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_commit_super(struct super_block *sb, int sync);
  66 static void ext4_mark_recovery_complete(struct super_block *sb,
  67                                         struct ext4_super_block *es);
  68 static void ext4_clear_journal_err(struct super_block *sb,
  69                                    struct ext4_super_block *es);
  70 static int ext4_sync_fs(struct super_block *sb, int wait);
  71 static const char *ext4_decode_error(struct super_block *sb, int errno,
  72                                      char nbuf[16]);
  73 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  75 static int ext4_unfreeze(struct super_block *sb);
  76 static void ext4_write_super(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  96 #else
  97 #define IS_EXT2_SB(sb) (0)
  98 #endif
  99
 100
 101 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 102 static struct file_system_type ext3_fs_type = {
 103         .owner          = THIS_MODULE,
 104         .name           = "ext3",
 105         .mount          = ext4_mount,
 106         .kill_sb        = kill_block_super,
 107         .fs_flags       = FS_REQUIRES_DEV,
 108 };
 109 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 110 #else
 111 #define IS_EXT3_SB(sb) (0)
 112 #endif
 113
 114 void *ext4_kvmalloc(size_t size, gfp_t flags)
 115 {
 116         void *ret;
 117
 118         ret = kmalloc(size, flags);
 119         if (!ret)
 120                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 121         return ret;
 122 }
 123
 124 void *ext4_kvzalloc(size_t size, gfp_t flags)
 125 {
 126         void *ret;
 127
 128         ret = kzalloc(size, flags);
 129         if (!ret)
 130                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 131         return ret;
 132 }
 133
 134 void ext4_kvfree(void *ptr)
 135 {
 136         if (is_vmalloc_addr(ptr))
 137                 vfree(ptr);
 138         else
 139                 kfree(ptr);
 140
 141 }
 142
 143 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 144                                struct ext4_group_desc *bg)
 145 {
 146         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 147                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 148                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 149 }
 150
 151 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 152                                struct ext4_group_desc *bg)
 153 {
 154         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 155                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 156                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 157 }
 158
 159 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 160                               struct ext4_group_desc *bg)
 161 {
 162         return le32_to_cpu(bg->bg_inode_table_lo) |
 163                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 164                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 165 }
 166
 167 __u32 ext4_free_group_clusters(struct super_block *sb,
 168                                struct ext4_group_desc *bg)
 169 {
 170         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 171                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 172                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 173 }
 174
 175 __u32 ext4_free_inodes_count(struct super_block *sb,
 176                               struct ext4_group_desc *bg)
 177 {
 178         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 179                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 180                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 181 }
 182
 183 __u32 ext4_used_dirs_count(struct super_block *sb,
 184                               struct ext4_group_desc *bg)
 185 {
 186         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 187                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 188                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 189 }
 190
 191 __u32 ext4_itable_unused_count(struct super_block *sb,
 192                               struct ext4_group_desc *bg)
 193 {
 194         return le16_to_cpu(bg->bg_itable_unused_lo) |
 195                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 196                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 197 }
 198
 199 void ext4_block_bitmap_set(struct super_block *sb,
 200                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 201 {
 202         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 203         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 204                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 205 }
 206
 207 void ext4_inode_bitmap_set(struct super_block *sb,
 208                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 209 {
 210         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 211         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 212                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 213 }
 214
 215 void ext4_inode_table_set(struct super_block *sb,
 216                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 217 {
 218         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 219         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 220                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 221 }
 222
 223 void ext4_free_group_clusters_set(struct super_block *sb,
 224                                   struct ext4_group_desc *bg, __u32 count)
 225 {
 226         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 227         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 228                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 229 }
 230
 231 void ext4_free_inodes_set(struct super_block *sb,
 232                           struct ext4_group_desc *bg, __u32 count)
 233 {
 234         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 235         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 236                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 237 }
 238
 239 void ext4_used_dirs_set(struct super_block *sb,
 240                           struct ext4_group_desc *bg, __u32 count)
 241 {
 242         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 243         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 244                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 245 }
 246
 247 void ext4_itable_unused_set(struct super_block *sb,
 248                           struct ext4_group_desc *bg, __u32 count)
 249 {
 250         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 251         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 252                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 253 }
 254
 255
 256 /* Just increment the non-pointer handle value */
 257 static handle_t *ext4_get_nojournal(void)
 258 {
 259         handle_t *handle = current->journal_info;
 260         unsigned long ref_cnt = (unsigned long)handle;
 261
 262         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 263
 264         ref_cnt++;
 265         handle = (handle_t *)ref_cnt;
 266
 267         current->journal_info = handle;
 268         return handle;
 269 }
 270
 271
 272 /* Decrement the non-pointer handle value */
 273 static void ext4_put_nojournal(handle_t *handle)
 274 {
 275         unsigned long ref_cnt = (unsigned long)handle;
 276
 277         BUG_ON(ref_cnt == 0);
 278
 279         ref_cnt--;
 280         handle = (handle_t *)ref_cnt;
 281
 282         current->journal_info = handle;
 283 }
 284
 285 /*
 286  * Wrappers for jbd2_journal_start/end.
 287  *
 288  * The only special thing we need to do here is to make sure that all
 289  * journal_end calls result in the superblock being marked dirty, so
 290  * that sync() will call the filesystem's write_super callback if
 291  * appropriate.
 292  *
 293  * To avoid j_barrier hold in userspace when a user calls freeze(),
 294  * ext4 prevents a new handle from being started by s_frozen, which
 295  * is in an upper layer.
 296  */
 297 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 298 {
 299         journal_t *journal;
 300         handle_t  *handle;
 301
 302         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 303         if (sb->s_flags & MS_RDONLY)
 304                 return ERR_PTR(-EROFS);
 305
 306         journal = EXT4_SB(sb)->s_journal;
 307         handle = ext4_journal_current_handle();
 308
 309         /*
 310          * If a handle has been started, it should be allowed to
 311          * finish, otherwise deadlock could happen between freeze
 312          * and others(e.g. truncate) due to the restart of the
 313          * journal handle if the filesystem is forzen and active
 314          * handles are not stopped.
 315          */
 316         if (!handle)
 317                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 318
 319         if (!journal)
 320                 return ext4_get_nojournal();
 321         /*
 322          * Special case here: if the journal has aborted behind our
 323          * backs (eg. EIO in the commit thread), then we still need to
 324          * take the FS itself readonly cleanly.
 325          */
 326         if (is_journal_aborted(journal)) {
 327                 ext4_abort(sb, "Detected aborted journal");
 328                 return ERR_PTR(-EROFS);
 329         }
 330         return jbd2_journal_start(journal, nblocks);
 331 }
 332
 333 /*
 334  * The only special thing we need to do here is to make sure that all
 335  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 336  * that sync() will call the filesystem's write_super callback if
 337  * appropriate.
 338  */
 339 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 340 {
 341         struct super_block *sb;
 342         int err;
 343         int rc;
 344
 345         if (!ext4_handle_valid(handle)) {
 346                 ext4_put_nojournal(handle);
 347                 return 0;
 348         }
 349         sb = handle->h_transaction->t_journal->j_private;
 350         err = handle->h_err;
 351         rc = jbd2_journal_stop(handle);
 352
 353         if (!err)
 354                 err = rc;
 355         if (err)
 356                 __ext4_std_error(sb, where, line, err);
 357         return err;
 358 }
 359
 360 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 361                                const char *err_fn, struct buffer_head *bh,
 362                                handle_t *handle, int err)
 363 {
 364         char nbuf[16];
 365         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 366
 367         BUG_ON(!ext4_handle_valid(handle));
 368
 369         if (bh)
 370                 BUFFER_TRACE(bh, "abort");
 371
 372         if (!handle->h_err)
 373                 handle->h_err = err;
 374
 375         if (is_handle_aborted(handle))
 376                 return;
 377
 378         printk(KERN_ERR "%s:%d: aborting transaction: %s in %s\n",
 379                caller, line, errstr, err_fn);
 380
 381         jbd2_journal_abort_handle(handle);
 382 }
 383
 384 static void __save_error_info(struct super_block *sb, const char *func,
 385                             unsigned int line)
 386 {
 387         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 388
 389         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 390         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 391         es->s_last_error_time = cpu_to_le32(get_seconds());
 392         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 393         es->s_last_error_line = cpu_to_le32(line);
 394         if (!es->s_first_error_time) {
 395                 es->s_first_error_time = es->s_last_error_time;
 396                 strncpy(es->s_first_error_func, func,
 397                         sizeof(es->s_first_error_func));
 398                 es->s_first_error_line = cpu_to_le32(line);
 399                 es->s_first_error_ino = es->s_last_error_ino;
 400                 es->s_first_error_block = es->s_last_error_block;
 401         }
 402         /*
 403          * Start the daily error reporting function if it hasn't been
 404          * started already
 405          */
 406         if (!es->s_error_count)
 407                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 408         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 409 }
 410
 411 static void save_error_info(struct super_block *sb, const char *func,
 412                             unsigned int line)
 413 {
 414         __save_error_info(sb, func, line);
 415         ext4_commit_super(sb, 1);
 416 }
 417
 418 /*
 419  * The del_gendisk() function uninitializes the disk-specific data
 420  * structures, including the bdi structure, without telling anyone
 421  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 422  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 423  * This is a kludge to prevent these oops until we can put in a proper
 424  * hook in del_gendisk() to inform the VFS and file system layers.
 425  */
 426 static int block_device_ejected(struct super_block *sb)
 427 {
 428         struct inode *bd_inode = sb->s_bdev->bd_inode;
 429         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 430
 431         return bdi->dev == NULL;
 432 }
 433
 434 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 435 {
 436         struct super_block              *sb = journal->j_private;
 437         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 438         int                             error = is_journal_aborted(journal);
 439         struct ext4_journal_cb_entry    *jce, *tmp;
 440
 441         spin_lock(&sbi->s_md_lock);
 442         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 443                 list_del_init(&jce->jce_list);
 444                 spin_unlock(&sbi->s_md_lock);
 445                 jce->jce_func(sb, jce, error);
 446                 spin_lock(&sbi->s_md_lock);
 447         }
 448         spin_unlock(&sbi->s_md_lock);
 449 }
 450
 451 /* Deal with the reporting of failure conditions on a filesystem such as
 452  * inconsistencies detected or read IO failures.
 453  *
 454  * On ext2, we can store the error state of the filesystem in the
 455  * superblock.  That is not possible on ext4, because we may have other
 456  * write ordering constraints on the superblock which prevent us from
 457  * writing it out straight away; and given that the journal is about to
 458  * be aborted, we can't rely on the current, or future, transactions to
 459  * write out the superblock safely.
 460  *
 461  * We'll just use the jbd2_journal_abort() error code to record an error in
 462  * the journal instead.  On recovery, the journal will complain about
 463  * that error until we've noted it down and cleared it.
 464  */
 465
 466 static void ext4_handle_error(struct super_block *sb)
 467 {
 468         if (sb->s_flags & MS_RDONLY)
 469                 return;
 470
 471         if (!test_opt(sb, ERRORS_CONT)) {
 472                 journal_t *journal = EXT4_SB(sb)->s_journal;
 473
 474                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 475                 if (journal)
 476                         jbd2_journal_abort(journal, -EIO);
 477         }
 478         if (test_opt(sb, ERRORS_RO)) {
 479                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 480                 sb->s_flags |= MS_RDONLY;
 481         }
 482         if (test_opt(sb, ERRORS_PANIC))
 483                 panic("EXT4-fs (device %s): panic forced after error\n",
 484                         sb->s_id);
 485 }
 486
 487 void __ext4_error(struct super_block *sb, const char *function,
 488                   unsigned int line, const char *fmt, ...)
 489 {
 490         struct va_format vaf;
 491         va_list args;
 492
 493         va_start(args, fmt);
 494         vaf.fmt = fmt;
 495         vaf.va = &args;
 496         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 497                sb->s_id, function, line, current->comm, &vaf);
 498         va_end(args);
 499
 500         ext4_handle_error(sb);
 501 }
 502
 503 void ext4_error_inode(struct inode *inode, const char *function,
 504                       unsigned int line, ext4_fsblk_t block,
 505                       const char *fmt, ...)
 506 {
 507         va_list args;
 508         struct va_format vaf;
 509         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 510
 511         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 512         es->s_last_error_block = cpu_to_le64(block);
 513         save_error_info(inode->i_sb, function, line);
 514         va_start(args, fmt);
 515         vaf.fmt = fmt;
 516         vaf.va = &args;
 517         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 518                inode->i_sb->s_id, function, line, inode->i_ino);
 519         if (block)
 520                 printk(KERN_CONT "block %llu: ", block);
 521         printk(KERN_CONT "comm %s: %pV\n", current->comm, &vaf);
 522         va_end(args);
 523
 524         ext4_handle_error(inode->i_sb);
 525 }
 526
 527 void ext4_error_file(struct file *file, const char *function,
 528                      unsigned int line, ext4_fsblk_t block,
 529                      const char *fmt, ...)
 530 {
 531         va_list args;
 532         struct va_format vaf;
 533         struct ext4_super_block *es;
 534         struct inode *inode = file->f_dentry->d_inode;
 535         char pathname[80], *path;
 536
 537         es = EXT4_SB(inode->i_sb)->s_es;
 538         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 539         save_error_info(inode->i_sb, function, line);
 540         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 541         if (IS_ERR(path))
 542                 path = "(unknown)";
 543         printk(KERN_CRIT
 544                "EXT4-fs error (device %s): %s:%d: inode #%lu: ",
 545                inode->i_sb->s_id, function, line, inode->i_ino);
 546         if (block)
 547                 printk(KERN_CONT "block %llu: ", block);
 548         va_start(args, fmt);
 549         vaf.fmt = fmt;
 550         vaf.va = &args;
 551         printk(KERN_CONT "comm %s: path %s: %pV\n", current->comm, path, &vaf);
 552         va_end(args);
 553
 554         ext4_handle_error(inode->i_sb);
 555 }
 556
 557 static const char *ext4_decode_error(struct super_block *sb, int errno,
 558                                      char nbuf[16])
 559 {
 560         char *errstr = NULL;
 561
 562         switch (errno) {
 563         case -EIO:
 564                 errstr = "IO failure";
 565                 break;
 566         case -ENOMEM:
 567                 errstr = "Out of memory";
 568                 break;
 569         case -EROFS:
 570                 if (!sb || (EXT4_SB(sb)->s_journal &&
 571                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 572                         errstr = "Journal has aborted";
 573                 else
 574                         errstr = "Readonly filesystem";
 575                 break;
 576         default:
 577                 /* If the caller passed in an extra buffer for unknown
 578                  * errors, textualise them now.  Else we just return
 579                  * NULL. */
 580                 if (nbuf) {
 581                         /* Check for truncated error codes... */
 582                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 583                                 errstr = nbuf;
 584                 }
 585                 break;
 586         }
 587
 588         return errstr;
 589 }
 590
 591 /* __ext4_std_error decodes expected errors from journaling functions
 592  * automatically and invokes the appropriate error response.  */
 593
 594 void __ext4_std_error(struct super_block *sb, const char *function,
 595                       unsigned int line, int errno)
 596 {
 597         char nbuf[16];
 598         const char *errstr;
 599
 600         /* Special case: if the error is EROFS, and we're not already
 601          * inside a transaction, then there's really no point in logging
 602          * an error. */
 603         if (errno == -EROFS && journal_current_handle() == NULL &&
 604             (sb->s_flags & MS_RDONLY))
 605                 return;
 606
 607         errstr = ext4_decode_error(sb, errno, nbuf);
 608         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 609                sb->s_id, function, line, errstr);
 610         save_error_info(sb, function, line);
 611
 612         ext4_handle_error(sb);
 613 }
 614
 615 /*
 616  * ext4_abort is a much stronger failure handler than ext4_error.  The
 617  * abort function may be used to deal with unrecoverable failures such
 618  * as journal IO errors or ENOMEM at a critical moment in log management.
 619  *
 620  * We unconditionally force the filesystem into an ABORT|READONLY state,
 621  * unless the error response on the fs has been set to panic in which
 622  * case we take the easy way out and panic immediately.
 623  */
 624
 625 void __ext4_abort(struct super_block *sb, const char *function,
 626                 unsigned int line, const char *fmt, ...)
 627 {
 628         va_list args;
 629
 630         save_error_info(sb, function, line);
 631         va_start(args, fmt);
 632         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 633                function, line);
 634         vprintk(fmt, args);
 635         printk("\n");
 636         va_end(args);
 637
 638         if ((sb->s_flags & MS_RDONLY) == 0) {
 639                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 640                 sb->s_flags |= MS_RDONLY;
 641                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 642                 if (EXT4_SB(sb)->s_journal)
 643                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 644                 save_error_info(sb, function, line);
 645         }
 646         if (test_opt(sb, ERRORS_PANIC))
 647                 panic("EXT4-fs panic from previous error\n");
 648 }
 649
 650 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 651 {
 652         struct va_format vaf;
 653         va_list args;
 654
 655         va_start(args, fmt);
 656         vaf.fmt = fmt;
 657         vaf.va = &args;
 658         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 659         va_end(args);
 660 }
 661
 662 void __ext4_warning(struct super_block *sb, const char *function,
 663                     unsigned int line, const char *fmt, ...)
 664 {
 665         struct va_format vaf;
 666         va_list args;
 667
 668         va_start(args, fmt);
 669         vaf.fmt = fmt;
 670         vaf.va = &args;
 671         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 672                sb->s_id, function, line, &vaf);
 673         va_end(args);
 674 }
 675
 676 void __ext4_grp_locked_error(const char *function, unsigned int line,
 677                              struct super_block *sb, ext4_group_t grp,
 678                              unsigned long ino, ext4_fsblk_t block,
 679                              const char *fmt, ...)
 680 __releases(bitlock)
 681 __acquires(bitlock)
 682 {
 683         struct va_format vaf;
 684         va_list args;
 685         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 686
 687         es->s_last_error_ino = cpu_to_le32(ino);
 688         es->s_last_error_block = cpu_to_le64(block);
 689         __save_error_info(sb, function, line);
 690
 691         va_start(args, fmt);
 692
 693         vaf.fmt = fmt;
 694         vaf.va = &args;
 695         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 696                sb->s_id, function, line, grp);
 697         if (ino)
 698                 printk(KERN_CONT "inode %lu: ", ino);
 699         if (block)
 700                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 701         printk(KERN_CONT "%pV\n", &vaf);
 702         va_end(args);
 703
 704         if (test_opt(sb, ERRORS_CONT)) {
 705                 ext4_commit_super(sb, 0);
 706                 return;
 707         }
 708
 709         ext4_unlock_group(sb, grp);
 710         ext4_handle_error(sb);
 711         /*
 712          * We only get here in the ERRORS_RO case; relocking the group
 713          * may be dangerous, but nothing bad will happen since the
 714          * filesystem will have already been marked read/only and the
 715          * journal has been aborted.  We return 1 as a hint to callers
 716          * who might what to use the return value from
 717          * ext4_grp_locked_error() to distinguish between the
 718          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 719          * aggressively from the ext4 function in question, with a
 720          * more appropriate error code.
 721          */
 722         ext4_lock_group(sb, grp);
 723         return;
 724 }
 725
 726 void ext4_update_dynamic_rev(struct super_block *sb)
 727 {
 728         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 729
 730         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 731                 return;
 732
 733         ext4_warning(sb,
 734                      "updating to rev %d because of new feature flag, "
 735                      "running e2fsck is recommended",
 736                      EXT4_DYNAMIC_REV);
 737
 738         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 739         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 740         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 741         /* leave es->s_feature_*compat flags alone */
 742         /* es->s_uuid will be set by e2fsck if empty */
 743
 744         /*
 745          * The rest of the superblock fields should be zero, and if not it
 746          * means they are likely already in use, so leave them alone.  We
 747          * can leave it up to e2fsck to clean up any inconsistencies there.
