNT_WRITEBACK_DATA:
3532                 if (!jbd2_journal_check_available_features
3533                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3534                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3535                                "requested data journaling mode");
3536                         goto failed_mount_wq;
3537                 }
3538         default:
3539                 break;
3540         }
3541         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3542
3543         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3544
3545         /*
3546          * The journal may have updated the bg summary counts, so we
3547          * need to update the global counters.
3548          */
3549         percpu_counter_set(&sbi->s_freeclusters_counter,
3550                            ext4_count_free_clusters(sb));
3551         percpu_counter_set(&sbi->s_freeinodes_counter,
3552                            ext4_count_free_inodes(sb));
3553         percpu_counter_set(&sbi->s_dirs_counter,
3554                            ext4_count_dirs(sb));
3555         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3556
3557 no_journal:
3558         /*
3559          * The maximum number of concurrent works can be high and
3560          * concurrency isn't really necessary.  Limit it to 1.
3561          */
3562         EXT4_SB(sb)->dio_unwritten_wq =
3563                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3564         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3565                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3566                 goto failed_mount_wq;
3567         }
3568
3569         /*
3570          * The jbd2_journal_load will have done any necessary log recovery,
3571          * so we can safely mount the rest of the filesystem now.
3572          */
3573
3574         root = ext4_iget(sb, EXT4_ROOT_INO);
3575         if (IS_ERR(root)) {
3576                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3577                 ret = PTR_ERR(root);
3578                 root = NULL;
3579                 goto failed_mount4;
3580         }
3581         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3582                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3583                 iput(root);
3584                 goto failed_mount4;
3585         }
3586         sb->s_root = d_alloc_root(root);
3587         if (!sb->s_root) {
3588                 iput(root);
3589                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3590                 ret = -ENOMEM;
3591                 goto failed_mount4;
3592         }
3593
3594         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3595
3596         /* determine the minimum size of new large inodes, if present */
3597         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3598                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3599                                                      EXT4_GOOD_OLD_INODE_SIZE;
3600                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3601                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3602                         if (sbi->s_want_extra_isize <
3603                             le16_to_cpu(es->s_want_extra_isize))
3604                                 sbi->s_want_extra_isize =
3605                                         le16_to_cpu(es->s_want_extra_isize);
3606                         if (sbi->s_want_extra_isize <
3607                             le16_to_cpu(es->s_min_extra_isize))
3608                                 sbi->s_want_extra_isize =
3609                                         le16_to_cpu(es->s_min_extra_isize);
3610                 }
3611         }
3612         /* Check if enough inode space is available */
3613         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3614                                                         sbi->s_inode_size) {
3615                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3616                                                        EXT4_GOOD_OLD_INODE_SIZE;
3617                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3618                          "available");
3619         }
3620
3621         err = ext4_setup_system_zone(sb);
3622         if (err) {
3623                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3624                          "zone (%d)", err);
3625                 goto failed_mount4a;
3626         }
3627
3628         ext4_ext_init(sb);
3629         err = ext4_mb_init(sb, needs_recovery);
3630         if (err) {
3631                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3632                          err);
3633                 goto failed_mount5;
3634         }
3635
3636         err = ext4_register_li_request(sb, first_not_zeroed);
3637         if (err)
3638                 goto failed_mount6;
3639
3640         sbi->s_kobj.kset = ext4_kset;
3641         init_completion(&sbi->s_kobj_unregister);
3642         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3643                                    "%s", sb->s_id);
3644         if (err)
3645                 goto failed_mount7;
3646
3647         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3648         ext4_orphan_cleanup(sb, es);
3649         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3650         if (needs_recovery) {
3651                 ext4_msg(sb, KERN_INFO, "recovery complete");
3652                 ext4_mark_recovery_complete(sb, es);
3653         }
3654         if (EXT4_SB(sb)->s_journal) {
3655                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3656                         descr = " journalled data mode";
3657                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3658                         descr = " ordered data mode";
3659                 else
3660                         descr = " writeback data mode";
3661         } else
3662                 descr = "out journal";
3663
3664         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3665                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3666                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3667
3668         if (es->s_error_count)
3669                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3670
3671         kfree(orig_data);
3672         return 0;
3673
3674 cantfind_ext4:
3675         if (!silent)
3676                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3677         goto failed_mount;
3678
3679 failed_mount7:
3680         ext4_unregister_li_request(sb);
3681 failed_mount6:
3682         ext4_mb_release(sb);
3683 failed_mount5:
3684         ext4_ext_release(sb);
3685         ext4_release_system_zone(sb);
3686 failed_mount4a:
3687         dput(sb->s_root);
3688         sb->s_root = NULL;
3689 failed_mount4:
3690         ext4_msg(sb, KERN_ERR, "mount failed");
3691         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3692 failed_mount_wq:
3693         if (sbi->s_journal) {
3694                 jbd2_journal_destroy(sbi->s_journal);
3695                 sbi->s_journal = NULL;
3696         }
3697 failed_mount3:
3698         del_timer(&sbi->s_err_report);
3699         if (sbi->s_flex_groups)
3700                 ext4_kvfree(sbi->s_flex_groups);
3701         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3702         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3703         percpu_counter_destroy(&sbi->s_dirs_counter);
3704         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3705         if (sbi->s_mmp_tsk)
3706                 kthread_stop(sbi->s_mmp_tsk);
3707 failed_mount2:
3708         for (i = 0; i < db_count; i++)
3709                 brelse(sbi->s_group_desc[i]);
3710         ext4_kvfree(sbi->s_group_desc);
3711 failed_mount:
3712         if (sbi->s_proc) {
3713                 remove_proc_entry("options", sbi->s_proc);
3714                 remove_proc_entry(sb->s_id, ext4_proc_root);
3715         }
3716 #ifdef CONFIG_QUOTA
3717         for (i = 0; i < MAXQUOTAS; i++)
3718                 kfree(sbi->s_qf_names[i]);
3719 #endif
3720         ext4_blkdev_remove(sbi);
3721         brelse(bh);
3722 out_fail:
3723         sb->s_fs_info = NULL;
3724         kfree(sbi->s_blockgroup_lock);
3725         kfree(sbi);
3726 out_free_orig:
3727         kfree(orig_data);
3728         return ret;
3729 }
3730
3731 /*
3732  * Setup any per-fs journal parameters now.  We'll do this both on
3733  * initial mount, once the journal has been initialised but before we've
3734  * done any recovery; and again on any subsequent remount.
