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 (sb->s_flags & MS_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_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_MASK);
1569                         set_opt(sb, ERRORS_PANIC);
1570                         break;
1571                 case Opt_err_ro:
1572                         clear_opt(sb, ERRORS_MASK);
1573                         set_opt(sb, ERRORS_RO);
1574                         break;
1575                 case Opt_err_cont:
1576                         clear_opt(sb, ERRORS_MASK);
1577                         set_opt(sb, ERRORS_CONT);
1578                         break;
1579                 case Opt_nouid32:
1580                         set_opt(sb, NO_UID32);
1581                         break;
1582                 case Opt_debug:
1583                         set_opt(sb, DEBUG);
1584                         break;
1585                 case Opt_oldalloc:
1586                         ext4_msg(sb, KERN_WARNING,
1587                                  "Ignoring deprecated oldalloc option");
1588                         break;
1589                 case Opt_orlov:
1590                         ext4_msg(sb, KERN_WARNING,
1591                                  "Ignoring deprecated orlov option");
1592                         break;
1593 #ifdef CONFIG_EXT4_FS_XATTR
1594                 case Opt_user_xattr:
1595                         set_opt(sb, XATTR_USER);
1596                         break;
1597                 case Opt_nouser_xattr:
1598                         clear_opt(sb, XATTR_USER);
1599                         break;
1600 #else
1601                 case Opt_user_xattr:
1602                 case Opt_nouser_xattr:
1603                         ext4_msg(sb, KERN_ERR, "(no)user_xattr options not supported");
1604                         break;
1605 #endif
1606 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1607                 case Opt_acl:
1608                         set_opt(sb, POSIX_ACL);
1609                         break;
1610                 case Opt_noacl:
1611                         clear_opt(sb, POSIX_ACL);
1612                         break;
1613 #else
1614                 case Opt_acl:
1615                 case Opt_noacl:
1616                         ext4_msg(sb, KERN_ERR, "(no)acl options not supported");
1617                         break;
1618 #endif
1619                 case Opt_journal_dev:
1620                         if (is_remount) {
1621                                 ext4_msg(sb, KERN_ERR,
1622                                         "Cannot specify journal on remount");
1623                                 return 0;
1624                         }
1625                         if (match_int(&args[0], &option))
1626                                 return 0;
1627                         *journal_devnum = option;
1628                         break;
1629                 case Opt_journal_checksum:
1630                         set_opt(sb, JOURNAL_CHECKSUM);
1631                         break;
1632                 case Opt_journal_async_commit:
1633                         set_opt(sb, JOURNAL_ASYNC_COMMIT);
1634                         set_opt(sb, JOURNAL_CHECKSUM);
1635                         break;
1636                 case Opt_noload:
1637                         set_opt(sb, NOLOAD);
1638                         break;
1639                 case Opt_commit:
1640                         if (match_int(&args[0], &option))
1641                                 return 0;
1642                         if (option < 0)
1643                                 return 0;
1644                         if (option == 0)
1645                                 option = JBD2_DEFAULT_MAX_COMMIT_AGE;
1646                         sbi->s_commit_interval = HZ * option;
1647                         break;
1648                 case Opt_max_batch_time:
1649                         if (match_int(&args[0], &option))
1650                                 return 0;
1651                         if (option < 0)
1652                                 return 0;
1653                         if (option == 0)
1654                                 option = EXT4_DEF_MAX_BATCH_TIME;
1655                         sbi->s_max_batch_time = option;
1656                         break;
1657                 case Opt_min_batch_time:
1658                         if (match_int(&args[0], &option))
1659                                 return 0;
1660                         if (option < 0)
1661                                 return 0;
1662                         sbi->s_min_batch_time = option;
1663                         break;
1664                 case Opt_data_journal:
1665                         data_opt = EXT4_MOUNT_JOURNAL_DATA;
1666                         goto datacheck;
1667                 case Opt_data_ordered:
1668                         data_opt = EXT4_MOUNT_ORDERED_DATA;
1669                         goto datacheck;
1670                 case Opt_data_writeback:
1671                         data_opt = EXT4_MOUNT_WRITEBACK_DATA;
1672                 datacheck:
1673                         if (is_remount) {
1674                                 if (!sbi->s_journal)
1675                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1676                                 else if (test_opt(sb, DATA_FLAGS) != data_opt) {
1677                                         ext4_msg(sb, KERN_ERR,
1678                                                 "Cannot change data mode on remount");
1679                                         return 0;
1680                                 }
1681                         } else {
1682                                 clear_opt(sb, DATA_FLAGS);
1683                                 sbi->s_mount_opt |= data_opt;
1684                         }
1685                         break;
1686                 case Opt_data_err_abort:
1687                         set_opt(sb, DATA_ERR_ABORT);
1688                         break;
1689                 case Opt_data_err_ignore:
1690                         clear_opt(sb, DATA_ERR_ABORT);
1691                         break;
1692 #ifdef CONFIG_QUOTA
1693                 case Opt_usrjquota:
1694                         if (!set_qf_name(sb, USRQUOTA, &args[0]))
1695                                 return 0;
1696                         break;
1697                 case Opt_grpjquota:
1698                         if (!set_qf_name(sb, GRPQUOTA, &args[0]))
1699                                 return 0;
1700                         break;
1701                 case Opt_offusrjquota:
1702                         if (!clear_qf_name(sb, USRQUOTA))
1703                                 return 0;
1704                         break;
1705                 case Opt_offgrpjquota:
1706                         if (!clear_qf_name(sb, GRPQUOTA))
1707                                 return 0;
1708                         break;
1709
1710                 case Opt_jqfmt_vfsold:
1711                         qfmt = QFMT_VFS_OLD;
1712                         goto set_qf_format;
1713                 case Opt_jqfmt_vfsv0:
1714                         qfmt = QFMT_VFS_V0;
1715                         goto set_qf_format;
1716                 case Opt_jqfmt_vfsv1:
1717                         qfmt = QFMT_VFS_V1;
1718 set_qf_format:
1719                         if (sb_any_quota_loaded(sb) &&
1720                             sbi->s_jquota_fmt != qfmt) {
1721                                 ext4_msg(sb, KERN_ERR, "Cannot change "
1722                                         "journaled quota options when "
1723                                         "quota turned on");
1724                                 return 0;
1725                         }
1726                         sbi->s_jquota_fmt = qfmt;
1727                         break;
1728                 case Opt_quota:
1729                 case Opt_usrquota:
1730                         set_opt(sb, QUOTA);
1731                         set_opt(sb, USRQUOTA);
1732                         break;
1733                 case Opt_grpquota:
1734                         set_opt(sb, QUOTA);
1735                         set_opt(sb, GRPQUOTA);
1736                         break;
1737                 case Opt_noquota:
1738                         if (sb_any_quota_loaded(sb)) {
1739                                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1740                                         "options when quota turned on");
1741                                 return 0;
1742                         }
1743                         clear_opt(sb, QUOTA);
1744                         clear_opt(sb, USRQUOTA);
1745                         clear_opt(sb, GRPQUOTA);
1746                         break;
1747 #else
1748                 case Opt_quota:
1749                 case Opt_usrquota:
1750                 case Opt_grpquota:
1751                         ext4_msg(sb, KERN_ERR,
1752                                 "quota options not supported");
1753                         break;
1754                 case Opt_usrjquota:
1755                 case Opt_grpjquota:
1756                 case Opt_offusrjquota:
1757                 case Opt_offgrpjquota:
1758                 case Opt_jqfmt_vfsold:
1759                 case Opt_jqfmt_vfsv0:
1760                 case Opt_jqfmt_vfsv1:
1761                         ext4_msg(sb, KERN_ERR,
1762                                 "journaled quota options not supported");
1763                         break;
1764                 case Opt_noquota:
1765                         break;
1766 #endif
1767                 case Opt_abort:
1768                         sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1769                         break;
1770                 case Opt_nobarrier:
1771                         clear_opt(sb, BARRIER);
1772                         break;
1773                 case Opt_barrier:
1774                         if (args[0].from) {
1775                                 if (match_int(&args[0], &option))
1776                                         return 0;
1777                         } else
1778                                 option = 1;     /* No argument, default to 1 */
1779                         if (option)
1780                                 set_opt(sb, BARRIER);
1781                         else
1782                                 clear_opt(sb, BARRIER);
1783                         break;
1784                 case Opt_nobh:
1785                         ext4_msg(sb, KERN_WARNING,
1786                                  "Ignoring deprecated nobh option");
1787                         break;
1788                 case Opt_bh:
1789                         ext4_msg(sb, KERN_WARNING,
1790                                  "Ignoring deprecated bh option");
1791                         break;
1792                 case Opt_i_version:
1793                         sb->s_flags |= MS_I_VERSION;
1794                         break;
1795                 case Opt_nodelalloc:
1796                         clear_opt(sb, DELALLOC);
1797                         clear_opt2(sb, EXPLICIT_DELALLOC);
1798                         break;
1799                 case Opt_mblk_io_submit:
1800                         set_opt(sb, MBLK_IO_SUBMIT);
1801                         break;
1802                 case Opt_nomblk_io_submit:
1803                         clear_opt(sb, MBLK_IO_SUBMIT);
1804                         break;
1805                 case Opt_stripe:
1806                         if (match_int(&args[0], &option))
1807                                 return 0;
1808                         if (option < 0)
1809                                 return 0;
1810                         sbi->s_stripe = option;
1811                         break;
1812                 case Opt_delalloc:
1813                         set_opt(sb, DELALLOC);
1814                         set_opt2(sb, EXPLICIT_DELALLOC);
1815                         break;
1816                 case Opt_block_validity:
1817                         set_opt(sb, BLOCK_VALIDITY);
1818                         break;
1819                 case Opt_noblock_validity:
1820                         clear_opt(sb, BLOCK_VALIDITY);
1821                         break;
1822                 case Opt_inode_readahead_blks:
1823                         if (match_int(&args[0], &option))
1824                                 return 0;
1825                         if (option < 0 || option > (1 << 30))
1826                                 return 0;
1827                         if (option && !is_power_of_2(option)) {
1828                                 ext4_msg(sb, KERN_ERR,
1829                                          "EXT4-fs: inode_readahead_blks"
1830                                          " must be a power of 2");
1831                                 return 0;
1832                         }
1833                         sbi->s_inode_readahead_blks = option;
1834                         break;
1835                 case Opt_journal_ioprio:
1836                         if (match_int(&args[0], &option))
1837                                 return 0;
1838                         if (option < 0 || option > 7)
1839                                 break;
1840                         *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE,
1841                                                             option);
1842                         break;
1843                 case Opt_noauto_da_alloc:
1844                         set_opt(sb, NO_AUTO_DA_ALLOC);
1845                         break;
1846                 case Opt_auto_da_alloc:
1847                         if (args[0].from) {
1848                                 if (match_int(&args[0], &option))
1849                                         return 0;