 748          */
 749 }
 750
 751 /*
 752  * Open the external journal device
 753  */
 754 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 755 {
 756         struct block_device *bdev;
 757         char b[BDEVNAME_SIZE];
 758
 759         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 760         if (IS_ERR(bdev))
 761                 goto fail;
 762         return bdev;
 763
 764 fail:
 765         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 766                         __bdevname(dev, b), PTR_ERR(bdev));
 767         return NULL;
 768 }
 769
 770 /*
 771  * Release the journal device
 772  */
 773 static int ext4_blkdev_put(struct block_device *bdev)
 774 {
 775         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 776 }
 777
 778 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 779 {
 780         struct block_device *bdev;
 781         int ret = -ENODEV;
 782
 783         bdev = sbi->journal_bdev;
 784         if (bdev) {
 785                 ret = ext4_blkdev_put(bdev);
 786                 sbi->journal_bdev = NULL;
 787         }
 788         return ret;
 789 }
 790
 791 static inline struct inode *orphan_list_entry(struct list_head *l)
 792 {
 793         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 794 }
 795
 796 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 797 {
 798         struct list_head *l;
 799
 800         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 801                  le32_to_cpu(sbi->s_es->s_last_orphan));
 802
 803         printk(KERN_ERR "sb_info orphan list:\n");
 804         list_for_each(l, &sbi->s_orphan) {
 805                 struct inode *inode = orphan_list_entry(l);
 806                 printk(KERN_ERR "  "
 807                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 808                        inode->i_sb->s_id, inode->i_ino, inode,
 809                        inode->i_mode, inode->i_nlink,
 810                        NEXT_ORPHAN(inode));
 811         }
 812 }
 813
 814 static void ext4_put_super(struct super_block *sb)
 815 {
 816         struct ext4_sb_info *sbi = EXT4_SB(sb);
 817         struct ext4_super_block *es = sbi->s_es;
 818         int i, err;
 819
 820         ext4_unregister_li_request(sb);
 821         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 822
 823         flush_workqueue(sbi->dio_unwritten_wq);
 824         destroy_workqueue(sbi->dio_unwritten_wq);
 825
 826         lock_super(sb);
 827         if (sb->s_dirt)
 828                 ext4_commit_super(sb, 1);
 829
 830         if (sbi->s_journal) {
 831                 err = jbd2_journal_destroy(sbi->s_journal);
 832                 sbi->s_journal = NULL;
 833                 if (err < 0)
 834                         ext4_abort(sb, "Couldn't clean up the journal");
 835         }
 836
 837         del_timer(&sbi->s_err_report);
 838         ext4_release_system_zone(sb);
 839         ext4_mb_release(sb);
 840         ext4_ext_release(sb);
 841         ext4_xattr_put_super(sb);
 842
 843         if (!(sb->s_flags & MS_RDONLY)) {
 844                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 845                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 846                 ext4_commit_super(sb, 1);
 847         }
 848         if (sbi->s_proc) {
 849                 remove_proc_entry(sb->s_id, ext4_proc_root);
 850         }
 851         kobject_del(&sbi->s_kobj);
 852
 853         for (i = 0; i < sbi->s_gdb_count; i++)
 854                 brelse(sbi->s_group_desc[i]);
 855         ext4_kvfree(sbi->s_group_desc);
 856         ext4_kvfree(sbi->s_flex_groups);
 857         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 858         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 859         percpu_counter_destroy(&sbi->s_dirs_counter);
 860         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 861         brelse(sbi->s_sbh);
 862 #ifdef CONFIG_QUOTA
 863         for (i = 0; i < MAXQUOTAS; i++)
 864                 kfree(sbi->s_qf_names[i]);
 865 #endif
 866
 867         /* Debugging code just in case the in-memory inode orphan list
 868          * isn't empty.  The on-disk one can be non-empty if we've
 869          * detected an error and taken the fs readonly, but the
 870          * in-memory list had better be clean by this point. */
 871         if (!list_empty(&sbi->s_orphan))
 872                 dump_orphan_list(sb, sbi);
 873         J_ASSERT(list_empty(&sbi->s_orphan));
 874
 875         invalidate_bdev(sb->s_bdev);
 876         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 877                 /*
 878                  * Invalidate the journal device's buffers.  We don't want them
 879                  * floating about in memory - the physical journal device may
 880                  * hotswapped, and it breaks the `ro-after' testing code.
 881                  */
 882                 sync_blockdev(sbi->journal_bdev);
 883                 invalidate_bdev(sbi->journal_bdev);
 884                 ext4_blkdev_remove(sbi);
 885         }
 886         if (sbi->s_mmp_tsk)
 887                 kthread_stop(sbi->s_mmp_tsk);
 888         sb->s_fs_info = NULL;
 889         /*
 890          * Now that we are completely done shutting down the
 891          * superblock, we need to actually destroy the kobject.
 892          */
 893         unlock_super(sb);
 894         kobject_put(&sbi->s_kobj);
 895         wait_for_completion(&sbi->s_kobj_unregister);
 896         kfree(sbi->s_blockgroup_lock);
 897         kfree(sbi);
 898 }
 899
 900 static struct kmem_cache *ext4_inode_cachep;
 901
 902 /*
 903  * Called inside transaction, so use GFP_NOFS
 904  */
 905 static struct inode *ext4_alloc_inode(struct super_block *sb)
 906 {
 907         struct ext4_inode_info *ei;
 908
 909         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 910         if (!ei)
 911                 return NULL;
 912
 913         ei->vfs_inode.i_version = 1;
 914         ei->vfs_inode.i_data.writeback_index = 0;
 915         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 916         INIT_LIST_HEAD(&ei->i_prealloc_list);
 917         spin_lock_init(&ei->i_prealloc_lock);
 918         ei->i_reserved_data_blocks = 0;
 919         ei->i_reserved_meta_blocks = 0;
 920         ei->i_allocated_meta_blocks = 0;
 921         ei->i_da_metadata_calc_len = 0;
 922         spin_lock_init(&(ei->i_block_reservation_lock));
 923 #ifdef CONFIG_QUOTA
 924         ei->i_reserved_quota = 0;
 925 #endif
 926         ei->jinode = NULL;
 927         INIT_LIST_HEAD(&ei->i_completed_io_list);
 928         spin_lock_init(&ei->i_completed_io_lock);
 929         ei->cur_aio_dio = NULL;
 930         ei->i_sync_tid = 0;
 931         ei->i_datasync_tid = 0;
 932         atomic_set(&ei->i_ioend_count, 0);
 933         atomic_set(&ei->i_aiodio_unwritten, 0);
 934
 935         return &ei->vfs_inode;
 936 }
 937
 938 static int ext4_drop_inode(struct inode *inode)
 939 {
 940         int drop = generic_drop_inode(inode);
 941
 942         trace_ext4_drop_inode(inode, drop);
 943         return drop;
 944 }
 945
 946 static void ext4_i_callback(struct rcu_head *head)
 947 {
 948         struct inode *inode = container_of(head, struct inode, i_rcu);
 949         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 950 }
 951
 952 static void ext4_destroy_inode(struct inode *inode)
 953 {
 954         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 955                 ext4_msg(inode->i_sb, KERN_ERR,
 956                          "Inode %lu (%p): orphan list check failed!",
 957                          inode->i_ino, EXT4_I(inode));
 958                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 959                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 960                                 true);
 961                 dump_stack();
 962         }
 963         call_rcu(&inode->i_rcu, ext4_i_callback);
 964 }
 965
 966 static void init_once(void *foo)
 967 {
 968         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 969
 970         INIT_LIST_HEAD(&ei->i_orphan);
 971 #ifdef CONFIG_EXT4_FS_XATTR
 972         init_rwsem(&ei->xattr_sem);
 973 #endif
 974         init_rwsem(&ei->i_data_sem);
 975         inode_init_once(&ei->vfs_inode);
 976 }
 977
 978 static int init_inodecache(void)
 979 {
 980         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 981                                              sizeof(struct ext4_inode_info),
 982                                              0, (SLAB_RECLAIM_ACCOUNT|
 983                                                 SLAB_MEM_SPREAD),
 984                                              init_once);
 985         if (ext4_inode_cachep == NULL)
 986                 return -ENOMEM;
 987         return 0;
 988 }
 989
 990 static void destroy_inodecache(void)
 991 {
 992         kmem_cache_destroy(ext4_inode_cachep);
 993 }
 994
 995 void ext4_clear_inode(struct inode *inode)
 996 {
 997         invalidate_inode_buffers(inode);
 998         end_writeback(inode);
 999         dquot_drop(inode);
1000         ext4_discard_preallocations(inode);
1001         if (EXT4_I(inode)->jinode) {
1002                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1003                                                EXT4_I(inode)->jinode);
1004                 jbd2_free_inode(EXT4_I(inode)->jinode);
1005                 EXT4_I(inode)->jinode = NULL;
1006         }
1007 }
1008
1009 static inline void ext4_show_quota_options(struct seq_file *seq,
1010                                            struct super_block *sb)
1011 {
1012 #if defined(CONFIG_QUOTA)
1013         struct ext4_sb_info *sbi = EXT4_SB(sb);
1014
1015         if (sbi->s_jquota_fmt) {
1016                 char *fmtname = "";
1017
1018                 switch (sbi->s_jquota_fmt) {
1019                 case QFMT_VFS_OLD:
1020                         fmtname = "vfsold";
1021                         break;
1022                 case QFMT_VFS_V0:
1023                         fmtname = "vfsv0";
1024                         break;
1025                 case QFMT_VFS_V1:
1026                         fmtname = "vfsv1";
1027                         break;
1028                 }
1029                 seq_printf(seq, ",jqfmt=%s", fmtname);
1030         }
1031
1032         if (sbi->s_qf_names[USRQUOTA])
1033                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1034
1035         if (sbi->s_qf_names[GRPQUOTA])
1036                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1037
1038         if (test_opt(sb, USRQUOTA))
1039                 seq_puts(seq, ",usrquota");
1040
1041         if (test_opt(sb, GRPQUOTA))
1042                 seq_puts(seq, ",grpquota");
1043 #endif
1044 }
1045
1046 /*
1047  * Show an option if
1048  *  - it's set to a non-default value OR
1049  *  - if the per-sb default is different from the global default
1050  */
1051 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1052 {
1053         int def_errors;
1054         unsigned long def_mount_opts;
1055         struct super_block *sb = root->d_sb;
1056         struct ext4_sb_info *sbi = EXT4_SB(sb);
1057         struct ext4_super_block *es = sbi->s_es;
1058
1059         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
1060         def_errors     = le16_to_cpu(es->s_errors);
1061
1062         if (sbi->s_sb_block != 1)
1063                 seq_printf(seq, ",sb=%llu", sbi->s_sb_block);
1064         if (test_opt(sb, MINIX_DF))
1065                 seq_puts(seq, ",minixdf");
1066         if (test_opt(sb, GRPID) && !(def_mount_opts & EXT4_DEFM_BSDGROUPS))
1067                 seq_puts(seq, ",grpid");
1068         if (!test_opt(sb, GRPID) && (def_mount_opts & EXT4_DEFM_BSDGROUPS))
1069                 seq_puts(seq, ",nogrpid");
1070         if (sbi->s_resuid != EXT4_DEF_RESUID ||
1071             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID) {
1072                 seq_printf(seq, ",resuid=%u", sbi->s_resuid);
1073         }
1074         if (sbi->s_resgid != EXT4_DEF_RESGID ||
1075             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID) {
1076                 seq_printf(seq, ",resgid=%u", sbi->s_resgid);
1077         }
1078         if (test_opt(sb, ERRORS_RO)) {
1079                 if (def_errors == EXT4_ERRORS_PANIC ||
1080                     def_errors == EXT4_ERRORS_CONTINUE) {
1081                         seq_puts(seq, ",errors=remount-ro");
1082                 }
1083         }
1084         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1085                 seq_puts(seq, ",errors=continue");
1086         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1087                 seq_puts(seq, ",errors=panic");
1088         if (test_opt(sb, NO_UID32) && !(def_mount_opts & EXT4_DEFM_UID16))
1089                 seq_puts(seq, ",nouid32");
1090         if (test_opt(sb, DEBUG) && !(def_mount_opts & EXT4_DEFM_DEBUG))
1091                 seq_puts(seq, ",debug");
1092 #ifdef CONFIG_EXT4_FS_XATTR
1093         if (test_opt(sb, XATTR_USER))
1094                 seq_puts(seq, ",user_xattr");
1095         if (!test_opt(sb, XATTR_USER))
1096                 seq_puts(seq, ",nouser_xattr");
1097 #endif
1098 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1099         if (test_opt(sb, POSIX_ACL) && !(def_mount_opts & EXT4_DEFM_ACL))
1100                 seq_puts(seq, ",acl");
1101         if (!test_opt(sb, POSIX_ACL) && (def_mount_opts & EXT4_DEFM_ACL))
1102                 seq_puts(seq, ",noacl");
1103 #endif
1104         if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
1105                 seq_printf(seq, ",commit=%u",
1106                            (unsigned) (sbi->s_commit_interval / HZ));
1107         }
1108         if (sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME) {
1109                 seq_printf(seq, ",min_batch_time=%u",
1110                            (unsigned) sbi->s_min_batch_time);
1111         }
1112         if (sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME) {
1113                 seq_printf(seq, ",max_batch_time=%u",
1114                            (unsigned) sbi->s_max_batch_time);
1115         }
1116
1117         /*
1118          * We're changing the default of barrier mount option, so
1119          * let's always display its mount state so it's clear what its
1120          * status is.
1121          */
1122         seq_puts(seq, ",barrier=");
1123         seq_puts(seq, test_opt(sb, BARRIER) ? "1" : "0");
1124         if (test_opt(sb, JOURNAL_ASYNC_COMMIT))
1125                 seq_puts(seq, ",journal_async_commit");
1126         else if (test_opt(sb, JOURNAL_CHECKSUM))
1127                 seq_puts(seq, ",journal_checksum");
1128         if (test_opt(sb, I_VERSION))
1129                 seq_puts(seq, ",i_version");
1130         if (!test_opt(sb, DELALLOC) &&
1131             !(def_mount_opts & EXT4_DEFM_NODELALLOC))
1132                 seq_puts(seq, ",nodelalloc");
1133
1134         if (!test_opt(sb, MBLK_IO_SUBMIT))
1135                 seq_puts(seq, ",nomblk_io_submit");
1136         if (sbi->s_stripe)
1137                 seq_printf(seq, ",stripe=%lu", sbi->s_stripe);
1138         /*
1139          * journal mode get enabled in different ways
1140          * So just print the value even if we didn't specify it
1141          */
1142         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1143                 seq_puts(seq, ",data=journal");
1144         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1145                 seq_puts(seq, ",data=ordered");
1146         else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1147                 seq_puts(seq, ",data=writeback");
1148
1149         if (sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1150                 seq_printf(seq, ",inode_readahead_blks=%u",
1151                            sbi->s_inode_readahead_blks);
1152
1153         if (test_opt(sb, DATA_ERR_ABORT))
1154                 seq_puts(seq, ",data_err=abort");
1155
1156         if (test_opt(sb, NO_AUTO_DA_ALLOC))
1157                 seq_puts(seq, ",noauto_da_alloc");
1158
1159         if (test_opt(sb, DISCARD) && !(def_mount_opts & EXT4_DEFM_DISCARD))
1160                 seq_puts(seq, ",discard");
1161
1162         if (test_opt(sb, NOLOAD))
1163                 seq_puts(seq, ",norecovery");
1164
1165         if (test_opt(sb, DIOREAD_NOLOCK))
1166                 seq_puts(seq, ",dioread_nolock");
1167
1168         if (test_opt(sb, BLOCK_VALIDITY) &&
1169             !(def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY))
1170                 seq_puts(seq, ",block_validity");
1171
1172         if (!test_opt(sb, INIT_INODE_TABLE))
1173                 seq_puts(seq, ",noinit_itable");
1174         else if (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)
1175                 seq_printf(seq, ",init_itable=%u",
1176                            (unsigned) sbi->s_li_wait_mult);
1177
1178         ext4_show_quota_options(seq, sb);
1179
1180         return 0;
1181 }
1182
1183 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1184                                         u64 ino, u32 generation)
1185 {
1186         struct inode *inode;
1187
1188         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1189                 return ERR_PTR(-ESTALE);
1190         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1191                 return ERR_PTR(-ESTALE);
1192
1193         /* iget isn't really right if the inode is currently unallocated!!
1194          *
1195          * ext4_read_inode will return a bad_inode if the inode had been
1196          * deleted, so we should be safe.
1197          *
1198          * Currently we don't know the generation for parent directory, so
1199          * a generation of 0 means "accept any"
1200          */
1201         inode = ext4_iget(sb, ino);
1202         if (IS_ERR(inode))
1203                 return ERR_CAST(inode);
1204         if (generation && inode->i_generation != generation) {
1205                 iput(inode);
1206                 return ERR_PTR(-ESTALE);
1207         }
1208
1209         return inode;
1210 }
1211
1212 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1213                                         int fh_len, int fh_type)
1214 {
1215         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1216                                     ext4_nfs_get_inode);
1217 }
1218
1219 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1220                                         int fh_len, int fh_type)
1221 {
1222         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1223                                     ext4_nfs_get_inode);
1224 }
1225
1226 /*
1227  * Try to release metadata pages (indirect blocks, directories) which are
1228  * mapped via the block device.  Since these pages could have journal heads
1229  * which would prevent try_to_free_buffers() from freeing them, we must use
1230  * jbd2 layer's try_to_free_buffers() function to release them.
