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 #ifdef CONFIG_QUOTA
1601         struct ext4_sb_info *sbi = EXT4_SB(sb);
1602 #endif
1603         char *p;
1604         substring_t args[MAX_OPT_ARGS];
1605         int token;
1606
1607         if (!options)
1608                 return 1;
1609
1610         while ((p = strsep(&options, ",")) != NULL) {
1611                 if (!*p)
1612                         continue;
1613                 /*
1614                  * Initialize args struct so we know whether arg was
1615                  * found; some options take optional arguments.
1616                  */
1617                 args[0].to = args[0].from = 0;
1618                 token = match_token(p, tokens, args);
1619                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1620                                      journal_ioprio, is_remount) < 0)
1621                         return 0;
1622         }
1623 #ifdef CONFIG_QUOTA
1624         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1625                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1626                         clear_opt(sb, USRQUOTA);
1627
1628                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1629                         clear_opt(sb, GRPQUOTA);
1630
1631                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1632                         ext4_msg(sb, KERN_ERR, "old and new quota "
1633                                         "format mixing");
1634                         return 0;
1635                 }
1636
1637                 if (!sbi->s_jquota_fmt) {
1638                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1639                                         "not specified");
1640                         return 0;
1641                 }
1642         } else {
1643                 if (sbi->s_jquota_fmt) {
1644                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1645                                         "specified with no journaling "
1646                                         "enabled");
1647                         return 0;
1648                 }
1649         }
1650 #endif
1651         return 1;
1652 }
1653
1654 static inline void ext4_show_quota_options(struct seq_file *seq,
1655                                            struct super_block *sb)
1656 {
1657 #if defined(CONFIG_QUOTA)
1658         struct ext4_sb_info *sbi = EXT4_SB(sb);
1659
1660         if (sbi->s_jquota_fmt) {
1661                 char *fmtname = "";
1662
1663                 switch (sbi->s_jquota_fmt) {
1664                 case QFMT_VFS_OLD:
1665                         fmtname = "vfsold";
1666                         break;
1667                 case QFMT_VFS_V0:
1668                         fmtname = "vfsv0";
1669                         break;
1670                 case QFMT_VFS_V1:
1671                         fmtname = "vfsv1";
1672                         break;
1673                 }
1674                 seq_printf(seq, ",jqfmt=%s", fmtname);
1675         }
1676
1677         if (sbi->s_qf_names[USRQUOTA])
1678                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1679
1680         if (sbi->s_qf_names[GRPQUOTA])
1681                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1682
1683         if (test_opt(sb, USRQUOTA))
1684                 seq_puts(seq, ",usrquota");
1685
1686         if (test_opt(sb, GRPQUOTA))
1687                 seq_puts(seq, ",grpquota");
1688 #endif
1689 }
1690
1691 static const char *token2str(int token)
1692 {
1693         static const struct match_token *t;
1694
1695         for (t = tokens; t->token != Opt_err; t++)
1696                 if (t->token == token && !strchr(t->pattern, '='))
1697                         break;
1698         return t->pattern;
1699 }
1700
1701 /*
1702  * Show an option if
1703  *  - it's set to a non-default value OR
1704  *  - if the per-sb default is different from the global default
1705  */
1706 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1707                               int nodefs)
1708 {
1709         struct ext4_sb_info *sbi = EXT4_SB(sb);
1710         struct ext4_super_block *es = sbi->s_es;
1711         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1712         const struct mount_opts *m;
1713         char sep = nodefs ? '\n' : ',';
1714
1715 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1716 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1717
1718         if (sbi->s_sb_block != 1)
1719                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1720
1721         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1722                 int want_set = m->flags & MOPT_SET;
1723                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1724                     (m->flags & MOPT_CLEAR_ERR))
1725                         continue;
1726                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1727                         continue; /* skip if same as the default */
1728                 if ((want_set &&
1729                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1730                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1731                         continue; /* select Opt_noFoo vs Opt_Foo */
1732                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1733         }
1734
1735         if (nodefs || sbi->s_resuid != EXT4_DEF_RESUID ||
1736             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1737                 SEQ_OPTS_PRINT("resuid=%u", sbi->s_resuid);
1738         if (nodefs || sbi->s_resgid != EXT4_DEF_RESGID ||
1739             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1740                 SEQ_OPTS_PRINT("resgid=%u", sbi->s_resgid);
1741         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1742         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1743                 SEQ_OPTS_PUTS("errors=remount-ro");
1744         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1745                 SEQ_OPTS_PUTS("errors=continue");
1746         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1747                 SEQ_OPTS_PUTS("errors=panic");
1748         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1749                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1750         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1751                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1752         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1753                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1754         if (sb->s_flags & MS_I_VERSION)
1755                 SEQ_OPTS_PUTS("i_version");
1756         if (nodefs || sbi->s_stripe)
1757                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1758         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1759                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1760                         SEQ_OPTS_PUTS("data=journal");
1761                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1762                         SEQ_OPTS_PUTS("data=ordered");
1763                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1764                         SEQ_OPTS_PUTS("data=writeback");
1765         }
1766         if (nodefs ||
1767             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1768                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1769                                sbi->s_inode_readahead_blks);
1770
1771         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1772                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1773                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1774
1775         ext4_show_quota_options(seq, sb);
1776         return 0;
1777 }
1778
1779 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1780 {
1781         return _ext4_show_options(seq, root->d_sb, 0);
1782 }
1783
1784 static int options_seq_show(struct seq_file *seq, void *offset)
1785 {
1786         struct super_block *sb = seq->private;
1787         int rc;
1788
1789         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1790         rc = _ext4_show_options(seq, sb, 1);
1791         seq_puts(seq, "\n");
1792         return rc;
1793 }
1794
1795 static int options_open_fs(struct inode *inode, struct file *file)
1796 {
1797         return single_open(file, options_seq_show, PDE(inode)->data);
1798 }
1799
1800 static const struct file_operations ext4_seq_options_fops = {
1801         .owner = THIS_MODULE,
1802         .open = options_open_fs,
1803         .read = seq_read,
1804         .llseek = seq_lseek,
1805         .release = single_release,
1806 };
1807
1808 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1809                             int read_only)
1810 {
1811         struct ext4_sb_info *sbi = EXT4_SB(sb);
1812         int res = 0;
1813
1814         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1815                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1816                          "forcing read-only mode");
1817                 res = MS_RDONLY;
1818         }
1819         if (read_only)
1820                 goto done;
1821         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1822                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1823                          "running e2fsck is recommended");
1824         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1825                 ext4_msg(sb, KERN_WARNING,
1826                          "warning: mounting fs with errors, "
1827                          "running e2fsck is recommended");
1828         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1829                  le16_to_cpu(es->s_mnt_count) >=
1830                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1831                 ext4_msg(sb, KERN_WARNING,
1832                          "warning: maximal mount count reached, "
1833                          "running e2fsck is recommended");
1834         else if (le32_to_cpu(es->s_checkinterval) &&
1835                 (le32_to_cpu(es->s_lastcheck) +
1836                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1837                 ext4_msg(sb, KERN_WARNING,
1838                          "warning: checktime reached, "
1839                          "running e2fsck is recommended");
1840         if (!sbi->s_journal)
1841                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1842         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1843                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1844         le16_add_cpu(&es->s_mnt_count, 1);
1845         es->s_mtime = cpu_to_le32(get_seconds());
1846         ext4_update_dynamic_rev(sb);
1847         if (sbi->s_journal)
1848                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1849
1850         ext4_commit_super(sb, 1);
1851 done:
1852         if (test_opt(sb, DEBUG))
1853                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1854                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1855                         sb->s_blocksize,
1856                         sbi->s_groups_count,
1857                         EXT4_BLOCKS_PER_GROUP(sb),
1858                         EXT4_INODES_PER_GROUP(sb),
1859                         sbi->s_mount_opt, sbi->s_mount_opt2);
1860
1861         cleancache_init_fs(sb);
1862         return res;
1863 }
1864
1865 static int ext4_fill_flex_info(struct super_block *sb)
1866 {
1867         struct ext4_sb_info *sbi = EXT4_SB(sb);
1868         struct ext4_group_desc *gdp = NULL;
1869         ext4_group_t flex_group_count;
1870         ext4_group_t flex_group;
1871         unsigned int groups_per_flex = 0;
1872         size_t size;
1873         int i;
1874
1875         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1876         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1877                 sbi->s_log_groups_per_flex = 0;
1878                 return 1;
1879         }
1880         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1881
1882         /* We allocate both existing and potentially added groups */
1883         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1884                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1885                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1886         size = flex_group_count * sizeof(struct flex_groups);
1887         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1888         if (sbi->s_flex_groups == NULL) {
1889                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1890                          flex_group_count);
1891                 goto failed;
1892         }
1893
1894         for (i = 0; i < sbi->s_groups_count; i++) {
1895                 gdp = ext4_get_group_desc(sb, i, NULL);
1896
1897                 flex_group = ext4_flex_group(sbi, i);
1898                 atomic_add(ext4_free_inodes_count(sb, gdp),
1899                            &sbi->s_flex_groups[flex_group].free_inodes);
1900                 atomic_add(ext4_free_group_clusters(sb, gdp),
1901                            &sbi->s_flex_groups[flex_group].free_clusters);
1902                 atomic_add(ext4_used_dirs_count(sb, gdp),
1903                            &sbi->s_flex_groups[flex_group].used_dirs);
1904         }
1905
1906         return 1;
1907 failed:
1908         return 0;
1909 }
1910
1911 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1912                             struct ext4_group_desc *gdp)
1913 {
1914         __u16 crc = 0;
1915
1916         if (sbi->s_es->s_feature_ro_compat &
1917             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
1918                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
1919                 __le32 le_group = cpu_to_le32(block_group);
1920
1921                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1922                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1923                 crc = crc16(crc, (__u8 *)gdp, offset);
1924                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
1925                 /* for checksum of struct ext4_group_desc do the rest...*/
1926                 if ((sbi->s_es->s_feature_incompat &
1927                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1928                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
1929                         crc = crc16(crc, (__u8 *)gdp + offset,
1930                                     le16_to_cpu(sbi->s_es->s_desc_size) -
1931                                         offset);
1932         }
1933
1934         return cpu_to_le16(crc);
1935 }
1936
1937 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
1938                                 struct ext4_group_desc *gdp)
1939 {
1940         if ((sbi->s_es->s_feature_ro_compat &
1941              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
1942             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
1943                 return 0;
1944
1945         return 1;
1946 }
1947
1948 /* Called at mount-time, super-block is locked */
1949 static int ext4_check_descriptors(struct super_block *sb,
1950                                   ext4_group_t *first_not_zeroed)
1951 {
1952         struct ext4_sb_info *sbi = EXT4_SB(sb);
1953         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1954         ext4_fsblk_t last_block;
1955         ext4_fsblk_t block_bitmap;
1956         ext4_fsblk_t inode_bitmap;
1957         ext4_fsblk_t inode_table;
1958         int flexbg_flag = 0;
1959         ext4_group_t i, grp = sbi->s_groups_count;
1960
1961         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
1962                 flexbg_flag = 1;
1963
1964         ext4_debug("Checking group descriptors");
1965
1966         for (i = 0; i < sbi->s_groups_count; i++) {
1967                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
1968
1969                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
1970                         last_block = ext4_blocks_count(sbi->s_es) - 1;
1971                 else
1972                         last_block = first_block +
1973                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
1974
1975                 if ((grp == sbi->s_groups_count) &&
1976                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
1977                         grp = i;
1978
1979                 block_bitmap = ext4_block_bitmap(sb, gdp);
1980                 if (block_bitmap < first_block || block_bitmap > last_block) {
1981                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1982                                "Block bitmap for group %u not in group "
1983                                "(block %llu)!", i, block_bitmap);
1984                         return 0;
1985                 }
1986                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
1987                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
1988                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1989                                "Inode bitmap for group %u not in group "
1990                                "(block %llu)!", i, inode_bitmap);
1991                         return 0;
1992                 }
1993                 inode_table = ext4_inode_table(sb, gdp);
1994                 if (inode_table < first_block ||
1995                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
1996                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
1997                                "Inode table for group %u not in group "
1998                                "(block %llu)!", i, inode_table);
1999                         return 0;
2000                 }
2001                 ext4_lock_group(sb, i);
2002                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2003                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2004                                  "Checksum for group %u failed (%u!=%u)",
2005                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2006                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2007                         if (!(sb->s_flags & MS_RDONLY)) {
2008                                 ext4_unlock_group(sb, i);
2009                                 return 0;
2010                         }
2011                 }
2012                 ext4_unlock_group(sb, i);
2013                 if (!flexbg_flag)
2014                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2015         }
2016         if (NULL != first_not_zeroed)
2017                 *first_not_zeroed = grp;
2018
2019         ext4_free_blocks_count_set(sbi->s_es,
2020                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2021         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2022         return 1;
2023 }
2024
2025 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2026  * the superblock) which were deleted from all directories, but held open by
2027  * a process at the time of a crash.  We walk the list and try to delete these
2028  * inodes at recovery time (only with a read-write filesystem).