3735  */
3736 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3737 {
3738         struct ext4_sb_info *sbi = EXT4_SB(sb);
3739
3740         journal->j_commit_interval = sbi->s_commit_interval;
3741         journal->j_min_batch_time = sbi->s_min_batch_time;
3742         journal->j_max_batch_time = sbi->s_max_batch_time;
3743
3744         write_lock(&journal->j_state_lock);
3745         if (test_opt(sb, BARRIER))
3746                 journal->j_flags |= JBD2_BARRIER;
3747         else
3748                 journal->j_flags &= ~JBD2_BARRIER;
3749         if (test_opt(sb, DATA_ERR_ABORT))
3750                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3751         else
3752                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3753         write_unlock(&journal->j_state_lock);
3754 }
3755
3756 static journal_t *ext4_get_journal(struct super_block *sb,
3757                                    unsigned int journal_inum)
3758 {
3759         struct inode *journal_inode;
3760         journal_t *journal;
3761
3762         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3763
3764         /* First, test for the existence of a valid inode on disk.  Bad
3765          * things happen if we iget() an unused inode, as the subsequent
3766          * iput() will try to delete it. */
3767
3768         journal_inode = ext4_iget(sb, journal_inum);
3769         if (IS_ERR(journal_inode)) {
3770                 ext4_msg(sb, KERN_ERR, "no journal found");
3771                 return NULL;
3772         }
3773         if (!journal_inode->i_nlink) {
3774                 make_bad_inode(journal_inode);
3775                 iput(journal_inode);
3776                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3777                 return NULL;
3778         }
3779
3780         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3781                   journal_inode, journal_inode->i_size);
3782         if (!S_ISREG(journal_inode->i_mode)) {
3783                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3784                 iput(journal_inode);
3785                 return NULL;
3786         }
3787
3788         journal = jbd2_journal_init_inode(journal_inode);
3789         if (!journal) {
3790                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3791                 iput(journal_inode);
3792                 return NULL;
3793         }
3794         journal->j_private = sb;
3795         ext4_init_journal_params(sb, journal);
3796         return journal;
3797 }
3798
3799 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3800                                        dev_t j_dev)
3801 {
3802         struct buffer_head *bh;
3803         journal_t *journal;
3804         ext4_fsblk_t start;
3805         ext4_fsblk_t len;
3806         int hblock, blocksize;
3807         ext4_fsblk_t sb_block;
3808         unsigned long offset;
3809         struct ext4_super_block *es;
3810         struct block_device *bdev;
3811
3812         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3813
3814         bdev = ext4_blkdev_get(j_dev, sb);
3815         if (bdev == NULL)
3816                 return NULL;
3817
3818         blocksize = sb->s_blocksize;
3819         hblock = bdev_logical_block_size(bdev);
3820         if (blocksize < hblock) {
3821                 ext4_msg(sb, KERN_ERR,
3822                         "blocksize too small for journal device");
3823                 goto out_bdev;
3824         }
3825
3826         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3827         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3828         set_blocksize(bdev, blocksize);
3829         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3830                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3831                        "external journal");
3832                 goto out_bdev;
3833         }
3834
3835         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3836         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3837             !(le32_to_cpu(es->s_feature_incompat) &
3838               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3839                 ext4_msg(sb, KERN_ERR, "external journal has "
3840                                         "bad superblock");
3841                 brelse(bh);
3842                 goto out_bdev;
3843         }
3844
3845         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3846                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3847                 brelse(bh);
3848                 goto out_bdev;
3849         }
3850
3851         len = ext4_blocks_count(es);
3852         start = sb_block + 1;
3853         brelse(bh);     /* we're done with the superblock */
3854
3855         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3856                                         start, len, blocksize);
3857         if (!journal) {
3858                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3859                 goto out_bdev;
3860         }
3861         journal->j_private = sb;
3862         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3863         wait_on_buffer(journal->j_sb_buffer);
3864         if (!buffer_uptodate(journal->j_sb_buffer)) {
3865                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3866                 goto out_journal;
3867         }
3868         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3869                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3870                                         "user (unsupported) - %d",
3871                         be32_to_cpu(journal->j_superblock->s_nr_users));
3872                 goto out_journal;
3873         }
3874         EXT4_SB(sb)->journal_bdev = bdev;
3875         ext4_init_journal_params(sb, journal);
3876         return journal;
3877
3878 out_journal:
3879         jbd2_journal_destroy(journal);
3880 out_bdev:
3881         ext4_blkdev_put(bdev);
3882         return NULL;
3883 }
3884
3885 static int ext4_load_journal(struct super_block *sb,
3886                              struct ext4_super_block *es,
3887                              unsigned long journal_devnum)
3888 {
3889         journal_t *journal;
3890         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3891         dev_t journal_dev;
3892         int err = 0;
3893         int really_read_only;
3894
3895         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3896
3897         if (journal_devnum &&
3898             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3899                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3900                         "numbers have changed");
3901                 journal_dev = new_decode_dev(journal_devnum);
3902         } else
3903                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3904
3905         really_read_only = bdev_read_only(sb->s_bdev);
3906
3907         /*
3908          * Are we loading a blank journal or performing recovery after a
3909          * crash?  For recovery, we need to check in advance whether we
3910          * can get read-write access to the device.