1231  */
1232 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1233                                  gfp_t wait)
1234 {
1235         journal_t *journal = EXT4_SB(sb)->s_journal;
1236
1237         WARN_ON(PageChecked(page));
1238         if (!page_has_buffers(page))
1239                 return 0;
1240         if (journal)
1241                 return jbd2_journal_try_to_free_buffers(journal, page,
1242                                                         wait & ~__GFP_WAIT);
1243         return try_to_free_buffers(page);
1244 }
1245
1246 #ifdef CONFIG_QUOTA
1247 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1248 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1249
1250 static int ext4_write_dquot(struct dquot *dquot);
1251 static int ext4_acquire_dquot(struct dquot *dquot);
1252 static int ext4_release_dquot(struct dquot *dquot);
1253 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1254 static int ext4_write_info(struct super_block *sb, int type);
1255 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1256                          struct path *path);
1257 static int ext4_quota_off(struct super_block *sb, int type);
1258 static int ext4_quota_on_mount(struct super_block *sb, int type);
1259 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1260                                size_t len, loff_t off);
1261 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1262                                 const char *data, size_t len, loff_t off);
1263
1264 static const struct dquot_operations ext4_quota_operations = {
1265         .get_reserved_space = ext4_get_reserved_space,
1266         .write_dquot    = ext4_write_dquot,
1267         .acquire_dquot  = ext4_acquire_dquot,
1268         .release_dquot  = ext4_release_dquot,
1269         .mark_dirty     = ext4_mark_dquot_dirty,
1270         .write_info     = ext4_write_info,
1271         .alloc_dquot    = dquot_alloc,
1272         .destroy_dquot  = dquot_destroy,
1273 };
1274
1275 static const struct quotactl_ops ext4_qctl_operations = {
1276         .quota_on       = ext4_quota_on,
1277         .quota_off      = ext4_quota_off,
1278         .quota_sync     = dquot_quota_sync,
1279         .get_info       = dquot_get_dqinfo,
1280         .set_info       = dquot_set_dqinfo,
1281         .get_dqblk      = dquot_get_dqblk,
1282         .set_dqblk      = dquot_set_dqblk
1283 };
1284 #endif
1285
1286 static const struct super_operations ext4_sops = {
1287         .alloc_inode    = ext4_alloc_inode,
1288         .destroy_inode  = ext4_destroy_inode,
1289         .write_inode    = ext4_write_inode,
1290         .dirty_inode    = ext4_dirty_inode,
1291         .drop_inode     = ext4_drop_inode,
1292         .evict_inode    = ext4_evict_inode,
1293         .put_super      = ext4_put_super,
1294         .sync_fs        = ext4_sync_fs,
1295         .freeze_fs      = ext4_freeze,
1296         .unfreeze_fs    = ext4_unfreeze,
1297         .statfs         = ext4_statfs,
1298         .remount_fs     = ext4_remount,
1299         .show_options   = ext4_show_options,
1300 #ifdef CONFIG_QUOTA
1301         .quota_read     = ext4_quota_read,
1302         .quota_write    = ext4_quota_write,
1303 #endif
1304         .bdev_try_to_free_page = bdev_try_to_free_page,
1305 };
1306
1307 static const struct super_operations ext4_nojournal_sops = {
1308         .alloc_inode    = ext4_alloc_inode,
1309         .destroy_inode  = ext4_destroy_inode,
1310         .write_inode    = ext4_write_inode,
1311         .dirty_inode    = ext4_dirty_inode,
1312         .drop_inode     = ext4_drop_inode,
1313         .evict_inode    = ext4_evict_inode,
1314         .write_super    = ext4_write_super,
1315         .put_super      = ext4_put_super,
1316         .statfs         = ext4_statfs,
1317         .remount_fs     = ext4_remount,
1318         .show_options   = ext4_show_options,
1319 #ifdef CONFIG_QUOTA
1320         .quota_read     = ext4_quota_read,
1321         .quota_write    = ext4_quota_write,
1322 #endif
1323         .bdev_try_to_free_page = bdev_try_to_free_page,
1324 };
1325
1326 static const struct export_operations ext4_export_ops = {
1327         .fh_to_dentry = ext4_fh_to_dentry,
1328         .fh_to_parent = ext4_fh_to_parent,
1329         .get_parent = ext4_get_parent,
1330 };
1331
1332 enum {
1333         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1334         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1335         Opt_nouid32, Opt_debug, Opt_oldalloc, Opt_orlov,
1336         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1337         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload, Opt_nobh, Opt_bh,
1338         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1339         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1340         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1341         Opt_data_err_abort, Opt_data_err_ignore,
1342         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1343         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1344         Opt_noquota, Opt_ignore, Opt_barrier, Opt_nobarrier, Opt_err,
1345         Opt_usrquota, Opt_grpquota, Opt_i_version,
1346         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1347         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1348         Opt_inode_readahead_blks, Opt_journal_ioprio,
1349         Opt_dioread_nolock, Opt_dioread_lock,
1350         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1351 };
1352
1353 static const match_table_t tokens = {
1354         {Opt_bsd_df, "bsddf"},
1355         {Opt_minix_df, "minixdf"},
1356         {Opt_grpid, "grpid"},
1357         {Opt_grpid, "bsdgroups"},
1358         {Opt_nogrpid, "nogrpid"},
1359         {Opt_nogrpid, "sysvgroups"},
1360         {Opt_resgid, "resgid=%u"},
1361         {Opt_resuid, "resuid=%u"},
1362         {Opt_sb, "sb=%u"},
1363         {Opt_err_cont, "errors=continue"},
1364         {Opt_err_panic, "errors=panic"},
1365         {Opt_err_ro, "errors=remount-ro"},
1366         {Opt_nouid32, "nouid32"},
1367         {Opt_debug, "debug"},
1368         {Opt_oldalloc, "oldalloc"},
1369         {Opt_orlov, "orlov"},
1370         {Opt_user_xattr, "user_xattr"},
1371         {Opt_nouser_xattr, "nouser_xattr"},
1372         {Opt_acl, "acl"},
1373         {Opt_noacl, "noacl"},
1374         {Opt_noload, "noload"},
1375         {Opt_noload, "norecovery"},
1376         {Opt_nobh, "nobh"},
1377         {Opt_bh, "bh"},
1378         {Opt_commit, "commit=%u"},
1379         {Opt_min_batch_time, "min_batch_time=%u"},
1380         {Opt_max_batch_time, "max_batch_time=%u"},
1381         {Opt_journal_dev, "journal_dev=%u"},
1382         {Opt_journal_checksum, "journal_checksum"},
1383         {Opt_journal_async_commit, "journal_async_commit"},
1384         {Opt_abort, "abort"},
1385         {Opt_data_journal, "data=journal"},
1386         {Opt_data_ordered, "data=ordered"},
1387         {Opt_data_writeback, "data=writeback"},
1388         {Opt_data_err_abort, "data_err=abort"},
1389         {Opt_data_err_ignore, "data_err=ignore"},
1390         {Opt_offusrjquota, "usrjquota="},
1391         {Opt_usrjquota, "usrjquota=%s"},
1392         {Opt_offgrpjquota, "grpjquota="},
1393         {Opt_grpjquota, "grpjquota=%s"},
1394         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1395         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1396         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1397         {Opt_grpquota, "grpquota"},
1398         {Opt_noquota, "noquota"},
1399         {Opt_quota, "quota"},
1400         {Opt_usrquota, "usrquota"},
1401         {Opt_barrier, "barrier=%u"},
1402         {Opt_barrier, "barrier"},
1403         {Opt_nobarrier, "nobarrier"},
1404         {Opt_i_version, "i_version"},
1405         {Opt_stripe, "stripe=%u"},
1406         {Opt_delalloc, "delalloc"},
1407         {Opt_nodelalloc, "nodelalloc"},
1408         {Opt_mblk_io_submit, "mblk_io_submit"},
1409         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1410         {Opt_block_validity, "block_validity"},
1411         {Opt_noblock_validity, "noblock_validity"},
1412         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1413         {Opt_journal_ioprio, "journal_ioprio=%u"},
1414         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1415         {Opt_auto_da_alloc, "auto_da_alloc"},
1416         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1417         {Opt_dioread_nolock, "dioread_nolock"},
1418         {Opt_dioread_lock, "dioread_lock"},
1419         {Opt_discard, "discard"},
1420         {Opt_nodiscard, "nodiscard"},
1421         {Opt_init_itable, "init_itable=%u"},
1422         {Opt_init_itable, "init_itable"},
1423         {Opt_noinit_itable, "noinit_itable"},
1424         {Opt_err, NULL},
1425 };
1426
1427 static ext4_fsblk_t get_sb_block(void **data)
1428 {
1429         ext4_fsblk_t    sb_block;
1430         char            *options = (char *) *data;
1431
1432         if (!options || strncmp(options, "sb=", 3) != 0)
1433                 return 1;       /* Default location */
1434
1435         options += 3;
1436         /* TODO: use simple_strtoll with >32bit ext4 */
1437         sb_block = simple_strtoul(options, &options, 0);
1438         if (*options && *options != ',') {
1439                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1440                        (char *) *data);
1441                 return 1;
1442         }
1443         if (*options == ',')
1444                 options++;
1445         *data = (void *) options;
1446
1447         return sb_block;
1448 }
1449
1450 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1451 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1452         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1453
1454 #ifdef CONFIG_QUOTA
1455 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1456 {
1457         struct ext4_sb_info *sbi = EXT4_SB(sb);
1458         char *qname;
1459
1460         if (sb_any_quota_loaded(sb) &&
1461                 !sbi->s_qf_names[qtype]) {
1462                 ext4_msg(sb, KERN_ERR,
1463                         "Cannot change journaled "
1464                         "quota options when quota turned on");
1465                 return 0;
1466         }
1467         qname = match_strdup(args);
1468         if (!qname) {
1469                 ext4_msg(sb, KERN_ERR,
1470                         "Not enough memory for storing quotafile name");
1471                 return 0;
1472         }
1473         if (sbi->s_qf_names[qtype] &&
1474                 strcmp(sbi->s_qf_names[qtype], qname)) {
1475                 ext4_msg(sb, KERN_ERR,
1476                         "%s quota file already specified", QTYPE2NAME(qtype));
1477                 kfree(qname);
1478                 return 0;
1479         }
1480         sbi->s_qf_names[qtype] = qname;
1481         if (strchr(sbi->s_qf_names[qtype], '/')) {
1482                 ext4_msg(sb, KERN_ERR,
1483                         "quotafile must be on filesystem root");
1484                 kfree(sbi->s_qf_names[qtype]);
1485                 sbi->s_qf_names[qtype] = NULL;
1486                 return 0;
1487         }
1488         set_opt(sb, QUOTA);
1489         return 1;
1490 }
1491
1492 static int clear_qf_name(struct super_block *sb, int qtype)
1493 {
1494
1495         struct ext4_sb_info *sbi = EXT4_SB(sb);
1496
1497         if (sb_any_quota_loaded(sb) &&
1498                 sbi->s_qf_names[qtype]) {
1499                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1500                         " when quota turned on");
1501                 return 0;
1502         }
1503         /*
1504          * The space will be released later when all options are confirmed
1505          * to be correct
1506          */
1507         sbi->s_qf_names[qtype] = NULL;
1508         return 1;
1509 }
1510 #endif
1511
1512 static int parse_options(char *options, struct super_block *sb,
1513                          unsigned long *journal_devnum,
1514                          unsigned int *journal_ioprio,
1515                          int is_remount)
1516 {
1517         struct ext4_sb_info *sbi = EXT4_SB(sb);
1518         char *p;
1519         substring_t args[MAX_OPT_ARGS];
1520         int data_opt = 0;
1521         int option;
1522 #ifdef CONFIG_QUOTA
1523         int qfmt;
1524 #endif
1525
1526         if (!options)
1527                 return 1;
1528
1529         while ((p = strsep(&options, ",")) != NULL) {
1530                 int token;
1531                 if (!*p)
1532                         continue;
1533
1534                 /*
1535                  * Initialize args struct so we know whether arg was
1536                  * found; some options take optional arguments.
1537                  */
1538                 args[0].to = args[0].from = NULL;
1539                 token = match_token(p, tokens, args);
1540                 switch (token) {
1541                 case Opt_bsd_df:
1542                         clear_opt(sb, MINIX_DF);
1543                         break;
1544                 case Opt_minix_df:
1545                         set_opt(sb, MINIX_DF);
1546                         break;
1547                 case Opt_grpid:
1548                         set_opt(sb, GRPID);
1549                         break;
1550                 case Opt_nogrpid:
1551                         clear_opt(sb, GRPID);
1552                         break;
1553                 case Opt_resuid:
1554                         if (match_int(&args[0], &option))
1555                                 return 0;
1556                         sbi->s_resuid = option;
1557                         break;
1558                 case Opt_resgid:
1559                         if (match_int(&args[0], &option))
1560                                 return 0;
1561                         sbi->s_resgid = option;
1562                         break;
1563                 case Opt_sb:
1564                         /* handled by get_sb_block() instead of here */
1565                         /* *sb_block = match_int(&args[0]); */
1566                         break;
1567                 case Opt_err_panic:
1568                         clear_opt(sb, ERRORS_CONT);
1569                         clear_opt(sb, ERRORS_RO);
1570                         set_opt(sb, ERRORS_PANIC);
1571                         break;
1572                 case Opt_err_ro:
1573                         clear_opt(sb, ERRORS_CONT);
1574                         clear_opt(sb, ERRORS_PANIC);
1575                         set_opt(sb, ERRORS_RO);
1576                         break;
1577                 case Opt_err_cont:
1578                         clear_opt(sb, ERRORS_RO);
1579                         clear_opt(sb, ERRORS_PANIC);
1580                         set_opt(sb, ERRORS_CONT);
1581                         break;
1582                 case Opt_nouid32:
1583                         set_opt(sb, NO_UID32);
1584                         break;
1585                 case Opt_debug:
1586                         set_opt(sb, DEBUG);
1587                         break;
1588                 case Opt_oldalloc:
1589                         ext4_msg(sb, KERN_WARNING,
1590                                  "Ignoring deprecated oldalloc option");
1591                         break;
1592                 case Opt_orlov:
1593                         ext4_msg(sb, KERN_WARNING,
1594                                  "Ignoring deprecated orlov option");
1595                         break;
1596 #ifdef CONFIG_EXT4_FS_XATTR
1597                 case Opt_user_xattr:
1598                         set_opt(sb, XATTR_USER);
1599                         break;
1600                 case Opt_nouser_xattr:
1601                         clear_opt(sb, XATTR_USER);
1602                         break;
1603 #else
1604                 case Opt_user_xattr:
1605                 case Opt_nouser_xattr:
1606                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1607                         break;
1608 #endif
1609 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1610                 case Opt_acl:
1611                         set_opt(sb, POSIX_ACL);
1612                         break;
1613                 case Opt_noacl:
1614                         clear_opt(sb, POSIX_ACL);
1615                         break;
1616 #else
1617                 case Opt_acl:
1618                 case Opt_noacl:
1619                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1620                         break;
1621 #endif
1622                 case Opt_journal_dev:
1623                         if (is_remount) {
1624                                 ext4_msg(sb, KERN_ERR,
1625                                         "Cannot specify journal on remount");
1626                                 return 0;
1627                         }
1628                         if (match_int(&args[0], &option))
1629                                 return 0;
1630                         *journal_devnum = option;
1631                         break;
1632                 case Opt_journal_checksum:
1633                         set_opt(sb, JOURNAL_CHECKSUM);
1634                         break;
1635                 case Opt_journal_async_commit:
1636                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1637                         set_opt(sb, JOURNAL_CHECKSUM);
1638                         break;
1639                 case Opt_noload:
1640                         set_opt(sb, NOLOAD);
1641                         break;
1642                 case Opt_commit:
1643                         if (match_int(&args[0], &option))
1644                                 return 0;
1645                         if (option < 0)
1646                                 return 0;
1647                         if (option == 0)
1648                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1649                         sbi->s_commit_interval = HZ * option;
1650                         break;
1651                 case Opt_max_batch_time:
1652                         if (match_int(&args[0], &option))
1653                                 return 0;
1654                         if (option < 0)
1655                                 return 0;
1656                         if (option == 0)
1657                                 option = EXT4_DEF_MAX_BATCH_TIME;
1658                         sbi->s_max_batch_time = option;
1659                         break;
1660                 case Opt_min_batch_time:
1661                         if (match_int(&args[0], &option))
1662                                 return 0;
1663                         if (option < 0)
1664                                 return 0;
1665                         sbi->s_min_batch_time = option;
1666                         break;
1667                 case Opt_data_journal:
1668                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1669                         goto datacheck;
1670                 case Opt_data_ordered:
1671                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1672                         goto datacheck;
1673                 case Opt_data_writeback:
1674                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1675                 datacheck:
1676                         if (is_remount) {
1677                                 if (!sbi->s_journal)
1678                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1679                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1680                                         ext4_msg(sb, KERN_ERR,
1681                                                 "Cannot change data mode on remount");
1682                                         return 0;
1683                                 }
1684                         } else {
1685                                 clear_opt(sb, DATA_FLAGS);
1686                                 sbi->s_mount_opt |= data_opt;
1687                         }
1688                         break;
1689                 case Opt_data_err_abort:
1690                         set_opt(sb, DATA_ERR_ABORT);
1691                         break;
1692                 case Opt_data_err_ignore:
1693                         clear_opt(sb, DATA_ERR_ABORT);
1694                         break;
1695 #ifdef CONFIG_QUOTA
1696                 case Opt_usrjquota:
1697                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1698                                 return 0;
1699                         break;
1700                 case Opt_grpjquota:
1701                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1702                                 return 0;
1703                         break;
1704                 case Opt_offusrjquota:
1705                         if (!clear_qf_name(sb, USRQUOTA))
1706                                 return 0;
1707                         break;
1708                 case Opt_offgrpjquota:
1709                         if (!clear_qf_name(sb, GRPQUOTA))
1710                                 return 0;
1711                         break;
1712
1713                 case Opt_jqfmt_vfsold:
1714                         qfmt = QFMT_VFS_OLD;
1715                         goto set_qf_format;
1716                 case Opt_jqfmt_vfsv0:
1717                         qfmt = QFMT_VFS_V0;
1718                         goto set_qf_format;
1719                 case Opt_jqfmt_vfsv1:
1720                         qfmt = QFMT_VFS_V1;
1721 set_qf_format:
1722                         if (sb_any_quota_loaded(sb) &&
1723                             sbi->s_jquota_fmt != qfmt) {
1724                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1725                                         "journaled quota options when "
1726                                         "quota turned on");
1727                                 return 0;
1728                         }
1729                         sbi->s_jquota_fmt = qfmt;
1730                         break;
1731                 case Opt_quota:
1732                 case Opt_usrquota:
1733                         set_opt(sb, QUOTA);
1734                         set_opt(sb, USRQUOTA);
1735                         break;
1736                 case Opt_grpquota:
1737                         set_opt(sb, QUOTA);
1738                         set_opt(sb, GRPQUOTA);
1739                         break;
1740                 case Opt_noquota:
1741                         if (sb_any_quota_loaded(sb)) {
1742                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1743                                         "options when quota turned on");
1744                                 return 0;
1745                         }
1746                         clear_opt(sb, QUOTA);
1747                         clear_opt(sb, USRQUOTA);
1748                         clear_opt(sb, GRPQUOTA);
1749                         break;
1750 #else
1751                 case Opt_quota:
1752                 case Opt_usrquota:
1753                 case Opt_grpquota:
1754                         ext4_msg(sb, KERN_ERR,
1755                                 "quota options not supported");
1756                         break;
1757                 case Opt_usrjquota:
1758                 case Opt_grpjquota:
1759                 case Opt_offusrjquota:
1760                 case Opt_offgrpjquota:
1761                 case Opt_jqfmt_vfsold:
1762                 case Opt_jqfmt_vfsv0:
1763                 case Opt_jqfmt_vfsv1:
1764                         ext4_msg(sb, KERN_ERR,
1765                                 "journaled quota options not supported");
1766                         break;
1767                 case Opt_noquota:
1768                         break;
1769 #endif
1770                 case Opt_abort:
1771                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1772                         break;
1773                 case Opt_nobarrier:
1774                         clear_opt(sb, BARRIER);
1775                         break;
1776                 case Opt_barrier:
1777                         if (args[0].from) {
1778                                 if (match_int(&args[0], &option))
1779                                         return 0;
1780                         } else
1781                                 option = 1;     /* No argument, default to 1 */
1782                         if (option)
1783                                 set_opt(sb, BARRIER);
1784                         else
1785                                 clear_opt(sb, BARRIER);
1786                         break;
1787                 case Opt_ignore:
1788                         break;
1789                 case Opt_nobh:
1790                         ext4_msg(sb, KERN_WARNING,
1791                                  "Ignoring deprecated nobh option");
1792                         break;
1793                 case Opt_bh:
1794                         ext4_msg(sb, KERN_WARNING,
1795                                  "Ignoring deprecated bh option");
1796                         break;
1797                 case Opt_i_version:
1798                         set_opt(sb, I_VERSION);
1799                         sb->s_flags |= MS_I_VERSION;
1800                         break;
1801                 case Opt_nodelalloc:
1802                         clear_opt(sb, DELALLOC);
1803                         clear_opt2(sb, EXPLICIT_DELALLOC);
1804                         break;
1805                 case Opt_mblk_io_submit:
1806                         set_opt(sb, MBLK_IO_SUBMIT);
1807                         break;
1808                 case Opt_nomblk_io_submit:
1809                         clear_opt(sb, MBLK_IO_SUBMIT);
1810                         break;
1811                 case Opt_stripe:
1812                         if (match_int(&args[0], &option))
1813                                 return 0;
1814                         if (option < 0)
1815                                 return 0;
1816                         sbi->s_stripe = option;
1817                         break;
1818                 case Opt_delalloc:
1819                         set_opt(sb, DELALLOC);
1820                         set_opt2(sb, EXPLICIT_DELALLOC);
1821                         break;
1822                 case Opt_block_validity:
1823                         set_opt(sb, BLOCK_VALIDITY);
1824                         break;
1825                 case Opt_noblock_validity:
1826                         clear_opt(sb, BLOCK_VALIDITY);
1827                         break;
1828                 case Opt_inode_readahead_blks:
1829                         if (match_int(&args[0], &option))
1830                                 return 0;
1831                         if (option < 0 || option > (1 << 30))
1832                                 return 0;
1833                         if (option && !is_power_of_2(option)) {
1834                                 ext4_msg(sb, KERN_ERR,
1835                                          "EXT4-fs: inode_readahead_blks"
1836                                          " must be a power of 2");
1837                                 return 0;
1838                         }
1839                         sbi->s_inode_readahead_blks = option;
1840                         break;
1841                 case Opt_journal_ioprio:
1842                         if (match_int(&args[0], &option))
1843                                 return 0;
1844                         if (option < 0 || option > 7)
1845                                 break;
1846                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1847                                                             option);
1848                         break;
1849                 case Opt_noauto_da_alloc:
1850                         set_opt(sb, NO_AUTO_DA_ALLOC);
1851                         break;
1852                 case Opt_auto_da_alloc:
1853                         if (args[0].from) {
1854                                 if (match_int(&args[0], &option))
1855                                         return 0;
1856                         } else
1857                                 option = 1;     /* No argument, default to 1 */
1858                         if (option)
1859                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1860                         else
1861                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1862                         break;
1863                 case Opt_discard:
1864                         set_opt(sb, DISCARD);
1865                         break;
1866                 case Opt_nodiscard:
1867                         clear_opt(sb, DISCARD);
1868                         break;
1869                 case Opt_dioread_nolock:
1870                         set_opt(sb, DIOREAD_NOLOCK);
1871                         break;
1872                 case Opt_dioread_lock:
1873                         clear_opt(sb, DIOREAD_NOLOCK);
1874                         break;
1875                 case Opt_init_itable:
1876                         set_opt(sb, INIT_INODE_TABLE);
1877                         if (args[0].from) {
1878                                 if (match_int(&args[0], &option))
1879                                         return 0;
1880                         } else
1881                                 option = EXT4_DEF_LI_WAIT_MULT;
1882                         if (option < 0)
1883                                 return 0;
1884                         sbi->s_li_wait_mult = option;
1885                         break;
1886                 case Opt_noinit_itable:
1887                         clear_opt(sb, INIT_INODE_TABLE);
1888                         break;
1889                 default:
1890                         ext4_msg(sb, KERN_ERR,
1891                                "Unrecognized mount option \"%s\" "
1892                                "or missing value", p);
1893                         return 0;
1894                 }
1895         }
1896 #ifdef CONFIG_QUOTA
1897         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1898                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1899                         clear_opt(sb, USRQUOTA);
1900
1901                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1902                         clear_opt(sb, GRPQUOTA);
1903
1904                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1905                         ext4_msg(sb, KERN_ERR, "old and new quota "
1906                                         "format mixing");
1907                         return 0;
1908                 }
1909
1910                 if (!sbi->s_jquota_fmt) {
1911                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1912                                         "not specified");
1913                         return 0;
1914                 }
1915         } else {
1916                 if (sbi->s_jquota_fmt) {
1917                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1918                                         "specified with no journaling "
1919                                         "enabled");
1920                         return 0;
1921                 }
1922         }
1923 #endif
1924         return 1;
1925 }
1926
1927 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1928                             int read_only)
1929 {
1930         struct ext4_sb_info *sbi = EXT4_SB(sb);
1931         int res = 0;
1932
1933         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1934                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1935                          "forcing read-only mode");
1936                 res = MS_RDONLY;
1937         }
1938         if (read_only)
1939                 goto done;
1940         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1941                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1942                          "running e2fsck is recommended");
1943         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1944                 ext4_msg(sb, KERN_WARNING,
1945                          "warning: mounting fs with errors, "
1946                          "running e2fsck is recommended");
1947         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1948                  le16_to_cpu(es->s_mnt_count) >=
1949                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1950                 ext4_msg(sb, KERN_WARNING,
1951                          "warning: maximal mount count reached, "
1952                          "running e2fsck is recommended");
1953         else if (le32_to_cpu(es->s_checkinterval) &&
1954                 (le32_to_cpu(es->s_lastcheck) +
1955                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1956                 ext4_msg(sb, KERN_WARNING,
1957                          "warning: checktime reached, "
1958                          "running e2fsck is recommended");
1959         if (!sbi->s_journal)
1960                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1961         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1962                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1963         le16_add_cpu(&es->s_mnt_count, 1);
1964         es->s_mtime = cpu_to_le32(get_seconds());
1965         ext4_update_dynamic_rev(sb);
1966         if (sbi->s_journal)
1967                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1968
1969         ext4_commit_super(sb, 1);
1970 done:
1971         if (test_opt(sb, DEBUG))
1972                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1973                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1974                         sb->s_blocksize,
1975                         sbi->s_groups_count,
1976                         EXT4_BLOCKS_PER_GROUP(sb),
1977                         EXT4_INODES_PER_GROUP(sb),
1978                         sbi->s_mount_opt, sbi->s_mount_opt2);
1979
1980         cleancache_init_fs(sb);
1981         return res;
1982 }
1983
1984 static int ext4_fill_flex_info(struct super_block *sb)
1985 {
1986         struct ext4_sb_info *sbi = EXT4_SB(sb);
1987         struct ext4_group_desc *gdp = NULL;
1988         ext4_group_t flex_group_count;
1989         ext4_group_t flex_group;
1990         unsigned int groups_per_flex = 0;
1991         size_t size;
1992         int i;
1993
1994         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1995         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1996                 sbi->s_log_groups_per_flex = 0;
1997                 return 1;
1998         }
1999         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
2000
2001         /* We allocate both existing and potentially added groups */
2002         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
2003                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
2004                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
2005         size = flex_group_count * sizeof(struct flex_groups);
2006         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2007         if (sbi->s_flex_groups == NULL) {
2008                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2009                          flex_group_count);
2010                 goto failed;
2011         }
2012
2013         for (i = 0; i < sbi->s_groups_count; i++) {
2014                 gdp = ext4_get_group_desc(sb, i, NULL);
2015
2016                 flex_group = ext4_flex_group(sbi, i);
2017                 atomic_add(ext4_free_inodes_count(sb, gdp),
2018                            &sbi->s_flex_groups[flex_group].free_inodes);
2019                 atomic_add(ext4_free_group_clusters(sb, gdp),
2020                            &sbi->s_flex_groups[flex_group].free_clusters);
2021                 atomic_add(ext4_used_dirs_count(sb, gdp),
2022                            &sbi->s_flex_groups[flex_group].used_dirs);
2023         }
2024
2025         return 1;
2026 failed:
2027         return 0;
2028 }
2029
2030 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2031                             struct ext4_group_desc *gdp)
2032 {
2033         __u16 crc = 0;
2034
2035         if (sbi->s_es->s_feature_ro_compat &
2036             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2037                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2038                 __le32 le_group = cpu_to_le32(block_group);
2039
2040                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2041                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2042                 crc = crc16(crc, (__u8 *)gdp, offset);
2043                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2044                 /* for checksum of struct ext4_group_desc do the rest...*/
2045                 if ((sbi->s_es->s_feature_incompat &
2046                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2047                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2048                         crc = crc16(crc, (__u8 *)gdp + offset,
2049                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2050                                         offset);
2051         }
2052
2053         return cpu_to_le16(crc);
2054 }
2055
2056 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2057                                 struct ext4_group_desc *gdp)
2058 {
2059         if ((sbi->s_es->s_feature_ro_compat &
2060              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2061             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2062                 return 0;
2063
2064         return 1;
2065 }
2066
2067 /* Called at mount-time, super-block is locked */
2068 static int ext4_check_descriptors(struct super_block *sb,
2069                                   ext4_group_t *first_not_zeroed)
2070 {
2071         struct ext4_sb_info *sbi = EXT4_SB(sb);
2072         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2073         ext4_fsblk_t last_block;
2074         ext4_fsblk_t block_bitmap;
2075         ext4_fsblk_t inode_bitmap;
2076         ext4_fsblk_t inode_table;
2077         int flexbg_flag = 0;
2078         ext4_group_t i, grp = sbi->s_groups_count;
2079
2080         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2081                 flexbg_flag = 1;
2082
2083         ext4_debug("Checking group descriptors");
2084
2085         for (i = 0; i < sbi->s_groups_count; i++) {
2086                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2087
2088                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2089                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2090                 else
2091                         last_block = first_block +
2092                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2093
2094                 if ((grp == sbi->s_groups_count) &&
2095                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2096                         grp = i;
2097
2098                 block_bitmap = ext4_block_bitmap(sb, gdp);
2099                 if (block_bitmap < first_block || block_bitmap > last_block) {
2100                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2101                                "Block bitmap for group %u not in group "
2102                                "(block %llu)!", i, block_bitmap);
2103                         return 0;
2104                 }
2105                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2106                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2107                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2108                                "Inode bitmap for group %u not in group "
2109                                "(block %llu)!", i, inode_bitmap);
2110                         return 0;
2111                 }
2112                 inode_table = ext4_inode_table(sb, gdp);
2113                 if (inode_table < first_block ||
2114                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2115                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2116                                "Inode table for group %u not in group "
2117                                "(block %llu)!", i, inode_table);
2118                         return 0;
2119                 }
2120                 ext4_lock_group(sb, i);
2121                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2122                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2123                                  "Checksum for group %u failed (%u!=%u)",
2124                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2125                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2126                         if (!(sb->s_flags & MS_RDONLY)) {
2127                                 ext4_unlock_group(sb, i);
2128                                 return 0;
2129                         }
2130                 }
2131                 ext4_unlock_group(sb, i);
2132                 if (!flexbg_flag)
2133                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2134         }
2135         if (NULL != first_not_zeroed)
2136                 *first_not_zeroed = grp;
2137
2138         ext4_free_blocks_count_set(sbi->s_es,
2139                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2140         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2141         return 1;
2142 }
2143
2144 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2145  * the superblock) which were deleted from all directories, but held open by
2146  * a process at the time of a crash.  We walk the list and try to delete these
2147  * inodes at recovery time (only with a read-write filesystem).
2148  *
2149  * In order to keep the orphan inode chain consistent during traversal (in
2150  * case of crash during recovery), we link each inode into the superblock
2151  * orphan list_head and handle it the same way as an inode deletion during
2152  * normal operation (which journals the operations for us).
2153  *
2154  * We only do an iget() and an iput() on each inode, which is very safe if we
2155  * accidentally point at an in-use or already deleted inode.  The worst that
2156  * can happen in this case is that we get a "bit already cleared" message from
2157  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2158  * e2fsck was run on this filesystem, and it must have already done the orphan
2159  * inode cleanup for us, so we can safely abort without any further action.
2160  */
2161 static void ext4_orphan_cleanup(struct super_block *sb,
2162                                 struct ext4_super_block *es)
2163 {
2164         unsigned int s_flags = sb->s_flags;
2165         int nr_orphans = 0, nr_truncates = 0;
2166 #ifdef CONFIG_QUOTA
2167         int i;
2168 #endif
2169         if (!es->s_last_orphan) {
2170                 jbd_debug(4, "no orphan inodes to clean up\n");
2171                 return;
2172         }
2173
2174         if (bdev_read_only(sb->s_bdev)) {
2175                 ext4_msg(sb, KERN_ERR, "write access "
2176                         "unavailable, skipping orphan cleanup");
2177                 return;
2178         }
2179
2180         /* Check if feature set would not allow a r/w mount */
2181         if (!ext4_feature_set_ok(sb, 0)) {
2182                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2183                          "unknown ROCOMPAT features");
2184                 return;
2185         }
2186
2187         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2188                 if (es->s_last_orphan)
2189                         jbd_debug(1, "Errors on filesystem, "
2190                                   "clearing orphan list.\n");
2191                 es->s_last_orphan = 0;
2192                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2193                 return;
2194         }
2195
2196         if (s_flags & MS_RDONLY) {
2197                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2198                 sb->s_flags &= ~MS_RDONLY;
2199         }
2200 #ifdef CONFIG_QUOTA
2201         /* Needed for iput() to work correctly and not trash data */
2202         sb->s_flags |= MS_ACTIVE;
2203         /* Turn on quotas so that they are updated correctly */
2204         for (i = 0; i < MAXQUOTAS; i++) {
2205                 if (EXT4_SB(sb)->s_qf_names[i]) {
2206                         int ret = ext4_quota_on_mount(sb, i);
2207                         if (ret < 0)
2208                                 ext4_msg(sb, KERN_ERR,
2209                                         "Cannot turn on journaled "
2210                                         "quota: error %d", ret);
2211                 }
2212         }
2213 #endif
2214
2215         while (es->s_last_orphan) {
2216                 struct inode *inode;
2217
2218                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2219                 if (IS_ERR(inode)) {
2220                         es->s_last_orphan = 0;
2221                         break;
2222                 }
2223
2224                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2225                 dquot_initialize(inode);
2226                 if (inode->i_nlink) {
2227                         ext4_msg(sb, KERN_DEBUG,
2228                                 "%s: truncating inode %lu to %lld bytes",
2229                                 __func__, inode->i_ino, inode->i_size);
2230                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2231                                   inode->i_ino, inode->i_size);
2232                         ext4_truncate(inode);
2233                         nr_truncates++;
2234                 } else {
2235                         ext4_msg(sb, KERN_DEBUG,
2236                                 "%s: deleting unreferenced inode %lu",
2237                                 __func__, inode->i_ino);
2238                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2239                                   inode->i_ino);
2240                         nr_orphans++;
2241                 }
2242                 iput(inode);  /* The delete magic happens here! */
2243         }
2244
2245 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2246
2247         if (nr_orphans)
2248                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2249                        PLURAL(nr_orphans));
2250         if (nr_truncates)
2251                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2252                        PLURAL(nr_truncates));
2253 #ifdef CONFIG_QUOTA
2254         /* Turn quotas off */
2255         for (i = 0; i < MAXQUOTAS; i++) {
2256                 if (sb_dqopt(sb)->files[i])
2257                         dquot_quota_off(sb, i);
2258         }
2259 #endif
2260         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2261 }
2262
2263 /*
2264  * Maximal extent format file size.
2265  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2266  * extent format containers, within a sector_t, and within i_blocks
2267  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2268  * so that won't be a limiting factor.
2269  *
2270  * However there is other limiting factor. We do store extents in the form
2271  * of starting block and length, hence the resulting length of the extent
2272  * covering maximum file size must fit into on-disk format containers as
2273  * well. Given that length is always by 1 unit bigger than max unit (because
2274  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2275  *
2276  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2277  */
2278 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2279 {
2280         loff_t res;
2281         loff_t upper_limit = MAX_LFS_FILESIZE;
2282
2283         /* small i_blocks in vfs inode? */
2284         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2285                 /*
2286                  * CONFIG_LBDAF is not enabled implies the inode
2287                  * i_block represent total blocks in 512 bytes
2288                  * 32 == size of vfs inode i_blocks * 8
2289                  */
2290                 upper_limit = (1LL << 32) - 1;
2291
2292                 /* total blocks in file system block size */
2293                 upper_limit >>= (blkbits - 9);
2294                 upper_limit <<= blkbits;
2295         }
2296
2297         /*
2298          * 32-bit extent-start container, ee_block. We lower the maxbytes
2299          * by one fs block, so ee_len can cover the extent of maximum file
2300          * size
2301          */
2302         res = (1LL << 32) - 1;
2303         res <<= blkbits;
2304
2305         /* Sanity check against vm- & vfs- imposed limits */
2306         if (res > upper_limit)
2307                 res = upper_limit;
2308
2309         return res;
2310 }
2311
2312 /*
2313  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2314  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2315  * We need to be 1 filesystem block less than the 2^48 sector limit.
2316  */
2317 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2318 {
2319         loff_t res = EXT4_NDIR_BLOCKS;
2320         int meta_blocks;
2321         loff_t upper_limit;
2322         /* This is calculated to be the largest file size for a dense, block
2323          * mapped file such that the file's total number of 512-byte sectors,
2324          * including data and all indirect blocks, does not exceed (2^48 - 1).
2325          *
2326          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2327          * number of 512-byte sectors of the file.
2328          */
2329
2330         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2331                 /*
2332                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2333                  * the inode i_block field represents total file blocks in
2334                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2335                  */
2336                 upper_limit = (1LL << 32) - 1;
2337
2338                 /* total blocks in file system block size */
2339                 upper_limit >>= (bits - 9);
2340
2341         } else {
2342                 /*
2343                  * We use 48 bit ext4_inode i_blocks
2344                  * With EXT4_HUGE_FILE_FL set the i_blocks
2345                  * represent total number of blocks in
2346                  * file system block size
2347                  */
2348                 upper_limit = (1LL << 48) - 1;
2349
2350         }
2351
2352         /* indirect blocks */
2353         meta_blocks = 1;
2354         /* double indirect blocks */
2355         meta_blocks += 1 + (1LL << (bits-2));
2356         /* tripple indirect blocks */
2357         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2358
2359         upper_limit -= meta_blocks;
2360         upper_limit <<= bits;
2361
2362         res += 1LL << (bits-2);
2363         res += 1LL << (2*(bits-2));
2364         res += 1LL << (3*(bits-2));
2365         res <<= bits;
2366         if (res > upper_limit)
2367                 res = upper_limit;
2368
2369         if (res > MAX_LFS_FILESIZE)
2370                 res = MAX_LFS_FILESIZE;
2371
2372         return res;
2373 }
2374
2375 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2376                                    ext4_fsblk_t logical_sb_block, int nr)
2377 {
2378         struct ext4_sb_info *sbi = EXT4_SB(sb);
2379         ext4_group_t bg, first_meta_bg;
2380         int has_super = 0;
2381
2382         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2383
2384         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2385             nr < first_meta_bg)
2386                 return logical_sb_block + nr + 1;
2387         bg = sbi->s_desc_per_block * nr;
2388         if (ext4_bg_has_super(sb, bg))
2389                 has_super = 1;
2390
2391         return (has_super + ext4_group_first_block_no(sb, bg));
2392 }
2393
2394 /**
2395  * ext4_get_stripe_size: Get the stripe size.
2396  * @sbi: In memory super block info
2397  *
2398  * If we have specified it via mount option, then
2399  * use the mount option value. If the value specified at mount time is
2400  * greater than the blocks per group use the super block value.
2401  * If the super block value is greater than blocks per group return 0.
2402  * Allocator needs it be less than blocks per group.
2403  *
2404  */
2405 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2406 {
2407         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2408         unsigned long stripe_width =
2409                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2410         int ret;
2411
2412         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2413                 ret = sbi->s_stripe;
2414         else if (stripe_width <= sbi->s_blocks_per_group)
2415                 ret = stripe_width;
2416         else if (stride <= sbi->s_blocks_per_group)
2417                 ret = stride;
2418         else
2419                 ret = 0;
2420
2421         /*
2422          * If the stripe width is 1, this makes no sense and
2423          * we set it to 0 to turn off stripe handling code.