1850                         } else
1851                                 option = 1;     /* No argument, default to 1 */
1852                         if (option)
1853                                 clear_opt(sb, NO_AUTO_DA_ALLOC);
1854                         else
1855                                 set_opt(sb,NO_AUTO_DA_ALLOC);
1856                         break;
1857                 case Opt_discard:
1858                         set_opt(sb, DISCARD);
1859                         break;
1860                 case Opt_nodiscard:
1861                         clear_opt(sb, DISCARD);
1862                         break;
1863                 case Opt_dioread_nolock:
1864                         set_opt(sb, DIOREAD_NOLOCK);
1865                         break;
1866                 case Opt_dioread_lock:
1867                         clear_opt(sb, DIOREAD_NOLOCK);
1868                         break;
1869                 case Opt_init_itable:
1870                         set_opt(sb, INIT_INODE_TABLE);
1871                         if (args[0].from) {
1872                                 if (match_int(&args[0], &option))
1873                                         return 0;
1874                         } else
1875                                 option = EXT4_DEF_LI_WAIT_MULT;
1876                         if (option < 0)
1877                                 return 0;
1878                         sbi->s_li_wait_mult = option;
1879                         break;
1880                 case Opt_noinit_itable:
1881                         clear_opt(sb, INIT_INODE_TABLE);
1882                         break;
1883                 default:
1884                         ext4_msg(sb, KERN_ERR,
1885                                "Unrecognized mount option \"%s\" "
1886                                "or missing value", p);
1887                         return 0;
1888                 }
1889         }
1890 #ifdef CONFIG_QUOTA
1891         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1892                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1893                         clear_opt(sb, USRQUOTA);
1894
1895                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1896                         clear_opt(sb, GRPQUOTA);
1897
1898                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1899                         ext4_msg(sb, KERN_ERR, "old and new quota "
1900                                         "format mixing");
1901                         return 0;
1902                 }
1903
1904                 if (!sbi->s_jquota_fmt) {
1905                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1906                                         "not specified");
1907                         return 0;
1908                 }
1909         } else {
1910                 if (sbi->s_jquota_fmt) {
1911                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1912                                         "specified with no journaling "
1913                                         "enabled");
1914                         return 0;
1915                 }
1916         }
1917 #endif
1918         return 1;
1919 }
1920
1921 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1922                             int read_only)
1923 {
1924         struct ext4_sb_info *sbi = EXT4_SB(sb);
1925         int res = 0;
1926
1927         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1928                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1929                          "forcing read-only mode");
1930                 res = MS_RDONLY;
1931         }
1932         if (read_only)
1933                 goto done;
1934         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1935                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1936                          "running e2fsck is recommended");
1937         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1938                 ext4_msg(sb, KERN_WARNING,
1939                          "warning: mounting fs with errors, "
1940                          "running e2fsck is recommended");
1941         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1942                  le16_to_cpu(es->s_mnt_count) >=
1943                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1944                 ext4_msg(sb, KERN_WARNING,
1945                          "warning: maximal mount count reached, "
1946                          "running e2fsck is recommended");
1947         else if (le32_to_cpu(es->s_checkinterval) &&
1948                 (le32_to_cpu(es->s_lastcheck) +
1949                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1950                 ext4_msg(sb, KERN_WARNING,
1951                          "warning: checktime reached, "
1952                          "running e2fsck is recommended");
1953         if (!sbi->s_journal)
1954                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1955         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1956                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1957         le16_add_cpu(&es->s_mnt_count, 1);
1958         es->s_mtime = cpu_to_le32(get_seconds());
1959         ext4_update_dynamic_rev(sb);
1960         if (sbi->s_journal)
1961                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1962
1963         ext4_commit_super(sb, 1);
1964 done:
1965         if (test_opt(sb, DEBUG))
1966                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1967                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1968                         sb->s_blocksize,
1969                         sbi->s_groups_count,
1970                         EXT4_BLOCKS_PER_GROUP(sb),
1971                         EXT4_INODES_PER_GROUP(sb),
1972                         sbi->s_mount_opt, sbi->s_mount_opt2);
1973
1974         cleancache_init_fs(sb);
1975         return res;
1976 }
1977
1978 static int ext4_fill_flex_info(struct super_block *sb)
1979 {
1980         struct ext4_sb_info *sbi = EXT4_SB(sb);
1981         struct ext4_group_desc *gdp = NULL;
1982         ext4_group_t flex_group_count;
1983         ext4_group_t flex_group;
1984         unsigned int groups_per_flex = 0;
1985         size_t size;
1986         int i;
1987
1988         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1989         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1990                 sbi->s_log_groups_per_flex = 0;
1991                 return 1;
1992         }
1993         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1994
1995         /* We allocate both existing and potentially added groups */
1996         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1997                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1998                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1999         size = flex_group_count * sizeof(struct flex_groups);
2000         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
2001         if (sbi->s_flex_groups == NULL) {
2002                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
2003                          flex_group_count);
2004                 goto failed;
2005         }
2006
2007         for (i = 0; i < sbi->s_groups_count; i++) {
2008                 gdp = ext4_get_group_desc(sb, i, NULL);
2009
2010                 flex_group = ext4_flex_group(sbi, i);
2011                 atomic_add(ext4_free_inodes_count(sb, gdp),
2012                            &sbi->s_flex_groups[flex_group].free_inodes);
2013                 atomic_add(ext4_free_group_clusters(sb, gdp),
2014                            &sbi->s_flex_groups[flex_group].free_clusters);
2015                 atomic_add(ext4_used_dirs_count(sb, gdp),
2016                            &sbi->s_flex_groups[flex_group].used_dirs);
2017         }
2018
2019         return 1;
2020 failed:
2021         return 0;
2022 }
2023
2024 __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
2025                             struct ext4_group_desc *gdp)
2026 {
2027         __u16 crc = 0;
2028
2029         if (sbi->s_es->s_feature_ro_compat &
2030             cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) {
2031                 int offset = offsetof(struct ext4_group_desc, bg_checksum);
2032                 __le32 le_group = cpu_to_le32(block_group);
2033
2034                 crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2035                 crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2036                 crc = crc16(crc, (__u8 *)gdp, offset);
2037                 offset += sizeof(gdp->bg_checksum); /* skip checksum */
2038                 /* for checksum of struct ext4_group_desc do the rest...*/
2039                 if ((sbi->s_es->s_feature_incompat &
2040                      cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2041                     offset < le16_to_cpu(sbi->s_es->s_desc_size))
2042                         crc = crc16(crc, (__u8 *)gdp + offset,
2043                                     le16_to_cpu(sbi->s_es->s_desc_size) -
2044                                         offset);
2045         }
2046
2047         return cpu_to_le16(crc);
2048 }
2049
2050 int ext4_group_desc_csum_verify(struct ext4_sb_info *sbi, __u32 block_group,
2051                                 struct ext4_group_desc *gdp)
2052 {
2053         if ((sbi->s_es->s_feature_ro_compat &
2054              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_GDT_CSUM)) &&
2055             (gdp->bg_checksum != ext4_group_desc_csum(sbi, block_group, gdp)))
2056                 return 0;
2057
2058         return 1;
2059 }
2060
2061 /* Called at mount-time, super-block is locked */
2062 static int ext4_check_descriptors(struct super_block *sb,
2063                                   ext4_group_t *first_not_zeroed)
2064 {
2065         struct ext4_sb_info *sbi = EXT4_SB(sb);
2066         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2067         ext4_fsblk_t last_block;
2068         ext4_fsblk_t block_bitmap;
2069         ext4_fsblk_t inode_bitmap;
2070         ext4_fsblk_t inode_table;
2071         int flexbg_flag = 0;
2072         ext4_group_t i, grp = sbi->s_groups_count;
2073
2074         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2075                 flexbg_flag = 1;
2076
2077         ext4_debug("Checking group descriptors");
2078
2079         for (i = 0; i < sbi->s_groups_count; i++) {
2080                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2081
2082                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2083                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2084                 else
2085                         last_block = first_block +
2086                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2087
2088                 if ((grp == sbi->s_groups_count) &&
2089                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2090                         grp = i;
2091
2092                 block_bitmap = ext4_block_bitmap(sb, gdp);
2093                 if (block_bitmap < first_block || block_bitmap > last_block) {
2094                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2095                                "Block bitmap for group %u not in group "
2096                                "(block %llu)!", i, block_bitmap);
2097                         return 0;
2098                 }
2099                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2100                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2101                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2102                                "Inode bitmap for group %u not in group "
2103                                "(block %llu)!", i, inode_bitmap);
2104                         return 0;
2105                 }
2106                 inode_table = ext4_inode_table(sb, gdp);
2107                 if (inode_table < first_block ||
2108                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2109                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2110                                "Inode table for group %u not in group "
2111                                "(block %llu)!", i, inode_table);
2112                         return 0;
2113                 }
2114                 ext4_lock_group(sb, i);
2115                 if (!ext4_group_desc_csum_verify(sbi, i, gdp)) {
2116                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2117                                  "Checksum for group %u failed (%u!=%u)",
2118                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2119                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2120                         if (!(sb->s_flags & MS_RDONLY)) {
2121                                 ext4_unlock_group(sb, i);
2122                                 return 0;
2123                         }
2124                 }
2125                 ext4_unlock_group(sb, i);
2126                 if (!flexbg_flag)
2127                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2128         }
2129         if (NULL != first_not_zeroed)
2130                 *first_not_zeroed = grp;
2131
2132         ext4_free_blocks_count_set(sbi->s_es,
2133                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2134         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2135         return 1;
2136 }
2137
2138 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2139  * the superblock) which were deleted from all directories, but held open by
2140  * a process at the time of a crash.  We walk the list and try to delete these
2141  * inodes at recovery time (only with a read-write filesystem).