3911          */
3912         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3913                 if (sb->s_flags & MS_RDONLY) {
3914                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3915                                         "required on readonly filesystem");
3916                         if (really_read_only) {
3917                                 ext4_msg(sb, KERN_ERR, "write access "
3918                                         "unavailable, cannot proceed");
3919                                 return -EROFS;
3920                         }
3921                         ext4_msg(sb, KERN_INFO, "write access will "
3922                                "be enabled during recovery");
3923                 }
3924         }
3925
3926         if (journal_inum && journal_dev) {
3927                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3928                        "and inode journals!");
3929                 return -EINVAL;
3930         }
3931
3932         if (journal_inum) {
3933                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3934                         return -EINVAL;
3935         } else {
3936                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3937                         return -EINVAL;
3938         }
3939
3940         if (!(journal->j_flags & JBD2_BARRIER))
3941                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3942
3943         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3944                 err = jbd2_journal_wipe(journal, !really_read_only);
3945         if (!err) {
3946                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3947                 if (save)
3948                         memcpy(save, ((char *) es) +
3949                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3950                 err = jbd2_journal_load(journal);
3951                 if (save)
3952                         memcpy(((char *) es) + EXT4_S_ERR_START,
3953                                save, EXT4_S_ERR_LEN);
3954                 kfree(save);
3955         }
3956
3957         if (err) {
3958                 ext4_msg(sb, KERN_ERR, "error loading journal");
3959                 jbd2_journal_destroy(journal);
3960                 return err;
3961         }
3962
3963         EXT4_SB(sb)->s_journal = journal;
3964         ext4_clear_journal_err(sb, es);
3965
3966         if (!really_read_only && journal_devnum &&
3967             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3968                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3969
3970                 /* Make sure we flush the recovery flag to disk. */
3971                 ext4_commit_super(sb, 1);
3972         }
3973
3974         return 0;
3975 }
3976
3977 static int ext4_commit_super(struct super_block *sb, int sync)
3978 {
3979         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3980         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3981         int error = 0;
3982
3983         if (!sbh || block_device_ejected(sb))
3984                 return error;
3985         if (buffer_write_io_error(sbh)) {
3986                 /*
3987                  * Oh, dear.  A previous attempt to write the
3988                  * superblock failed.  This could happen because the
3989                  * USB device was yanked out.  Or it could happen to
3990                  * be a transient write error and maybe the block will
3991                  * be remapped.  Nothing we can do but to retry the
3992                  * write and hope for the best.
3993                  */
3994                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3995                        "superblock detected");
3996                 clear_buffer_write_io_error(sbh);
3997                 set_buffer_uptodate(sbh);
3998         }
3999         /*
4000          * If the file system is mounted read-only, don't update the
4001          * superblock write time.  This avoids updating the superblock
4002          * write time when we are mounting the root file system
4003          * read/only but we need to replay the journal; at that point,
4004          * for people who are east of GMT and who make their clock
4005          * tick in localtime for Windows bug-for-bug compatibility,
4006          * the clock is set in the future, and this will cause e2fsck
4007          * to complain and force a full file system check.
4008          */
4009         if (!(sb->s_flags & MS_RDONLY))
4010                 es->s_wtime = cpu_to_le32(get_seconds());
4011         if (sb->s_bdev->bd_part)
4012                 es->s_kbytes_written =
4013                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4014                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4015                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4016         else
4017                 es->s_kbytes_written =
4018                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4019         ext4_free_blocks_count_set(es,
4020                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4021                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4022         es->s_free_inodes_count =
4023                 cpu_to_le32(percpu_counter_sum_positive(
4024                                 &EXT4_SB(sb)->s_freeinodes_counter));
4025         sb->s_dirt = 0;
4026         BUFFER_TRACE(sbh, "marking dirty");
4027         mark_buffer_dirty(sbh);
4028         if (sync) {
4029                 error = sync_dirty_buffer(sbh);
4030                 if (error)
4031                         return error;
4032
4033                 error = buffer_write_io_error(sbh);
4034                 if (error) {
4035                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4036                                "superblock");
4037                         clear_buffer_write_io_error(sbh);
4038                         set_buffer_uptodate(sbh);
4039                 }
4040         }
4041         return error;
4042 }
4043
4044 /*
4045  * Have we just finished recovery?  If so, and if we are mounting (or
4046  * remounting) the filesystem readonly, then we will end up with a
4047  * consistent fs on disk.  Record that fact.
4048  */
4049 static void ext4_mark_recovery_complete(struct super_block *sb,
4050                                         struct ext4_super_block *es)
4051 {
4052         journal_t *journal = EXT4_SB(sb)->s_journal;
4053
4054         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4055                 BUG_ON(journal != NULL);
4056                 return;
4057         }
4058         jbd2_journal_lock_updates(journal);
4059         if (jbd2_journal_flush(journal) < 0)
4060                 goto out;
4061
4062         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4063             sb->s_flags & MS_RDONLY) {
4064                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4065                 ext4_commit_super(sb, 1);
4066         }
4067
4068 out:
4069         jbd2_journal_unlock_updates(journal);
4070 }
4071
4072 /*
4073  * If we are mounting (or read-write remounting) a filesystem whose journal
4074  * has recorded an error from a previous lifetime, move that error to the
4075  * main filesystem now.