2424          */
2425         if (ret <= 1)
2426                 ret = 0;
2427
2428         return ret;
2429 }
2430
2431 /* sysfs supprt */
2432
2433 struct ext4_attr {
2434         struct attribute attr;
2435         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2436         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2437                          const char *, size_t);
2438         int offset;
2439 };
2440
2441 static int parse_strtoul(const char *buf,
2442                 unsigned long max, unsigned long *value)
2443 {
2444         char *endp;
2445
2446         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2447         endp = skip_spaces(endp);
2448         if (*endp || *value > max)
2449                 return -EINVAL;
2450
2451         return 0;
2452 }
2453
2454 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2455                                               struct ext4_sb_info *sbi,
2456                                               char *buf)
2457 {
2458         return snprintf(buf, PAGE_SIZE, "%llu\n",
2459                 (s64) EXT4_C2B(sbi,
2460                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2461 }
2462
2463 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2464                                          struct ext4_sb_info *sbi, char *buf)
2465 {
2466         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2467
2468         if (!sb->s_bdev->bd_part)
2469                 return snprintf(buf, PAGE_SIZE, "0\n");
2470         return snprintf(buf, PAGE_SIZE, "%lu\n",
2471                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2472                          sbi->s_sectors_written_start) >> 1);
2473 }
2474
2475 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2476                                           struct ext4_sb_info *sbi, char *buf)
2477 {
2478         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2479
2480         if (!sb->s_bdev->bd_part)
2481                 return snprintf(buf, PAGE_SIZE, "0\n");
2482         return snprintf(buf, PAGE_SIZE, "%llu\n",
2483                         (unsigned long long)(sbi->s_kbytes_written +
2484                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2485                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2486 }
2487
2488 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2489                                       struct ext4_sb_info *sbi, char *buf)
2490 {
2491         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2492 }
2493
2494 static ssize_t extent_cache_misses_show(struct ext4_attr *a,
2495                                         struct ext4_sb_info *sbi, char *buf)
2496 {
2497         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_misses);
2498 }
2499
2500 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2501                                           struct ext4_sb_info *sbi,
2502                                           const char *buf, size_t count)
2503 {
2504         unsigned long t;
2505
2506         if (parse_strtoul(buf, 0x40000000, &t))
2507                 return -EINVAL;
2508
2509         if (t && !is_power_of_2(t))
2510                 return -EINVAL;
2511
2512         sbi->s_inode_readahead_blks = t;
2513         return count;
2514 }
2515
2516 static ssize_t sbi_ui_show(struct ext4_attr *a,
2517                            struct ext4_sb_info *sbi, char *buf)
2518 {
2519         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2520
2521         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2522 }
2523
2524 static ssize_t sbi_ui_store(struct ext4_attr *a,
2525                             struct ext4_sb_info *sbi,
2526                             const char *buf, size_t count)
2527 {
2528         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2529         unsigned long t;
2530
2531         if (parse_strtoul(buf, 0xffffffff, &t))
2532                 return -EINVAL;
2533         *ui = t;
2534         return count;
2535 }
2536
2537 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2538 static struct ext4_attr ext4_attr_##_name = {                   \
2539         .attr = {.name = __stringify(_name), .mode = _mode },   \
2540         .show   = _show,                                        \
2541         .store  = _store,                                       \
2542         .offset = offsetof(struct ext4_sb_info, _elname),       \
2543 }
2544 #define EXT4_ATTR(name, mode, show, store) \
2545 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2546
2547 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2548 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2549 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2550 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2551         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2552 #define ATTR_LIST(name) &ext4_attr_##name.attr
2553
2554 EXT4_RO_ATTR(delayed_allocation_blocks);
2555 EXT4_RO_ATTR(session_write_kbytes);
2556 EXT4_RO_ATTR(lifetime_write_kbytes);
2557 EXT4_RO_ATTR(extent_cache_hits);
2558 EXT4_RO_ATTR(extent_cache_misses);
2559 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2560                  inode_readahead_blks_store, s_inode_readahead_blks);
2561 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2562 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2563 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2564 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2565 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2566 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2567 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2568 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2569
2570 static struct attribute *ext4_attrs[] = {
2571         ATTR_LIST(delayed_allocation_blocks),
2572         ATTR_LIST(session_write_kbytes),
2573         ATTR_LIST(lifetime_write_kbytes),
2574         ATTR_LIST(extent_cache_hits),
2575         ATTR_LIST(extent_cache_misses),
2576         ATTR_LIST(inode_readahead_blks),
2577         ATTR_LIST(inode_goal),
2578         ATTR_LIST(mb_stats),
2579         ATTR_LIST(mb_max_to_scan),
2580         ATTR_LIST(mb_min_to_scan),
2581         ATTR_LIST(mb_order2_req),
2582         ATTR_LIST(mb_stream_req),
2583         ATTR_LIST(mb_group_prealloc),
2584         ATTR_LIST(max_writeback_mb_bump),
2585         NULL,
2586 };
2587
2588 /* Features this copy of ext4 supports */
2589 EXT4_INFO_ATTR(lazy_itable_init);
2590 EXT4_INFO_ATTR(batched_discard);
2591
2592 static struct attribute *ext4_feat_attrs[] = {
2593         ATTR_LIST(lazy_itable_init),
2594         ATTR_LIST(batched_discard),
2595         NULL,
2596 };
2597
2598 static ssize_t ext4_attr_show(struct kobject *kobj,
2599                               struct attribute *attr, char *buf)
2600 {
2601         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2602                                                 s_kobj);
2603         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2604
2605         return a->show ? a->show(a, sbi, buf) : 0;
2606 }
2607
2608 static ssize_t ext4_attr_store(struct kobject *kobj,
2609                                struct attribute *attr,
2610                                const char *buf, size_t len)
2611 {
2612         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2613                                                 s_kobj);
2614         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2615
2616         return a->store ? a->store(a, sbi, buf, len) : 0;
2617 }
2618
2619 static void ext4_sb_release(struct kobject *kobj)
2620 {
2621         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2622                                                 s_kobj);
2623         complete(&sbi->s_kobj_unregister);
2624 }
2625
2626 static const struct sysfs_ops ext4_attr_ops = {
2627         .show   = ext4_attr_show,
2628         .store  = ext4_attr_store,
2629 };
2630
2631 static struct kobj_type ext4_ktype = {
2632         .default_attrs  = ext4_attrs,
2633         .sysfs_ops      = &ext4_attr_ops,
2634         .release        = ext4_sb_release,
2635 };
2636
2637 static void ext4_feat_release(struct kobject *kobj)
2638 {
2639         complete(&ext4_feat->f_kobj_unregister);
2640 }
2641
2642 static struct kobj_type ext4_feat_ktype = {
2643         .default_attrs  = ext4_feat_attrs,
2644         .sysfs_ops      = &ext4_attr_ops,
2645         .release        = ext4_feat_release,
2646 };
2647
2648 /*
2649  * Check whether this filesystem can be mounted based on
2650  * the features present and the RDONLY/RDWR mount requested.
2651  * Returns 1 if this filesystem can be mounted as requested,
2652  * 0 if it cannot be.
2653  */
2654 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2655 {
2656         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2657                 ext4_msg(sb, KERN_ERR,
2658                         "Couldn't mount because of "
2659                         "unsupported optional features (%x)",
2660                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2661                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2662                 return 0;
2663         }
2664
2665         if (readonly)
2666                 return 1;
2667
2668         /* Check that feature set is OK for a read-write mount */
2669         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2670                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2671                          "unsupported optional features (%x)",
2672                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2673                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2674                 return 0;
2675         }
2676         /*
2677          * Large file size enabled file system can only be mounted
2678          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2679          */
2680         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2681                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2682                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2683                                  "cannot be mounted RDWR without "
2684                                  "CONFIG_LBDAF");
2685                         return 0;
2686                 }
2687         }
2688         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2689             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2690                 ext4_msg(sb, KERN_ERR,
2691                          "Can't support bigalloc feature without "
2692                          "extents feature\n");
2693                 return 0;
2694         }
2695         return 1;
2696 }
2697
2698 /*
2699  * This function is called once a day if we have errors logged
2700  * on the file system
2701  */
2702 static void print_daily_error_info(unsigned long arg)
2703 {
2704         struct super_block *sb = (struct super_block *) arg;
2705         struct ext4_sb_info *sbi;
2706         struct ext4_super_block *es;
2707
2708         sbi = EXT4_SB(sb);
2709         es = sbi->s_es;
2710
2711         if (es->s_error_count)
2712                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2713                          le32_to_cpu(es->s_error_count));
2714         if (es->s_first_error_time) {
2715                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2716                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2717                        (int) sizeof(es->s_first_error_func),
2718                        es->s_first_error_func,
2719                        le32_to_cpu(es->s_first_error_line));
2720                 if (es->s_first_error_ino)
2721                         printk(": inode %u",
2722                                le32_to_cpu(es->s_first_error_ino));
2723                 if (es->s_first_error_block)
2724                         printk(": block %llu", (unsigned long long)
2725                                le64_to_cpu(es->s_first_error_block));
2726                 printk("\n");
2727         }
2728         if (es->s_last_error_time) {
2729                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2730                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2731                        (int) sizeof(es->s_last_error_func),
2732                        es->s_last_error_func,
2733                        le32_to_cpu(es->s_last_error_line));
2734                 if (es->s_last_error_ino)
2735                         printk(": inode %u",
2736                                le32_to_cpu(es->s_last_error_ino));
2737                 if (es->s_last_error_block)
2738                         printk(": block %llu", (unsigned long long)
2739                                le64_to_cpu(es->s_last_error_block));
2740                 printk("\n");
2741         }
2742         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2743 }
2744
2745 /* Find next suitable group and run ext4_init_inode_table */
2746 static int ext4_run_li_request(struct ext4_li_request *elr)
2747 {
2748         struct ext4_group_desc *gdp = NULL;
2749         ext4_group_t group, ngroups;
2750         struct super_block *sb;
2751         unsigned long timeout = 0;
2752         int ret = 0;
2753
2754         sb = elr->lr_super;
2755         ngroups = EXT4_SB(sb)->s_groups_count;
2756
2757         for (group = elr->lr_next_group; group < ngroups; group++) {
2758                 gdp = ext4_get_group_desc(sb, group, NULL);
2759                 if (!gdp) {
2760                         ret = 1;
2761                         break;
2762                 }
2763
2764                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2765                         break;
2766         }
2767
2768         if (group == ngroups)
2769                 ret = 1;
2770
2771         if (!ret) {
2772                 timeout = jiffies;
2773                 ret = ext4_init_inode_table(sb, group,
2774                                             elr->lr_timeout ? 0 : 1);
2775                 if (elr->lr_timeout == 0) {
2776                         timeout = (jiffies - timeout) *
2777                                   elr->lr_sbi->s_li_wait_mult;
2778                         elr->lr_timeout = timeout;
2779                 }
2780                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2781                 elr->lr_next_group = group + 1;
2782         }
2783
2784         return ret;
2785 }
2786
2787 /*
2788  * Remove lr_request from the list_request and free the
2789  * request structure. Should be called with li_list_mtx held
2790  */
2791 static void ext4_remove_li_request(struct ext4_li_request *elr)
2792 {
2793         struct ext4_sb_info *sbi;
2794
2795         if (!elr)
2796                 return;
2797
2798         sbi = elr->lr_sbi;
2799
2800         list_del(&elr->lr_request);
2801         sbi->s_li_request = NULL;
2802         kfree(elr);
2803 }
2804
2805 static void ext4_unregister_li_request(struct super_block *sb)
2806 {
2807         mutex_lock(&ext4_li_mtx);
2808         if (!ext4_li_info) {
2809                 mutex_unlock(&ext4_li_mtx);
2810                 return;
2811         }
2812
2813         mutex_lock(&ext4_li_info->li_list_mtx);
2814         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2815         mutex_unlock(&ext4_li_info->li_list_mtx);
2816         mutex_unlock(&ext4_li_mtx);
2817 }
2818
2819 static struct task_struct *ext4_lazyinit_task;
2820
2821 /*
2822  * This is the function where ext4lazyinit thread lives. It walks
2823  * through the request list searching for next scheduled filesystem.
2824  * When such a fs is found, run the lazy initialization request
2825  * (ext4_rn_li_request) and keep track of the time spend in this
2826  * function. Based on that time we compute next schedule time of
2827  * the request. When walking through the list is complete, compute
2828  * next waking time and put itself into sleep.
2829  */
2830 static int ext4_lazyinit_thread(void *arg)
2831 {
2832         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2833         struct list_head *pos, *n;
2834         struct ext4_li_request *elr;
2835         unsigned long next_wakeup, cur;
2836
2837         BUG_ON(NULL == eli);
2838
2839 cont_thread:
2840         while (true) {
2841                 next_wakeup = MAX_JIFFY_OFFSET;
2842
2843                 mutex_lock(&eli->li_list_mtx);
2844                 if (list_empty(&eli->li_request_list)) {
2845                         mutex_unlock(&eli->li_list_mtx);
2846                         goto exit_thread;
2847                 }
2848
2849                 list_for_each_safe(pos, n, &eli->li_request_list) {
2850                         elr = list_entry(pos, struct ext4_li_request,
2851                                          lr_request);
2852
2853                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2854                                 if (ext4_run_li_request(elr) != 0) {
2855                                         /* error, remove the lazy_init job */
2856                                         ext4_remove_li_request(elr);
2857                                         continue;
2858                                 }
2859                         }
2860
2861                         if (time_before(elr->lr_next_sched, next_wakeup))
2862                                 next_wakeup = elr->lr_next_sched;
2863                 }
2864                 mutex_unlock(&eli->li_list_mtx);
2865
2866                 try_to_freeze();
2867
2868                 cur = jiffies;
2869                 if ((time_after_eq(cur, next_wakeup)) ||
2870                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2871                         cond_resched();
2872                         continue;
2873                 }
2874
2875                 schedule_timeout_interruptible(next_wakeup - cur);
2876
2877                 if (kthread_should_stop()) {
2878                         ext4_clear_request_list();
2879                         goto exit_thread;
2880                 }
2881         }
2882
2883 exit_thread:
2884         /*
2885          * It looks like the request list is empty, but we need
2886          * to check it under the li_list_mtx lock, to prevent any
2887          * additions into it, and of course we should lock ext4_li_mtx
2888          * to atomically free the list and ext4_li_info, because at
2889          * this point another ext4 filesystem could be registering
2890          * new one.
2891          */
2892         mutex_lock(&ext4_li_mtx);
2893         mutex_lock(&eli->li_list_mtx);
2894         if (!list_empty(&eli->li_request_list)) {
2895                 mutex_unlock(&eli->li_list_mtx);
2896                 mutex_unlock(&ext4_li_mtx);
2897                 goto cont_thread;
2898         }
2899         mutex_unlock(&eli->li_list_mtx);
2900         kfree(ext4_li_info);
2901         ext4_li_info = NULL;
2902         mutex_unlock(&ext4_li_mtx);
2903
2904         return 0;
2905 }
2906
2907 static void ext4_clear_request_list(void)
2908 {
2909         struct list_head *pos, *n;
2910         struct ext4_li_request *elr;
2911
2912         mutex_lock(&ext4_li_info->li_list_mtx);
2913         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2914                 elr = list_entry(pos, struct ext4_li_request,
2915                                  lr_request);
2916                 ext4_remove_li_request(elr);
2917         }
2918         mutex_unlock(&ext4_li_info->li_list_mtx);
2919 }
2920
2921 static int ext4_run_lazyinit_thread(void)
2922 {
2923         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2924                                          ext4_li_info, "ext4lazyinit");
2925         if (IS_ERR(ext4_lazyinit_task)) {
2926                 int err = PTR_ERR(ext4_lazyinit_task);
2927                 ext4_clear_request_list();
2928                 kfree(ext4_li_info);
2929                 ext4_li_info = NULL;
2930                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2931                                  "initialization thread\n",
2932                                  err);
2933                 return err;
2934         }
2935         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2936         return 0;
2937 }
2938
2939 /*
2940  * Check whether it make sense to run itable init. thread or not.
2941  * If there is at least one uninitialized inode table, return
2942  * corresponding group number, else the loop goes through all
2943  * groups and return total number of groups.
2944  */
2945 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2946 {
2947         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2948         struct ext4_group_desc *gdp = NULL;
2949
2950         for (group = 0; group < ngroups; group++) {
2951                 gdp = ext4_get_group_desc(sb, group, NULL);
2952                 if (!gdp)
2953                         continue;
2954
2955                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2956                         break;
2957         }
2958
2959         return group;
2960 }
2961
2962 static int ext4_li_info_new(void)
2963 {
2964         struct ext4_lazy_init *eli = NULL;
2965
2966         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2967         if (!eli)
2968                 return -ENOMEM;
2969
2970         INIT_LIST_HEAD(&eli->li_request_list);
2971         mutex_init(&eli->li_list_mtx);
2972
2973         eli->li_state |= EXT4_LAZYINIT_QUIT;
2974
2975         ext4_li_info = eli;
2976
2977         return 0;
2978 }
2979
2980 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2981                                             ext4_group_t start)
2982 {
2983         struct ext4_sb_info *sbi = EXT4_SB(sb);
2984         struct ext4_li_request *elr;
2985         unsigned long rnd;
2986
2987         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2988         if (!elr)
2989                 return NULL;
2990
2991         elr->lr_super = sb;
2992         elr->lr_sbi = sbi;
2993         elr->lr_next_group = start;
2994
2995         /*
2996          * Randomize first schedule time of the request to
2997          * spread the inode table initialization requests
2998          * better.
2999          */
3000         get_random_bytes(&rnd, sizeof(rnd));
3001         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3002                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3003
3004         return elr;
3005 }
3006
3007 static int ext4_register_li_request(struct super_block *sb,
3008                                     ext4_group_t first_not_zeroed)
3009 {
3010         struct ext4_sb_info *sbi = EXT4_SB(sb);
3011         struct ext4_li_request *elr;
3012         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3013         int ret = 0;
3014
3015         if (sbi->s_li_request != NULL) {
3016                 /*
3017                  * Reset timeout so it can be computed again, because
3018                  * s_li_wait_mult might have changed.
3019                  */
3020                 sbi->s_li_request->lr_timeout = 0;
3021                 return 0;
3022         }
3023
3024         if (first_not_zeroed == ngroups ||
3025             (sb->s_flags & MS_RDONLY) ||
3026             !test_opt(sb, INIT_INODE_TABLE))
3027                 return 0;
3028
3029         elr = ext4_li_request_new(sb, first_not_zeroed);
3030         if (!elr)
3031                 return -ENOMEM;
3032
3033         mutex_lock(&ext4_li_mtx);
3034
3035         if (NULL == ext4_li_info) {
3036                 ret = ext4_li_info_new();
3037                 if (ret)
3038                         goto out;
3039         }
3040
3041         mutex_lock(&ext4_li_info->li_list_mtx);
3042         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3043         mutex_unlock(&ext4_li_info->li_list_mtx);
3044
3045         sbi->s_li_request = elr;
3046         /*
3047          * set elr to NULL here since it has been inserted to
3048          * the request_list and the removal and free of it is
3049          * handled by ext4_clear_request_list from now on.
3050          */
3051         elr = NULL;
3052
3053         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3054                 ret = ext4_run_lazyinit_thread();
3055                 if (ret)
3056                         goto out;
3057         }
3058 out:
3059         mutex_unlock(&ext4_li_mtx);
3060         if (ret)
3061                 kfree(elr);
3062         return ret;
3063 }
3064
3065 /*
3066  * We do not need to lock anything since this is called on
3067  * module unload.
3068  */
3069 static void ext4_destroy_lazyinit_thread(void)
3070 {
3071         /*
3072          * If thread exited earlier
3073          * there's nothing to be done.
3074          */
3075         if (!ext4_li_info || !ext4_lazyinit_task)
3076                 return;
3077
3078         kthread_stop(ext4_lazyinit_task);
3079 }
3080
3081 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3082 {
3083         char *orig_data = kstrdup(data, GFP_KERNEL);
3084         struct buffer_head *bh;
3085         struct ext4_super_block *es = NULL;
3086         struct ext4_sb_info *sbi;
3087         ext4_fsblk_t block;
3088         ext4_fsblk_t sb_block = get_sb_block(&data);
3089         ext4_fsblk_t logical_sb_block;
3090         unsigned long offset = 0;
3091         unsigned long journal_devnum = 0;
3092         unsigned long def_mount_opts;
3093         struct inode *root;
3094         char *cp;
3095         const char *descr;
3096         int ret = -ENOMEM;
3097         int blocksize, clustersize;
3098         unsigned int db_count;
3099         unsigned int i;
3100         int needs_recovery, has_huge_files, has_bigalloc;
3101         __u64 blocks_count;
3102         int err;
3103         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3104         ext4_group_t first_not_zeroed;
3105
3106         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3107         if (!sbi)
3108                 goto out_free_orig;
3109
3110         sbi->s_blockgroup_lock =
3111                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3112         if (!sbi->s_blockgroup_lock) {
3113                 kfree(sbi);
3114                 goto out_free_orig;
3115         }
3116         sb->s_fs_info = sbi;
3117         sbi->s_mount_opt = 0;
3118         sbi->s_resuid = EXT4_DEF_RESUID;
3119         sbi->s_resgid = EXT4_DEF_RESGID;
3120         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3121         sbi->s_sb_block = sb_block;
3122         if (sb->s_bdev->bd_part)
3123                 sbi->s_sectors_written_start =
3124                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3125
3126         /* Cleanup superblock name */
3127         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3128                 *cp = '!';
3129
3130         ret = -EINVAL;
3131         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3132         if (!blocksize) {
3133                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3134                 goto out_fail;
3135         }
3136
3137         /*
3138          * The ext4 superblock will not be buffer aligned for other than 1kB
3139          * block sizes.  We need to calculate the offset from buffer start.