2142  *
2143  * In order to keep the orphan inode chain consistent during traversal (in
2144  * case of crash during recovery), we link each inode into the superblock
2145  * orphan list_head and handle it the same way as an inode deletion during
2146  * normal operation (which journals the operations for us).
2147  *
2148  * We only do an iget() and an iput() on each inode, which is very safe if we
2149  * accidentally point at an in-use or already deleted inode.  The worst that
2150  * can happen in this case is that we get a "bit already cleared" message from
2151  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2152  * e2fsck was run on this filesystem, and it must have already done the orphan
2153  * inode cleanup for us, so we can safely abort without any further action.
2154  */
2155 static void ext4_orphan_cleanup(struct super_block *sb,
2156                                 struct ext4_super_block *es)
2157 {
2158         unsigned int s_flags = sb->s_flags;
2159         int nr_orphans = 0, nr_truncates = 0;
2160 #ifdef CONFIG_QUOTA
2161         int i;
2162 #endif
2163         if (!es->s_last_orphan) {
2164                 jbd_debug(4, "no orphan inodes to clean up\n");
2165                 return;
2166         }
2167
2168         if (bdev_read_only(sb->s_bdev)) {
2169                 ext4_msg(sb, KERN_ERR, "write access "
2170                         "unavailable, skipping orphan cleanup");
2171                 return;
2172         }
2173
2174         /* Check if feature set would not allow a r/w mount */
2175         if (!ext4_feature_set_ok(sb, 0)) {
2176                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2177                          "unknown ROCOMPAT features");
2178                 return;
2179         }
2180
2181         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2182                 if (es->s_last_orphan)
2183                         jbd_debug(1, "Errors on filesystem, "
2184                                   "clearing orphan list.\n");
2185                 es->s_last_orphan = 0;
2186                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2187                 return;
2188         }
2189
2190         if (s_flags & MS_RDONLY) {
2191                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2192                 sb->s_flags &= ~MS_RDONLY;
2193         }
2194 #ifdef CONFIG_QUOTA
2195         /* Needed for iput() to work correctly and not trash data */
2196         sb->s_flags |= MS_ACTIVE;
2197         /* Turn on quotas so that they are updated correctly */
2198         for (i = 0; i < MAXQUOTAS; i++) {
2199                 if (EXT4_SB(sb)->s_qf_names[i]) {
2200                         int ret = ext4_quota_on_mount(sb, i);
2201                         if (ret < 0)
2202                                 ext4_msg(sb, KERN_ERR,
2203                                         "Cannot turn on journaled "
2204                                         "quota: error %d", ret);
2205                 }
2206         }
2207 #endif
2208
2209         while (es->s_last_orphan) {
2210                 struct inode *inode;
2211
2212                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2213                 if (IS_ERR(inode)) {
2214                         es->s_last_orphan = 0;
2215                         break;
2216                 }
2217
2218                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2219                 dquot_initialize(inode);
2220                 if (inode->i_nlink) {
2221                         ext4_msg(sb, KERN_DEBUG,
2222                                 "%s: truncating inode %lu to %lld bytes",
2223                                 __func__, inode->i_ino, inode->i_size);
2224                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2225                                   inode->i_ino, inode->i_size);
2226                         ext4_truncate(inode);
2227                         nr_truncates++;
2228                 } else {
2229                         ext4_msg(sb, KERN_DEBUG,
2230                                 "%s: deleting unreferenced inode %lu",
2231                                 __func__, inode->i_ino);
2232                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2233                                   inode->i_ino);
2234                         nr_orphans++;
2235                 }
2236                 iput(inode);  /* The delete magic happens here! */
2237         }
2238
2239 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2240
2241         if (nr_orphans)
2242                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2243                        PLURAL(nr_orphans));
2244         if (nr_truncates)
2245                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2246                        PLURAL(nr_truncates));
2247 #ifdef CONFIG_QUOTA
2248         /* Turn quotas off */
2249         for (i = 0; i < MAXQUOTAS; i++) {
2250                 if (sb_dqopt(sb)->files[i])
2251                         dquot_quota_off(sb, i);
2252         }
2253 #endif
2254         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2255 }
2256
2257 /*
2258  * Maximal extent format file size.
2259  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2260  * extent format containers, within a sector_t, and within i_blocks
2261  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2262  * so that won't be a limiting factor.
2263  *
2264  * However there is other limiting factor. We do store extents in the form
2265  * of starting block and length, hence the resulting length of the extent
2266  * covering maximum file size must fit into on-disk format containers as
2267  * well. Given that length is always by 1 unit bigger than max unit (because
2268  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2269  *
2270  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2271  */
2272 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2273 {
2274         loff_t res;
2275         loff_t upper_limit = MAX_LFS_FILESIZE;
2276
2277         /* small i_blocks in vfs inode? */
2278         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2279                 /*
2280                  * CONFIG_LBDAF is not enabled implies the inode
2281                  * i_block represent total blocks in 512 bytes
2282                  * 32 == size of vfs inode i_blocks * 8
2283                  */
2284                 upper_limit = (1LL << 32) - 1;
2285
2286                 /* total blocks in file system block size */
2287                 upper_limit >>= (blkbits - 9);
2288                 upper_limit <<= blkbits;
2289         }
2290
2291         /*
2292          * 32-bit extent-start container, ee_block. We lower the maxbytes
2293          * by one fs block, so ee_len can cover the extent of maximum file
2294          * size
2295          */
2296         res = (1LL << 32) - 1;
2297         res <<= blkbits;
2298
2299         /* Sanity check against vm- & vfs- imposed limits */
2300         if (res > upper_limit)
2301                 res = upper_limit;
2302
2303         return res;
2304 }
2305
2306 /*
2307  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2308  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2309  * We need to be 1 filesystem block less than the 2^48 sector limit.
2310  */
2311 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2312 {
2313         loff_t res = EXT4_NDIR_BLOCKS;
2314         int meta_blocks;
2315         loff_t upper_limit;
2316         /* This is calculated to be the largest file size for a dense, block
2317          * mapped file such that the file's total number of 512-byte sectors,
2318          * including data and all indirect blocks, does not exceed (2^48 - 1).
2319          *
2320          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2321          * number of 512-byte sectors of the file.
2322          */
2323
2324         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2325                 /*
2326                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2327                  * the inode i_block field represents total file blocks in
2328                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2329                  */
2330                 upper_limit = (1LL << 32) - 1;
2331
2332                 /* total blocks in file system block size */
2333                 upper_limit >>= (bits - 9);
2334
2335         } else {
2336                 /*
2337                  * We use 48 bit ext4_inode i_blocks
2338                  * With EXT4_HUGE_FILE_FL set the i_blocks
2339                  * represent total number of blocks in
2340                  * file system block size
2341                  */
2342                 upper_limit = (1LL << 48) - 1;
2343
2344         }
2345
2346         /* indirect blocks */
2347         meta_blocks = 1;
2348         /* double indirect blocks */
2349         meta_blocks += 1 + (1LL << (bits-2));
2350         /* tripple indirect blocks */
2351         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2352
2353         upper_limit -= meta_blocks;
2354         upper_limit <<= bits;
2355
2356         res += 1LL << (bits-2);
2357         res += 1LL << (2*(bits-2));
2358         res += 1LL << (3*(bits-2));
2359         res <<= bits;
2360         if (res > upper_limit)
2361                 res = upper_limit;
2362
2363         if (res > MAX_LFS_FILESIZE)
2364                 res = MAX_LFS_FILESIZE;
2365
2366         return res;
2367 }
2368
2369 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2370                                    ext4_fsblk_t logical_sb_block, int nr)
2371 {
2372         struct ext4_sb_info *sbi = EXT4_SB(sb);
2373         ext4_group_t bg, first_meta_bg;
2374         int has_super = 0;
2375
2376         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2377
2378         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2379             nr < first_meta_bg)
2380                 return logical_sb_block + nr + 1;
2381         bg = sbi->s_desc_per_block * nr;
2382         if (ext4_bg_has_super(sb, bg))
2383                 has_super = 1;
2384
2385         return (has_super + ext4_group_first_block_no(sb, bg));
2386 }
2387
2388 /**
2389  * ext4_get_stripe_size: Get the stripe size.
