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