3140          */
3141         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3142                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3143                 offset = do_div(logical_sb_block, blocksize);
3144         } else {
3145                 logical_sb_block = sb_block;
3146         }
3147
3148         if (!(bh = sb_bread(sb, logical_sb_block))) {
3149                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3150                 goto out_fail;
3151         }
3152         /*
3153          * Note: s_es must be initialized as soon as possible because
3154          *       some ext4 macro-instructions depend on its value
3155          */
3156         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3157         sbi->s_es = es;
3158         sb->s_magic = le16_to_cpu(es->s_magic);
3159         if (sb->s_magic != EXT4_SUPER_MAGIC)
3160                 goto cantfind_ext4;
3161         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3162
3163         /* Set defaults before we parse the mount options */
3164         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3165         set_opt(sb, INIT_INODE_TABLE);
3166         if (def_mount_opts & EXT4_DEFM_DEBUG)
3167                 set_opt(sb, DEBUG);
3168         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3169                 set_opt(sb, GRPID);
3170         if (def_mount_opts & EXT4_DEFM_UID16)
3171                 set_opt(sb, NO_UID32);
3172         /* xattr user namespace & acls are now defaulted on */
3173 #ifdef CONFIG_EXT4_FS_XATTR
3174         set_opt(sb, XATTR_USER);
3175 #endif
3176 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3177         set_opt(sb, POSIX_ACL);
3178 #endif
3179         set_opt(sb, MBLK_IO_SUBMIT);
3180         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3181                 set_opt(sb, JOURNAL_DATA);
3182         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3183                 set_opt(sb, ORDERED_DATA);
3184         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3185                 set_opt(sb, WRITEBACK_DATA);
3186
3187         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3188                 set_opt(sb, ERRORS_PANIC);
3189         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3190                 set_opt(sb, ERRORS_CONT);
3191         else
3192                 set_opt(sb, ERRORS_RO);
3193         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3194                 set_opt(sb, BLOCK_VALIDITY);
3195         if (def_mount_opts & EXT4_DEFM_DISCARD)
3196                 set_opt(sb, DISCARD);
3197
3198         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3199         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3200         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3201         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3202         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3203
3204         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3205                 set_opt(sb, BARRIER);
3206
3207         /*
3208          * enable delayed allocation by default
3209          * Use -o nodelalloc to turn it off
3210          */
3211         if (!IS_EXT3_SB(sb) &&
3212             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3213                 set_opt(sb, DELALLOC);
3214
3215         /*
3216          * set default s_li_wait_mult for lazyinit, for the case there is
3217          * no mount option specified.
3218          */
3219         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3220
3221         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3222                            &journal_devnum, &journal_ioprio, 0)) {
3223                 ext4_msg(sb, KERN_WARNING,
3224                          "failed to parse options in superblock: %s",
3225                          sbi->s_es->s_mount_opts);
3226         }
3227         if (!parse_options((char *) data, sb, &journal_devnum,
3228                            &journal_ioprio, 0))
3229                 goto failed_mount;
3230
3231         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3232                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3233                             "with data=journal disables delayed "
3234                             "allocation and O_DIRECT support!\n");
3235                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3236                         ext4_msg(sb, KERN_ERR, "can't mount with "
3237                                  "both data=journal and delalloc");
3238                         goto failed_mount;
3239                 }
3240                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3241                         ext4_msg(sb, KERN_ERR, "can't mount with "
3242                                  "both data=journal and delalloc");
3243                         goto failed_mount;
3244                 }
3245                 if (test_opt(sb, DELALLOC))
3246                         clear_opt(sb, DELALLOC);
3247         }
3248
3249         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3250         if (test_opt(sb, DIOREAD_NOLOCK)) {
3251                 if (blocksize < PAGE_SIZE) {
3252                         ext4_msg(sb, KERN_ERR, "can't mount with "
3253                                  "dioread_nolock if block size != PAGE_SIZE");
3254                         goto failed_mount;
3255                 }
3256         }
3257
3258         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3259                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3260
3261         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3262             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3263              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3264              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3265                 ext4_msg(sb, KERN_WARNING,
3266                        "feature flags set on rev 0 fs, "
3267                        "running e2fsck is recommended");
3268
3269         if (IS_EXT2_SB(sb)) {
3270                 if (ext2_feature_set_ok(sb))
3271                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3272                                  "using the ext4 subsystem");
3273                 else {
3274                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3275                                  "to feature incompatibilities");
3276                         goto failed_mount;
3277                 }
3278         }
3279
3280         if (IS_EXT3_SB(sb)) {
3281                 if (ext3_feature_set_ok(sb))
3282                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3283                                  "using the ext4 subsystem");
3284                 else {
3285                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3286                                  "to feature incompatibilities");
3287                         goto failed_mount;
3288                 }
3289         }
3290
3291         /*
3292          * Check feature flags regardless of the revision level, since we
3293          * previously didn't change the revision level when setting the flags,
3294          * so there is a chance incompat flags are set on a rev 0 filesystem.
3295          */
3296         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3297                 goto failed_mount;
3298
3299         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3300             blocksize > EXT4_MAX_BLOCK_SIZE) {
3301                 ext4_msg(sb, KERN_ERR,
3302                        "Unsupported filesystem blocksize %d", blocksize);
3303                 goto failed_mount;
3304         }
3305
3306         if (sb->s_blocksize != blocksize) {
3307                 /* Validate the filesystem blocksize */
3308                 if (!sb_set_blocksize(sb, blocksize)) {
3309                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3310                                         blocksize);
3311                         goto failed_mount;
3312                 }
3313
3314                 brelse(bh);
3315                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3316                 offset = do_div(logical_sb_block, blocksize);
3317                 bh = sb_bread(sb, logical_sb_block);
3318                 if (!bh) {
3319                         ext4_msg(sb, KERN_ERR,
3320                                "Can't read superblock on 2nd try");
3321                         goto failed_mount;
3322                 }
3323                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3324                 sbi->s_es = es;
3325                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3326                         ext4_msg(sb, KERN_ERR,
3327                                "Magic mismatch, very weird!");
3328                         goto failed_mount;
3329                 }
3330         }
3331
3332         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3333                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3334         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3335                                                       has_huge_files);
3336         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3337
3338         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3339                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3340                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3341         } else {
3342                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3343                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3344                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3345                     (!is_power_of_2(sbi->s_inode_size)) ||
3346                     (sbi->s_inode_size > blocksize)) {
3347                         ext4_msg(sb, KERN_ERR,
3348                                "unsupported inode size: %d",
3349                                sbi->s_inode_size);
3350                         goto failed_mount;
3351                 }
3352                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3353                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3354         }
3355
3356         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3357         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3358                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3359                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3360                     !is_power_of_2(sbi->s_desc_size)) {
3361                         ext4_msg(sb, KERN_ERR,
3362                                "unsupported descriptor size %lu",
3363                                sbi->s_desc_size);
3364                         goto failed_mount;
3365                 }
3366         } else
3367                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3368
3369         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3370         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3371         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3372                 goto cantfind_ext4;
3373
3374         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3375         if (sbi->s_inodes_per_block == 0)
3376                 goto cantfind_ext4;
3377         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3378                                         sbi->s_inodes_per_block;
3379         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3380         sbi->s_sbh = bh;
3381         sbi->s_mount_state = le16_to_cpu(es->s_state);
3382         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3383         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3384
3385         for (i = 0; i < 4; i++)
3386                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3387         sbi->s_def_hash_version = es->s_def_hash_version;
3388         i = le32_to_cpu(es->s_flags);
3389         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3390                 sbi->s_hash_unsigned = 3;
3391         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3392 #ifdef __CHAR_UNSIGNED__
3393                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3394                 sbi->s_hash_unsigned = 3;
3395 #else
3396                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3397 #endif
3398                 sb->s_dirt = 1;
3399         }
3400
3401         /* Handle clustersize */
3402         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3403         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3404                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3405         if (has_bigalloc) {
3406                 if (clustersize < blocksize) {
3407                         ext4_msg(sb, KERN_ERR,
3408                                  "cluster size (%d) smaller than "
3409                                  "block size (%d)", clustersize, blocksize);
3410                         goto failed_mount;
3411                 }
3412                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3413                         le32_to_cpu(es->s_log_block_size);
3414                 sbi->s_clusters_per_group =
3415                         le32_to_cpu(es->s_clusters_per_group);
3416                 if (sbi->s_clusters_per_group > blocksize * 8) {
3417                         ext4_msg(sb, KERN_ERR,
3418                                  "#clusters per group too big: %lu",
3419                                  sbi->s_clusters_per_group);
3420                         goto failed_mount;
3421                 }
3422                 if (sbi->s_blocks_per_group !=
3423                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3424                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3425                                  "clusters per group (%lu) inconsistent",
3426                                  sbi->s_blocks_per_group,
3427                                  sbi->s_clusters_per_group);
3428                         goto failed_mount;
3429                 }
3430         } else {
3431                 if (clustersize != blocksize) {
3432                         ext4_warning(sb, "fragment/cluster size (%d) != "
3433                                      "block size (%d)", clustersize,
3434                                      blocksize);
3435                         clustersize = blocksize;
3436                 }
3437                 if (sbi->s_blocks_per_group > blocksize * 8) {
3438                         ext4_msg(sb, KERN_ERR,
3439                                  "#blocks per group too big: %lu",
3440                                  sbi->s_blocks_per_group);
3441                         goto failed_mount;
3442                 }
3443                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3444                 sbi->s_cluster_bits = 0;
3445         }
3446         sbi->s_cluster_ratio = clustersize / blocksize;
3447
3448         if (sbi->s_inodes_per_group > blocksize * 8) {
3449                 ext4_msg(sb, KERN_ERR,
3450                        "#inodes per group too big: %lu",
3451                        sbi->s_inodes_per_group);
3452                 goto failed_mount;
3453         }
3454
3455         /*
3456          * Test whether we have more sectors than will fit in sector_t,
3457          * and whether the max offset is addressable by the page cache.
3458          */
3459         err = generic_check_addressable(sb->s_blocksize_bits,
3460                                         ext4_blocks_count(es));
3461         if (err) {
3462                 ext4_msg(sb, KERN_ERR, "filesystem"
3463                          " too large to mount safely on this system");
3464                 if (sizeof(sector_t) < 8)
3465                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3466                 ret = err;
3467                 goto failed_mount;
3468         }
3469
3470         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3471                 goto cantfind_ext4;
3472
3473         /* check blocks count against device size */
3474         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3475         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3476                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3477                        "exceeds size of device (%llu blocks)",
3478                        ext4_blocks_count(es), blocks_count);
3479                 goto failed_mount;
3480         }
3481
3482         /*
3483          * It makes no sense for the first data block to be beyond the end
3484          * of the filesystem.
3485          */
3486         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3487                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3488                          "block %u is beyond end of filesystem (%llu)",
3489                          le32_to_cpu(es->s_first_data_block),
3490                          ext4_blocks_count(es));
3491                 goto failed_mount;
3492         }
3493         blocks_count = (ext4_blocks_count(es) -
3494                         le32_to_cpu(es->s_first_data_block) +
3495                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3496         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3497         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3498                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3499                        "(block count %llu, first data block %u, "
3500                        "blocks per group %lu)", sbi->s_groups_count,
3501                        ext4_blocks_count(es),
3502                        le32_to_cpu(es->s_first_data_block),
3503                        EXT4_BLOCKS_PER_GROUP(sb));
3504                 goto failed_mount;
3505         }
3506         sbi->s_groups_count = blocks_count;
3507         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3508                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3509         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3510                    EXT4_DESC_PER_BLOCK(sb);
3511         sbi->s_group_desc = ext4_kvmalloc(db_count *
3512                                           sizeof(struct buffer_head *),
3513                                           GFP_KERNEL);
3514         if (sbi->s_group_desc == NULL) {
3515                 ext4_msg(sb, KERN_ERR, "not enough memory");
3516                 goto failed_mount;
3517         }
3518
3519         if (ext4_proc_root)
3520                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3521
3522         bgl_lock_init(sbi->s_blockgroup_lock);
3523
3524         for (i = 0; i < db_count; i++) {
3525                 block = descriptor_loc(sb, logical_sb_block, i);
3526                 sbi->s_group_desc[i] = sb_bread(sb, block);
3527                 if (!sbi->s_group_desc[i]) {
3528                         ext4_msg(sb, KERN_ERR,
3529                                "can't read group descriptor %d", i);
3530                         db_count = i;
3531                         goto failed_mount2;
3532                 }
3533         }
3534         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3535                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3536                 goto failed_mount2;
3537         }
3538         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3539                 if (!ext4_fill_flex_info(sb)) {
3540                         ext4_msg(sb, KERN_ERR,
3541                                "unable to initialize "
3542                                "flex_bg meta info!");
3543                         goto failed_mount2;
3544                 }
3545
3546         sbi->s_gdb_count = db_count;
3547         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3548         spin_lock_init(&sbi->s_next_gen_lock);
3549
3550         init_timer(&sbi->s_err_report);
3551         sbi->s_err_report.function = print_daily_error_info;
3552         sbi->s_err_report.data = (unsigned long) sb;
3553
3554         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3555                         ext4_count_free_clusters(sb));
3556         if (!err) {
3557                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3558                                 ext4_count_free_inodes(sb));
3559         }
3560         if (!err) {
3561                 err = percpu_counter_init(&sbi->s_dirs_counter,
3562                                 ext4_count_dirs(sb));
3563         }
3564         if (!err) {
3565                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3566         }
3567         if (err) {
3568                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3569                 goto failed_mount3;
3570         }
3571
3572         sbi->s_stripe = ext4_get_stripe_size(sbi);
3573         sbi->s_max_writeback_mb_bump = 128;
3574
3575         /*
3576          * set up enough so that it can read an inode
3577          */
3578         if (!test_opt(sb, NOLOAD) &&
3579             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3580                 sb->s_op = &ext4_sops;
3581         else
3582                 sb->s_op = &ext4_nojournal_sops;
3583         sb->s_export_op = &ext4_export_ops;
3584         sb->s_xattr = ext4_xattr_handlers;
3585 #ifdef CONFIG_QUOTA
3586         sb->s_qcop = &ext4_qctl_operations;
3587         sb->dq_op = &ext4_quota_operations;
3588 #endif
3589         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3590
3591         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3592         mutex_init(&sbi->s_orphan_lock);
3593         sbi->s_resize_flags = 0;
3594
3595         sb->s_root = NULL;
3596
3597         needs_recovery = (es->s_last_orphan != 0 ||
3598                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3599                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3600
3601         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3602             !(sb->s_flags & MS_RDONLY))
3603                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3604                         goto failed_mount3;
3605
3606         /*
3607          * The first inode we look at is the journal inode.  Don't try
3608          * root first: it may be modified in the journal!
3609          */
3610         if (!test_opt(sb, NOLOAD) &&
3611             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3612                 if (ext4_load_journal(sb, es, journal_devnum))
3613                         goto failed_mount3;
3614         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3615               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3616                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3617                        "suppressed and not mounted read-only");
3618                 goto failed_mount_wq;
3619         } else {
3620                 clear_opt(sb, DATA_FLAGS);
3621                 sbi->s_journal = NULL;
3622                 needs_recovery = 0;
3623                 goto no_journal;
3624         }
3625
3626         if (ext4_blocks_count(es) > 0xffffffffULL &&
3627             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3628                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3629                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3630                 goto failed_mount_wq;
3631         }
3632
3633         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3634                 jbd2_journal_set_features(sbi->s_journal,
3635                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3636                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3637         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3638                 jbd2_journal_set_features(sbi->s_journal,
3639                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3640                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3641                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3642         } else {
3643                 jbd2_journal_clear_features(sbi->s_journal,
3644                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3645                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3646         }
3647
3648         /* We have now updated the journal if required, so we can
3649          * validate the data journaling mode. */
3650         switch (test_opt(sb, DATA_FLAGS)) {
3651         case 0:
3652                 /* No mode set, assume a default based on the journal
3653                  * capabilities: ORDERED_DATA if the journal can
3654                  * cope, else JOURNAL_DATA
3655                  */
3656                 if (jbd2_journal_check_available_features
3657                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3658                         set_opt(sb, ORDERED_DATA);
3659                 else
3660                         set_opt(sb, JOURNAL_DATA);
3661                 break;
3662
3663         case EXT4_MOUNT_ORDERED_DATA:
3664         case EXT4_MOUNT_WRITEBACK_DATA:
3665                 if (!jbd2_journal_check_available_features
3666                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3667                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3668                                "requested data journaling mode");
3669                         goto failed_mount_wq;
3670                 }
3671         default:
3672                 break;
3673         }
3674         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3675
3676         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3677
3678         /*
3679          * The journal may have updated the bg summary counts, so we
3680          * need to update the global counters.
3681          */
3682         percpu_counter_set(&sbi->s_freeclusters_counter,
3683                            ext4_count_free_clusters(sb));
3684         percpu_counter_set(&sbi->s_freeinodes_counter,
3685                            ext4_count_free_inodes(sb));
3686         percpu_counter_set(&sbi->s_dirs_counter,
3687                            ext4_count_dirs(sb));
3688         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3689
3690 no_journal:
3691         /*
3692          * The maximum number of concurrent works can be high and
3693          * concurrency isn't really necessary.  Limit it to 1.
3694          */
3695         EXT4_SB(sb)->dio_unwritten_wq =
3696                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3697         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3698                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3699                 goto failed_mount_wq;
3700         }
3701
3702         /*
3703          * The jbd2_journal_load will have done any necessary log recovery,
3704          * so we can safely mount the rest of the filesystem now.