4076  */
4077 static void ext4_clear_journal_err(struct super_block *sb,
4078                                    struct ext4_super_block *es)
4079 {
4080         journal_t *journal;
4081         int j_errno;
4082         const char *errstr;
4083
4084         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4085
4086         journal = EXT4_SB(sb)->s_journal;
4087
4088         /*
4089          * Now check for any error status which may have been recorded in the
4090          * journal by a prior ext4_error() or ext4_abort()
4091          */
4092
4093         j_errno = jbd2_journal_errno(journal);
4094         if (j_errno) {
4095                 char nbuf[16];
4096
4097                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4098                 ext4_warning(sb, "Filesystem error recorded "
4099                              "from previous mount: %s", errstr);
4100                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4101
4102                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4103                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4104                 ext4_commit_super(sb, 1);
4105
4106                 jbd2_journal_clear_err(journal);
4107         }
4108 }
4109
4110 /*
4111  * Force the running and committing transactions to commit,
4112  * and wait on the commit.
4113  */
4114 int ext4_force_commit(struct super_block *sb)
4115 {
4116         journal_t *journal;
4117         int ret = 0;
4118
4119         if (sb->s_flags & MS_RDONLY)
4120                 return 0;
4121
4122         journal = EXT4_SB(sb)->s_journal;
4123         if (journal) {
4124                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4125                 ret = ext4_journal_force_commit(journal);
4126         }
4127
4128         return ret;
4129 }
4130
4131 static void ext4_write_super(struct super_block *sb)
4132 {
4133         lock_super(sb);
4134         ext4_commit_super(sb, 1);
4135         unlock_super(sb);
4136 }
4137
4138 static int ext4_sync_fs(struct super_block *sb, int wait)
4139 {
4140         int ret = 0;
4141         tid_t target;
4142         struct ext4_sb_info *sbi = EXT4_SB(sb);
4143
4144         trace_ext4_sync_fs(sb, wait);
4145         flush_workqueue(sbi->dio_unwritten_wq);
4146         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4147                 if (wait)
4148                         jbd2_log_wait_commit(sbi->s_journal, target);
4149         }
4150         return ret;
4151 }
4152
4153 /*
4154  * LVM calls this function before a (read-only) snapshot is created.  This
4155  * gives us a chance to flush the journal completely and mark the fs clean.
4156  *
4157  * Note that only this function cannot bring a filesystem to be in a clean
4158  * state independently, because ext4 prevents a new handle from being started
4159  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4160  * the upper layer.
4161  */
4162 static int ext4_freeze(struct super_block *sb)
4163 {
4164         int error = 0;
4165         journal_t *journal;
4166
4167         if (sb->s_flags & MS_RDONLY)
4168                 return 0;
4169
4170         journal = EXT4_SB(sb)->s_journal;
4171
4172         /* Now we set up the journal barrier. */
4173         jbd2_journal_lock_updates(journal);
4174
4175         /*
4176          * Don't clear the needs_recovery flag if we failed to flush
4177          * the journal.
4178          */
4179         error = jbd2_journal_flush(journal);
4180         if (error < 0)
4181                 goto out;
4182
4183         /* Journal blocked and flushed, clear needs_recovery flag. */
4184         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4185         error = ext4_commit_super(sb, 1);
4186 out:
4187         /* we rely on s_frozen to stop further updates */
4188         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4189         return error;
4190 }
4191
4192 /*
4193  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4194  * flag here, even though the filesystem is not technically dirty yet.
4195  */
4196 static int ext4_unfreeze(struct super_block *sb)
4197 {
4198         if (sb->s_flags & MS_RDONLY)
4199                 return 0;
4200
4201         lock_super(sb);
4202         /* Reset the needs_recovery flag before the fs is unlocked. */
4203         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4204         ext4_commit_super(sb, 1);
4205         unlock_super(sb);
4206         return 0;
4207 }
4208
4209 /*
4210  * Structure to save mount options for ext4_remount's benefit
4211  */
4212 struct ext4_mount_options {
4213         unsigned long s_mount_opt;
4214         unsigned long s_mount_opt2;
4215         uid_t s_resuid;
4216         gid_t s_resgid;
4217         unsigned long s_commit_interval;
4218         u32 s_min_batch_time, s_max_batch_time;
4219 #ifdef CONFIG_QUOTA
4220         int s_jquota_fmt;
4221         char *s_qf_names[MAXQUOTAS];
4222 #endif
4223 };
4224
4225 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4226 {
4227         struct ext4_super_block *es;
4228         struct ext4_sb_info *sbi = EXT4_SB(sb);
4229         unsigned long old_sb_flags;
4230         struct ext4_mount_options old_opts;
4231         int enable_quota = 0;
4232         ext4_group_t g;
4233         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4234         int err = 0;
4235 #ifdef CONFIG_QUOTA
4236         int i;
4237 #endif
4238         char *orig_data = kstrdup(data, GFP_KERNEL);
4239
4240         /* Store the original options */
4241         lock_super(sb);
4242         old_sb_flags = sb->s_flags;
4243         old_opts.s_mount_opt = sbi->s_mount_opt;
4244         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4245         old_opts.s_resuid = sbi->s_resuid;
4246         old_opts.s_resgid = sbi->s_resgid;
4247         old_opts.s_commit_interval = sbi->s_commit_interval;
4248         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4249         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4250 #ifdef CONFIG_QUOTA
4251         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4252         for (i = 0; i < MAXQUOTAS; i++)
4253                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4254 #endif
4255         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4256                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4257
4258         /*
4259          * Allow the "check" option to be passed as a remount option.