2029  *
2030  * In order to keep the orphan inode chain consistent during traversal (in
2031  * case of crash during recovery), we link each inode into the superblock
2032  * orphan list_head and handle it the same way as an inode deletion during
2033  * normal operation (which journals the operations for us).
2034  *
2035  * We only do an iget() and an iput() on each inode, which is very safe if we
2036  * accidentally point at an in-use or already deleted inode.  The worst that
2037  * can happen in this case is that we get a "bit already cleared" message from
2038  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2039  * e2fsck was run on this filesystem, and it must have already done the orphan
2040  * inode cleanup for us, so we can safely abort without any further action.
2041  */
2042 static void ext4_orphan_cleanup(struct super_block *sb,
2043                                 struct ext4_super_block *es)
2044 {
2045         unsigned int s_flags = sb->s_flags;
2046         int nr_orphans = 0, nr_truncates = 0;
2047 #ifdef CONFIG_QUOTA
2048         int i;
2049 #endif
2050         if (!es->s_last_orphan) {
2051                 jbd_debug(4, "no orphan inodes to clean up\n");
2052                 return;
2053         }
2054
2055         if (bdev_read_only(sb->s_bdev)) {
2056                 ext4_msg(sb, KERN_ERR, "write access "
2057                         "unavailable, skipping orphan cleanup");
2058                 return;
2059         }
2060
2061         /* Check if feature set would not allow a r/w mount */
2062         if (!ext4_feature_set_ok(sb, 0)) {
2063                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2064                          "unknown ROCOMPAT features");
2065                 return;
2066         }
2067
2068         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2069                 if (es->s_last_orphan)
2070                         jbd_debug(1, "Errors on filesystem, "
2071                                   "clearing orphan list.\n");
2072                 es->s_last_orphan = 0;
2073                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2074                 return;
2075         }
2076
2077         if (s_flags & MS_RDONLY) {
2078                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2079                 sb->s_flags &= ~MS_RDONLY;
2080         }
2081 #ifdef CONFIG_QUOTA
2082         /* Needed for iput() to work correctly and not trash data */
2083         sb->s_flags |= MS_ACTIVE;
2084         /* Turn on quotas so that they are updated correctly */
2085         for (i = 0; i < MAXQUOTAS; i++) {
2086                 if (EXT4_SB(sb)->s_qf_names[i]) {
2087                         int ret = ext4_quota_on_mount(sb, i);
2088                         if (ret < 0)
2089                                 ext4_msg(sb, KERN_ERR,
2090                                         "Cannot turn on journaled "
2091                                         "quota: error %d", ret);
2092                 }
2093         }
2094 #endif
2095
2096         while (es->s_last_orphan) {
2097                 struct inode *inode;
2098
2099                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2100                 if (IS_ERR(inode)) {
2101                         es->s_last_orphan = 0;
2102                         break;
2103                 }
2104
2105                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2106                 dquot_initialize(inode);
2107                 if (inode->i_nlink) {
2108                         ext4_msg(sb, KERN_DEBUG,
2109                                 "%s: truncating inode %lu to %lld bytes",
2110                                 __func__, inode->i_ino, inode->i_size);
2111                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2112                                   inode->i_ino, inode->i_size);
2113                         ext4_truncate(inode);
2114                         nr_truncates++;
2115                 } else {
2116                         ext4_msg(sb, KERN_DEBUG,
2117                                 "%s: deleting unreferenced inode %lu",
2118                                 __func__, inode->i_ino);
2119                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2120                                   inode->i_ino);
2121                         nr_orphans++;
2122                 }
2123                 iput(inode);  /* The delete magic happens here! */
2124         }
2125
2126 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2127
2128         if (nr_orphans)
2129                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2130                        PLURAL(nr_orphans));
2131         if (nr_truncates)
2132                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2133                        PLURAL(nr_truncates));
2134 #ifdef CONFIG_QUOTA
2135         /* Turn quotas off */
2136         for (i = 0; i < MAXQUOTAS; i++) {
2137                 if (sb_dqopt(sb)->files[i])
2138                         dquot_quota_off(sb, i);
2139         }
2140 #endif
2141         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2142 }
2143
2144 /*
2145  * Maximal extent format file size.
2146  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2147  * extent format containers, within a sector_t, and within i_blocks
2148  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2149  * so that won't be a limiting factor.
2150  *
2151  * However there is other limiting factor. We do store extents in the form
2152  * of starting block and length, hence the resulting length of the extent
2153  * covering maximum file size must fit into on-disk format containers as
2154  * well. Given that length is always by 1 unit bigger than max unit (because
2155  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2156  *
2157  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2158  */
2159 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2160 {
2161         loff_t res;
2162         loff_t upper_limit = MAX_LFS_FILESIZE;
2163
2164         /* small i_blocks in vfs inode? */
2165         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2166                 /*
2167                  * CONFIG_LBDAF is not enabled implies the inode
2168                  * i_block represent total blocks in 512 bytes
2169                  * 32 == size of vfs inode i_blocks * 8
2170                  */
2171                 upper_limit = (1LL << 32) - 1;
2172
2173                 /* total blocks in file system block size */
2174                 upper_limit >>= (blkbits - 9);
2175                 upper_limit <<= blkbits;
2176         }
2177
2178         /*
2179          * 32-bit extent-start container, ee_block. We lower the maxbytes
2180          * by one fs block, so ee_len can cover the extent of maximum file
2181          * size
2182          */
2183         res = (1LL << 32) - 1;
2184         res <<= blkbits;
2185
2186         /* Sanity check against vm- & vfs- imposed limits */
2187         if (res > upper_limit)
2188                 res = upper_limit;
2189
2190         return res;
2191 }
2192
2193 /*
2194  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2195  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2196  * We need to be 1 filesystem block less than the 2^48 sector limit.
2197  */
2198 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2199 {
2200         loff_t res = EXT4_NDIR_BLOCKS;
2201         int meta_blocks;
2202         loff_t upper_limit;
2203         /* This is calculated to be the largest file size for a dense, block
2204          * mapped file such that the file's total number of 512-byte sectors,
2205          * including data and all indirect blocks, does not exceed (2^48 - 1).
2206          *
2207          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2208          * number of 512-byte sectors of the file.
2209          */
2210
2211         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2212                 /*
2213                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2214                  * the inode i_block field represents total file blocks in
2215                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2216                  */
2217                 upper_limit = (1LL << 32) - 1;
2218
2219                 /* total blocks in file system block size */
2220                 upper_limit >>= (bits - 9);
2221
2222         } else {
2223                 /*
2224                  * We use 48 bit ext4_inode i_blocks
2225                  * With EXT4_HUGE_FILE_FL set the i_blocks
2226                  * represent total number of blocks in
2227                  * file system block size
2228                  */
2229                 upper_limit = (1LL << 48) - 1;
2230
2231         }
2232
2233         /* indirect blocks */
2234         meta_blocks = 1;
2235         /* double indirect blocks */
2236         meta_blocks += 1 + (1LL << (bits-2));
2237         /* tripple indirect blocks */
2238         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2239
2240         upper_limit -= meta_blocks;
2241         upper_limit <<= bits;
2242
2243         res += 1LL << (bits-2);
2244         res += 1LL << (2*(bits-2));
2245         res += 1LL << (3*(bits-2));
2246         res <<= bits;
2247         if (res > upper_limit)
2248                 res = upper_limit;
2249
2250         if (res > MAX_LFS_FILESIZE)
2251                 res = MAX_LFS_FILESIZE;
2252
2253         return res;
2254 }
2255
2256 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2257                                    ext4_fsblk_t logical_sb_block, int nr)
2258 {
2259         struct ext4_sb_info *sbi = EXT4_SB(sb);
2260         ext4_group_t bg, first_meta_bg;
2261         int has_super = 0;
2262
2263         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2264
2265         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2266             nr < first_meta_bg)
2267                 return logical_sb_block + nr + 1;
2268         bg = sbi->s_desc_per_block * nr;
2269         if (ext4_bg_has_super(sb, bg))
2270                 has_super = 1;
2271
2272         return (has_super + ext4_group_first_block_no(sb, bg));
2273 }
2274
2275 /**
2276  * ext4_get_stripe_size: Get the stripe size.
2277  * @sbi: In memory super block info
2278  *
2279  * If we have specified it via mount option, then
2280  * use the mount option value. If the value specified at mount time is
2281  * greater than the blocks per group use the super block value.
2282  * If the super block value is greater than blocks per group return 0.
2283  * Allocator needs it be less than blocks per group.
2284  *
2285  */
2286 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2287 {
2288         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2289         unsigned long stripe_width =
2290                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2291         int ret;
2292
2293         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2294                 ret = sbi->s_stripe;
2295         else if (stripe_width <= sbi->s_blocks_per_group)
2296                 ret = stripe_width;
2297         else if (stride <= sbi->s_blocks_per_group)
2298                 ret = stride;
2299         else
2300                 ret = 0;
2301
2302         /*
2303          * If the stripe width is 1, this makes no sense and
2304          * we set it to 0 to turn off stripe handling code.
2305          */
2306         if (ret <= 1)
2307                 ret = 0;
2308
2309         return ret;
2310 }
2311
2312 /* sysfs supprt */
2313
2314 struct ext4_attr {
2315         struct attribute attr;
2316         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2317         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2318                          const char *, size_t);
2319         int offset;
2320 };
2321
2322 static int parse_strtoul(const char *buf,
2323                 unsigned long max, unsigned long *value)
2324 {
2325         char *endp;
2326
2327         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2328         endp = skip_spaces(endp);
2329         if (*endp || *value > max)
2330                 return -EINVAL;
2331
2332         return 0;
2333 }
2334
2335 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2336                                               struct ext4_sb_info *sbi,
2337                                               char *buf)
2338 {
2339         return snprintf(buf, PAGE_SIZE, "%llu\n",
2340                 (s64) EXT4_C2B(sbi,
2341                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2342 }
2343
2344 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2345                                          struct ext4_sb_info *sbi, char *buf)
2346 {
2347         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2348
2349         if (!sb->s_bdev->bd_part)
2350                 return snprintf(buf, PAGE_SIZE, "0\n");
2351         return snprintf(buf, PAGE_SIZE, "%lu\n",
2352                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2353                          sbi->s_sectors_written_start) >> 1);
2354 }
2355
2356 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2357                                           struct ext4_sb_info *sbi, char *buf)
2358 {
2359         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2360
2361         if (!sb->s_bdev->bd_part)
2362                 return snprintf(buf, PAGE_SIZE, "0\n");
2363         return snprintf(buf, PAGE_SIZE, "%llu\n",
2364                         (unsigned long long)(sbi->s_kbytes_written +
2365                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2366                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2367 }
2368
2369 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2370                                           struct ext4_sb_info *sbi,
2371                                           const char *buf, size_t count)
2372 {
2373         unsigned long t;
2374
2375         if (parse_strtoul(buf, 0x40000000, &t))
2376                 return -EINVAL;
2377
2378         if (t && !is_power_of_2(t))
2379                 return -EINVAL;
2380
2381         sbi->s_inode_readahead_blks = t;
2382         return count;
2383 }
2384
2385 static ssize_t sbi_ui_show(struct ext4_attr *a,
2386                            struct ext4_sb_info *sbi, char *buf)
2387 {
2388         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2389
2390         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2391 }
2392
2393 static ssize_t sbi_ui_store(struct ext4_attr *a,
2394                             struct ext4_sb_info *sbi,
2395                             const char *buf, size_t count)
2396 {
2397         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2398         unsigned long t;
2399
2400         if (parse_strtoul(buf, 0xffffffff, &t))
2401                 return -EINVAL;
2402         *ui = t;
2403         return count;
2404 }
2405
2406 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2407 static struct ext4_attr ext4_attr_##_name = {                   \
2408         .attr = {.name = __stringify(_name), .mode = _mode },   \
2409         .show   = _show,                                        \
2410         .store  = _store,                                       \
2411         .offset = offsetof(struct ext4_sb_info, _elname),       \
2412 }
2413 #define EXT4_ATTR(name, mode, show, store) \
2414 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2415
2416 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2417 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2418 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2419 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2420         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2421 #define ATTR_LIST(name) &ext4_attr_##name.attr
2422
2423 EXT4_RO_ATTR(delayed_allocation_blocks);
2424 EXT4_RO_ATTR(session_write_kbytes);
2425 EXT4_RO_ATTR(lifetime_write_kbytes);
2426 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2427                  inode_readahead_blks_store, s_inode_readahead_blks);
2428 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2429 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2430 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2431 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2432 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2433 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2434 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2435 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2436
2437 static struct attribute *ext4_attrs[] = {
2438         ATTR_LIST(delayed_allocation_blocks),
2439         ATTR_LIST(session_write_kbytes),
2440         ATTR_LIST(lifetime_write_kbytes),
2441         ATTR_LIST(inode_readahead_blks),
2442         ATTR_LIST(inode_goal),
2443         ATTR_LIST(mb_stats),
2444         ATTR_LIST(mb_max_to_scan),
2445         ATTR_LIST(mb_min_to_scan),
2446         ATTR_LIST(mb_order2_req),
2447         ATTR_LIST(mb_stream_req),
2448         ATTR_LIST(mb_group_prealloc),
2449         ATTR_LIST(max_writeback_mb_bump),
2450         NULL,
2451 };
2452
2453 /* Features this copy of ext4 supports */
2454 EXT4_INFO_ATTR(lazy_itable_init);
2455 EXT4_INFO_ATTR(batched_discard);
2456
2457 static struct attribute *ext4_feat_attrs[] = {
2458         ATTR_LIST(lazy_itable_init),
2459         ATTR_LIST(batched_discard),
2460         NULL,
2461 };
2462
2463 static ssize_t ext4_attr_show(struct kobject *kobj,
2464                               struct attribute *attr, char *buf)
2465 {
2466         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2467                                                 s_kobj);
2468         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2469
2470         return a->show ? a->show(a, sbi, buf) : 0;
2471 }
2472
2473 static ssize_t ext4_attr_store(struct kobject *kobj,
2474                                struct attribute *attr,
2475                                const char *buf, size_t len)
2476 {
2477         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2478                                                 s_kobj);
2479         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2480
2481         return a->store ? a->store(a, sbi, buf, len) : 0;
2482 }
2483
2484 static void ext4_sb_release(struct kobject *kobj)
2485 {
2486         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2487                                                 s_kobj);
2488         complete(&sbi->s_kobj_unregister);
2489 }
2490
2491 static const struct sysfs_ops ext4_attr_ops = {
2492         .show   = ext4_attr_show,
2493         .store  = ext4_attr_store,
2494 };
2495
2496 static struct kobj_type ext4_ktype = {
2497         .default_attrs  = ext4_attrs,
2498         .sysfs_ops      = &ext4_attr_ops,
2499         .release        = ext4_sb_release,
2500 };
2501
2502 static void ext4_feat_release(struct kobject *kobj)
2503 {
2504         complete(&ext4_feat->f_kobj_unregister);
2505 }
2506
2507 static struct kobj_type ext4_feat_ktype = {
2508         .default_attrs  = ext4_feat_attrs,
2509         .sysfs_ops      = &ext4_attr_ops,
2510         .release        = ext4_feat_release,
2511 };
2512
2513 /*
2514  * Check whether this filesystem can be mounted based on
2515  * the features present and the RDONLY/RDWR mount requested.