3705          */
3706
3707         root = ext4_iget(sb, EXT4_ROOT_INO);
3708         if (IS_ERR(root)) {
3709                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3710                 ret = PTR_ERR(root);
3711                 root = NULL;
3712                 goto failed_mount4;
3713         }
3714         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3715                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3716                 iput(root);
3717                 goto failed_mount4;
3718         }
3719         sb->s_root = d_alloc_root(root);
3720         if (!sb->s_root) {
3721                 iput(root);
3722                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3723                 ret = -ENOMEM;
3724                 goto failed_mount4;
3725         }
3726
3727         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3728
3729         /* determine the minimum size of new large inodes, if present */
3730         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3731                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3732                                                      EXT4_GOOD_OLD_INODE_SIZE;
3733                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3734                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3735                         if (sbi->s_want_extra_isize <
3736                             le16_to_cpu(es->s_want_extra_isize))
3737                                 sbi->s_want_extra_isize =
3738                                         le16_to_cpu(es->s_want_extra_isize);
3739                         if (sbi->s_want_extra_isize <
3740                             le16_to_cpu(es->s_min_extra_isize))
3741                                 sbi->s_want_extra_isize =
3742                                         le16_to_cpu(es->s_min_extra_isize);
3743                 }
3744         }
3745         /* Check if enough inode space is available */
3746         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3747                                                         sbi->s_inode_size) {
3748                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3749                                                        EXT4_GOOD_OLD_INODE_SIZE;
3750                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3751                          "available");
3752         }
3753
3754         err = ext4_setup_system_zone(sb);
3755         if (err) {
3756                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3757                          "zone (%d)", err);
3758                 goto failed_mount4a;
3759         }
3760
3761         ext4_ext_init(sb);
3762         err = ext4_mb_init(sb, needs_recovery);
3763         if (err) {
3764                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3765                          err);
3766                 goto failed_mount5;
3767         }
3768
3769         err = ext4_register_li_request(sb, first_not_zeroed);
3770         if (err)
3771                 goto failed_mount6;
3772
3773         sbi->s_kobj.kset = ext4_kset;
3774         init_completion(&sbi->s_kobj_unregister);
3775         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3776                                    "%s", sb->s_id);
3777         if (err)
3778                 goto failed_mount7;
3779
3780         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3781         ext4_orphan_cleanup(sb, es);
3782         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3783         if (needs_recovery) {
3784                 ext4_msg(sb, KERN_INFO, "recovery complete");
3785                 ext4_mark_recovery_complete(sb, es);
3786         }
3787         if (EXT4_SB(sb)->s_journal) {
3788                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3789                         descr = " journalled data mode";
3790                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3791                         descr = " ordered data mode";
3792                 else
3793                         descr = " writeback data mode";
3794         } else
3795                 descr = "out journal";
3796
3797         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3798                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3799                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3800
3801         if (es->s_error_count)
3802                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3803
3804         kfree(orig_data);
3805         return 0;
3806
3807 cantfind_ext4:
3808         if (!silent)
3809                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3810         goto failed_mount;
3811
3812 failed_mount7:
3813         ext4_unregister_li_request(sb);
3814 failed_mount6:
3815         ext4_mb_release(sb);
3816 failed_mount5:
3817         ext4_ext_release(sb);
3818         ext4_release_system_zone(sb);
3819 failed_mount4a:
3820         dput(sb->s_root);
3821         sb->s_root = NULL;
3822 failed_mount4:
3823         ext4_msg(sb, KERN_ERR, "mount failed");
3824         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3825 failed_mount_wq:
3826         if (sbi->s_journal) {
3827                 jbd2_journal_destroy(sbi->s_journal);
3828                 sbi->s_journal = NULL;
3829         }
3830 failed_mount3:
3831         del_timer(&sbi->s_err_report);
3832         if (sbi->s_flex_groups)
3833                 ext4_kvfree(sbi->s_flex_groups);
3834         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3835         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3836         percpu_counter_destroy(&sbi->s_dirs_counter);
3837         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3838         if (sbi->s_mmp_tsk)
3839                 kthread_stop(sbi->s_mmp_tsk);
3840 failed_mount2:
3841         for (i = 0; i < db_count; i++)
3842                 brelse(sbi->s_group_desc[i]);
3843         ext4_kvfree(sbi->s_group_desc);
3844 failed_mount:
3845         if (sbi->s_proc) {
3846                 remove_proc_entry(sb->s_id, ext4_proc_root);
3847         }
3848 #ifdef CONFIG_QUOTA
3849         for (i = 0; i < MAXQUOTAS; i++)
3850                 kfree(sbi->s_qf_names[i]);
3851 #endif
3852         ext4_blkdev_remove(sbi);
3853         brelse(bh);
3854 out_fail:
3855         sb->s_fs_info = NULL;
3856         kfree(sbi->s_blockgroup_lock);
3857         kfree(sbi);
3858 out_free_orig:
3859         kfree(orig_data);
3860         return ret;
3861 }
3862
3863 /*
3864  * Setup any per-fs journal parameters now.  We'll do this both on
3865  * initial mount, once the journal has been initialised but before we've
3866  * done any recovery; and again on any subsequent remount.
3867  */
3868 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3869 {
3870         struct ext4_sb_info *sbi = EXT4_SB(sb);
3871
3872         journal->j_commit_interval = sbi->s_commit_interval;
3873         journal->j_min_batch_time = sbi->s_min_batch_time;
3874         journal->j_max_batch_time = sbi->s_max_batch_time;
3875
3876         write_lock(&journal->j_state_lock);
3877         if (test_opt(sb, BARRIER))
3878                 journal->j_flags |= JBD2_BARRIER;
3879         else
3880                 journal->j_flags &= ~JBD2_BARRIER;
3881         if (test_opt(sb, DATA_ERR_ABORT))
3882                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3883         else
3884                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3885         write_unlock(&journal->j_state_lock);
3886 }
3887
3888 static journal_t *ext4_get_journal(struct super_block *sb,
3889                                    unsigned int journal_inum)
3890 {
3891         struct inode *journal_inode;
3892         journal_t *journal;
3893
3894         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3895
3896         /* First, test for the existence of a valid inode on disk.  Bad
3897          * things happen if we iget() an unused inode, as the subsequent
3898          * iput() will try to delete it. */
3899
3900         journal_inode = ext4_iget(sb, journal_inum);
3901         if (IS_ERR(journal_inode)) {
3902                 ext4_msg(sb, KERN_ERR, "no journal found");
3903                 return NULL;
3904         }
3905         if (!journal_inode->i_nlink) {
3906                 make_bad_inode(journal_inode);
3907                 iput(journal_inode);
3908                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3909                 return NULL;
3910         }
3911
3912         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3913                   journal_inode, journal_inode->i_size);
3914         if (!S_ISREG(journal_inode->i_mode)) {
3915                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3916                 iput(journal_inode);
3917                 return NULL;
3918         }
3919
3920         journal = jbd2_journal_init_inode(journal_inode);
3921         if (!journal) {
3922                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3923                 iput(journal_inode);
3924                 return NULL;
3925         }
3926         journal->j_private = sb;
3927         ext4_init_journal_params(sb, journal);
3928         return journal;
3929 }
3930
3931 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3932                                        dev_t j_dev)
3933 {
3934         struct buffer_head *bh;
3935         journal_t *journal;
3936         ext4_fsblk_t start;
3937         ext4_fsblk_t len;
3938         int hblock, blocksize;
3939         ext4_fsblk_t sb_block;
3940         unsigned long offset;
3941         struct ext4_super_block *es;
3942         struct block_device *bdev;
3943
3944         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3945
3946         bdev = ext4_blkdev_get(j_dev, sb);
3947         if (bdev == NULL)
3948                 return NULL;
3949
3950         blocksize = sb->s_blocksize;
3951         hblock = bdev_logical_block_size(bdev);
3952         if (blocksize < hblock) {
3953                 ext4_msg(sb, KERN_ERR,
3954                         "blocksize too small for journal device");
3955                 goto out_bdev;
3956         }
3957
3958         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3959         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3960         set_blocksize(bdev, blocksize);
3961         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3962                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3963                        "external journal");
3964                 goto out_bdev;
3965         }
3966
3967         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3968         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3969             !(le32_to_cpu(es->s_feature_incompat) &
3970               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3971                 ext4_msg(sb, KERN_ERR, "external journal has "
3972                                         "bad superblock");
3973                 brelse(bh);
3974                 goto out_bdev;
3975         }
3976
3977         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3978                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3979                 brelse(bh);
3980                 goto out_bdev;
3981         }
3982
3983         len = ext4_blocks_count(es);
3984         start = sb_block + 1;
3985         brelse(bh);     /* we're done with the superblock */
3986
3987         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3988                                         start, len, blocksize);
3989         if (!journal) {
3990                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3991                 goto out_bdev;
3992         }
3993         journal->j_private = sb;
3994         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3995         wait_on_buffer(journal->j_sb_buffer);
3996         if (!buffer_uptodate(journal->j_sb_buffer)) {
3997                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3998                 goto out_journal;
3999         }
4000         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4001                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4002                                         "user (unsupported) - %d",
4003                         be32_to_cpu(journal->j_superblock->s_nr_users));
4004                 goto out_journal;
4005         }
4006         EXT4_SB(sb)->journal_bdev = bdev;
4007         ext4_init_journal_params(sb, journal);
4008         return journal;
4009
4010 out_journal:
4011         jbd2_journal_destroy(journal);
4012 out_bdev:
4013         ext4_blkdev_put(bdev);
4014         return NULL;
4015 }
4016
4017 static int ext4_load_journal(struct super_block *sb,
4018                              struct ext4_super_block *es,
4019                              unsigned long journal_devnum)
4020 {
4021         journal_t *journal;
4022         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4023         dev_t journal_dev;
4024         int err = 0;
4025         int really_read_only;
4026
4027         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4028
4029         if (journal_devnum &&
4030             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4031                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4032                         "numbers have changed");
4033                 journal_dev = new_decode_dev(journal_devnum);
4034         } else
4035                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4036
4037         really_read_only = bdev_read_only(sb->s_bdev);
4038
4039         /*
4040          * Are we loading a blank journal or performing recovery after a
4041          * crash?  For recovery, we need to check in advance whether we
4042          * can get read-write access to the device.
4043          */
4044         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4045                 if (sb->s_flags & MS_RDONLY) {
4046                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4047                                         "required on readonly filesystem");
4048                         if (really_read_only) {
4049                                 ext4_msg(sb, KERN_ERR, "write access "
4050                                         "unavailable, cannot proceed");
4051                                 return -EROFS;
4052                         }
4053                         ext4_msg(sb, KERN_INFO, "write access will "
4054                                "be enabled during recovery");
4055                 }
4056         }
4057
4058         if (journal_inum && journal_dev) {
4059                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4060                        "and inode journals!");
4061                 return -EINVAL;
4062         }
4063
4064         if (journal_inum) {
4065                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4066                         return -EINVAL;
4067         } else {
4068                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4069                         return -EINVAL;
4070         }
4071
4072         if (!(journal->j_flags & JBD2_BARRIER))
4073                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4074
4075         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4076                 err = jbd2_journal_wipe(journal, !really_read_only);
4077         if (!err) {
4078                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4079                 if (save)
4080                         memcpy(save, ((char *) es) +
4081                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4082                 err = jbd2_journal_load(journal);
4083                 if (save)
4084                         memcpy(((char *) es) + EXT4_S_ERR_START,
4085                                save, EXT4_S_ERR_LEN);
4086                 kfree(save);
4087         }
4088
4089         if (err) {
4090                 ext4_msg(sb, KERN_ERR, "error loading journal");
4091                 jbd2_journal_destroy(journal);
4092                 return err;
4093         }
4094
4095         EXT4_SB(sb)->s_journal = journal;
4096         ext4_clear_journal_err(sb, es);
4097
4098         if (!really_read_only && journal_devnum &&
4099             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4100                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4101
4102                 /* Make sure we flush the recovery flag to disk. */
4103                 ext4_commit_super(sb, 1);
4104         }
4105
4106         return 0;
4107 }
4108
4109 static int ext4_commit_super(struct super_block *sb, int sync)
4110 {
4111         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4112         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4113         int error = 0;
4114
4115         if (!sbh || block_device_ejected(sb))
4116                 return error;
4117         if (buffer_write_io_error(sbh)) {
4118                 /*
4119                  * Oh, dear.  A previous attempt to write the
4120                  * superblock failed.  This could happen because the
4121                  * USB device was yanked out.  Or it could happen to
4122                  * be a transient write error and maybe the block will
4123                  * be remapped.  Nothing we can do but to retry the
4124                  * write and hope for the best.
4125                  */
4126                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4127                        "superblock detected");
4128                 clear_buffer_write_io_error(sbh);
4129                 set_buffer_uptodate(sbh);
4130         }
4131         /*
4132          * If the file system is mounted read-only, don't update the
4133          * superblock write time.  This avoids updating the superblock
4134          * write time when we are mounting the root file system
4135          * read/only but we need to replay the journal; at that point,
4136          * for people who are east of GMT and who make their clock
4137          * tick in localtime for Windows bug-for-bug compatibility,
4138          * the clock is set in the future, and this will cause e2fsck
4139          * to complain and force a full file system check.
4140          */
4141         if (!(sb->s_flags & MS_RDONLY))
4142                 es->s_wtime = cpu_to_le32(get_seconds());
4143         if (sb->s_bdev->bd_part)
4144                 es->s_kbytes_written =
4145                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4146                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4147                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4148         else
4149                 es->s_kbytes_written =
4150                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4151         ext4_free_blocks_count_set(es,
4152                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4153                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4154         es->s_free_inodes_count =
4155                 cpu_to_le32(percpu_counter_sum_positive(
4156                                 &EXT4_SB(sb)->s_freeinodes_counter));
4157         sb->s_dirt = 0;
4158         BUFFER_TRACE(sbh, "marking dirty");
4159         mark_buffer_dirty(sbh);
4160         if (sync) {
4161                 error = sync_dirty_buffer(sbh);
4162                 if (error)
4163                         return error;
4164
4165                 error = buffer_write_io_error(sbh);
4166                 if (error) {
4167                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4168                                "superblock");
4169                         clear_buffer_write_io_error(sbh);
4170                         set_buffer_uptodate(sbh);
4171                 }
4172         }
4173         return error;
4174 }
4175
4176 /*
4177  * Have we just finished recovery?  If so, and if we are mounting (or
4178  * remounting) the filesystem readonly, then we will end up with a
4179  * consistent fs on disk.  Record that fact.
4180  */
4181 static void ext4_mark_recovery_complete(struct super_block *sb,
4182                                         struct ext4_super_block *es)
4183 {
4184         journal_t *journal = EXT4_SB(sb)->s_journal;
4185
4186         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4187                 BUG_ON(journal != NULL);
4188                 return;
4189         }
4190         jbd2_journal_lock_updates(journal);
4191         if (jbd2_journal_flush(journal) < 0)
4192                 goto out;
4193
4194         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4195             sb->s_flags & MS_RDONLY) {
4196                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4197                 ext4_commit_super(sb, 1);
4198         }
4199
4200 out:
4201         jbd2_journal_unlock_updates(journal);
4202 }
4203
4204 /*
4205  * If we are mounting (or read-write remounting) a filesystem whose journal
4206  * has recorded an error from a previous lifetime, move that error to the
4207  * main filesystem now.
4208  */
4209 static void ext4_clear_journal_err(struct super_block *sb,
4210                                    struct ext4_super_block *es)
4211 {
4212         journal_t *journal;
4213         int j_errno;
4214         const char *errstr;
4215
4216         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4217
4218         journal = EXT4_SB(sb)->s_journal;
4219
4220         /*
4221          * Now check for any error status which may have been recorded in the
4222          * journal by a prior ext4_error() or ext4_abort()
4223          */
4224
4225         j_errno = jbd2_journal_errno(journal);
4226         if (j_errno) {
4227                 char nbuf[16];
4228
4229                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4230                 ext4_warning(sb, "Filesystem error recorded "
4231                              "from previous mount: %s", errstr);
4232                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4233
4234                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4235                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4236                 ext4_commit_super(sb, 1);
4237
4238                 jbd2_journal_clear_err(journal);
4239         }
4240 }
4241
4242 /*
4243  * Force the running and committing transactions to commit,
4244  * and wait on the commit.
4245  */
4246 int ext4_force_commit(struct super_block *sb)
4247 {
4248         journal_t *journal;
4249         int ret = 0;
4250
4251         if (sb->s_flags & MS_RDONLY)
4252                 return 0;
4253
4254         journal = EXT4_SB(sb)->s_journal;
4255         if (journal) {
4256                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4257                 ret = ext4_journal_force_commit(journal);
4258         }
4259
4260         return ret;
4261 }
4262
4263 static void ext4_write_super(struct super_block *sb)
4264 {
4265         lock_super(sb);
4266         ext4_commit_super(sb, 1);
4267         unlock_super(sb);
4268 }
4269
4270 static int ext4_sync_fs(struct super_block *sb, int wait)
4271 {
4272         int ret = 0;
4273         tid_t target;
4274         struct ext4_sb_info *sbi = EXT4_SB(sb);
4275
4276         trace_ext4_sync_fs(sb, wait);
4277         flush_workqueue(sbi->dio_unwritten_wq);
4278         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4279                 if (wait)
4280                         jbd2_log_wait_commit(sbi->s_journal, target);
4281         }
4282         return ret;
4283 }
4284
4285 /*
4286  * LVM calls this function before a (read-only) snapshot is created.  This
4287  * gives us a chance to flush the journal completely and mark the fs clean.
4288  *
4289  * Note that only this function cannot bring a filesystem to be in a clean
4290  * state independently, because ext4 prevents a new handle from being started
4291  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4292  * the upper layer.
4293  */
4294 static int ext4_freeze(struct super_block *sb)
4295 {
4296         int error = 0;
4297         journal_t *journal;
4298
4299         if (sb->s_flags & MS_RDONLY)
4300                 return 0;
4301
4302         journal = EXT4_SB(sb)->s_journal;
4303
4304         /* Now we set up the journal barrier. */
4305         jbd2_journal_lock_updates(journal);
4306
4307         /*
4308          * Don't clear the needs_recovery flag if we failed to flush
4309          * the journal.
4310          */
4311         error = jbd2_journal_flush(journal);
4312         if (error < 0)
4313                 goto out;
4314
4315         /* Journal blocked and flushed, clear needs_recovery flag. */
4316         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4317         error = ext4_commit_super(sb, 1);
4318 out:
4319         /* we rely on s_frozen to stop further updates */
4320         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4321         return error;
4322 }
4323
4324 /*
4325  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4326  * flag here, even though the filesystem is not technically dirty yet.
4327  */
4328 static int ext4_unfreeze(struct super_block *sb)
4329 {
4330         if (sb->s_flags & MS_RDONLY)
4331                 return 0;
4332
4333         lock_super(sb);
4334         /* Reset the needs_recovery flag before the fs is unlocked. */
4335         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4336         ext4_commit_super(sb, 1);
4337         unlock_super(sb);
4338         return 0;
4339 }
4340
4341 /*
4342  * Structure to save mount options for ext4_remount's benefit
4343  */
4344 struct ext4_mount_options {
4345         unsigned long s_mount_opt;
4346         unsigned long s_mount_opt2;
4347         uid_t s_resuid;
4348         gid_t s_resgid;
4349         unsigned long s_commit_interval;
4350         u32 s_min_batch_time, s_max_batch_time;
4351 #ifdef CONFIG_QUOTA
4352         int s_jquota_fmt;
4353         char *s_qf_names[MAXQUOTAS];
4354 #endif
4355 };
4356
4357 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4358 {
4359         struct ext4_super_block *es;
4360         struct ext4_sb_info *sbi = EXT4_SB(sb);
4361         unsigned long old_sb_flags;
4362         struct ext4_mount_options old_opts;
4363         int enable_quota = 0;
4364         ext4_group_t g;
4365         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4366         int err = 0;
4367 #ifdef CONFIG_QUOTA
4368         int i;
4369 #endif
4370         char *orig_data = kstrdup(data, GFP_KERNEL);
4371
4372         /* Store the original options */
4373         lock_super(sb);
4374         old_sb_flags = sb->s_flags;
4375         old_opts.s_mount_opt = sbi->s_mount_opt;
4376         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4377         old_opts.s_resuid = sbi->s_resuid;
4378         old_opts.s_resgid = sbi->s_resgid;
4379         old_opts.s_commit_interval = sbi->s_commit_interval;
4380         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4381         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4382 #ifdef CONFIG_QUOTA
4383         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4384         for (i = 0; i < MAXQUOTAS; i++)
4385                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4386 #endif
4387         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4388                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4389
4390         /*
4391          * Allow the "check" option to be passed as a remount option.
4392          */
4393         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4394                 err = -EINVAL;
4395                 goto restore_opts;
4396         }
4397
4398         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4399                 ext4_abort(sb, "Abort forced by user");
4400
4401         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4402                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4403
4404         es = sbi->s_es;
4405
4406         if (sbi->s_journal) {
4407                 ext4_init_journal_params(sb, sbi->s_journal);
4408                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4409         }
4410
4411         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4412                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4413                         err = -EROFS;
4414                         goto restore_opts;
4415                 }
4416
4417                 if (*flags & MS_RDONLY) {
4418                         err = dquot_suspend(sb, -1);
4419                         if (err < 0)
4420                                 goto restore_opts;
4421
4422                         /*
4423                          * First of all, the unconditional stuff we have to do
4424                          * to disable replay of the journal when we next remount
4425                          */
4426                         sb->s_flags |= MS_RDONLY;
4427
4428                         /*
4429                          * OK, test if we are remounting a valid rw partition
4430                          * readonly, and if so set the rdonly flag and then
4431                          * mark the partition as valid again.
4432                          */
4433                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4434                             (sbi->s_mount_state & EXT4_VALID_FS))
4435                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4436
4437                         if (sbi->s_journal)
4438                                 ext4_mark_recovery_complete(sb, es);
4439                 } else {
4440                         /* Make sure we can mount this feature set readwrite */
4441                         if (!ext4_feature_set_ok(sb, 0)) {
4442                                 err = -EROFS;
4443                                 goto restore_opts;
4444                         }
4445                         /*
4446                          * Make sure the group descriptor checksums
4447                          * are sane.  If they aren't, refuse to remount r/w.