2390  * @sbi: In memory super block info
2391  *
2392  * If we have specified it via mount option, then
2393  * use the mount option value. If the value specified at mount time is
2394  * greater than the blocks per group use the super block value.
2395  * If the super block value is greater than blocks per group return 0.
2396  * Allocator needs it be less than blocks per group.
2397  *
2398  */
2399 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2400 {
2401         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2402         unsigned long stripe_width =
2403                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2404         int ret;
2405
2406         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2407                 ret = sbi->s_stripe;
2408         else if (stripe_width <= sbi->s_blocks_per_group)
2409                 ret = stripe_width;
2410         else if (stride <= sbi->s_blocks_per_group)
2411                 ret = stride;
2412         else
2413                 ret = 0;
2414
2415         /*
2416          * If the stripe width is 1, this makes no sense and
2417          * we set it to 0 to turn off stripe handling code.
2418          */
2419         if (ret <= 1)
2420                 ret = 0;
2421
2422         return ret;
2423 }
2424
2425 /* sysfs supprt */
2426
2427 struct ext4_attr {
2428         struct attribute attr;
2429         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2430         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2431                          const char *, size_t);
2432         int offset;
2433 };
2434
2435 static int parse_strtoul(const char *buf,
2436                 unsigned long max, unsigned long *value)
2437 {
2438         char *endp;
2439
2440         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2441         endp = skip_spaces(endp);
2442         if (*endp || *value > max)
2443                 return -EINVAL;
2444
2445         return 0;
2446 }
2447
2448 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2449                                               struct ext4_sb_info *sbi,
2450                                               char *buf)
2451 {
2452         return snprintf(buf, PAGE_SIZE, "%llu\n",
2453                 (s64) EXT4_C2B(sbi,
2454                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2455 }
2456
2457 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2458                                          struct ext4_sb_info *sbi, char *buf)
2459 {
2460         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2461
2462         if (!sb->s_bdev->bd_part)
2463                 return snprintf(buf, PAGE_SIZE, "0\n");
2464         return snprintf(buf, PAGE_SIZE, "%lu\n",
2465                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2466                          sbi->s_sectors_written_start) >> 1);
2467 }
2468
2469 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2470                                           struct ext4_sb_info *sbi, char *buf)
2471 {
2472         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2473
2474         if (!sb->s_bdev->bd_part)
2475                 return snprintf(buf, PAGE_SIZE, "0\n");
2476         return snprintf(buf, PAGE_SIZE, "%llu\n",
2477                         (unsigned long long)(sbi->s_kbytes_written +
2478                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2479                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2480 }
2481
2482 static ssize_t extent_cache_hits_show(struct ext4_attr *a,
2483                                       struct ext4_sb_info *sbi, char *buf)
2484 {
2485         return snprintf(buf, PAGE_SIZE, "%lu\n", sbi->extent_cache_hits);
2486 }
2487
2488 static ssize_t extent_cache_misses_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_misses);
2492 }
2493
2494 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2495                                           struct ext4_sb_info *sbi,
2496                                           const char *buf, size_t count)
2497 {
2498         unsigned long t;
2499
2500         if (parse_strtoul(buf, 0x40000000, &t))
2501                 return -EINVAL;
2502
2503         if (t && !is_power_of_2(t))
2504                 return -EINVAL;
2505
2506         sbi->s_inode_readahead_blks = t;
2507         return count;
2508 }
2509
2510 static ssize_t sbi_ui_show(struct ext4_attr *a,
2511                            struct ext4_sb_info *sbi, char *buf)
2512 {
2513         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2514
2515         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2516 }
2517
2518 static ssize_t sbi_ui_store(struct ext4_attr *a,
2519                             struct ext4_sb_info *sbi,
2520                             const char *buf, size_t count)
2521 {
2522         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2523         unsigned long t;
2524
2525         if (parse_strtoul(buf, 0xffffffff, &t))
2526                 return -EINVAL;
2527         *ui = t;
2528         return count;
2529 }
2530
2531 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2532 static struct ext4_attr ext4_attr_##_name = {                   \
2533         .attr = {.name = __stringify(_name), .mode = _mode },   \
2534         .show   = _show,                                        \
2535         .store  = _store,                                       \
2536         .offset = offsetof(struct ext4_sb_info, _elname),       \
2537 }
2538 #define EXT4_ATTR(name, mode, show, store) \
2539 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2540
2541 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2542 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2543 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2544 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2545         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2546 #define ATTR_LIST(name) &ext4_attr_##name.attr
2547
2548 EXT4_RO_ATTR(delayed_allocation_blocks);
2549 EXT4_RO_ATTR(session_write_kbytes);
2550 EXT4_RO_ATTR(lifetime_write_kbytes);
2551 EXT4_RO_ATTR(extent_cache_hits);
2552 EXT4_RO_ATTR(extent_cache_misses);
2553 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2554                  inode_readahead_blks_store, s_inode_readahead_blks);
2555 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2556 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2557 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2558 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2559 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2560 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2561 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2562 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2563
2564 static struct attribute *ext4_attrs[] = {
2565         ATTR_LIST(delayed_allocation_blocks),
2566         ATTR_LIST(session_write_kbytes),
2567         ATTR_LIST(lifetime_write_kbytes),
2568         ATTR_LIST(extent_cache_hits),
2569         ATTR_LIST(extent_cache_misses),
2570         ATTR_LIST(inode_readahead_blks),
2571         ATTR_LIST(inode_goal),
2572         ATTR_LIST(mb_stats),
2573         ATTR_LIST(mb_max_to_scan),
2574         ATTR_LIST(mb_min_to_scan),
2575         ATTR_LIST(mb_order2_req),
2576         ATTR_LIST(mb_stream_req),
2577         ATTR_LIST(mb_group_prealloc),
2578         ATTR_LIST(max_writeback_mb_bump),
2579         NULL,
2580 };
2581
2582 /* Features this copy of ext4 supports */
2583 EXT4_INFO_ATTR(lazy_itable_init);
2584 EXT4_INFO_ATTR(batched_discard);
2585
2586 static struct attribute *ext4_feat_attrs[] = {
2587         ATTR_LIST(lazy_itable_init),
2588         ATTR_LIST(batched_discard),
2589         NULL,
2590 };
2591
2592 static ssize_t ext4_attr_show(struct kobject *kobj,
2593                               struct attribute *attr, char *buf)
2594 {
2595         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2596                                                 s_kobj);
2597         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2598
2599         return a->show ? a->show(a, sbi, buf) : 0;
2600 }
2601
2602 static ssize_t ext4_attr_store(struct kobject *kobj,
2603                                struct attribute *attr,
2604                                const char *buf, size_t len)
2605 {
2606         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2607                                                 s_kobj);
2608         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2609
2610         return a->store ? a->store(a, sbi, buf, len) : 0;
2611 }
2612
2613 static void ext4_sb_release(struct kobject *kobj)
2614 {
2615         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2616                                                 s_kobj);
2617         complete(&sbi->s_kobj_unregister);
2618 }
2619
2620 static const struct sysfs_ops ext4_attr_ops = {
2621         .show   = ext4_attr_show,
2622         .store  = ext4_attr_store,
2623 };
2624
2625 static struct kobj_type ext4_ktype = {
2626         .default_attrs  = ext4_attrs,
2627         .sysfs_ops      = &ext4_attr_ops,
2628         .release        = ext4_sb_release,
2629 };
2630
2631 static void ext4_feat_release(struct kobject *kobj)
2632 {
2633         complete(&ext4_feat->f_kobj_unregister);
2634 }
2635
2636 static struct kobj_type ext4_feat_ktype = {
2637         .default_attrs  = ext4_feat_attrs,
2638         .sysfs_ops      = &ext4_attr_ops,
2639         .release        = ext4_feat_release,
2640 };
2641
2642 /*
2643  * Check whether this filesystem can be mounted based on
2644  * the features present and the RDONLY/RDWR mount requested.
2645  * Returns 1 if this filesystem can be mounted as requested,
2646  * 0 if it cannot be.