4260          */
4261         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4262                 err = -EINVAL;
4263                 goto restore_opts;
4264         }
4265
4266         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4267                 ext4_abort(sb, "Abort forced by user");
4268
4269         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4270                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4271
4272         es = sbi->s_es;
4273
4274         if (sbi->s_journal) {
4275                 ext4_init_journal_params(sb, sbi->s_journal);
4276                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4277         }
4278
4279         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4280                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4281                         err = -EROFS;
4282                         goto restore_opts;
4283                 }
4284
4285                 if (*flags & MS_RDONLY) {
4286                         err = dquot_suspend(sb, -1);
4287                         if (err < 0)
4288                                 goto restore_opts;
4289
4290                         /*
4291                          * First of all, the unconditional stuff we have to do
4292                          * to disable replay of the journal when we next remount
4293                          */
4294                         sb->s_flags |= MS_RDONLY;
4295
4296                         /*
4297                          * OK, test if we are remounting a valid rw partition
4298                          * readonly, and if so set the rdonly flag and then
4299                          * mark the partition as valid again.
4300                          */
4301                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4302                             (sbi->s_mount_state & EXT4_VALID_FS))
4303                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4304
4305                         if (sbi->s_journal)
4306                                 ext4_mark_recovery_complete(sb, es);
4307                 } else {
4308                         /* Make sure we can mount this feature set readwrite */
4309                         if (!ext4_feature_set_ok(sb, 0)) {
4310                                 err = -EROFS;
4311                                 goto restore_opts;
4312                         }
4313                         /*
4314                          * Make sure the group descriptor checksums
4315                          * are sane.  If they aren't, refuse to remount r/w.
4316                          */
4317                         for (g = 0; g < sbi->s_groups_count; g++) {
4318                                 struct ext4_group_desc *gdp =
4319                                         ext4_get_group_desc(sb, g, NULL);
4320
4321                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4322                                         ext4_msg(sb, KERN_ERR,
4323                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4324                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4325                                                le16_to_cpu(gdp->bg_checksum));
4326                                         err = -EINVAL;
4327                                         goto restore_opts;
4328                                 }
4329                         }
4330
4331                         /*
4332                          * If we have an unprocessed orphan list hanging
4333                          * around from a previously readonly bdev mount,
4334                          * require a full umount/remount for now.
4335                          */
4336                         if (es->s_last_orphan) {
4337                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4338                                        "remount RDWR because of unprocessed "
4339                                        "orphan inode list.  Please "
4340                                        "umount/remount instead");
4341                                 err = -EINVAL;
4342                                 goto restore_opts;
4343                         }
4344
4345                         /*
4346                          * Mounting a RDONLY partition read-write, so reread
4347                          * and store the current valid flag.  (It may have
4348                          * been changed by e2fsck since we originally mounted
4349                          * the partition.)
4350                          */
4351                         if (sbi->s_journal)
4352                                 ext4_clear_journal_err(sb, es);
4353                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4354                         if (!ext4_setup_super(sb, es, 0))
4355                                 sb->s_flags &= ~MS_RDONLY;
4356                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4357                                                      EXT4_FEATURE_INCOMPAT_MMP))
4358                                 if (ext4_multi_mount_protect(sb,
4359                                                 le64_to_cpu(es->s_mmp_block))) {
4360                                         err = -EROFS;
4361                                         goto restore_opts;
4362                                 }
4363                         enable_quota = 1;
4364                 }
4365         }
4366
4367         /*
4368          * Reinitialize lazy itable initialization thread based on
4369          * current settings
4370          */
4371         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4372                 ext4_unregister_li_request(sb);
4373         else {
4374                 ext4_group_t first_not_zeroed;
4375                 first_not_zeroed = ext4_has_uninit_itable(sb);
4376                 ext4_register_li_request(sb, first_not_zeroed);
4377         }
4378
4379         ext4_setup_system_zone(sb);
4380         if (sbi->s_journal == NULL)
4381                 ext4_commit_super(sb, 1);
4382
4383 #ifdef CONFIG_QUOTA
4384         /* Release old quota file names */
4385         for (i = 0; i < MAXQUOTAS; i++)
4386                 if (old_opts.s_qf_names[i] &&
4387                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4388                         kfree(old_opts.s_qf_names[i]);
4389 #endif
4390         unlock_super(sb);
4391         if (enable_quota)
4392                 dquot_resume(sb, -1);
4393
4394         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4395         kfree(orig_data);
4396         return 0;
4397
4398 restore_opts:
4399         sb->s_flags = old_sb_flags;
4400         sbi->s_mount_opt = old_opts.s_mount_opt;
4401         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4402         sbi->s_resuid = old_opts.s_resuid;
4403         sbi->s_resgid = old_opts.s_resgid;
4404         sbi->s_commit_interval = old_opts.s_commit_interval;
4405         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4406         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4407 #ifdef CONFIG_QUOTA
4408         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4409         for (i = 0; i < MAXQUOTAS; i++) {
4410                 if (sbi->s_qf_names[i] &&
4411                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4412                         kfree(sbi->s_qf_names[i]);
4413                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4414         }
4415 #endif
4416         unlock_super(sb);
4417         kfree(orig_data);
4418         return err;
4419 }
4420
4421 /*
4422  * Note: calculating the overhead so we can be compatible with
4423  * historical BSD practice is quite difficult in the face of
4424  * clusters/bigalloc.  This is because multiple metadata blocks from
4425  * different block group can end up in the same allocation cluster.