4448                          */
4449                         for (g = 0; g < sbi->s_groups_count; g++) {
4450                                 struct ext4_group_desc *gdp =
4451                                         ext4_get_group_desc(sb, g, NULL);
4452
4453                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4454                                         ext4_msg(sb, KERN_ERR,
4455                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4456                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4457                                                le16_to_cpu(gdp->bg_checksum));
4458                                         err = -EINVAL;
4459                                         goto restore_opts;
4460                                 }
4461                         }
4462
4463                         /*
4464                          * If we have an unprocessed orphan list hanging
4465                          * around from a previously readonly bdev mount,
4466                          * require a full umount/remount for now.
4467                          */
4468                         if (es->s_last_orphan) {
4469                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4470                                        "remount RDWR because of unprocessed "
4471                                        "orphan inode list.  Please "
4472                                        "umount/remount instead");
4473                                 err = -EINVAL;
4474                                 goto restore_opts;
4475                         }
4476
4477                         /*
4478                          * Mounting a RDONLY partition read-write, so reread
4479                          * and store the current valid flag.  (It may have
4480                          * been changed by e2fsck since we originally mounted
4481                          * the partition.)
4482                          */
4483                         if (sbi->s_journal)
4484                                 ext4_clear_journal_err(sb, es);
4485                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4486                         if (!ext4_setup_super(sb, es, 0))
4487                                 sb->s_flags &= ~MS_RDONLY;
4488                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4489                                                      EXT4_FEATURE_INCOMPAT_MMP))
4490                                 if (ext4_multi_mount_protect(sb,
4491                                                 le64_to_cpu(es->s_mmp_block))) {
4492                                         err = -EROFS;
4493                                         goto restore_opts;
4494                                 }
4495                         enable_quota = 1;
4496                 }
4497         }
4498
4499         /*
4500          * Reinitialize lazy itable initialization thread based on
4501          * current settings
4502          */
4503         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4504                 ext4_unregister_li_request(sb);
4505         else {
4506                 ext4_group_t first_not_zeroed;
4507                 first_not_zeroed = ext4_has_uninit_itable(sb);
4508                 ext4_register_li_request(sb, first_not_zeroed);
4509         }
4510
4511         ext4_setup_system_zone(sb);
4512         if (sbi->s_journal == NULL)
4513                 ext4_commit_super(sb, 1);
4514
4515 #ifdef CONFIG_QUOTA
4516         /* Release old quota file names */
4517         for (i = 0; i < MAXQUOTAS; i++)
4518                 if (old_opts.s_qf_names[i] &&
4519                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4520                         kfree(old_opts.s_qf_names[i]);
4521 #endif
4522         unlock_super(sb);
4523         if (enable_quota)
4524                 dquot_resume(sb, -1);
4525
4526         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4527         kfree(orig_data);
4528         return 0;
4529
4530 restore_opts:
4531         sb->s_flags = old_sb_flags;
4532         sbi->s_mount_opt = old_opts.s_mount_opt;
4533         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4534         sbi->s_resuid = old_opts.s_resuid;
4535         sbi->s_resgid = old_opts.s_resgid;
4536         sbi->s_commit_interval = old_opts.s_commit_interval;
4537         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4538         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4539 #ifdef CONFIG_QUOTA
4540         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4541         for (i = 0; i < MAXQUOTAS; i++) {
4542                 if (sbi->s_qf_names[i] &&
4543                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4544                         kfree(sbi->s_qf_names[i]);
4545                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4546         }
4547 #endif
4548         unlock_super(sb);
4549         kfree(orig_data);
4550         return err;
4551 }
4552
4553 /*
4554  * Note: calculating the overhead so we can be compatible with
4555  * historical BSD practice is quite difficult in the face of
4556  * clusters/bigalloc.  This is because multiple metadata blocks from
4557  * different block group can end up in the same allocation cluster.
4558  * Calculating the exact overhead in the face of clustered allocation
4559  * requires either O(all block bitmaps) in memory or O(number of block
4560  * groups**2) in time.  We will still calculate the superblock for
4561  * older file systems --- and if we come across with a bigalloc file
4562  * system with zero in s_overhead_clusters the estimate will be close to
4563  * correct especially for very large cluster sizes --- but for newer
4564  * file systems, it's better to calculate this figure once at mkfs
4565  * time, and store it in the superblock.  If the superblock value is
4566  * present (even for non-bigalloc file systems), we will use it.
4567  */
4568 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4569 {
4570         struct super_block *sb = dentry->d_sb;
4571         struct ext4_sb_info *sbi = EXT4_SB(sb);
4572         struct ext4_super_block *es = sbi->s_es;
4573         struct ext4_group_desc *gdp;
4574         u64 fsid;
4575         s64 bfree;
4576
4577         if (test_opt(sb, MINIX_DF)) {
4578                 sbi->s_overhead_last = 0;
4579         } else if (es->s_overhead_clusters) {
4580                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4581         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4582                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4583                 ext4_fsblk_t overhead = 0;
4584
4585                 /*
4586                  * Compute the overhead (FS structures).  This is constant
4587                  * for a given filesystem unless the number of block groups
4588                  * changes so we cache the previous value until it does.
4589                  */
4590
4591                 /*
4592                  * All of the blocks before first_data_block are
4593                  * overhead
4594                  */
4595                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4596
4597                 /*
4598                  * Add the overhead found in each block group
4599                  */
4600                 for (i = 0; i < ngroups; i++) {
4601                         gdp = ext4_get_group_desc(sb, i, NULL);
4602                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4603                         cond_resched();
4604                 }
4605                 sbi->s_overhead_last = overhead;
4606                 smp_wmb();
4607                 sbi->s_blocks_last = ext4_blocks_count(es);
4608         }
4609
4610         buf->f_type = EXT4_SUPER_MAGIC;
4611         buf->f_bsize = sb->s_blocksize;
4612         buf->f_blocks = (ext4_blocks_count(es) -
4613                          EXT4_C2B(sbi, sbi->s_overhead_last));
4614         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4615                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4616         /* prevent underflow in case that few free space is available */
4617         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4618         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4619         if (buf->f_bfree < ext4_r_blocks_count(es))
4620                 buf->f_bavail = 0;
4621         buf->f_files = le32_to_cpu(es->s_inodes_count);
4622         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4623         buf->f_namelen = EXT4_NAME_LEN;
4624         fsid = le64_to_cpup((void *)es->s_uuid) ^
4625                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4626         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4627         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4628
4629         return 0;
4630 }
4631
4632 /* Helper function for writing quotas on sync - we need to start transaction
4633  * before quota file is locked for write. Otherwise the are possible deadlocks:
4634  * Process 1                         Process 2
4635  * ext4_create()                     quota_sync()
4636  *   jbd2_journal_start()                  write_dquot()
4637  *   dquot_initialize()                         down(dqio_mutex)
4638  *     down(dqio_mutex)                    jbd2_journal_start()
4639  *
4640  */
4641
4642 #ifdef CONFIG_QUOTA
4643
4644 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4645 {
4646         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4647 }
4648
4649 static int ext4_write_dquot(struct dquot *dquot)
4650 {
4651         int ret, err;
4652         handle_t *handle;
4653         struct inode *inode;
4654
4655         inode = dquot_to_inode(dquot);
4656         handle = ext4_journal_start(inode,
4657                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4658         if (IS_ERR(handle))
4659                 return PTR_ERR(handle);
4660         ret = dquot_commit(dquot);
4661         err = ext4_journal_stop(handle);
4662         if (!ret)
4663                 ret = err;
4664         return ret;
4665 }
4666
4667 static int ext4_acquire_dquot(struct dquot *dquot)
4668 {
4669         int ret, err;
4670         handle_t *handle;
4671
4672         handle = ext4_journal_start(dquot_to_inode(dquot),
4673                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4674         if (IS_ERR(handle))
4675                 return PTR_ERR(handle);
4676         ret = dquot_acquire(dquot);
4677         err = ext4_journal_stop(handle);
4678         if (!ret)
4679                 ret = err;
4680         return ret;
4681 }
4682
4683 static int ext4_release_dquot(struct dquot *dquot)
4684 {
4685         int ret, err;
4686         handle_t *handle;
4687
4688         handle = ext4_journal_start(dquot_to_inode(dquot),
4689                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4690         if (IS_ERR(handle)) {
4691                 /* Release dquot anyway to avoid endless cycle in dqput() */
4692                 dquot_release(dquot);
4693                 return PTR_ERR(handle);
4694         }
4695         ret = dquot_release(dquot);
4696         err = ext4_journal_stop(handle);
4697         if (!ret)
4698                 ret = err;
4699         return ret;
4700 }
4701
4702 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4703 {
4704         /* Are we journaling quotas? */
4705         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4706             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4707                 dquot_mark_dquot_dirty(dquot);
4708                 return ext4_write_dquot(dquot);
4709         } else {
4710                 return dquot_mark_dquot_dirty(dquot);
4711         }
4712 }
4713
4714 static int ext4_write_info(struct super_block *sb, int type)
4715 {
4716         int ret, err;
4717         handle_t *handle;
4718
4719         /* Data block + inode block */
4720         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4721         if (IS_ERR(handle))
4722                 return PTR_ERR(handle);
4723         ret = dquot_commit_info(sb, type);
4724         err = ext4_journal_stop(handle);
4725         if (!ret)
4726                 ret = err;
4727         return ret;
4728 }
4729
4730 /*
4731  * Turn on quotas during mount time - we need to find
4732  * the quota file and such...
4733  */
4734 static int ext4_quota_on_mount(struct super_block *sb, int type)
4735 {
4736         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4737                                         EXT4_SB(sb)->s_jquota_fmt, type);
4738 }
4739
4740 /*
4741  * Standard function to be called on quota_on
4742  */
4743 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4744                          struct path *path)
4745 {
4746         int err;
4747
4748         if (!test_opt(sb, QUOTA))
4749                 return -EINVAL;
4750
4751         /* Quotafile not on the same filesystem? */
4752         if (path->dentry->d_sb != sb)
4753                 return -EXDEV;
4754         /* Journaling quota? */
4755         if (EXT4_SB(sb)->s_qf_names[type]) {
4756                 /* Quotafile not in fs root? */
4757                 if (path->dentry->d_parent != sb->s_root)
4758                         ext4_msg(sb, KERN_WARNING,
4759                                 "Quota file not on filesystem root. "
4760                                 "Journaled quota will not work");
4761         }
4762
4763         /*
4764          * When we journal data on quota file, we have to flush journal to see
4765          * all updates to the file when we bypass pagecache...
4766          */
4767         if (EXT4_SB(sb)->s_journal &&
4768             ext4_should_journal_data(path->dentry->d_inode)) {
4769                 /*
4770                  * We don't need to lock updates but journal_flush() could
4771                  * otherwise be livelocked...
4772                  */
4773                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4774                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4775                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4776                 if (err)
4777                         return err;
4778         }
4779
4780         return dquot_quota_on(sb, type, format_id, path);
4781 }
4782
4783 static int ext4_quota_off(struct super_block *sb, int type)
4784 {
4785         struct inode *inode = sb_dqopt(sb)->files[type];
4786         handle_t *handle;
4787
4788         /* Force all delayed allocation blocks to be allocated.
4789          * Caller already holds s_umount sem */
4790         if (test_opt(sb, DELALLOC))
4791                 sync_filesystem(sb);
4792
4793         if (!inode)
4794                 goto out;
4795
4796         /* Update modification times of quota files when userspace can
4797          * start looking at them */
4798         handle = ext4_journal_start(inode, 1);
4799         if (IS_ERR(handle))
4800                 goto out;
4801         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4802         ext4_mark_inode_dirty(handle, inode);
4803         ext4_journal_stop(handle);
4804
4805 out:
4806         return dquot_quota_off(sb, type);
4807 }
4808
4809 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4810  * acquiring the locks... As quota files are never truncated and quota code
4811  * itself serializes the operations (and no one else should touch the files)
4812  * we don't have to be afraid of races */
4813 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4814                                size_t len, loff_t off)
4815 {
4816         struct inode *inode = sb_dqopt(sb)->files[type];
4817         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4818         int err = 0;
4819         int offset = off & (sb->s_blocksize - 1);
4820         int tocopy;
4821         size_t toread;
4822         struct buffer_head *bh;
4823         loff_t i_size = i_size_read(inode);
4824
4825         if (off > i_size)
4826                 return 0;
4827         if (off+len > i_size)
4828                 len = i_size-off;
4829         toread = len;
4830         while (toread > 0) {
4831                 tocopy = sb->s_blocksize - offset < toread ?
4832                                 sb->s_blocksize - offset : toread;
4833                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4834                 if (err)
4835                         return err;
4836                 if (!bh)        /* A hole? */
4837                         memset(data, 0, tocopy);
4838                 else
4839                         memcpy(data, bh->b_data+offset, tocopy);
4840                 brelse(bh);
4841                 offset = 0;
4842                 toread -= tocopy;
4843                 data += tocopy;
4844                 blk++;
4845         }
4846         return len;
4847 }
4848
4849 /* Write to quotafile (we know the transaction is already started and has
4850  * enough credits) */
4851 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4852                                 const char *data, size_t len, loff_t off)
4853 {
4854         struct inode *inode = sb_dqopt(sb)->files[type];
4855         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4856         int err = 0;
4857         int offset = off & (sb->s_blocksize - 1);
4858         struct buffer_head *bh;
4859         handle_t *handle = journal_current_handle();
4860
4861         if (EXT4_SB(sb)->s_journal && !handle) {
4862                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4863                         " cancelled because transaction is not started",
4864                         (unsigned long long)off, (unsigned long long)len);
4865                 return -EIO;
4866         }
4867         /*
4868          * Since we account only one data block in transaction credits,
4869          * then it is impossible to cross a block boundary.
4870          */
4871         if (sb->s_blocksize - offset < len) {
4872                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4873                         " cancelled because not block aligned",
4874                         (unsigned long long)off, (unsigned long long)len);
4875                 return -EIO;
4876         }
4877
4878         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4879         bh = ext4_bread(handle, inode, blk, 1, &err);
4880         if (!bh)
4881                 goto out;
4882         err = ext4_journal_get_write_access(handle, bh);
4883         if (err) {
4884                 brelse(bh);
4885                 goto out;
4886         }
4887         lock_buffer(bh);
4888         memcpy(bh->b_data+offset, data, len);
4889         flush_dcache_page(bh->b_page);
4890         unlock_buffer(bh);
4891         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4892         brelse(bh);
4893 out:
4894         if (err) {
4895                 mutex_unlock(&inode->i_mutex);
4896                 return err;
4897         }
4898         if (inode->i_size < off + len) {
4899                 i_size_write(inode, off + len);
4900                 EXT4_I(inode)->i_disksize = inode->i_size;
4901                 ext4_mark_inode_dirty(handle, inode);
4902         }
4903         mutex_unlock(&inode->i_mutex);
4904         return len;
4905 }
4906
4907 #endif
4908
4909 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4910                        const char *dev_name, void *data)
4911 {
4912         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4913 }
4914
4915 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4916 static inline void register_as_ext2(void)
4917 {
4918         int err = register_filesystem(&ext2_fs_type);
4919         if (err)
4920                 printk(KERN_WARNING
4921                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4922 }
4923
4924 static inline void unregister_as_ext2(void)
4925 {
4926         unregister_filesystem(&ext2_fs_type);
4927 }
4928
4929 static inline int ext2_feature_set_ok(struct super_block *sb)
4930 {
4931         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4932                 return 0;
4933         if (sb->s_flags & MS_RDONLY)
4934                 return 1;
4935         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4936                 return 0;
4937         return 1;
4938 }
4939 MODULE_ALIAS("ext2");
4940 #else
4941 static inline void register_as_ext2(void) { }
4942 static inline void unregister_as_ext2(void) { }
4943 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4944 #endif
4945
4946 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4947 static inline void register_as_ext3(void)
4948 {
4949         int err = register_filesystem(&ext3_fs_type);
4950         if (err)
4951                 printk(KERN_WARNING
4952                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4953 }
4954
4955 static inline void unregister_as_ext3(void)
4956 {
4957         unregister_filesystem(&ext3_fs_type);
4958 }
4959
4960 static inline int ext3_feature_set_ok(struct super_block *sb)
4961 {
4962         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4963                 return 0;
4964         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4965                 return 0;
4966         if (sb->s_flags & MS_RDONLY)
4967                 return 1;
4968         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4969                 return 0;
4970         return 1;
4971 }
4972 MODULE_ALIAS("ext3");
4973 #else
4974 static inline void register_as_ext3(void) { }
4975 static inline void unregister_as_ext3(void) { }
4976 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4977 #endif
4978
4979 static struct file_system_type ext4_fs_type = {
4980         .owner          = THIS_MODULE,
4981         .name           = "ext4",
4982         .mount          = ext4_mount,
4983         .kill_sb        = kill_block_super,
4984         .fs_flags       = FS_REQUIRES_DEV,
4985 };
4986
4987 static int __init ext4_init_feat_adverts(void)
4988 {
4989         struct ext4_features *ef;
4990         int ret = -ENOMEM;
4991
4992         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4993         if (!ef)
4994                 goto out;
4995
4996         ef->f_kobj.kset = ext4_kset;
4997         init_completion(&ef->f_kobj_unregister);
4998         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4999                                    "features");
5000         if (ret) {
5001                 kfree(ef);
5002                 goto out;
5003         }
5004
5005         ext4_feat = ef;
5006         ret = 0;
5007 out:
5008         return ret;
5009 }
5010
5011 static void ext4_exit_feat_adverts(void)
5012 {
5013         kobject_put(&ext4_feat->f_kobj);
5014         wait_for_completion(&ext4_feat->f_kobj_unregister);
5015         kfree(ext4_feat);
5016 }
5017
5018 /* Shared across all ext4 file systems */
5019 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5020 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5021
5022 static int __init ext4_init_fs(void)
5023 {
5024         int i, err;
5025
5026         ext4_check_flag_values();
5027
5028         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5029                 mutex_init(&ext4__aio_mutex[i]);
5030                 init_waitqueue_head(&ext4__ioend_wq[i]);
5031         }
5032
5033         err = ext4_init_pageio();
5034         if (err)
5035                 return err;
5036         err = ext4_init_system_zone();
5037         if (err)
5038                 goto out6;
5039         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5040         if (!ext4_kset)
5041                 goto out5;
5042         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5043
5044         err = ext4_init_feat_adverts();
5045         if (err)
5046                 goto out4;
5047
5048         err = ext4_init_mballoc();
5049         if (err)
5050                 goto out3;
5051
5052         err = ext4_init_xattr();
5053         if (err)
5054                 goto out2;
5055         err = init_inodecache();
5056         if (err)
5057                 goto out1;
5058         register_as_ext3();
5059         register_as_ext2();
5060         err = register_filesystem(&ext4_fs_type);
5061         if (err)
5062                 goto out;
5063
5064         ext4_li_info = NULL;
5065         mutex_init(&ext4_li_mtx);
5066         return 0;
5067 out:
5068         unregister_as_ext2();
5069         unregister_as_ext3();
5070         destroy_inodecache();
5071 out1:
5072         ext4_exit_xattr();
5073 out2:
5074         ext4_exit_mballoc();
5075 out3:
5076         ext4_exit_feat_adverts();
5077 out4:
5078         if (ext4_proc_root)
5079                 remove_proc_entry("fs/ext4", NULL);
5080         kset_unregister(ext4_kset);
5081 out5:
5082         ext4_exit_system_zone();
5083 out6:
5084         ext4_exit_pageio();
5085         return err;
5086 }
5087
5088 static void __exit ext4_exit_fs(void)
5089 {
5090         ext4_destroy_lazyinit_thread();
5091         unregister_as_ext2();
5092         unregister_as_ext3();
5093         unregister_filesystem(&ext4_fs_type);
5094         destroy_inodecache();
5095         ext4_exit_xattr();
5096         ext4_exit_mballoc();
5097         ext4_exit_feat_adverts();
5098         remove_proc_entry("fs/ext4", NULL);
5099         kset_unregister(ext4_kset);
5100         ext4_exit_system_zone();
5101         ext4_exit_pageio();
5102 }
5103
5104 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5105 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5106 MODULE_LICENSE("GPL");
5107 module_init(ext4_init_fs)
5108 module_exit(ext4_exit_fs)