2647  */
2648 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2649 {
2650         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2651                 ext4_msg(sb, KERN_ERR,
2652                         "Couldn't mount because of "
2653                         "unsupported optional features (%x)",
2654                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2655                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2656                 return 0;
2657         }
2658
2659         if (readonly)
2660                 return 1;
2661
2662         /* Check that feature set is OK for a read-write mount */
2663         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2664                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2665                          "unsupported optional features (%x)",
2666                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2667                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2668                 return 0;
2669         }
2670         /*
2671          * Large file size enabled file system can only be mounted
2672          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2673          */
2674         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2675                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2676                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2677                                  "cannot be mounted RDWR without "
2678                                  "CONFIG_LBDAF");
2679                         return 0;
2680                 }
2681         }
2682         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2683             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2684                 ext4_msg(sb, KERN_ERR,
2685                          "Can't support bigalloc feature without "
2686                          "extents feature\n");
2687                 return 0;
2688         }
2689         return 1;
2690 }
2691
2692 /*
2693  * This function is called once a day if we have errors logged
2694  * on the file system
2695  */
2696 static void print_daily_error_info(unsigned long arg)
2697 {
2698         struct super_block *sb = (struct super_block *) arg;
2699         struct ext4_sb_info *sbi;
2700         struct ext4_super_block *es;
2701
2702         sbi = EXT4_SB(sb);
2703         es = sbi->s_es;
2704
2705         if (es->s_error_count)
2706                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2707                          le32_to_cpu(es->s_error_count));
2708         if (es->s_first_error_time) {
2709                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2710                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2711                        (int) sizeof(es->s_first_error_func),
2712                        es->s_first_error_func,
2713                        le32_to_cpu(es->s_first_error_line));
2714                 if (es->s_first_error_ino)
2715                         printk(": inode %u",
2716                                le32_to_cpu(es->s_first_error_ino));
2717                 if (es->s_first_error_block)
2718                         printk(": block %llu", (unsigned long long)
2719                                le64_to_cpu(es->s_first_error_block));
2720                 printk("\n");
2721         }
2722         if (es->s_last_error_time) {
2723                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2724                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2725                        (int) sizeof(es->s_last_error_func),
2726                        es->s_last_error_func,
2727                        le32_to_cpu(es->s_last_error_line));
2728                 if (es->s_last_error_ino)
2729                         printk(": inode %u",
2730                                le32_to_cpu(es->s_last_error_ino));
2731                 if (es->s_last_error_block)
2732                         printk(": block %llu", (unsigned long long)
2733                                le64_to_cpu(es->s_last_error_block));
2734                 printk("\n");
2735         }
2736         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2737 }
2738
2739 /* Find next suitable group and run ext4_init_inode_table */
2740 static int ext4_run_li_request(struct ext4_li_request *elr)
2741 {
2742         struct ext4_group_desc *gdp = NULL;
2743         ext4_group_t group, ngroups;
2744         struct super_block *sb;
2745         unsigned long timeout = 0;
2746         int ret = 0;
2747
2748         sb = elr->lr_super;
2749         ngroups = EXT4_SB(sb)->s_groups_count;
2750
2751         for (group = elr->lr_next_group; group < ngroups; group++) {
2752                 gdp = ext4_get_group_desc(sb, group, NULL);
2753                 if (!gdp) {
2754                         ret = 1;
2755                         break;
2756                 }
2757
2758                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2759                         break;
2760         }
2761
2762         if (group == ngroups)
2763                 ret = 1;
2764
2765         if (!ret) {
2766                 timeout = jiffies;
2767                 ret = ext4_init_inode_table(sb, group,
2768                                             elr->lr_timeout ? 0 : 1);
2769                 if (elr->lr_timeout == 0) {
2770                         timeout = (jiffies - timeout) *
2771                                   elr->lr_sbi->s_li_wait_mult;
2772                         elr->lr_timeout = timeout;
2773                 }
2774                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2775                 elr->lr_next_group = group + 1;
2776         }
2777
2778         return ret;
2779 }
2780
2781 /*
2782  * Remove lr_request from the list_request and free the
2783  * request structure. Should be called with li_list_mtx held
2784  */
2785 static void ext4_remove_li_request(struct ext4_li_request *elr)
2786 {
2787         struct ext4_sb_info *sbi;
2788
2789         if (!elr)
2790                 return;
2791
2792         sbi = elr->lr_sbi;
2793
2794         list_del(&elr->lr_request);
2795         sbi->s_li_request = NULL;
2796         kfree(elr);
2797 }
2798
2799 static void ext4_unregister_li_request(struct super_block *sb)
2800 {
2801         mutex_lock(&ext4_li_mtx);
2802         if (!ext4_li_info) {
2803                 mutex_unlock(&ext4_li_mtx);
2804                 return;
2805         }
2806
2807         mutex_lock(&ext4_li_info->li_list_mtx);
2808         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2809         mutex_unlock(&ext4_li_info->li_list_mtx);
2810         mutex_unlock(&ext4_li_mtx);
2811 }
2812
2813 static struct task_struct *ext4_lazyinit_task;
2814
2815 /*
2816  * This is the function where ext4lazyinit thread lives. It walks
2817  * through the request list searching for next scheduled filesystem.
2818  * When such a fs is found, run the lazy initialization request
2819  * (ext4_rn_li_request) and keep track of the time spend in this
2820  * function. Based on that time we compute next schedule time of
2821  * the request. When walking through the list is complete, compute
2822  * next waking time and put itself into sleep.
2823  */
2824 static int ext4_lazyinit_thread(void *arg)
2825 {
2826         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2827         struct list_head *pos, *n;
2828         struct ext4_li_request *elr;
2829         unsigned long next_wakeup, cur;
2830
2831         BUG_ON(NULL == eli);
2832
2833 cont_thread:
2834         while (true) {
2835                 next_wakeup = MAX_JIFFY_OFFSET;
2836
2837                 mutex_lock(&eli->li_list_mtx);
2838                 if (list_empty(&eli->li_request_list)) {
2839                         mutex_unlock(&eli->li_list_mtx);
2840                         goto exit_thread;
2841                 }
2842
2843                 list_for_each_safe(pos, n, &eli->li_request_list) {
2844                         elr = list_entry(pos, struct ext4_li_request,
2845                                          lr_request);
2846
2847                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2848                                 if (ext4_run_li_request(elr) != 0) {
2849                                         /* error, remove the lazy_init job */
2850                                         ext4_remove_li_request(elr);
2851                                         continue;
2852                                 }
2853                         }
2854
2855                         if (time_before(elr->lr_next_sched, next_wakeup))
2856                                 next_wakeup = elr->lr_next_sched;
2857                 }
2858                 mutex_unlock(&eli->li_list_mtx);
2859
2860                 try_to_freeze();
2861
2862                 cur = jiffies;
2863                 if ((time_after_eq(cur, next_wakeup)) ||
2864                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2865                         cond_resched();
2866                         continue;
2867                 }
2868
2869                 schedule_timeout_interruptible(next_wakeup - cur);
2870
2871                 if (kthread_should_stop()) {
2872                         ext4_clear_request_list();
2873                         goto exit_thread;
2874                 }
2875         }
2876
2877 exit_thread:
2878         /*
2879          * It looks like the request list is empty, but we need
2880          * to check it under the li_list_mtx lock, to prevent any
2881          * additions into it, and of course we should lock ext4_li_mtx
2882          * to atomically free the list and ext4_li_info, because at
2883          * this point another ext4 filesystem could be registering
2884          * new one.
2885          */
2886         mutex_lock(&ext4_li_mtx);
2887         mutex_lock(&eli->li_list_mtx);
2888         if (!list_empty(&eli->li_request_list)) {
2889                 mutex_unlock(&eli->li_list_mtx);
2890                 mutex_unlock(&ext4_li_mtx);
2891                 goto cont_thread;
2892         }
2893         mutex_unlock(&eli->li_list_mtx);
2894         kfree(ext4_li_info);
2895         ext4_li_info = NULL;
2896         mutex_unlock(&ext4_li_mtx);
2897
2898         return 0;
2899 }
2900
2901 static void ext4_clear_request_list(void)
2902 {
2903         struct list_head *pos, *n;
2904         struct ext4_li_request *elr;
2905
2906         mutex_lock(&ext4_li_info->li_list_mtx);
2907         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2908                 elr = list_entry(pos, struct ext4_li_request,
2909                                  lr_request);
2910                 ext4_remove_li_request(elr);
2911         }
2912         mutex_unlock(&ext4_li_info->li_list_mtx);
2913 }
2914
2915 static int ext4_run_lazyinit_thread(void)
2916 {
2917         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2918                                          ext4_li_info, "ext4lazyinit");
2919         if (IS_ERR(ext4_lazyinit_task)) {
2920                 int err = PTR_ERR(ext4_lazyinit_task);
2921                 ext4_clear_request_list();
2922                 kfree(ext4_li_info);
2923                 ext4_li_info = NULL;
2924                 printk(KERN_CRIT "EXT4: error %d creating inode table "
2925                                  "initialization thread\n",
2926                                  err);
2927                 return err;
2928         }
2929         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2930         return 0;
2931 }
2932
2933 /*
2934  * Check whether it make sense to run itable init. thread or not.
2935  * If there is at least one uninitialized inode table, return
2936  * corresponding group number, else the loop goes through all
2937  * groups and return total number of groups.