4426  * Calculating the exact overhead in the face of clustered allocation
4427  * requires either O(all block bitmaps) in memory or O(number of block
4428  * groups**2) in time.  We will still calculate the superblock for
4429  * older file systems --- and if we come across with a bigalloc file
4430  * system with zero in s_overhead_clusters the estimate will be close to
4431  * correct especially for very large cluster sizes --- but for newer
4432  * file systems, it's better to calculate this figure once at mkfs
4433  * time, and store it in the superblock.  If the superblock value is
4434  * present (even for non-bigalloc file systems), we will use it.
4435  */
4436 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4437 {
4438         struct super_block *sb = dentry->d_sb;
4439         struct ext4_sb_info *sbi = EXT4_SB(sb);
4440         struct ext4_super_block *es = sbi->s_es;
4441         struct ext4_group_desc *gdp;
4442         u64 fsid;
4443         s64 bfree;
4444
4445         if (test_opt(sb, MINIX_DF)) {
4446                 sbi->s_overhead_last = 0;
4447         } else if (es->s_overhead_clusters) {
4448                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4449         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4450                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4451                 ext4_fsblk_t overhead = 0;
4452
4453                 /*
4454                  * Compute the overhead (FS structures).  This is constant
4455                  * for a given filesystem unless the number of block groups
4456                  * changes so we cache the previous value until it does.
4457                  */
4458
4459                 /*
4460                  * All of the blocks before first_data_block are
4461                  * overhead
4462                  */
4463                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4464
4465                 /*
4466                  * Add the overhead found in each block group
4467                  */
4468                 for (i = 0; i < ngroups; i++) {
4469                         gdp = ext4_get_group_desc(sb, i, NULL);
4470                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4471                         cond_resched();
4472                 }
4473                 sbi->s_overhead_last = overhead;
4474                 smp_wmb();
4475                 sbi->s_blocks_last = ext4_blocks_count(es);
4476         }
4477
4478         buf->f_type = EXT4_SUPER_MAGIC;
4479         buf->f_bsize = sb->s_blocksize;
4480         buf->f_blocks = (ext4_blocks_count(es) -
4481                          EXT4_C2B(sbi, sbi->s_overhead_last));
4482         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4483                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4484         /* prevent underflow in case that few free space is available */
4485         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4486         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4487         if (buf->f_bfree < ext4_r_blocks_count(es))
4488                 buf->f_bavail = 0;
4489         buf->f_files = le32_to_cpu(es->s_inodes_count);
4490         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4491         buf->f_namelen = EXT4_NAME_LEN;
4492         fsid = le64_to_cpup((void *)es->s_uuid) ^
4493                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4494         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4495         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4496
4497         return 0;
4498 }
4499
4500 /* Helper function for writing quotas on sync - we need to start transaction
4501  * before quota file is locked for write. Otherwise the are possible deadlocks:
4502  * Process 1                         Process 2
4503  * ext4_create()                     quota_sync()
4504  *   jbd2_journal_start()                  write_dquot()
4505  *   dquot_initialize()                         down(dqio_mutex)
4506  *     down(dqio_mutex)                    jbd2_journal_start()
4507  *
4508  */
4509
4510 #ifdef CONFIG_QUOTA
4511
4512 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4513 {
4514         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4515 }
4516
4517 static int ext4_write_dquot(struct dquot *dquot)
4518 {
4519         int ret, err;
4520         handle_t *handle;
4521         struct inode *inode;
4522
4523         inode = dquot_to_inode(dquot);
4524         handle = ext4_journal_start(inode,
4525                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4526         if (IS_ERR(handle))
4527                 return PTR_ERR(handle);
4528         ret = dquot_commit(dquot);
4529         err = ext4_journal_stop(handle);
4530         if (!ret)
4531                 ret = err;
4532         return ret;
4533 }
4534
4535 static int ext4_acquire_dquot(struct dquot *dquot)
4536 {
4537         int ret, err;
4538         handle_t *handle;
4539
4540         handle = ext4_journal_start(dquot_to_inode(dquot),
4541                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4542         if (IS_ERR(handle))
4543                 return PTR_ERR(handle);
4544         ret = dquot_acquire(dquot);
4545         err = ext4_journal_stop(handle);
4546         if (!ret)
4547                 ret = err;
4548         return ret;
4549 }
4550
4551 static int ext4_release_dquot(struct dquot *dquot)
4552 {
4553         int ret, err;
4554         handle_t *handle;
4555
4556         handle = ext4_journal_start(dquot_to_inode(dquot),
4557                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4558         if (IS_ERR(handle)) {
4559                 /* Release dquot anyway to avoid endless cycle in dqput() */
4560                 dquot_release(dquot);
4561                 return PTR_ERR(handle);
4562         }
4563         ret = dquot_release(dquot);
4564         err = ext4_journal_stop(handle);
4565         if (!ret)
4566                 ret = err;
4567         return ret;
4568 }
4569
4570 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4571 {
4572         /* Are we journaling quotas? */
4573         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4574             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4575                 dquot_mark_dquot_dirty(dquot);
4576                 return ext4_write_dquot(dquot);
4577         } else {
4578                 return dquot_mark_dquot_dirty(dquot);
4579         }
4580 }
4581
4582 static int ext4_write_info(struct super_block *sb, int type)
4583 {
4584         int ret, err;
4585         handle_t *handle;
4586
4587         /* Data block + inode block */
4588         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4589         if (IS_ERR(handle))
4590                 return PTR_ERR(handle);
4591         ret = dquot_commit_info(sb, type);
4592         err = ext4_journal_stop(handle);
4593         if (!ret)
4594                 ret = err;
4595         return ret;
4596 }
4597
4598 /*
4599  * Turn on quotas during mount time - we need to find
4600  * the quota file and such...