2516  * Returns 1 if this filesystem can be mounted as requested,
2517  * 0 if it cannot be.
2518  */
2519 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2520 {
2521         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2522                 ext4_msg(sb, KERN_ERR,
2523                         "Couldn't mount because of "
2524                         "unsupported optional features (%x)",
2525                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2526                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2527                 return 0;
2528         }
2529
2530         if (readonly)
2531                 return 1;
2532
2533         /* Check that feature set is OK for a read-write mount */
2534         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2535                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2536                          "unsupported optional features (%x)",
2537                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2538                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2539                 return 0;
2540         }
2541         /*
2542          * Large file size enabled file system can only be mounted
2543          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2544          */
2545         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2546                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2547                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2548                                  "cannot be mounted RDWR without "
2549                                  "CONFIG_LBDAF");
2550                         return 0;
2551                 }
2552         }
2553         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2554             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2555                 ext4_msg(sb, KERN_ERR,
2556                          "Can't support bigalloc feature without "
2557                          "extents feature\n");
2558                 return 0;
2559         }
2560         return 1;
2561 }
2562
2563 /*
2564  * This function is called once a day if we have errors logged
2565  * on the file system
2566  */
2567 static void print_daily_error_info(unsigned long arg)
2568 {
2569         struct super_block *sb = (struct super_block *) arg;
2570         struct ext4_sb_info *sbi;
2571         struct ext4_super_block *es;
2572
2573         sbi = EXT4_SB(sb);
2574         es = sbi->s_es;
2575
2576         if (es->s_error_count)
2577                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2578                          le32_to_cpu(es->s_error_count));
2579         if (es->s_first_error_time) {
2580                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2581                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2582                        (int) sizeof(es->s_first_error_func),
2583                        es->s_first_error_func,
2584                        le32_to_cpu(es->s_first_error_line));
2585                 if (es->s_first_error_ino)
2586                         printk(": inode %u",
2587                                le32_to_cpu(es->s_first_error_ino));
2588                 if (es->s_first_error_block)
2589                         printk(": block %llu", (unsigned long long)
2590                                le64_to_cpu(es->s_first_error_block));
2591                 printk("\n");
2592         }
2593         if (es->s_last_error_time) {
2594                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2595                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2596                        (int) sizeof(es->s_last_error_func),
2597                        es->s_last_error_func,
2598                        le32_to_cpu(es->s_last_error_line));
2599                 if (es->s_last_error_ino)
2600                         printk(": inode %u",
2601                                le32_to_cpu(es->s_last_error_ino));
2602                 if (es->s_last_error_block)
2603                         printk(": block %llu", (unsigned long long)
2604                                le64_to_cpu(es->s_last_error_block));
2605                 printk("\n");
2606         }
2607         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2608 }
2609
2610 /* Find next suitable group and run ext4_init_inode_table */
2611 static int ext4_run_li_request(struct ext4_li_request *elr)
2612 {
2613         struct ext4_group_desc *gdp = NULL;
2614         ext4_group_t group, ngroups;
2615         struct super_block *sb;
2616         unsigned long timeout = 0;
2617         int ret = 0;
2618
2619         sb = elr->lr_super;
2620         ngroups = EXT4_SB(sb)->s_groups_count;
2621
2622         for (group = elr->lr_next_group; group < ngroups; group++) {
2623                 gdp = ext4_get_group_desc(sb, group, NULL);
2624                 if (!gdp) {
2625                         ret = 1;
2626                         break;
2627                 }
2628
2629                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2630                         break;
2631         }
2632
2633         if (group == ngroups)
2634                 ret = 1;
2635
2636         if (!ret) {
2637                 timeout = jiffies;
2638                 ret = ext4_init_inode_table(sb, group,
2639                                             elr->lr_timeout ? 0 : 1);
2640                 if (elr->lr_timeout == 0) {
2641                         timeout = (jiffies - timeout) *
2642                                   elr->lr_sbi->s_li_wait_mult;
2643                         elr->lr_timeout = timeout;
2644                 }
2645                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2646                 elr->lr_next_group = group + 1;
2647         }
2648
2649         return ret;
2650 }
2651
2652 /*
2653  * Remove lr_request from the list_request and free the
2654  * request structure. Should be called with li_list_mtx held
2655  */
2656 static void ext4_remove_li_request(struct ext4_li_request *elr)
2657 {
2658         struct ext4_sb_info *sbi;
2659
2660         if (!elr)
2661                 return;
2662
2663         sbi = elr->lr_sbi;
2664
2665         list_del(&elr->lr_request);
2666         sbi->s_li_request = NULL;
2667         kfree(elr);
2668 }
2669
2670 static void ext4_unregister_li_request(struct super_block *sb)
2671 {
2672         mutex_lock(&ext4_li_mtx);
2673         if (!ext4_li_info) {
2674                 mutex_unlock(&ext4_li_mtx);
2675                 return;
2676         }
2677
2678         mutex_lock(&ext4_li_info->li_list_mtx);
2679         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2680         mutex_unlock(&ext4_li_info->li_list_mtx);
2681         mutex_unlock(&ext4_li_mtx);
2682 }
2683
2684 static struct task_struct *ext4_lazyinit_task;
2685
2686 /*
2687  * This is the function where ext4lazyinit thread lives. It walks
2688  * through the request list searching for next scheduled filesystem.
2689  * When such a fs is found, run the lazy initialization request
2690  * (ext4_rn_li_request) and keep track of the time spend in this
2691  * function. Based on that time we compute next schedule time of
2692  * the request. When walking through the list is complete, compute
2693  * next waking time and put itself into sleep.
2694  */
2695 static int ext4_lazyinit_thread(void *arg)
2696 {
2697         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2698         struct list_head *pos, *n;
2699         struct ext4_li_request *elr;
2700         unsigned long next_wakeup, cur;
2701
2702         BUG_ON(NULL == eli);
2703
2704 cont_thread:
2705         while (true) {
2706                 next_wakeup = MAX_JIFFY_OFFSET;
2707
2708                 mutex_lock(&eli->li_list_mtx);
2709                 if (list_empty(&eli->li_request_list)) {
2710                         mutex_unlock(&eli->li_list_mtx);
2711                         goto exit_thread;
2712                 }
2713
2714                 list_for_each_safe(pos, n, &eli->li_request_list) {
2715                         elr = list_entry(pos, struct ext4_li_request,
2716                                          lr_request);
2717
2718                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2719                                 if (ext4_run_li_request(elr) != 0) {
2720                                         /* error, remove the lazy_init job */
2721                                         ext4_remove_li_request(elr);
2722                                         continue;
2723                                 }
2724                         }
2725
2726                         if (time_before(elr->lr_next_sched, next_wakeup))
2727                                 next_wakeup = elr->lr_next_sched;
2728                 }
2729                 mutex_unlock(&eli->li_list_mtx);
2730
2731                 try_to_freeze();
2732
2733                 cur = jiffies;
2734                 if ((time_after_eq(cur, next_wakeup)) ||
2735                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2736                         cond_resched();
2737                         continue;
2738                 }
2739
2740                 schedule_timeout_interruptible(next_wakeup - cur);
2741
2742                 if (kthread_should_stop()) {
2743                         ext4_clear_request_list();
2744                         goto exit_thread;
2745                 }
2746         }
2747
2748 exit_thread:
2749         /*
2750          * It looks like the request list is empty, but we need
2751          * to check it under the li_list_mtx lock, to prevent any
2752          * additions into it, and of course we should lock ext4_li_mtx
2753          * to atomically free the list and ext4_li_info, because at
2754          * this point another ext4 filesystem could be registering
2755          * new one.
2756          */
2757         mutex_lock(&ext4_li_mtx);
2758         mutex_lock(&eli->li_list_mtx);
2759         if (!list_empty(&eli->li_request_list)) {
2760                 mutex_unlock(&eli->li_list_mtx);
2761                 mutex_unlock(&ext4_li_mtx);
2762                 goto cont_thread;
2763         }
2764         mutex_unlock(&eli->li_list_mtx);
2765         kfree(ext4_li_info);
2766         ext4_li_info = NULL;
2767         mutex_unlock(&ext4_li_mtx);
2768
2769         return 0;
2770 }
2771
2772 static void ext4_clear_request_list(void)
2773 {
2774         struct list_head *pos, *n;
2775         struct ext4_li_request *elr;
2776
2777         mutex_lock(&ext4_li_info->li_list_mtx);
2778         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2779                 elr = list_entry(pos, struct ext4_li_request,
2780                                  lr_request);
2781                 ext4_remove_li_request(elr);
2782         }
2783         mutex_unlock(&ext4_li_info->li_list_mtx);
2784 }
2785
2786 static int ext4_run_lazyinit_thread(void)
2787 {
2788         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2789                                          ext4_li_info, "ext4lazyinit");
2790         if (IS_ERR(ext4_lazyinit_task)) {
2791                 int err = PTR_ERR(ext4_lazyinit_task);
2792                 ext4_clear_request_list();
2793                 kfree(ext4_li_info);
2794                 ext4_li_info = NULL;
2795                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2796                                  "initialization thread\n",
2797                                  err);
2798                 return err;
2799         }
2800         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2801         return 0;
2802 }
2803
2804 /*
2805  * Check whether it make sense to run itable init. thread or not.
2806  * If there is at least one uninitialized inode table, return
2807  * corresponding group number, else the loop goes through all
2808  * groups and return total number of groups.
2809  */
2810 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2811 {
2812         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2813         struct ext4_group_desc *gdp = NULL;
2814
2815         for (group = 0; group < ngroups; group++) {
2816                 gdp = ext4_get_group_desc(sb, group, NULL);
2817                 if (!gdp)
2818                         continue;
2819
2820                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2821                         break;
2822         }
2823
2824         return group;
2825 }
2826
2827 static int ext4_li_info_new(void)
2828 {
2829         struct ext4_lazy_init *eli = NULL;
2830
2831         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2832         if (!eli)
2833                 return -ENOMEM;
2834
2835         INIT_LIST_HEAD(&eli->li_request_list);
2836         mutex_init(&eli->li_list_mtx);
2837
2838         eli->li_state |= EXT4_LAZYINIT_QUIT;
2839
2840         ext4_li_info = eli;
2841
2842         return 0;
2843 }
2844
2845 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2846                                             ext4_group_t start)
2847 {
2848         struct ext4_sb_info *sbi = EXT4_SB(sb);
2849         struct ext4_li_request *elr;
2850         unsigned long rnd;
2851
2852         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2853         if (!elr)
2854                 return NULL;
2855
2856         elr->lr_super = sb;
2857         elr->lr_sbi = sbi;
2858         elr->lr_next_group = start;
2859
2860         /*
2861          * Randomize first schedule time of the request to
2862          * spread the inode table initialization requests
2863          * better.