2938  */
2939 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2940 {
2941         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2942         struct ext4_group_desc *gdp = NULL;
2943
2944         for (group = 0; group < ngroups; group++) {
2945                 gdp = ext4_get_group_desc(sb, group, NULL);
2946                 if (!gdp)
2947                         continue;
2948
2949                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2950                         break;
2951         }
2952
2953         return group;
2954 }
2955
2956 static int ext4_li_info_new(void)
2957 {
2958         struct ext4_lazy_init *eli = NULL;
2959
2960         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2961         if (!eli)
2962                 return -ENOMEM;
2963
2964         INIT_LIST_HEAD(&eli->li_request_list);
2965         mutex_init(&eli->li_list_mtx);
2966
2967         eli->li_state |= EXT4_LAZYINIT_QUIT;
2968
2969         ext4_li_info = eli;
2970
2971         return 0;
2972 }
2973
2974 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2975                                             ext4_group_t start)
2976 {
2977         struct ext4_sb_info *sbi = EXT4_SB(sb);
2978         struct ext4_li_request *elr;
2979         unsigned long rnd;
2980
2981         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2982         if (!elr)
2983                 return NULL;
2984
2985         elr->lr_super = sb;
2986         elr->lr_sbi = sbi;
2987         elr->lr_next_group = start;
2988
2989         /*
2990          * Randomize first schedule time of the request to
2991          * spread the inode table initialization requests
2992          * better.
2993          */
2994         get_random_bytes(&rnd, sizeof(rnd));
2995         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2996                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2997
2998         return elr;
2999 }
3000
3001 static int ext4_register_li_request(struct super_block *sb,
3002                                     ext4_group_t first_not_zeroed)
3003 {
3004         struct ext4_sb_info *sbi = EXT4_SB(sb);
3005         struct ext4_li_request *elr;
3006         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3007         int ret = 0;
3008
3009         if (sbi->s_li_request != NULL) {
3010                 /*
3011                  * Reset timeout so it can be computed again, because
3012                  * s_li_wait_mult might have changed.
3013                  */
3014                 sbi->s_li_request->lr_timeout = 0;
3015                 return 0;
3016         }
3017
3018         if (first_not_zeroed == ngroups ||
3019             (sb->s_flags & MS_RDONLY) ||
3020             !test_opt(sb, INIT_INODE_TABLE))
3021                 return 0;
3022
3023         elr = ext4_li_request_new(sb, first_not_zeroed);
3024         if (!elr)
3025                 return -ENOMEM;
3026
3027         mutex_lock(&ext4_li_mtx);
3028
3029         if (NULL == ext4_li_info) {
3030                 ret = ext4_li_info_new();
3031                 if (ret)
3032                         goto out;
3033         }
3034
3035         mutex_lock(&ext4_li_info->li_list_mtx);
3036         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3037         mutex_unlock(&ext4_li_info->li_list_mtx);
3038
3039         sbi->s_li_request = elr;
3040         /*
3041          * set elr to NULL here since it has been inserted to
3042          * the request_list and the removal and free of it is
3043          * handled by ext4_clear_request_list from now on.
3044          */
3045         elr = NULL;
3046
3047         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3048                 ret = ext4_run_lazyinit_thread();
3049                 if (ret)
3050                         goto out;
3051         }
3052 out:
3053         mutex_unlock(&ext4_li_mtx);
3054         if (ret)
3055                 kfree(elr);
3056         return ret;
3057 }
3058
3059 /*
3060  * We do not need to lock anything since this is called on
3061  * module unload.
3062  */
3063 static void ext4_destroy_lazyinit_thread(void)
3064 {
3065         /*
3066          * If thread exited earlier
3067          * there's nothing to be done.
3068          */
3069         if (!ext4_li_info || !ext4_lazyinit_task)
3070                 return;
3071
3072         kthread_stop(ext4_lazyinit_task);
3073 }
3074
3075 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3076 {
3077         char *orig_data = kstrdup(data, GFP_KERNEL);
3078         struct buffer_head *bh;
3079         struct ext4_super_block *es = NULL;
3080         struct ext4_sb_info *sbi;
3081         ext4_fsblk_t block;
3082         ext4_fsblk_t sb_block = get_sb_block(&data);
3083         ext4_fsblk_t logical_sb_block;
3084         unsigned long offset = 0;
3085         unsigned long journal_devnum = 0;
3086         unsigned long def_mount_opts;
3087         struct inode *root;
3088         char *cp;
3089         const char *descr;
3090         int ret = -ENOMEM;
3091         int blocksize, clustersize;
3092         unsigned int db_count;
3093         unsigned int i;
3094         int needs_recovery, has_huge_files, has_bigalloc;
3095         __u64 blocks_count;
3096         int err;
3097         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3098         ext4_group_t first_not_zeroed;
3099
3100         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3101         if (!sbi)
3102                 goto out_free_orig;
3103
3104         sbi->s_blockgroup_lock =
3105                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3106         if (!sbi->s_blockgroup_lock) {
3107                 kfree(sbi);
3108                 goto out_free_orig;
3109         }
3110         sb->s_fs_info = sbi;
3111         sbi->s_mount_opt = 0;
3112         sbi->s_resuid = EXT4_DEF_RESUID;
3113         sbi->s_resgid = EXT4_DEF_RESGID;
3114         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3115         sbi->s_sb_block = sb_block;
3116         if (sb->s_bdev->bd_part)
3117                 sbi->s_sectors_written_start =
3118                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3119
3120         /* Cleanup superblock name */
3121         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3122                 *cp = '!';
3123
3124         ret = -EINVAL;
3125         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3126         if (!blocksize) {
3127                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3128                 goto out_fail;
3129         }
3130
3131         /*
3132          * The ext4 superblock will not be buffer aligned for other than 1kB
3133          * block sizes.  We need to calculate the offset from buffer start.
3134          */
3135         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3136                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3137                 offset = do_div(logical_sb_block, blocksize);
3138         } else {
3139                 logical_sb_block = sb_block;
3140         }
3141
3142         if (!(bh = sb_bread(sb, logical_sb_block))) {
3143                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3144                 goto out_fail;
3145         }
3146         /*
3147          * Note: s_es must be initialized as soon as possible because
3148          *       some ext4 macro-instructions depend on its value
3149          */
3150         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3151         sbi->s_es = es;
3152         sb->s_magic = le16_to_cpu(es->s_magic);
3153         if (sb->s_magic != EXT4_SUPER_MAGIC)
3154                 goto cantfind_ext4;
3155         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3156
3157         /* Set defaults before we parse the mount options */
3158         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3159         set_opt(sb, INIT_INODE_TABLE);
3160         if (def_mount_opts & EXT4_DEFM_DEBUG)
3161                 set_opt(sb, DEBUG);
3162         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3163                 set_opt(sb, GRPID);
3164         if (def_mount_opts & EXT4_DEFM_UID16)
3165                 set_opt(sb, NO_UID32);
3166         /* xattr user namespace & acls are now defaulted on */
3167 #ifdef CONFIG_EXT4_FS_XATTR
3168         set_opt(sb, XATTR_USER);
3169 #endif
3170 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3171         set_opt(sb, POSIX_ACL);
3172 #endif
3173         set_opt(sb, MBLK_IO_SUBMIT);
3174         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3175                 set_opt(sb, JOURNAL_DATA);
3176         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3177                 set_opt(sb, ORDERED_DATA);
3178         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3179                 set_opt(sb, WRITEBACK_DATA);
3180
3181         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3182                 set_opt(sb, ERRORS_PANIC);
3183         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3184                 set_opt(sb, ERRORS_CONT);
3185         else
3186                 set_opt(sb, ERRORS_RO);
3187         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3188                 set_opt(sb, BLOCK_VALIDITY);
3189         if (def_mount_opts & EXT4_DEFM_DISCARD)
3190                 set_opt(sb, DISCARD);
3191
3192         sbi->s_resuid = le16_to_cpu(es->s_def_resuid);
3193         sbi->s_resgid = le16_to_cpu(es->s_def_resgid);
3194         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3195         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3196         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3197
3198         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3199                 set_opt(sb, BARRIER);
3200
3201         /*
3202          * enable delayed allocation by default
3203          * Use -o nodelalloc to turn it off
3204          */
3205         if (!IS_EXT3_SB(sb) &&
3206             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3207                 set_opt(sb, DELALLOC);
3208
3209         /*
3210          * set default s_li_wait_mult for lazyinit, for the case there is
3211          * no mount option specified.
3212          */
3213         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3214
3215         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3216                            &journal_devnum, &journal_ioprio, 0)) {
3217                 ext4_msg(sb, KERN_WARNING,
3218                          "failed to parse options in superblock: %s",
3219                          sbi->s_es->s_mount_opts);
3220         }
3221         if (!parse_options((char *) data, sb, &journal_devnum,
3222                            &journal_ioprio, 0))
3223                 goto failed_mount;
3224
3225         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3226                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3227                             "with data=journal disables delayed "
3228                             "allocation and O_DIRECT support!\n");
3229                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3230                         ext4_msg(sb, KERN_ERR, "can't mount with "
3231                                  "both data=journal and delalloc");
3232                         goto failed_mount;
3233                 }
3234                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3235                         ext4_msg(sb, KERN_ERR, "can't mount with "
3236                                  "both data=journal and delalloc");
3237                         goto failed_mount;
3238                 }
3239                 if (test_opt(sb, DELALLOC))
3240                         clear_opt(sb, DELALLOC);
3241         }
3242
3243         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3244         if (test_opt(sb, DIOREAD_NOLOCK)) {
3245                 if (blocksize < PAGE_SIZE) {
3246                         ext4_msg(sb, KERN_ERR, "can't mount with "
3247                                  "dioread_nolock if block size != PAGE_SIZE");
3248                         goto failed_mount;
3249                 }
3250         }
3251
3252         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3253                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3254
3255         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3256             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3257              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3258              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3259                 ext4_msg(sb, KERN_WARNING,
3260                        "feature flags set on rev 0 fs, "
3261                        "running e2fsck is recommended");
3262
3263         if (IS_EXT2_SB(sb)) {
3264                 if (ext2_feature_set_ok(sb))
3265                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3266                                  "using the ext4 subsystem");
3267                 else {
3268                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3269                                  "to feature incompatibilities");
3270                         goto failed_mount;
3271                 }
3272         }
3273
3274         if (IS_EXT3_SB(sb)) {
3275                 if (ext3_feature_set_ok(sb))
3276                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3277                                  "using the ext4 subsystem");
3278                 else {
3279                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3280                                  "to feature incompatibilities");
3281                         goto failed_mount;
3282                 }
3283         }
3284
3285         /*
3286          * Check feature flags regardless of the revision level, since we
3287          * previously didn't change the revision level when setting the flags,
3288          * so there is a chance incompat flags are set on a rev 0 filesystem.
