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