2864          */
2865         get_random_bytes(&rnd, sizeof(rnd));
2866         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2867                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2868
2869         return elr;
2870 }
2871
2872 static int ext4_register_li_request(struct super_block *sb,
2873                                     ext4_group_t first_not_zeroed)
2874 {
2875         struct ext4_sb_info *sbi = EXT4_SB(sb);
2876         struct ext4_li_request *elr;
2877         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2878         int ret = 0;
2879
2880         if (sbi->s_li_request != NULL) {
2881                 /*
2882                  * Reset timeout so it can be computed again, because
2883                  * s_li_wait_mult might have changed.
2884                  */
2885                 sbi->s_li_request->lr_timeout = 0;
2886                 return 0;
2887         }
2888
2889         if (first_not_zeroed == ngroups ||
2890             (sb->s_flags & MS_RDONLY) ||
2891             !test_opt(sb, INIT_INODE_TABLE))
2892                 return 0;
2893
2894         elr = ext4_li_request_new(sb, first_not_zeroed);
2895         if (!elr)
2896                 return -ENOMEM;
2897
2898         mutex_lock(&ext4_li_mtx);
2899
2900         if (NULL == ext4_li_info) {
2901                 ret = ext4_li_info_new();
2902                 if (ret)
2903                         goto out;
2904         }
2905
2906         mutex_lock(&ext4_li_info->li_list_mtx);
2907         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2908         mutex_unlock(&ext4_li_info->li_list_mtx);
2909
2910         sbi->s_li_request = elr;
2911         /*
2912          * set elr to NULL here since it has been inserted to
2913          * the request_list and the removal and free of it is
2914          * handled by ext4_clear_request_list from now on.
2915          */
2916         elr = NULL;
2917
2918         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2919                 ret = ext4_run_lazyinit_thread();
2920                 if (ret)
2921                         goto out;
2922         }
2923 out:
2924         mutex_unlock(&ext4_li_mtx);
2925         if (ret)
2926                 kfree(elr);
2927         return ret;
2928 }
2929
2930 /*
2931  * We do not need to lock anything since this is called on
2932  * module unload.
2933  */
2934 static void ext4_destroy_lazyinit_thread(void)
2935 {
2936         /*
2937          * If thread exited earlier
2938          * there's nothing to be done.
2939          */
2940         if (!ext4_li_info || !ext4_lazyinit_task)
2941                 return;
2942
2943         kthread_stop(ext4_lazyinit_task);
2944 }
2945
2946 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
2947 {
2948         char *orig_data = kstrdup(data, GFP_KERNEL);
2949         struct buffer_head *bh;
2950         struct ext4_super_block *es = NULL;
2951         struct ext4_sb_info *sbi;
2952         ext4_fsblk_t block;
2953         ext4_fsblk_t sb_block = get_sb_block(&data);
2954         ext4_fsblk_t logical_sb_block;
2955         unsigned long offset = 0;
2956         unsigned long journal_devnum = 0;
2957         unsigned long def_mount_opts;
2958         struct inode *root;
2959         char *cp;
2960         const char *descr;
2961         int ret = -ENOMEM;
2962         int blocksize, clustersize;
2963         unsigned int db_count;
2964         unsigned int i;
2965         int needs_recovery, has_huge_files, has_bigalloc;
2966         __u64 blocks_count;
2967         int err;
2968         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
2969         ext4_group_t first_not_zeroed;
2970
2971         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
2972         if (!sbi)
2973                 goto out_free_orig;
2974
2975         sbi->s_blockgroup_lock =
2976                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
2977         if (!sbi->s_blockgroup_lock) {
2978                 kfree(sbi);
2979                 goto out_free_orig;
2980         }
2981         sb->s_fs_info = sbi;
2982         sbi->s_mount_opt = 0;
2983         sbi->s_resuid = EXT4_DEF_RESUID;
2984         sbi->s_resgid = EXT4_DEF_RESGID;
2985         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
2986         sbi->s_sb_block = sb_block;
2987         if (sb->s_bdev->bd_part)
2988                 sbi->s_sectors_written_start =
2989                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
2990
2991         /* Cleanup superblock name */
2992         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
2993                 *cp = '!';
2994
2995         ret = -EINVAL;
2996         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
2997         if (!blocksize) {
2998                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
2999                 goto out_fail;
3000         }
3001
3002         /*
3003          * The ext4 superblock will not be buffer aligned for other than 1kB
3004          * block sizes.  We need to calculate the offset from buffer start.
3005          */
3006         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3007                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3008                 offset = do_div(logical_sb_block, blocksize);
3009         } else {
3010                 logical_sb_block = sb_block;
3011         }
3012
3013         if (!(bh = sb_bread(sb, logical_sb_block))) {
3014                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3015                 goto out_fail;
3016         }
3017         /*
3018          * Note: s_es must be initialized as soon as possible because
3019          *       some ext4 macro-instructions depend on its value
3020          */
3021         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3022         sbi->s_es = es;
3023         sb->s_magic = le16_to_cpu(es->s_magic);
3024         if (sb->s_magic != EXT4_SUPER_MAGIC)
3025                 goto cantfind_ext4;
3026         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3027
3028         /* Set defaults before we parse the mount options */
3029         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3030         set_opt(sb, INIT_INODE_TABLE);
3031         if (def_mount_opts & EXT4_DEFM_DEBUG)
3032                 set_opt(sb, DEBUG);
3033         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3034                 set_opt(sb, GRPID);
3035         if (def_mount_opts & EXT4_DEFM_UID16)
3036                 set_opt(sb, NO_UID32);
3037         /* xattr user namespace & acls are now defaulted on */
3038 #ifdef CONFIG_EXT4_FS_XATTR
3039         set_opt(sb, XATTR_USER);
3040 #endif
3041 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3042         set_opt(sb, POSIX_ACL);
3043 #endif
3044         set_opt(sb, MBLK_IO_SUBMIT);
3045         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3046                 set_opt(sb, JOURNAL_DATA);
3047         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3048                 set_opt(sb, ORDERED_DATA);
3049         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3050                 set_opt(sb, WRITEBACK_DATA);
3051
3052         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3053                 set_opt(sb, ERRORS_PANIC);
3054         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3055                 set_opt(sb, ERRORS_CONT);
3056         else
3057                 set_opt(sb, ERRORS_RO);
3058         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3059                 set_opt(sb, BLOCK_VALIDITY);
3060         if (def_mount_opts & EXT4_DEFM_DISCARD)
3061                 set_opt(sb, DISCARD);
3062
3063         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3064         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3065         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3066         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3067         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3068
3069         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3070                 set_opt(sb, BARRIER);
3071
3072         /*
3073          * enable delayed allocation by default
3074          * Use -o nodelalloc to turn it off
3075          */
3076         if (!IS_EXT3_SB(sb) &&
3077             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3078                 set_opt(sb, DELALLOC);
3079
3080         /*
3081          * set default s_li_wait_mult for lazyinit, for the case there is
3082          * no mount option specified.
3083          */
3084         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3085
3086         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3087                            &journal_devnum, &journal_ioprio, 0)) {
3088                 ext4_msg(sb, KERN_WARNING,
3089                          "failed to parse options in superblock: %s",
3090                          sbi->s_es->s_mount_opts);
3091         }
3092         sbi->s_def_mount_opt = sbi->s_mount_opt;
3093         if (!parse_options((char *) data, sb, &journal_devnum,
3094                            &journal_ioprio, 0))
3095                 goto failed_mount;
3096
3097         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3098                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3099                             "with data=journal disables delayed "
3100                             "allocation and O_DIRECT support!\n");
3101                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3102                         ext4_msg(sb, KERN_ERR, "can't mount with "
3103                                  "both data=journal and delalloc");
3104                         goto failed_mount;
3105                 }
3106                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3107                         ext4_msg(sb, KERN_ERR, "can't mount with "
3108                                  "both data=journal and delalloc");
3109                         goto failed_mount;
3110                 }
3111                 if (test_opt(sb, DELALLOC))
3112                         clear_opt(sb, DELALLOC);
3113         }
3114
3115         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3116         if (test_opt(sb, DIOREAD_NOLOCK)) {
3117                 if (blocksize < PAGE_SIZE) {
3118                         ext4_msg(sb, KERN_ERR, "can't mount with "
3119                                  "dioread_nolock if block size != PAGE_SIZE");
3120                         goto failed_mount;
3121                 }
3122         }
3123
3124         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3125                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3126
3127         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3128             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3129              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3130              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3131                 ext4_msg(sb, KERN_WARNING,
3132                        "feature flags set on rev 0 fs, "
3133                        "running e2fsck is recommended");
3134
3135         if (IS_EXT2_SB(sb)) {
3136                 if (ext2_feature_set_ok(sb))
3137                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3138                                  "using the ext4 subsystem");
3139                 else {
3140                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3141                                  "to feature incompatibilities");
3142                         goto failed_mount;
3143                 }
3144         }
3145
3146         if (IS_EXT3_SB(sb)) {
3147                 if (ext3_feature_set_ok(sb))
3148                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3149                                  "using the ext4 subsystem");
3150                 else {
3151                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3152                                  "to feature incompatibilities");
3153                         goto failed_mount;
3154                 }
3155         }
3156
3157         /*
3158          * Check feature flags regardless of the revision level, since we
3159          * previously didn't change the revision level when setting the flags,
3160          * so there is a chance incompat flags are set on a rev 0 filesystem.
3161          */
3162         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3163                 goto failed_mount;
3164
3165         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3166             blocksize > EXT4_MAX_BLOCK_SIZE) {
3167                 ext4_msg(sb, KERN_ERR,
3168                        "Unsupported filesystem blocksize %d", blocksize);
3169                 goto failed_mount;
3170         }
3171
3172         if (sb->s_blocksize != blocksize) {
3173                 /* Validate the filesystem blocksize */
3174                 if (!sb_set_blocksize(sb, blocksize)) {
3175                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3176                                         blocksize);
3177                         goto failed_mount;
3178                 }
3179
3180                 brelse(bh);
3181                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3182                 offset = do_div(logical_sb_block, blocksize);
3183                 bh = sb_bread(sb, logical_sb_block);
3184                 if (!bh) {
3185                         ext4_msg(sb, KERN_ERR,
3186                                "Can't read superblock on 2nd try");
3187                         goto failed_mount;
3188                 }
3189                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3190                 sbi->s_es = es;
3191                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3192                         ext4_msg(sb, KERN_ERR,
3193                                "Magic mismatch, very weird!");
3194                         goto failed_mount;
3195                 }
3196         }
3197
3198         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3199                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3200         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3201                                                       has_huge_files);
3202         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3203
3204         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3205                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3206                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3207         } else {
3208                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3209                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3210                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3211                     (!is_power_of_2(sbi->s_inode_size)) ||
3212                     (sbi->s_inode_size > blocksize)) {
3213                         ext4_msg(sb, KERN_ERR,
3214                                "unsupported inode size: %d",
3215                                sbi->s_inode_size);
3216                         goto failed_mount;
3217                 }
3218                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3219                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3220         }
3221
3222         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3223         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3224                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3225                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3226                     !is_power_of_2(sbi->s_desc_size)) {
3227                         ext4_msg(sb, KERN_ERR,
3228                                "unsupported descriptor size %lu",
3229                                sbi->s_desc_size);
3230                         goto failed_mount;
3231                 }
3232         } else
3233                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3234
3235         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3236         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3237         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3238                 goto cantfind_ext4;
3239
3240         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3241         if (sbi->s_inodes_per_block == 0)
3242                 goto cantfind_ext4;
3243         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3244                                         sbi->s_inodes_per_block;
3245         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3246         sbi->s_sbh = bh;
3247         sbi->s_mount_state = le16_to_cpu(es->s_state);
3248         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3249         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3250
3251         for (i = 0; i < 4; i++)
3252                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3253         sbi->s_def_hash_version = es->s_def_hash_version;
3254         i = le32_to_cpu(es->s_flags);
3255         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3256                 sbi->s_hash_unsigned = 3;
3257         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3258 #ifdef __CHAR_UNSIGNED__
3259                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3260                 sbi->s_hash_unsigned = 3;
3261 #else
3262                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3263 #endif
3264         }
3265
3266         /* Handle clustersize */
3267         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3268         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3269                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3270         if (has_bigalloc) {
3271                 if (clustersize < blocksize) {
3272                         ext4_msg(sb, KERN_ERR,
3273                                  "cluster size (%d) smaller than "
3274                                  "block size (%d)", clustersize, blocksize);
3275                         goto failed_mount;
3276                 }
3277                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3278                         le32_to_cpu(es->s_log_block_size);
3279                 sbi->s_clusters_per_group =
3280                         le32_to_cpu(es->s_clusters_per_group);
3281                 if (sbi->s_clusters_per_group > blocksize * 8) {
3282                         ext4_msg(sb, KERN_ERR,
3283                                  "#clusters per group too big: %lu",
3284                                  sbi->s_clusters_per_group);
3285                         goto failed_mount;
3286                 }
3287                 if (sbi->s_blocks_per_group !=
3288                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3289                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3290                                  "clusters per group (%lu) inconsistent",
3291                                  sbi->s_blocks_per_group,
3292                                  sbi->s_clusters_per_group);
3293                         goto failed_mount;
3294                 }
3295         } else {
3296                 if (clustersize != blocksize) {
3297                         ext4_warning(sb, "fragment/cluster size (%d) != "
3298                                      "block size (%d)", clustersize,
3299                                      blocksize);
3300                         clustersize = blocksize;
3301                 }
3302                 if (sbi->s_blocks_per_group > blocksize * 8) {
3303                         ext4_msg(sb, KERN_ERR,
3304                                  "#blocks per group too big: %lu",
3305                                  sbi->s_blocks_per_group);
3306                         goto failed_mount;
3307                 }
3308                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3309                 sbi->s_cluster_bits = 0;
3310         }
3311         sbi->s_cluster_ratio = clustersize / blocksize;
3312
3313         if (sbi->s_inodes_per_group > blocksize * 8) {
3314                 ext4_msg(sb, KERN_ERR,
3315                        "#inodes per group too big: %lu",
3316                        sbi->s_inodes_per_group);
3317                 goto failed_mount;
3318         }
3319
3320         /*
3321          * Test whether we have more sectors than will fit in sector_t,
3322          * and whether the max offset is addressable by the page cache.