4601  */
4602 static int ext4_quota_on_mount(struct super_block *sb, int type)
4603 {
4604         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4605                                         EXT4_SB(sb)->s_jquota_fmt, type);
4606 }
4607
4608 /*
4609  * Standard function to be called on quota_on
4610  */
4611 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4612                          struct path *path)
4613 {
4614         int err;
4615
4616         if (!test_opt(sb, QUOTA))
4617                 return -EINVAL;
4618
4619         /* Quotafile not on the same filesystem? */
4620         if (path->dentry->d_sb != sb)
4621                 return -EXDEV;
4622         /* Journaling quota? */
4623         if (EXT4_SB(sb)->s_qf_names[type]) {
4624                 /* Quotafile not in fs root? */
4625                 if (path->dentry->d_parent != sb->s_root)
4626                         ext4_msg(sb, KERN_WARNING,
4627                                 "Quota file not on filesystem root. "
4628                                 "Journaled quota will not work");
4629         }
4630
4631         /*
4632          * When we journal data on quota file, we have to flush journal to see
4633          * all updates to the file when we bypass pagecache...
4634          */
4635         if (EXT4_SB(sb)->s_journal &&
4636             ext4_should_journal_data(path->dentry->d_inode)) {
4637                 /*
4638                  * We don't need to lock updates but journal_flush() could
4639                  * otherwise be livelocked...
4640                  */
4641                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4642                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4643                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4644                 if (err)
4645                         return err;
4646         }
4647
4648         return dquot_quota_on(sb, type, format_id, path);
4649 }
4650
4651 static int ext4_quota_off(struct super_block *sb, int type)
4652 {
4653         struct inode *inode = sb_dqopt(sb)->files[type];
4654         handle_t *handle;
4655
4656         /* Force all delayed allocation blocks to be allocated.
4657          * Caller already holds s_umount sem */
4658         if (test_opt(sb, DELALLOC))
4659                 sync_filesystem(sb);
4660
4661         if (!inode)
4662                 goto out;
4663
4664         /* Update modification times of quota files when userspace can
4665          * start looking at them */
4666         handle = ext4_journal_start(inode, 1);
4667         if (IS_ERR(handle))
4668                 goto out;
4669         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4670         ext4_mark_inode_dirty(handle, inode);
4671         ext4_journal_stop(handle);
4672
4673 out:
4674         return dquot_quota_off(sb, type);
4675 }
4676
4677 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4678  * acquiring the locks... As quota files are never truncated and quota code
4679  * itself serializes the operations (and no one else should touch the files)
4680  * we don't have to be afraid of races */
4681 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4682                                size_t len, loff_t off)
4683 {
4684         struct inode *inode = sb_dqopt(sb)->files[type];
4685         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4686         int err = 0;
4687         int offset = off & (sb->s_blocksize - 1);
4688         int tocopy;
4689         size_t toread;
4690         struct buffer_head *bh;
4691         loff_t i_size = i_size_read(inode);
4692
4693         if (off > i_size)
4694                 return 0;
4695         if (off+len > i_size)
4696                 len = i_size-off;
4697         toread = len;
4698         while (toread > 0) {
4699                 tocopy = sb->s_blocksize - offset < toread ?
4700                                 sb->s_blocksize - offset : toread;
4701                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4702                 if (err)
4703                         return err;
4704                 if (!bh)        /* A hole? */
4705                         memset(data, 0, tocopy);
4706                 else
4707                         memcpy(data, bh->b_data+offset, tocopy);
4708                 brelse(bh);
4709                 offset = 0;
4710                 toread -= tocopy;
4711                 data += tocopy;
4712                 blk++;
4713         }
4714         return len;
4715 }
4716
4717 /* Write to quotafile (we know the transaction is already started and has
4718  * enough credits) */
4719 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4720                                 const char *data, size_t len, loff_t off)
4721 {
4722         struct inode *inode = sb_dqopt(sb)->files[type];
4723         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4724         int err = 0;
4725         int offset = off & (sb->s_blocksize - 1);
4726         struct buffer_head *bh;
4727         handle_t *handle = journal_current_handle();
4728
4729         if (EXT4_SB(sb)->s_journal && !handle) {
4730                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4731                         " cancelled because transaction is not started",
4732                         (unsigned long long)off, (unsigned long long)len);
4733                 return -EIO;
4734         }
4735         /*
4736          * Since we account only one data block in transaction credits,
4737          * then it is impossible to cross a block boundary.