3289          */
3290         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3291                 goto failed_mount;
3292
3293         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3294             blocksize > EXT4_MAX_BLOCK_SIZE) {
3295                 ext4_msg(sb, KERN_ERR,
3296                        "Unsupported filesystem blocksize %d", blocksize);
3297                 goto failed_mount;
3298         }
3299
3300         if (sb->s_blocksize != blocksize) {
3301                 /* Validate the filesystem blocksize */
3302                 if (!sb_set_blocksize(sb, blocksize)) {
3303                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3304                                         blocksize);
3305                         goto failed_mount;
3306                 }
3307
3308                 brelse(bh);
3309                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3310                 offset = do_div(logical_sb_block, blocksize);
3311                 bh = sb_bread(sb, logical_sb_block);
3312                 if (!bh) {
3313                         ext4_msg(sb, KERN_ERR,
3314                                "Can't read superblock on 2nd try");
3315                         goto failed_mount;
3316                 }
3317                 es = (struct ext4_super_block *)(((char *)bh->b_data) + offset);
3318                 sbi->s_es = es;
3319                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3320                         ext4_msg(sb, KERN_ERR,
3321                                "Magic mismatch, very weird!");
3322                         goto failed_mount;
3323                 }
3324         }
3325
3326         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3327                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3328         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3329                                                       has_huge_files);
3330         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3331
3332         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3333                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3334                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3335         } else {
3336                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3337                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3338                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3339                     (!is_power_of_2(sbi->s_inode_size)) ||
3340                     (sbi->s_inode_size > blocksize)) {
3341                         ext4_msg(sb, KERN_ERR,
3342                                "unsupported inode size: %d",
3343                                sbi->s_inode_size);
3344                         goto failed_mount;
3345                 }
3346                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3347                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3348         }
3349
3350         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3351         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3352                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3353                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3354                     !is_power_of_2(sbi->s_desc_size)) {
3355                         ext4_msg(sb, KERN_ERR,
3356                                "unsupported descriptor size %lu",
3357                                sbi->s_desc_size);
3358                         goto failed_mount;
3359                 }
3360         } else
3361                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3362
3363         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3364         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3365         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3366                 goto cantfind_ext4;
3367
3368         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3369         if (sbi->s_inodes_per_block == 0)
3370                 goto cantfind_ext4;
3371         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3372                                         sbi->s_inodes_per_block;
3373         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3374         sbi->s_sbh = bh;
3375         sbi->s_mount_state = le16_to_cpu(es->s_state);
3376         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3377         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3378
3379         for (i = 0; i < 4; i++)
3380                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3381         sbi->s_def_hash_version = es->s_def_hash_version;
3382         i = le32_to_cpu(es->s_flags);
3383         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3384                 sbi->s_hash_unsigned = 3;
3385         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3386 #ifdef __CHAR_UNSIGNED__
3387                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3388                 sbi->s_hash_unsigned = 3;
3389 #else
3390                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3391 #endif
3392                 sb->s_dirt = 1;
3393         }
3394
3395         /* Handle clustersize */
3396         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3397         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3398                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3399         if (has_bigalloc) {
3400                 if (clustersize < blocksize) {
3401                         ext4_msg(sb, KERN_ERR,
3402                                  "cluster size (%d) smaller than "
3403                                  "block size (%d)", clustersize, blocksize);
3404                         goto failed_mount;
3405                 }
3406                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3407                         le32_to_cpu(es->s_log_block_size);
3408                 sbi->s_clusters_per_group =
3409                         le32_to_cpu(es->s_clusters_per_group);
3410                 if (sbi->s_clusters_per_group > blocksize * 8) {
3411                         ext4_msg(sb, KERN_ERR,
3412                                  "#clusters per group too big: %lu",
3413                                  sbi->s_clusters_per_group);
3414                         goto failed_mount;
3415                 }
3416                 if (sbi->s_blocks_per_group !=
3417                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3418                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3419                                  "clusters per group (%lu) inconsistent",
3420                                  sbi->s_blocks_per_group,
3421                                  sbi->s_clusters_per_group);
3422                         goto failed_mount;
3423                 }
3424         } else {
3425                 if (clustersize != blocksize) {
3426                         ext4_warning(sb, "fragment/cluster size (%d) != "
3427                                      "block size (%d)", clustersize,
3428                                      blocksize);
3429                         clustersize = blocksize;
3430                 }
3431                 if (sbi->s_blocks_per_group > blocksize * 8) {
3432                         ext4_msg(sb, KERN_ERR,
3433                                  "#blocks per group too big: %lu",
3434                                  sbi->s_blocks_per_group);
3435                         goto failed_mount;
3436                 }
3437                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3438                 sbi->s_cluster_bits = 0;
3439         }
3440         sbi->s_cluster_ratio = clustersize / blocksize;
3441
3442         if (sbi->s_inodes_per_group > blocksize * 8) {
3443                 ext4_msg(sb, KERN_ERR,
3444                        "#inodes per group too big: %lu",
3445                        sbi->s_inodes_per_group);
3446                 goto failed_mount;
3447         }
3448
3449         /*
3450          * Test whether we have more sectors than will fit in sector_t,
3451          * and whether the max offset is addressable by the page cache.
3452          */
3453         err = generic_check_addressable(sb->s_blocksize_bits,
3454                                         ext4_blocks_count(es));
3455         if (err) {
3456                 ext4_msg(sb, KERN_ERR, "filesystem"
3457                          " too large to mount safely on this system");
3458                 if (sizeof(sector_t) < 8)
3459                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3460                 ret = err;
3461                 goto failed_mount;
3462         }
3463
3464         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3465                 goto cantfind_ext4;
3466
3467         /* check blocks count against device size */
3468         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3469         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3470                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3471                        "exceeds size of device (%llu blocks)",
3472                        ext4_blocks_count(es), blocks_count);
3473                 goto failed_mount;
3474         }
3475
3476         /*
3477          * It makes no sense for the first data block to be beyond the end
3478          * of the filesystem.
3479          */
3480         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3481                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3482                          "block %u is beyond end of filesystem (%llu)",
3483                          le32_to_cpu(es->s_first_data_block),
3484                          ext4_blocks_count(es));
3485                 goto failed_mount;
3486         }
3487         blocks_count = (ext4_blocks_count(es) -
3488                         le32_to_cpu(es->s_first_data_block) +
3489                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3490         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3491         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3492                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3493                        "(block count %llu, first data block %u, "
3494                        "blocks per group %lu)", sbi->s_groups_count,
3495                        ext4_blocks_count(es),
3496                        le32_to_cpu(es->s_first_data_block),
3497                        EXT4_BLOCKS_PER_GROUP(sb));
3498                 goto failed_mount;
3499         }
3500         sbi->s_groups_count = blocks_count;
3501         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3502                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3503         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3504                    EXT4_DESC_PER_BLOCK(sb);
3505         sbi->s_group_desc = ext4_kvmalloc(db_count *
3506                                           sizeof(struct buffer_head *),
3507                                           GFP_KERNEL);
3508         if (sbi->s_group_desc == NULL) {
3509                 ext4_msg(sb, KERN_ERR, "not enough memory");
3510                 goto failed_mount;
3511         }
3512
3513         if (ext4_proc_root)
3514                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3515
3516         bgl_lock_init(sbi->s_blockgroup_lock);
3517
3518         for (i = 0; i < db_count; i++) {
3519                 block = descriptor_loc(sb, logical_sb_block, i);
3520                 sbi->s_group_desc[i] = sb_bread(sb, block);
3521                 if (!sbi->s_group_desc[i]) {
3522                         ext4_msg(sb, KERN_ERR,
3523                                "can't read group descriptor %d", i);
3524                         db_count = i;
3525                         goto failed_mount2;
3526                 }
3527         }
3528         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3529                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3530                 goto failed_mount2;
3531         }
3532         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3533                 if (!ext4_fill_flex_info(sb)) {
3534                         ext4_msg(sb, KERN_ERR,
3535                                "unable to initialize "
3536                                "flex_bg meta info!");
3537                         goto failed_mount2;
3538                 }
3539
3540         sbi->s_gdb_count = db_count;
3541         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3542         spin_lock_init(&sbi->s_next_gen_lock);
3543
3544         init_timer(&sbi->s_err_report);
3545         sbi->s_err_report.function = print_daily_error_info;
3546         sbi->s_err_report.data = (unsigned long) sb;
3547
3548         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3549                         ext4_count_free_clusters(sb));
3550         if (!err) {
3551                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3552                                 ext4_count_free_inodes(sb));
3553         }
3554         if (!err) {
3555                 err = percpu_counter_init(&sbi->s_dirs_counter,
3556                                 ext4_count_dirs(sb));
3557         }
3558         if (!err) {
3559                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3560         }
3561         if (err) {
3562                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3563                 goto failed_mount3;
3564         }
3565
3566         sbi->s_stripe = ext4_get_stripe_size(sbi);
3567         sbi->s_max_writeback_mb_bump = 128;
3568
3569         /*
3570          * set up enough so that it can read an inode
3571          */
3572         if (!test_opt(sb, NOLOAD) &&
3573             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3574                 sb->s_op = &ext4_sops;
3575         else
3576                 sb->s_op = &ext4_nojournal_sops;
3577         sb->s_export_op = &ext4_export_ops;
3578         sb->s_xattr = ext4_xattr_handlers;
3579 #ifdef CONFIG_QUOTA
3580         sb->s_qcop = &ext4_qctl_operations;
3581         sb->dq_op = &ext4_quota_operations;
3582 #endif
3583         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3584
3585         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3586         mutex_init(&sbi->s_orphan_lock);
3587         sbi->s_resize_flags = 0;
3588
3589         sb->s_root = NULL;
3590
3591         needs_recovery = (es->s_last_orphan != 0 ||
3592                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3593                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3594
3595         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3596             !(sb->s_flags & MS_RDONLY))
3597                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3598                         goto failed_mount3;
3599
3600         /*
3601          * The first inode we look at is the journal inode.  Don't try
3602          * root first: it may be modified in the journal!