3323          */
3324         err = generic_check_addressable(sb->s_blocksize_bits,
3325                                         ext4_blocks_count(es));
3326         if (err) {
3327                 ext4_msg(sb, KERN_ERR, "filesystem"
3328                          " too large to mount safely on this system");
3329                 if (sizeof(sector_t) < 8)
3330                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3331                 ret = err;
3332                 goto failed_mount;
3333         }
3334
3335         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3336                 goto cantfind_ext4;
3337
3338         /* check blocks count against device size */
3339         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3340         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3341                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3342                        "exceeds size of device (%llu blocks)",
3343                        ext4_blocks_count(es), blocks_count);
3344                 goto failed_mount;
3345         }
3346
3347         /*
3348          * It makes no sense for the first data block to be beyond the end
3349          * of the filesystem.
3350          */
3351         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3352                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3353                          "block %u is beyond end of filesystem (%llu)",
3354                          le32_to_cpu(es->s_first_data_block),
3355                          ext4_blocks_count(es));
3356                 goto failed_mount;
3357         }
3358         blocks_count = (ext4_blocks_count(es) -
3359                         le32_to_cpu(es->s_first_data_block) +
3360                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3361         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3362         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3363                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3364                        "(block count %llu, first data block %u, "
3365                        "blocks per group %lu)", sbi->s_groups_count,
3366                        ext4_blocks_count(es),
3367                        le32_to_cpu(es->s_first_data_block),
3368                        EXT4_BLOCKS_PER_GROUP(sb));
3369                 goto failed_mount;
3370         }
3371         sbi->s_groups_count = blocks_count;
3372         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3373                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3374         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3375                    EXT4_DESC_PER_BLOCK(sb);
3376         sbi->s_group_desc = ext4_kvmalloc(db_count *
3377                                           sizeof(struct buffer_head *),
3378                                           GFP_KERNEL);
3379         if (sbi->s_group_desc == NULL) {
3380                 ext4_msg(sb, KERN_ERR, "not enough memory");
3381                 goto failed_mount;
3382         }
3383
3384         if (ext4_proc_root)
3385                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3386
3387         if (sbi->s_proc)
3388                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3389                                  &ext4_seq_options_fops, sb);
3390
3391         bgl_lock_init(sbi->s_blockgroup_lock);
3392
3393         for (i = 0; i < db_count; i++) {
3394                 block = descriptor_loc(sb, logical_sb_block, i);
3395                 sbi->s_group_desc[i] = sb_bread(sb, block);
3396                 if (!sbi->s_group_desc[i]) {
3397                         ext4_msg(sb, KERN_ERR,
3398                                "can't read group descriptor %d", i);
3399                         db_count = i;
3400                         goto failed_mount2;
3401                 }
3402         }
3403         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3404                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3405                 goto failed_mount2;
3406         }
3407         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3408                 if (!ext4_fill_flex_info(sb)) {
3409                         ext4_msg(sb, KERN_ERR,
3410                                "unable to initialize "
3411                                "flex_bg meta info!");
3412                         goto failed_mount2;
3413                 }
3414
3415         sbi->s_gdb_count = db_count;
3416         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3417         spin_lock_init(&sbi->s_next_gen_lock);
3418
3419         init_timer(&sbi->s_err_report);
3420         sbi->s_err_report.function = print_daily_error_info;
3421         sbi->s_err_report.data = (unsigned long) sb;
3422
3423         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3424                         ext4_count_free_clusters(sb));
3425         if (!err) {
3426                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3427                                 ext4_count_free_inodes(sb));
3428         }
3429         if (!err) {
3430                 err = percpu_counter_init(&sbi->s_dirs_counter,
3431                                 ext4_count_dirs(sb));
3432         }
3433         if (!err) {
3434                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3435         }
3436         if (err) {
3437                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3438                 goto failed_mount3;
3439         }
3440
3441         sbi->s_stripe = ext4_get_stripe_size(sbi);
3442         sbi->s_max_writeback_mb_bump = 128;
3443
3444         /*
3445          * set up enough so that it can read an inode
3446          */
3447         if (!test_opt(sb, NOLOAD) &&
3448             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3449                 sb->s_op = &ext4_sops;
3450         else
3451                 sb->s_op = &ext4_nojournal_sops;
3452         sb->s_export_op = &ext4_export_ops;
3453         sb->s_xattr = ext4_xattr_handlers;
3454 #ifdef CONFIG_QUOTA
3455         sb->s_qcop = &ext4_qctl_operations;
3456         sb->dq_op = &ext4_quota_operations;
3457 #endif
3458         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3459
3460         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3461         mutex_init(&sbi->s_orphan_lock);
3462         sbi->s_resize_flags = 0;
3463
3464         sb->s_root = NULL;
3465
3466         needs_recovery = (es->s_last_orphan != 0 ||
3467                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3468                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3469
3470         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3471             !(sb->s_flags & MS_RDONLY))
3472                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3473                         goto failed_mount3;
3474
3475         /*
3476          * The first inode we look at is the journal inode.  Don't try
3477          * root first: it may be modified in the journal!
3478          */
3479         if (!test_opt(sb, NOLOAD) &&
3480             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3481                 if (ext4_load_journal(sb, es, journal_devnum))
3482                         goto failed_mount3;
3483         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3484               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3485                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3486                        "suppressed and not mounted read-only");
3487                 goto failed_mount_wq;
3488         } else {
3489                 clear_opt(sb, DATA_FLAGS);
3490                 sbi->s_journal = NULL;
3491                 needs_recovery = 0;
3492                 goto no_journal;
3493         }
3494
3495         if (ext4_blocks_count(es) > 0xffffffffULL &&
3496             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3497                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3498                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3499                 goto failed_mount_wq;
3500         }
3501
3502         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3503                 jbd2_journal_set_features(sbi->s_journal,
3504                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3505                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3506         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3507                 jbd2_journal_set_features(sbi->s_journal,
3508                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3509                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3510                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3511         } else {
3512                 jbd2_journal_clear_features(sbi->s_journal,
3513                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3514                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3515         }
3516
3517         /* We have now updated the journal if required, so we can
3518          * validate the data journaling mode. */
3519         switch (test_opt(sb, DATA_FLAGS)) {
3520         case 0:
3521                 /* No mode set, assume a default based on the journal
3522                  * capabilities: ORDERED_DATA if the journal can
3523                  * cope, else JOURNAL_DATA
3524                  */
3525                 if (jbd2_journal_check_available_features
3526                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3527                         set_opt(sb, ORDERED_DATA);
3528                 else
3529                         set_opt(sb, JOURNAL_DATA);
3530                 break;
3531
3532         case EXT4_MOUNT_ORDERED_DATA:
3533         case EXT4_MOUNT_WRITEBACK_DATA:
3534                 if (!jbd2_journal_check_available_features
3535                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3536                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3537                                "requested data journaling mode");
3538                         goto failed_mount_wq;
3539                 }
3540         default:
3541                 break;
3542         }
3543         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3544
3545         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3546
3547         /*
3548          * The journal may have updated the bg summary counts, so we
3549          * need to update the global counters.
3550          */
3551         percpu_counter_set(&sbi->s_freeclusters_counter,
3552                            ext4_count_free_clusters(sb));
3553         percpu_counter_set(&sbi->s_freeinodes_counter,
3554                            ext4_count_free_inodes(sb));
3555         percpu_counter_set(&sbi->s_dirs_counter,
3556                            ext4_count_dirs(sb));
3557         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3558
3559 no_journal:
3560         /*
3561          * The maximum number of concurrent works can be high and
3562          * concurrency isn't really necessary.  Limit it to 1.
3563          */
3564         EXT4_SB(sb)->dio_unwritten_wq =
3565                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3566         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3567                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3568                 goto failed_mount_wq;
3569         }
3570
3571         /*
3572          * The jbd2_journal_load will have done any necessary log recovery,
3573          * so we can safely mount the rest of the filesystem now.
3574          */
3575
3576         root = ext4_iget(sb, EXT4_ROOT_INO);
3577         if (IS_ERR(root)) {
3578                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3579                 ret = PTR_ERR(root);
3580                 root = NULL;
3581                 goto failed_mount4;
3582         }
3583         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3584                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3585                 iput(root);
3586                 goto failed_mount4;
3587         }
3588         sb->s_root = d_make_root(root);
3589         if (!sb->s_root) {
3590                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3591                 ret = -ENOMEM;
3592                 goto failed_mount4;
3593         }
3594
3595         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3596
3597         /* determine the minimum size of new large inodes, if present */
3598         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3599                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3600                                                      EXT4_GOOD_OLD_INODE_SIZE;
3601                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3602                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3603                         if (sbi->s_want_extra_isize <
3604                             le16_to_cpu(es->s_want_extra_isize))
3605                                 sbi->s_want_extra_isize =
3606                                         le16_to_cpu(es->s_want_extra_isize);
3607                         if (sbi->s_want_extra_isize <
3608                             le16_to_cpu(es->s_min_extra_isize))
3609                                 sbi->s_want_extra_isize =
3610                                         le16_to_cpu(es->s_min_extra_isize);
3611                 }
3612         }
3613         /* Check if enough inode space is available */
3614         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3615                                                         sbi->s_inode_size) {
3616                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3617                                                        EXT4_GOOD_OLD_INODE_SIZE;
3618                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3619                          "available");
3620         }
3621
3622         err = ext4_setup_system_zone(sb);
3623         if (err) {
3624                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3625                          "zone (%d)", err);
3626                 goto failed_mount4a;
3627         }
3628
3629         ext4_ext_init(sb);
3630         err = ext4_mb_init(sb, needs_recovery);
3631         if (err) {
3632                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3633                          err);
3634                 goto failed_mount5;
3635         }
3636
3637         err = ext4_register_li_request(sb, first_not_zeroed);
3638         if (err)
3639                 goto failed_mount6;
3640
3641         sbi->s_kobj.kset = ext4_kset;
3642         init_completion(&sbi->s_kobj_unregister);
3643         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3644                                    "%s", sb->s_id);
3645         if (err)
3646                 goto failed_mount7;
3647
3648         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3649         ext4_orphan_cleanup(sb, es);
3650         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3651         if (needs_recovery) {
3652                 ext4_msg(sb, KERN_INFO, "recovery complete");
3653                 ext4_mark_recovery_complete(sb, es);
3654         }
3655         if (EXT4_SB(sb)->s_journal) {
3656                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3657                         descr = " journalled data mode";
3658                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3659                         descr = " ordered data mode";
3660                 else
3661                         descr = " writeback data mode";
3662         } else
3663                 descr = "out journal";
3664
3665         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3666                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3667                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3668
3669         if (es->s_error_count)
3670                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3671
3672         kfree(orig_data);
3673         return 0;
3674
3675 cantfind_ext4:
3676         if (!silent)
3677                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3678         goto failed_mount;
3679
3680 failed_mount7:
3681         ext4_unregister_li_request(sb);
3682 failed_mount6:
3683         ext4_mb_release(sb);
3684 failed_mount5:
3685         ext4_ext_release(sb);
3686         ext4_release_system_zone(sb);
3687 failed_mount4a:
3688         dput(sb->s_root);
3689         sb->s_root = NULL;
3690 failed_mount4:
3691         ext4_msg(sb, KERN_ERR, "mount failed");
3692         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3693 failed_mount_wq:
3694         if (sbi->s_journal) {
3695                 jbd2_journal_destroy(sbi->s_journal);
3696                 sbi->s_journal = NULL;
3697         }
3698 failed_mount3:
3699         del_timer(&sbi->s_err_report);
3700         if (sbi->s_flex_groups)
3701                 ext4_kvfree(sbi->s_flex_groups);
3702         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3703         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3704         percpu_counter_destroy(&sbi->s_dirs_counter);
3705         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3706         if (sbi->s_mmp_tsk)
3707                 kthread_stop(sbi->s_mmp_tsk);
3708 failed_mount2:
3709         for (i = 0; i < db_count; i++)
3710                 brelse(sbi->s_group_desc[i]);
3711         ext4_kvfree(sbi->s_group_desc);
3712 failed_mount:
3713         if (sbi->s_proc) {
3714                 remove_proc_entry("options", sbi->s_proc);
3715                 remove_proc_entry(sb->s_id, ext4_proc_root);
3716         }
3717 #ifdef CONFIG_QUOTA
3718         for (i = 0; i < MAXQUOTAS; i++)
3719                 kfree(sbi->s_qf_names[i]);
3720 #endif
3721         ext4_blkdev_remove(sbi);
3722         brelse(bh);
3723 out_fail:
3724         sb->s_fs_info = NULL;
3725         kfree(sbi->s_blockgroup_lock);
3726         kfree(sbi);
3727 out_free_orig:
3728         kfree(orig_data);
3729         return ret;
3730 }
3731
3732 /*
3733  * Setup any per-fs journal parameters now.  We'll do this both on
3734  * initial mount, once the journal has been initialised but before we've
3735  * done any recovery; and again on any subsequent remount.