4738          */
4739         if (sb->s_blocksize - offset < len) {
4740                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4741                         " cancelled because not block aligned",
4742                         (unsigned long long)off, (unsigned long long)len);
4743                 return -EIO;
4744         }
4745
4746         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4747         bh = ext4_bread(handle, inode, blk, 1, &err);
4748         if (!bh)
4749                 goto out;
4750         err = ext4_journal_get_write_access(handle, bh);
4751         if (err) {
4752                 brelse(bh);
4753                 goto out;
4754         }
4755         lock_buffer(bh);
4756         memcpy(bh->b_data+offset, data, len);
4757         flush_dcache_page(bh->b_page);
4758         unlock_buffer(bh);
4759         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4760         brelse(bh);
4761 out:
4762         if (err) {
4763                 mutex_unlock(&inode->i_mutex);
4764                 return err;
4765         }
4766         if (inode->i_size < off + len) {
4767                 i_size_write(inode, off + len);
4768                 EXT4_I(inode)->i_disksize = inode->i_size;
4769                 ext4_mark_inode_dirty(handle, inode);
4770         }
4771         mutex_unlock(&inode->i_mutex);
4772         return len;
4773 }
4774
4775 #endif
4776
4777 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4778                        const char *dev_name, void *data)
4779 {
4780         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4781 }
4782
4783 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4784 static inline void register_as_ext2(void)
4785 {
4786         int err = register_filesystem(&ext2_fs_type);
4787         if (err)
4788                 printk(KERN_WARNING
4789                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4790 }
4791
4792 static inline void unregister_as_ext2(void)
4793 {
4794         unregister_filesystem(&ext2_fs_type);
4795 }
4796
4797 static inline int ext2_feature_set_ok(struct super_block *sb)
4798 {
4799         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4800                 return 0;
4801         if (sb->s_flags & MS_RDONLY)
4802                 return 1;
4803         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4804                 return 0;
4805         return 1;
4806 }
4807 MODULE_ALIAS("ext2");
4808 #else
4809 static inline void register_as_ext2(void) { }
4810 static inline void unregister_as_ext2(void) { }
4811 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4812 #endif
4813
4814 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4815 static inline void register_as_ext3(void)
4816 {
4817         int err = register_filesystem(&ext3_fs_type);
4818         if (err)
4819                 printk(KERN_WARNING
4820                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4821 }
4822
4823 static inline void unregister_as_ext3(void)
4824 {
4825         unregister_filesystem(&ext3_fs_type);
4826 }
4827
4828 static inline int ext3_feature_set_ok(struct super_block *sb)
4829 {
4830         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4831                 return 0;
4832         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4833                 return 0;
4834         if (sb->s_flags & MS_RDONLY)
4835                 return 1;
4836         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4837                 return 0;
4838         return 1;
4839 }
4840 MODULE_ALIAS("ext3");
4841 #else
4842 static inline void register_as_ext3(void) { }
4843 static inline void unregister_as_ext3(void) { }
4844 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4845 #endif
4846
4847 static struct file_system_type ext4_fs_type = {
4848         .owner          = THIS_MODULE,
4849         .name           = "ext4",
4850         .mount          = ext4_mount,
4851         .kill_sb        = kill_block_super,
4852         .fs_flags       = FS_REQUIRES_DEV,
4853 };
4854
4855 static int __init ext4_init_feat_adverts(void)
4856 {
4857         struct ext4_features *ef;
4858         int ret = -ENOMEM;
4859
4860         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4861         if (!ef)
4862                 goto out;
4863
4864         ef->f_kobj.kset = ext4_kset;
4865         init_completion(&ef->f_kobj_unregister);
4866         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4867                                    "features");
4868         if (ret) {
4869                 kfree(ef);
4870                 goto out;
4871         }
4872
4873         ext4_feat = ef;
4874         ret = 0;
4875 out:
4876         return ret;
4877 }
4878
4879 static void ext4_exit_feat_adverts(void)
4880 {
4881         kobject_put(&ext4_feat->f_kobj);
4882         wait_for_completion(&ext4_feat->f_kobj_unregister);
4883         kfree(ext4_feat);
4884 }
4885
4886 /* Shared across all ext4 file systems */
4887 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4888 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4889
4890 static int __init ext4_init_fs(void)
4891 {
4892         int i, err;
4893
4894         ext4_check_flag_values();
4895
4896         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4897                 mutex_init(&ext4__aio_mutex[i]);
4898                 init_waitqueue_head(&ext4__ioend_wq[i]);
4899         }
4900
4901         err = ext4_init_pageio();
4902         if (err)
4903                 return err;
4904         err = ext4_init_system_zone();
4905         if (err)
4906                 goto out6;
4907         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4908         if (!ext4_kset)
4909                 goto out5;
4910         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4911
4912         err = ext4_init_feat_adverts();
4913         if (err)
4914                 goto out4;
4915
4916         err = ext4_init_mballoc();
4917         if (err)
4918                 goto out3;
4919
4920         err = ext4_init_xattr();
4921         if (err)
4922                 goto out2;
4923         err = init_inodecache();
4924         if (err)
4925                 goto out1;
4926         register_as_ext3();
4927         register_as_ext2();
4928         err = register_filesystem(&ext4_fs_type);
4929         if (err)
4930                 goto out;
4931
4932         ext4_li_info = NULL;
4933         mutex_init(&ext4_li_mtx);
4934         return 0;
4935 out:
4936         unregister_as_ext2();
4937         unregister_as_ext3();
4938         destroy_inodecache();
4939 out1:
4940         ext4_exit_xattr();
4941 out2:
4942         ext4_exit_mballoc();
4943 out3:
4944         ext4_exit_feat_adverts();
4945 out4:
4946         if (ext4_proc_root)
4947                 remove_proc_entry("fs/ext4", NULL);
4948         kset_unregister(ext4_kset);
4949 out5:
4950         ext4_exit_system_zone();
4951 out6:
4952         ext4_exit_pageio();
4953         return err;
4954 }
4955
4956 static void __exit ext4_exit_fs(void)
4957 {
4958         ext4_destroy_lazyinit_thread();
4959         unregister_as_ext2();
4960         unregister_as_ext3();
4961         unregister_filesystem(&ext4_fs_type);
4962         destroy_inodecache();
4963         ext4_exit_xattr();
4964         ext4_exit_mballoc();
4965         ext4_exit_feat_adverts();
4966         remove_proc_entry("fs/ext4", NULL);
4967         kset_unregister(ext4_kset);
4968         ext4_exit_system_zone();
4969         ext4_exit_pageio();
4970 }
4971
4972 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4973 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4974 MODULE_LICENSE("GPL");
4975 module_init(ext4_init_fs)
4976 module_exit(ext4_exit_fs)