3736  */
3737 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3738 {
3739         struct ext4_sb_info *sbi = EXT4_SB(sb);
3740
3741         journal->j_commit_interval = sbi->s_commit_interval;
3742         journal->j_min_batch_time = sbi->s_min_batch_time;
3743         journal->j_max_batch_time = sbi->s_max_batch_time;
3744
3745         write_lock(&journal->j_state_lock);
3746         if (test_opt(sb, BARRIER))
3747                 journal->j_flags |= JBD2_BARRIER;
3748         else
3749                 journal->j_flags &= ~JBD2_BARRIER;
3750         if (test_opt(sb, DATA_ERR_ABORT))
3751                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3752         else
3753                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3754         write_unlock(&journal->j_state_lock);
3755 }
3756
3757 static journal_t *ext4_get_journal(struct super_block *sb,
3758                                    unsigned int journal_inum)
3759 {
3760         struct inode *journal_inode;
3761         journal_t *journal;
3762
3763         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3764
3765         /* First, test for the existence of a valid inode on disk.  Bad
3766          * things happen if we iget() an unused inode, as the subsequent
3767          * iput() will try to delete it. */
3768
3769         journal_inode = ext4_iget(sb, journal_inum);
3770         if (IS_ERR(journal_inode)) {
3771                 ext4_msg(sb, KERN_ERR, "no journal found");
3772                 return NULL;
3773         }
3774         if (!journal_inode->i_nlink) {
3775                 make_bad_inode(journal_inode);
3776                 iput(journal_inode);
3777                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3778                 return NULL;
3779         }
3780
3781         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3782                   journal_inode, journal_inode->i_size);
3783         if (!S_ISREG(journal_inode->i_mode)) {
3784                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3785                 iput(journal_inode);
3786                 return NULL;
3787         }
3788
3789         journal = jbd2_journal_init_inode(journal_inode);
3790         if (!journal) {
3791                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3792                 iput(journal_inode);
3793                 return NULL;
3794         }
3795         journal->j_private = sb;
3796         ext4_init_journal_params(sb, journal);
3797         return journal;
3798 }
3799
3800 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3801                                        dev_t j_dev)
3802 {
3803         struct buffer_head *bh;
3804         journal_t *journal;
3805         ext4_fsblk_t start;
3806         ext4_fsblk_t len;
3807         int hblock, blocksize;
3808         ext4_fsblk_t sb_block;
3809         unsigned long offset;
3810         struct ext4_super_block *es;
3811         struct block_device *bdev;
3812
3813         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3814
3815         bdev = ext4_blkdev_get(j_dev, sb);
3816         if (bdev == NULL)
3817                 return NULL;
3818
3819         blocksize = sb->s_blocksize;
3820         hblock = bdev_logical_block_size(bdev);
3821         if (blocksize < hblock) {
3822                 ext4_msg(sb, KERN_ERR,
3823                         "blocksize too small for journal device");
3824                 goto out_bdev;
3825         }
3826
3827         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3828         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3829         set_blocksize(bdev, blocksize);
3830         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3831                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3832                        "external journal");
3833                 goto out_bdev;
3834         }
3835
3836         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3837         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3838             !(le32_to_cpu(es->s_feature_incompat) &
3839               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3840                 ext4_msg(sb, KERN_ERR, "external journal has "
3841                                         "bad superblock");
3842                 brelse(bh);
3843                 goto out_bdev;
3844         }
3845
3846         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3847                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3848                 brelse(bh);
3849                 goto out_bdev;
3850         }
3851
3852         len = ext4_blocks_count(es);
3853         start = sb_block + 1;
3854         brelse(bh);     /* we're done with the superblock */
3855
3856         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3857                                         start, len, blocksize);
3858         if (!journal) {
3859                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3860                 goto out_bdev;
3861         }
3862         journal->j_private = sb;
3863         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3864         wait_on_buffer(journal->j_sb_buffer);
3865         if (!buffer_uptodate(journal->j_sb_buffer)) {
3866                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3867                 goto out_journal;
3868         }
3869         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3870                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3871                                         "user (unsupported) - %d",
3872                         be32_to_cpu(journal->j_superblock->s_nr_users));
3873                 goto out_journal;
3874         }
3875         EXT4_SB(sb)->journal_bdev = bdev;
3876         ext4_init_journal_params(sb, journal);
3877         return journal;
3878
3879 out_journal:
3880         jbd2_journal_destroy(journal);
3881 out_bdev:
3882         ext4_blkdev_put(bdev);
3883         return NULL;
3884 }
3885
3886 static int ext4_load_journal(struct super_block *sb,
3887                              struct ext4_super_block *es,
3888                              unsigned long journal_devnum)
3889 {
3890         journal_t *journal;
3891         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
3892         dev_t journal_dev;
3893         int err = 0;
3894         int really_read_only;
3895
3896         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3897
3898         if (journal_devnum &&
3899             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3900                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
3901                         "numbers have changed");
3902                 journal_dev = new_decode_dev(journal_devnum);
3903         } else
3904                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
3905
3906         really_read_only = bdev_read_only(sb->s_bdev);
3907
3908         /*
3909          * Are we loading a blank journal or performing recovery after a
3910          * crash?  For recovery, we need to check in advance whether we
3911          * can get read-write access to the device.
3912          */
3913         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3914                 if (sb->s_flags & MS_RDONLY) {
3915                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
3916                                         "required on readonly filesystem");
3917                         if (really_read_only) {
3918                                 ext4_msg(sb, KERN_ERR, "write access "
3919                                         "unavailable, cannot proceed");
3920                                 return -EROFS;
3921                         }
3922                         ext4_msg(sb, KERN_INFO, "write access will "
3923                                "be enabled during recovery");
3924                 }
3925         }
3926
3927         if (journal_inum && journal_dev) {
3928                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
3929                        "and inode journals!");
3930                 return -EINVAL;
3931         }
3932
3933         if (journal_inum) {
3934                 if (!(journal = ext4_get_journal(sb, journal_inum)))
3935                         return -EINVAL;
3936         } else {
3937                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
3938                         return -EINVAL;
3939         }
3940
3941         if (!(journal->j_flags & JBD2_BARRIER))
3942                 ext4_msg(sb, KERN_INFO, "barriers disabled");
3943
3944         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
3945                 err = jbd2_journal_wipe(journal, !really_read_only);
3946         if (!err) {
3947                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
3948                 if (save)
3949                         memcpy(save, ((char *) es) +
3950                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
3951                 err = jbd2_journal_load(journal);
3952                 if (save)
3953                         memcpy(((char *) es) + EXT4_S_ERR_START,
3954                                save, EXT4_S_ERR_LEN);
3955                 kfree(save);
3956         }
3957
3958         if (err) {
3959                 ext4_msg(sb, KERN_ERR, "error loading journal");
3960                 jbd2_journal_destroy(journal);
3961                 return err;
3962         }
3963
3964         EXT4_SB(sb)->s_journal = journal;
3965         ext4_clear_journal_err(sb, es);
3966
3967         if (!really_read_only && journal_devnum &&
3968             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
3969                 es->s_journal_dev = cpu_to_le32(journal_devnum);
3970
3971                 /* Make sure we flush the recovery flag to disk. */
3972                 ext4_commit_super(sb, 1);
3973         }
3974
3975         return 0;
3976 }
3977
3978 static int ext4_commit_super(struct super_block *sb, int sync)
3979 {
3980         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
3981         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
3982         int error = 0;
3983
3984         if (!sbh || block_device_ejected(sb))
3985                 return error;
3986         if (buffer_write_io_error(sbh)) {
3987                 /*
3988                  * Oh, dear.  A previous attempt to write the
3989                  * superblock failed.  This could happen because the
3990                  * USB device was yanked out.  Or it could happen to
3991                  * be a transient write error and maybe the block will
3992                  * be remapped.  Nothing we can do but to retry the
3993                  * write and hope for the best.
3994                  */
3995                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
3996                        "superblock detected");
3997                 clear_buffer_write_io_error(sbh);
3998                 set_buffer_uptodate(sbh);
3999         }
4000         /*
4001          * If the file system is mounted read-only, don't update the
4002          * superblock write time.  This avoids updating the superblock
4003          * write time when we are mounting the root file system
4004          * read/only but we need to replay the journal; at that point,
4005          * for people who are east of GMT and who make their clock
4006          * tick in localtime for Windows bug-for-bug compatibility,
4007          * the clock is set in the future, and this will cause e2fsck
4008          * to complain and force a full file system check.
4009          */
4010         if (!(sb->s_flags & MS_RDONLY))
4011                 es->s_wtime = cpu_to_le32(get_seconds());
4012         if (sb->s_bdev->bd_part)
4013                 es->s_kbytes_written =
4014                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4015                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4016                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4017         else
4018                 es->s_kbytes_written =
4019                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4020         ext4_free_blocks_count_set(es,
4021                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4022                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4023         es->s_free_inodes_count =
4024                 cpu_to_le32(percpu_counter_sum_positive(
4025                                 &EXT4_SB(sb)->s_freeinodes_counter));
4026         sb->s_dirt = 0;
4027         BUFFER_TRACE(sbh, "marking dirty");
4028         mark_buffer_dirty(sbh);
4029         if (sync) {
4030                 error = sync_dirty_buffer(sbh);
4031                 if (error)
4032                         return error;
4033
4034                 error = buffer_write_io_error(sbh);
4035                 if (error) {
4036                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4037                                "superblock");
4038                         clear_buffer_write_io_error(sbh);
4039                         set_buffer_uptodate(sbh);
4040                 }
4041         }
4042         return error;
4043 }
4044
4045 /*
4046  * Have we just finished recovery?  If so, and if we are mounting (or
4047  * remounting) the filesystem readonly, then we will end up with a
4048  * consistent fs on disk.  Record that fact.
4049  */
4050 static void ext4_mark_recovery_complete(struct super_block *sb,
4051                                         struct ext4_super_block *es)
4052 {
4053         journal_t *journal = EXT4_SB(sb)->s_journal;
4054
4055         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4056                 BUG_ON(journal != NULL);
4057                 return;
4058         }
4059         jbd2_journal_lock_updates(journal);
4060         if (jbd2_journal_flush(journal) < 0)
4061                 goto out;
4062
4063         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4064             sb->s_flags & MS_RDONLY) {
4065                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4066                 ext4_commit_super(sb, 1);
4067         }
4068
4069 out:
4070         jbd2_journal_unlock_updates(journal);
4071 }
4072
4073 /*
4074  * If we are mounting (or read-write remounting) a filesystem whose journal
4075  * has recorded an error from a previous lifetime, move that error to the
4076  * main filesystem now.
4077  */
4078 static void ext4_clear_journal_err(struct super_block *sb,
4079                                    struct ext4_super_block *es)
4080 {
4081         journal_t *journal;
4082         int j_errno;
4083         const char *errstr;
4084
4085         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4086
4087         journal = EXT4_SB(sb)->s_journal;
4088
4089         /*
4090          * Now check for any error status which may have been recorded in the
4091          * journal by a prior ext4_error() or ext4_abort()
4092          */
4093
4094         j_errno = jbd2_journal_errno(journal);
4095         if (j_errno) {
4096                 char nbuf[16];
4097
4098                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4099                 ext4_warning(sb, "Filesystem error recorded "
4100                              "from previous mount: %s", errstr);
4101                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4102
4103                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4104                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4105                 ext4_commit_super(sb, 1);
4106
4107                 jbd2_journal_clear_err(journal);
4108         }
4109 }
4110
4111 /*
4112  * Force the running and committing transactions to commit,
4113  * and wait on the commit.
4114  */
4115 int ext4_force_commit(struct super_block *sb)
4116 {
4117         journal_t *journal;
4118         int ret = 0;
4119
4120         if (sb->s_flags & MS_RDONLY)
4121                 return 0;
4122
4123         journal = EXT4_SB(sb)->s_journal;
4124         if (journal) {
4125                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4126                 ret = ext4_journal_force_commit(journal);
4127         }
4128
4129         return ret;
4130 }
4131
4132 static void ext4_write_super(struct super_block *sb)
4133 {
4134         lock_super(sb);
4135         ext4_commit_super(sb, 1);
4136         unlock_super(sb);
4137 }
4138
4139 static int ext4_sync_fs(struct super_block *sb, int wait)
4140 {
4141         int ret = 0;
4142         tid_t target;
4143         struct ext4_sb_info *sbi = EXT4_SB(sb);
4144
4145         trace_ext4_sync_fs(sb, wait);
4146         flush_workqueue(sbi->dio_unwritten_wq);
4147         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4148                 if (wait)
4149                         jbd2_log_wait_commit(sbi->s_journal, target);
4150         }
4151         return ret;
4152 }
4153
4154 /*
4155  * LVM calls this function before a (read-only) snapshot is created.  This
4156  * gives us a chance to flush the journal completely and mark the fs clean.
4157  *
4158  * Note that only this function cannot bring a filesystem to be in a clean
4159  * state independently, because ext4 prevents a new handle from being started
4160  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4161  * the upper layer.
4162  */
4163 static int ext4_freeze(struct super_block *sb)
4164 {
4165         int error = 0;
4166         journal_t *journal;
4167
4168         if (sb->s_flags & MS_RDONLY)
4169                 return 0;
4170
4171         journal = EXT4_SB(sb)->s_journal;
4172
4173         /* Now we set up the journal barrier. */
4174         jbd2_journal_lock_updates(journal);
4175
4176         /*
4177          * Don't clear the needs_recovery flag if we failed to flush
4178          * the journal.
4179          */
4180         error = jbd2_journal_flush(journal);
4181         if (error < 0)
4182                 goto out;
4183
4184         /* Journal blocked and flushed, clear needs_recovery flag. */
4185         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4186         error = ext4_commit_super(sb, 1);
4187 out:
4188         /* we rely on s_frozen to stop further updates */
4189         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4190         return error;
4191 }
4192
4193 /*
4194  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4195  * flag here, even though the filesystem is not technically dirty yet.
4196  */
4197 static int ext4_unfreeze(struct super_block *sb)
4198 {
4199         if (sb->s_flags & MS_RDONLY)
4200                 return 0;
4201
4202         lock_super(sb);
4203         /* Reset the needs_recovery flag before the fs is unlocked. */
4204         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4205         ext4_commit_super(sb, 1);
4206         unlock_super(sb);
4207         return 0;
4208 }
4209
4210 /*
4211  * Structure to save mount options for ext4_remount's benefit
4212  */
4213 struct ext4_mount_options {
4214         unsigned long s_mount_opt;
4215         unsigned long s_mount_opt2;
4216         uid_t s_resuid;
4217         gid_t s_resgid;
4218         unsigned long s_commit_interval;
4219         u32 s_min_batch_time, s_max_batch_time;
4220 #ifdef CONFIG_QUOTA
4221         int s_jquota_fmt;
4222         char *s_qf_names[MAXQUOTAS];
4223 #endif
4224 };
4225
4226 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4227 {
4228         struct ext4_super_block *es;
4229         struct ext4_sb_info *sbi = EXT4_SB(sb);
4230         unsigned long old_sb_flags;
4231         struct ext4_mount_options old_opts;
4232         int enable_quota = 0;
4233         ext4_group_t g;
4234         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4235         int err = 0;
4236 #ifdef CONFIG_QUOTA
4237         int i;
4238 #endif
4239         char *orig_data = kstrdup(data, GFP_KERNEL);
4240
4241         /* Store the original options */
4242         lock_super(sb);
4243         old_sb_flags = sb->s_flags;
4244         old_opts.s_mount_opt = sbi->s_mount_opt;
4245         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4246         old_opts.s_resuid = sbi->s_resuid;
4247         old_opts.s_resgid = sbi->s_resgid;
4248         old_opts.s_commit_interval = sbi->s_commit_interval;
4249         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4250         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4251 #ifdef CONFIG_QUOTA
4252         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4253         for (i = 0; i < MAXQUOTAS; i++)
4254                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4255 #endif
4256         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4257                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4258
4259         /*
4260          * Allow the "check" option to be passed as a remount option.
4261          */
4262         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4263                 err = -EINVAL;
4264                 goto restore_opts;
4265         }
4266
4267         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4268                 ext4_abort(sb, "Abort forced by user");
4269
4270         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4271                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4272
4273         es = sbi->s_es;
4274
4275         if (sbi->s_journal) {
4276                 ext4_init_journal_params(sb, sbi->s_journal);
4277                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4278         }
4279
4280         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4281                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4282                         err = -EROFS;
4283                         goto restore_opts;
4284                 }
4285
4286                 if (*flags & MS_RDONLY) {
4287                         err = dquot_suspend(sb, -1);
4288                         if (err < 0)
4289                                 goto restore_opts;
4290
4291                         /*
4292                          * First of all, the unconditional stuff we have to do
4293                          * to disable replay of the journal when we next remount
4294                          */
4295                         sb->s_flags |= MS_RDONLY;
4296
4297                         /*
4298                          * OK, test if we are remounting a valid rw partition
4299                          * readonly, and if so set the rdonly flag and then
4300                          * mark the partition as valid again.
4301                          */
4302                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4303                             (sbi->s_mount_state & EXT4_VALID_FS))
4304                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4305
4306                         if (sbi->s_journal)
4307                                 ext4_mark_recovery_complete(sb, es);
4308                 } else {
4309                         /* Make sure we can mount this feature set readwrite */
4310                         if (!ext4_feature_set_ok(sb, 0)) {
4311                                 err = -EROFS;
4312                                 goto restore_opts;
4313                         }
4314                         /*
4315                          * Make sure the group descriptor checksums
4316                          * are sane.  If they aren't, refuse to remount r/w.
4317                          */
4318                         for (g = 0; g < sbi->s_groups_count; g++) {
4319                                 struct ext4_group_desc *gdp =
4320                                         ext4_get_group_desc(sb, g, NULL);
4321
4322                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4323                                         ext4_msg(sb, KERN_ERR,
4324                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4325                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4326                                                le16_to_cpu(gdp->bg_checksum));
4327                                         err = -EINVAL;
4328                                         goto restore_opts;
4329                                 }
4330                         }
4331
4332                         /*
4333                          * If we have an unprocessed orphan list hanging
4334                          * around from a previously readonly bdev mount,
4335                          * require a full umount/remount for now.
4336                          */
4337                         if (es->s_last_orphan) {
4338                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4339                                        "remount RDWR because of unprocessed "
4340                                        "orphan inode list.  Please "
4341                                        "umount/remount instead");
4342                                 err = -EINVAL;
4343                                 goto restore_opts;
4344                         }
4345
4346                         /*
4347                          * Mounting a RDONLY partition read-write, so reread
4348                          * and store the current valid flag.  (It may have
4349                          * been changed by e2fsck since we originally mounted
4350                          * the partition.)
4351                          */
4352                         if (sbi->s_journal)
4353                                 ext4_clear_journal_err(sb, es);
4354                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4355                         if (!ext4_setup_super(sb, es, 0))
4356                                 sb->s_flags &= ~MS_RDONLY;
4357                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4358                                                      EXT4_FEATURE_INCOMPAT_MMP))
4359                                 if (ext4_multi_mount_protect(sb,
4360                                                 le64_to_cpu(es->s_mmp_block))) {
4361                                         err = -EROFS;
4362                                         goto restore_opts;
4363                                 }
4364                         enable_quota = 1;
4365                 }
4366         }
4367
4368         /*
4369          * Reinitialize lazy itable initialization thread based on
4370          * current settings
4371          */
4372         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4373                 ext4_unregister_li_request(sb);
4374         else {
4375                 ext4_group_t first_not_zeroed;
4376                 first_not_zeroed = ext4_has_uninit_itable(sb);
4377                 ext4_register_li_request(sb, first_not_zeroed);
4378         }
4379
4380         ext4_setup_system_zone(sb);
4381         if (sbi->s_journal == NULL)
4382                 ext4_commit_super(sb, 1);
4383
4384 #ifdef CONFIG_QUOTA
4385         /* Release old quota file names */
4386         for (i = 0; i < MAXQUOTAS; i++)
4387                 if (old_opts.s_qf_names[i] &&
4388                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4389                         kfree(old_opts.s_qf_names[i]);
4390 #endif
4391         unlock_super(sb);
4392         if (enable_quota)
4393                 dquot_resume(sb, -1);
4394
4395         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4396         kfree(orig_data);
4397         return 0;
4398
4399 restore_opts:
4400         sb->s_flags = old_sb_flags;
4401         sbi->s_mount_opt = old_opts.s_mount_opt;
4402         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4403         sbi->s_resuid = old_opts.s_resuid;
4404         sbi->s_resgid = old_opts.s_resgid;
4405         sbi->s_commit_interval = old_opts.s_commit_interval;
4406         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4407         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4408 #ifdef CONFIG_QUOTA
4409         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4410         for (i = 0; i < MAXQUOTAS; i++) {
4411                 if (sbi->s_qf_names[i] &&
4412                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4413                         kfree(sbi->s_qf_names[i]);
4414                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4415         }
4416 #endif
4417         unlock_super(sb);
4418         kfree(orig_data);
4419         return err;
4420 }
4421
4422 /*
4423  * Note: calculating the overhead so we can be compatible with
4424  * historical BSD practice is quite difficult in the face of
4425  * clusters/bigalloc.  This is because multiple metadata blocks from
4426  * different block group can end up in the same allocation cluster.
4427  * Calculating the exact overhead in the face of clustered allocation
4428  * requires either O(all block bitmaps) in memory or O(number of block
4429  * groups**2) in time.  We will still calculate the superblock for
4430  * older file systems --- and if we come across with a bigalloc file
4431  * system with zero in s_overhead_clusters the estimate will be close to
4432  * correct especially for very large cluster sizes --- but for newer
4433  * file systems, it's better to calculate this figure once at mkfs
4434  * time, and store it in the superblock.  If the superblock value is
4435  * present (even for non-bigalloc file systems), we will use it.
4436  */
4437 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4438 {
4439         struct super_block *sb = dentry->d_sb;
4440         struct ext4_sb_info *sbi = EXT4_SB(sb);
4441         struct ext4_super_block *es = sbi->s_es;
4442         struct ext4_group_desc *gdp;
4443         u64 fsid;
4444         s64 bfree;
4445
4446         if (test_opt(sb, MINIX_DF)) {
4447                 sbi->s_overhead_last = 0;
4448         } else if (es->s_overhead_clusters) {
4449                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4450         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4451                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4452                 ext4_fsblk_t overhead = 0;
4453
4454                 /*
4455                  * Compute the overhead (FS structures).  This is constant
4456                  * for a given filesystem unless the number of block groups
4457                  * changes so we cache the previous value until it does.
4458                  */
4459
4460                 /*
4461                  * All of the blocks before first_data_block are
4462                  * overhead
4463                  */
4464                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4465
4466                 /*
4467                  * Add the overhead found in each block group
4468                  */
4469                 for (i = 0; i < ngroups; i++) {
4470                         gdp = ext4_get_group_desc(sb, i, NULL);
4471                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4472                         cond_resched();
4473                 }
4474                 sbi->s_overhead_last = overhead;
4475                 smp_wmb();
4476                 sbi->s_blocks_last = ext4_blocks_count(es);
4477         }
4478
4479         buf->f_type = EXT4_SUPER_MAGIC;
4480         buf->f_bsize = sb->s_blocksize;
4481         buf->f_blocks = (ext4_blocks_count(es) -
4482                          EXT4_C2B(sbi, sbi->s_overhead_last));
4483         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4484                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4485         /* prevent underflow in case that few free space is available */
4486         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4487         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4488         if (buf->f_bfree < ext4_r_blocks_count(es))
4489                 buf->f_bavail = 0;
4490         buf->f_files = le32_to_cpu(es->s_inodes_count);
4491         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4492         buf->f_namelen = EXT4_NAME_LEN;
4493         fsid = le64_to_cpup((void *)es->s_uuid) ^
4494                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4495         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4496         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4497
4498         return 0;
4499 }
4500
4501 /* Helper function for writing quotas on sync - we need to start transaction
4502  * before quota file is locked for write. Otherwise the are possible deadlocks:
4503  * Process 1                         Process 2
4504  * ext4_create()                     quota_sync()
4505  *   jbd2_journal_start()                  write_dquot()
4506  *   dquot_initialize()                         down(dqio_mutex)
4507  *     down(dqio_mutex)                    jbd2_journal_start()
4508  *
4509  */
4510
4511 #ifdef CONFIG_QUOTA
4512
4513 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4514 {
4515         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4516 }
4517
4518 static int ext4_write_dquot(struct dquot *dquot)
4519 {
4520         int ret, err;
4521         handle_t *handle;
4522         struct inode *inode;
4523
4524         inode = dquot_to_inode(dquot);
4525         handle = ext4_journal_start(inode,
4526                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4527         if (IS_ERR(handle))
4528                 return PTR_ERR(handle);
4529         ret = dquot_commit(dquot);
4530         err = ext4_journal_stop(handle);
4531         if (!ret)
4532                 ret = err;
4533         return ret;
4534 }
4535
4536 static int ext4_acquire_dquot(struct dquot *dquot)
4537 {
4538         int ret, err;
4539         handle_t *handle;
4540
4541         handle = ext4_journal_start(dquot_to_inode(dquot),
4542                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4543         if (IS_ERR(handle))
4544                 return PTR_ERR(handle);
4545         ret = dquot_acquire(dquot);
4546         err = ext4_journal_stop(handle);
4547         if (!ret)
4548                 ret = err;
4549         return ret;
4550 }
4551
4552 static int ext4_release_dquot(struct dquot *dquot)
4553 {
4554         int ret, err;
4555         handle_t *handle;
4556
4557         handle = ext4_journal_start(dquot_to_inode(dquot),
4558                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4559         if (IS_ERR(handle)) {
4560                 /* Release dquot anyway to avoid endless cycle in dqput() */
4561                 dquot_release(dquot);
4562                 return PTR_ERR(handle);
4563         }
4564         ret = dquot_release(dquot);
4565         err = ext4_journal_stop(handle);
4566         if (!ret)
4567                 ret = err;
4568         return ret;
4569 }
4570
4571 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4572 {
4573         /* Are we journaling quotas? */
4574         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4575             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4576                 dquot_mark_dquot_dirty(dquot);
4577                 return ext4_write_dquot(dquot);
4578         } else {
4579                 return dquot_mark_dquot_dirty(dquot);
4580         }
4581 }
4582
4583 static int ext4_write_info(struct super_block *sb, int type)
4584 {
4585         int ret, err;
4586         handle_t *handle;
4587
4588         /* Data block + inode block */
4589         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4590         if (IS_ERR(handle))
4591                 return PTR_ERR(handle);
4592         ret = dquot_commit_info(sb, type);
4593         err = ext4_journal_stop(handle);
4594         if (!ret)
4595                 ret = err;
4596         return ret;
4597 }
4598
4599 /*
4600  * Turn on quotas during mount time - we need to find
4601  * the quota file and such...
4602  */
4603 static int ext4_quota_on_mount(struct super_block *sb, int type)
4604 {
4605         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4606                                         EXT4_SB(sb)->s_jquota_fmt, type);
4607 }
4608
4609 /*
4610  * Standard function to be called on quota_on
4611  */
4612 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4613                          struct path *path)
4614 {
4615         int err;
4616
4617         if (!test_opt(sb, QUOTA))
4618                 return -EINVAL;
4619
4620         /* Quotafile not on the same filesystem? */
4621         if (path->dentry->d_sb != sb)
4622                 return -EXDEV;
4623         /* Journaling quota? */
4624         if (EXT4_SB(sb)->s_qf_names[type]) {
4625                 /* Quotafile not in fs root? */
4626                 if (path->dentry->d_parent != sb->s_root)
4627                         ext4_msg(sb, KERN_WARNING,
4628                                 "Quota file not on filesystem root. "
4629                                 "Journaled quota will not work");
4630         }
4631
4632         /*
4633          * When we journal data on quota file, we have to flush journal to see
4634          * all updates to the file when we bypass pagecache...
4635          */
4636         if (EXT4_SB(sb)->s_journal &&
4637             ext4_should_journal_data(path->dentry->d_inode)) {
4638                 /*
4639                  * We don't need to lock updates but journal_flush() could
4640                  * otherwise be livelocked...
4641                  */
4642                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4643                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4644                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4645                 if (err)
4646                         return err;
4647         }
4648
4649         return dquot_quota_on(sb, type, format_id, path);
4650 }
4651
4652 static int ext4_quota_off(struct super_block *sb, int type)
4653 {
4654         struct inode *inode = sb_dqopt(sb)->files[type];
4655         handle_t *handle;
4656
4657         /* Force all delayed allocation blocks to be allocated.
4658          * Caller already holds s_umount sem */
4659         if (test_opt(sb, DELALLOC))
4660                 sync_filesystem(sb);
4661
4662         if (!inode)
4663                 goto out;
4664
4665         /* Update modification times of quota files when userspace can
4666          * start looking at them */
4667         handle = ext4_journal_start(inode, 1);
4668         if (IS_ERR(handle))
4669                 goto out;
4670         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4671         ext4_mark_inode_dirty(handle, inode);
4672         ext4_journal_stop(handle);
4673
4674 out:
4675         return dquot_quota_off(sb, type);
4676 }
4677
4678 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4679  * acquiring the locks... As quota files are never truncated and quota code
4680  * itself serializes the operations (and no one else should touch the files)
4681  * we don't have to be afraid of races */
4682 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4683                                size_t len, loff_t off)
4684 {
4685         struct inode *inode = sb_dqopt(sb)->files[type];
4686         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4687         int err = 0;
4688         int offset = off & (sb->s_blocksize - 1);
4689         int tocopy;
4690         size_t toread;
4691         struct buffer_head *bh;
4692         loff_t i_size = i_size_read(inode);
4693
4694         if (off > i_size)
4695                 return 0;
4696         if (off+len > i_size)
4697                 len = i_size-off;
4698         toread = len;
4699         while (toread > 0) {
4700                 tocopy = sb->s_blocksize - offset < toread ?
4701                                 sb->s_blocksize - offset : toread;
4702                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4703                 if (err)
4704                         return err;
4705                 if (!bh)        /* A hole? */
4706                         memset(data, 0, tocopy);
4707                 else
4708                         memcpy(data, bh->b_data+offset, tocopy);
4709                 brelse(bh);
4710                 offset = 0;
4711                 toread -= tocopy;
4712                 data += tocopy;
4713                 blk++;
4714         }
4715         return len;
4716 }
4717
4718 /* Write to quotafile (we know the transaction is already started and has
4719  * enough credits) */
4720 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4721                                 const char *data, size_t len, loff_t off)
4722 {
4723         struct inode *inode = sb_dqopt(sb)->files[type];
4724         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4725         int err = 0;
4726         int offset = off & (sb->s_blocksize - 1);
4727         struct buffer_head *bh;
4728         handle_t *handle = journal_current_handle();
4729
4730         if (EXT4_SB(sb)->s_journal && !handle) {
4731                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4732                         " cancelled because transaction is not started",
4733                         (unsigned long long)off, (unsigned long long)len);
4734                 return -EIO;
4735         }
4736         /*
4737          * Since we account only one data block in transaction credits,
4738          * then it is impossible to cross a block boundary.
4739          */
4740         if (sb->s_blocksize - offset < len) {
4741                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4742                         " cancelled because not block aligned",
4743                         (unsigned long long)off, (unsigned long long)len);
4744                 return -EIO;
4745         }
4746
4747         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4748         bh = ext4_bread(handle, inode, blk, 1, &err);
4749         if (!bh)
4750                 goto out;
4751         err = ext4_journal_get_write_access(handle, bh);
4752         if (err) {
4753                 brelse(bh);
4754                 goto out;
4755         }
4756         lock_buffer(bh);
4757         memcpy(bh->b_data+offset, data, len);
4758         flush_dcache_page(bh->b_page);
4759         unlock_buffer(bh);
4760         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4761         brelse(bh);
4762 out:
4763         if (err) {
4764                 mutex_unlock(&inode->i_mutex);
4765                 return err;
4766         }
4767         if (inode->i_size < off + len) {
4768                 i_size_write(inode, off + len);
4769                 EXT4_I(inode)->i_disksize = inode->i_size;
4770                 ext4_mark_inode_dirty(handle, inode);
4771         }
4772         mutex_unlock(&inode->i_mutex);
4773         return len;
4774 }
4775
4776 #endif
4777
4778 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4779                        const char *dev_name, void *data)
4780 {
4781         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4782 }
4783
4784 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4785 static inline void register_as_ext2(void)
4786 {
4787         int err = register_filesystem(&ext2_fs_type);
4788         if (err)
4789                 printk(KERN_WARNING
4790                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4791 }
4792
4793 static inline void unregister_as_ext2(void)
4794 {
4795         unregister_filesystem(&ext2_fs_type);
4796 }
4797
4798 static inline int ext2_feature_set_ok(struct super_block *sb)
4799 {
4800         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4801                 return 0;
4802         if (sb->s_flags & MS_RDONLY)
4803                 return 1;
4804         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4805                 return 0;
4806         return 1;
4807 }
4808 MODULE_ALIAS("ext2");
4809 #else
4810 static inline void register_as_ext2(void) { }
4811 static inline void unregister_as_ext2(void) { }
4812 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4813 #endif
4814
4815 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4816 static inline void register_as_ext3(void)
4817 {
4818         int err = register_filesystem(&ext3_fs_type);
4819         if (err)
4820                 printk(KERN_WARNING
4821                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4822 }
4823
4824 static inline void unregister_as_ext3(void)
4825 {
4826         unregister_filesystem(&ext3_fs_type);
4827 }
4828
4829 static inline int ext3_feature_set_ok(struct super_block *sb)
4830 {
4831         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4832                 return 0;
4833         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4834                 return 0;
4835         if (sb->s_flags & MS_RDONLY)
4836                 return 1;
4837         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4838                 return 0;
4839         return 1;
4840 }
4841 MODULE_ALIAS("ext3");
4842 #else
4843 static inline void register_as_ext3(void) { }
4844 static inline void unregister_as_ext3(void) { }
4845 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4846 #endif
4847
4848 static struct file_system_type ext4_fs_type = {
4849         .owner          = THIS_MODULE,
4850         .name           = "ext4",
4851         .mount          = ext4_mount,
4852         .kill_sb        = kill_block_super,
4853         .fs_flags       = FS_REQUIRES_DEV,
4854 };
4855
4856 static int __init ext4_init_feat_adverts(void)
4857 {
4858         struct ext4_features *ef;
4859         int ret = -ENOMEM;
4860
4861         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4862         if (!ef)
4863                 goto out;
4864
4865         ef->f_kobj.kset = ext4_kset;
4866         init_completion(&ef->f_kobj_unregister);
4867         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4868                                    "features");
4869         if (ret) {
4870                 kfree(ef);
4871                 goto out;
4872         }
4873
4874         ext4_feat = ef;
4875         ret = 0;
4876 out:
4877         return ret;
4878 }
4879
4880 static void ext4_exit_feat_adverts(void)
4881 {
4882         kobject_put(&ext4_feat->f_kobj);
4883         wait_for_completion(&ext4_feat->f_kobj_unregister);
4884         kfree(ext4_feat);
4885 }
4886
4887 /* Shared across all ext4 file systems */
4888 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
4889 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
4890
4891 static int __init ext4_init_fs(void)
4892 {
4893         int i, err;
4894
4895         ext4_li_info = NULL;
4896         mutex_init(&ext4_li_mtx);
4897
4898         ext4_check_flag_values();
4899
4900         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
4901                 mutex_init(&ext4__aio_mutex[i]);
4902                 init_waitqueue_head(&ext4__ioend_wq[i]);
4903         }
4904
4905         err = ext4_init_pageio();
4906         if (err)
4907                 return err;
4908         err = ext4_init_system_zone();
4909         if (err)
4910                 goto out6;
4911         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
4912         if (!ext4_kset)
4913                 goto out5;
4914         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
4915
4916         err = ext4_init_feat_adverts();
4917         if (err)
4918                 goto out4;
4919
4920         err = ext4_init_mballoc();
4921         if (err)
4922                 goto out3;
4923
4924         err = ext4_init_xattr();
4925         if (err)
4926                 goto out2;
4927         err = init_inodecache();
4928         if (err)
4929                 goto out1;
4930         register_as_ext3();
4931         register_as_ext2();
4932         err = register_filesystem(&ext4_fs_type);
4933         if (err)
4934                 goto out;
4935
4936         return 0;
4937 out:
4938         unregister_as_ext2();
4939         unregister_as_ext3();
4940         destroy_inodecache();
4941 out1:
4942         ext4_exit_xattr();
4943 out2:
4944         ext4_exit_mballoc();
4945 out3:
4946         ext4_exit_feat_adverts();
4947 out4:
4948         if (ext4_proc_root)
4949                 remove_proc_entry("fs/ext4", NULL);
4950         kset_unregister(ext4_kset);
4951 out5:
4952         ext4_exit_system_zone();
4953 out6:
4954         ext4_exit_pageio();
4955         return err;
4956 }
4957
4958 static void __exit ext4_exit_fs(void)
4959 {
4960         ext4_destroy_lazyinit_thread();
4961         unregister_as_ext2();
4962         unregister_as_ext3();
4963         unregister_filesystem(&ext4_fs_type);
4964         destroy_inodecache();
4965         ext4_exit_xattr();
4966         ext4_exit_mballoc();
4967         ext4_exit_feat_adverts();
4968         remove_proc_entry("fs/ext4", NULL);
4969         kset_unregister(ext4_kset);
4970         ext4_exit_system_zone();
4971         ext4_exit_pageio();
4972 }
4973
4974 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
4975 MODULE_DESCRIPTION("Fourth Extended Filesystem");
4976 MODULE_LICENSE("GPL");
4977 module_init(ext4_init_fs)
4978 module_exit(ext4_exit_fs)