3603          */
3604         if (!test_opt(sb, NOLOAD) &&
3605             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3606                 if (ext4_load_journal(sb, es, journal_devnum))
3607                         goto failed_mount3;
3608         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3609               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3610                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3611                        "suppressed and not mounted read-only");
3612                 goto failed_mount_wq;
3613         } else {
3614                 clear_opt(sb, DATA_FLAGS);
3615                 sbi->s_journal = NULL;
3616                 needs_recovery = 0;
3617                 goto no_journal;
3618         }
3619
3620         if (ext4_blocks_count(es) > 0xffffffffULL &&
3621             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3622                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3623                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3624                 goto failed_mount_wq;
3625         }
3626
3627         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3628                 jbd2_journal_set_features(sbi->s_journal,
3629                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3630                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3631         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3632                 jbd2_journal_set_features(sbi->s_journal,
3633                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0, 0);
3634                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3635                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3636         } else {
3637                 jbd2_journal_clear_features(sbi->s_journal,
3638                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3639                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3640         }
3641
3642         /* We have now updated the journal if required, so we can
3643          * validate the data journaling mode. */
3644         switch (test_opt(sb, DATA_FLAGS)) {
3645         case 0:
3646                 /* No mode set, assume a default based on the journal
3647                  * capabilities: ORDERED_DATA if the journal can
3648                  * cope, else JOURNAL_DATA
3649                  */
3650                 if (jbd2_journal_check_available_features
3651                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3652                         set_opt(sb, ORDERED_DATA);
3653                 else
3654                         set_opt(sb, JOURNAL_DATA);
3655                 break;
3656
3657         case EXT4_MOUNT_ORDERED_DATA:
3658         case EXT4_MOUNT_WRITEBACK_DATA:
3659                 if (!jbd2_journal_check_available_features
3660                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3661                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3662                                "requested data journaling mode");
3663                         goto failed_mount_wq;
3664                 }
3665         default:
3666                 break;
3667         }
3668         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3669
3670         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3671
3672         /*
3673          * The journal may have updated the bg summary counts, so we
3674          * need to update the global counters.
3675          */
3676         percpu_counter_set(&sbi->s_freeclusters_counter,
3677                            ext4_count_free_clusters(sb));
3678         percpu_counter_set(&sbi->s_freeinodes_counter,
3679                            ext4_count_free_inodes(sb));
3680         percpu_counter_set(&sbi->s_dirs_counter,
3681                            ext4_count_dirs(sb));
3682         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3683
3684 no_journal:
3685         /*
3686          * The maximum number of concurrent works can be high and
3687          * concurrency isn't really necessary.  Limit it to 1.
3688          */
3689         EXT4_SB(sb)->dio_unwritten_wq =
3690                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3691         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3692                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3693                 goto failed_mount_wq;
3694         }
3695
3696         /*
3697          * The jbd2_journal_load will have done any necessary log recovery,
3698          * so we can safely mount the rest of the filesystem now.
3699          */
3700
3701         root = ext4_iget(sb, EXT4_ROOT_INO);
3702         if (IS_ERR(root)) {
3703                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3704                 ret = PTR_ERR(root);
3705                 root = NULL;
3706                 goto failed_mount4;
3707         }
3708         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3709                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3710                 iput(root);
3711                 goto failed_mount4;
3712         }
3713         sb->s_root = d_alloc_root(root);
3714         if (!sb->s_root) {
3715                 iput(root);
3716                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3717                 ret = -ENOMEM;
3718                 goto failed_mount4;
3719         }
3720
3721         ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY);
3722
3723         /* determine the minimum size of new large inodes, if present */
3724         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3725                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3726                                                      EXT4_GOOD_OLD_INODE_SIZE;
3727                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3728                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3729                         if (sbi->s_want_extra_isize <
3730                             le16_to_cpu(es->s_want_extra_isize))
3731                                 sbi->s_want_extra_isize =
3732                                         le16_to_cpu(es->s_want_extra_isize);
3733                         if (sbi->s_want_extra_isize <
3734                             le16_to_cpu(es->s_min_extra_isize))
3735                                 sbi->s_want_extra_isize =
3736                                         le16_to_cpu(es->s_min_extra_isize);
3737                 }
3738         }
3739         /* Check if enough inode space is available */
3740         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3741                                                         sbi->s_inode_size) {
3742                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3743                                                        EXT4_GOOD_OLD_INODE_SIZE;
3744                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3745                          "available");
3746         }
3747
3748         err = ext4_setup_system_zone(sb);
3749         if (err) {
3750                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3751                          "zone (%d)", err);
3752                 goto failed_mount4a;
3753         }
3754
3755         ext4_ext_init(sb);
3756         err = ext4_mb_init(sb, needs_recovery);
3757         if (err) {
3758                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3759                          err);
3760                 goto failed_mount5;
3761         }
3762
3763         err = ext4_register_li_request(sb, first_not_zeroed);
3764         if (err)
3765                 goto failed_mount6;
3766
3767         sbi->s_kobj.kset = ext4_kset;
3768         init_completion(&sbi->s_kobj_unregister);
3769         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3770                                    "%s", sb->s_id);
3771         if (err)
3772                 goto failed_mount7;
3773
3774         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3775         ext4_orphan_cleanup(sb, es);
3776         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3777         if (needs_recovery) {
3778                 ext4_msg(sb, KERN_INFO, "recovery complete");
3779                 ext4_mark_recovery_complete(sb, es);
3780         }
3781         if (EXT4_SB(sb)->s_journal) {
3782                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3783                         descr = " journalled data mode";
3784                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3785                         descr = " ordered data mode";
3786                 else
3787                         descr = " writeback data mode";
3788         } else
3789                 descr = "out journal";
3790
3791         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3792                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3793                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3794
3795         if (es->s_error_count)
3796                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3797
3798         kfree(orig_data);
3799         return 0;
3800
3801 cantfind_ext4:
3802         if (!silent)
3803                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3804         goto failed_mount;
3805
3806 failed_mount7:
3807         ext4_unregister_li_request(sb);
3808 failed_mount6:
3809         ext4_mb_release(sb);
3810 failed_mount5:
3811         ext4_ext_release(sb);
3812         ext4_release_system_zone(sb);
3813 failed_mount4a:
3814         dput(sb->s_root);
3815         sb->s_root = NULL;
3816 failed_mount4:
3817         ext4_msg(sb, KERN_ERR, "mount failed");
3818         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3819 failed_mount_wq:
3820         if (sbi->s_journal) {
3821                 jbd2_journal_destroy(sbi->s_journal);
3822                 sbi->s_journal = NULL;
3823         }
3824 failed_mount3:
3825         del_timer(&sbi->s_err_report);
3826         if (sbi->s_flex_groups)
3827                 ext4_kvfree(sbi->s_flex_groups);
3828         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3829         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3830         percpu_counter_destroy(&sbi->s_dirs_counter);
3831         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3832         if (sbi->s_mmp_tsk)
3833                 kthread_stop(sbi->s_mmp_tsk);
3834 failed_mount2:
3835         for (i = 0; i < db_count; i++)
3836                 brelse(sbi->s_group_desc[i]);
3837         ext4_kvfree(sbi->s_group_desc);
3838 failed_mount:
3839         if (sbi->s_proc) {
3840                 remove_proc_entry(sb->s_id, ext4_proc_root);
3841         }
3842 #ifdef CONFIG_QUOTA
3843         for (i = 0; i < MAXQUOTAS; i++)
3844                 kfree(sbi->s_qf_names[i]);
3845 #endif
3846         ext4_blkdev_remove(sbi);
3847         brelse(bh);
3848 out_fail:
3849         sb->s_fs_info = NULL;
3850         kfree(sbi->s_blockgroup_lock);
3851         kfree(sbi);
3852 out_free_orig:
3853         kfree(orig_data);
3854         return ret;
3855 }
3856
3857 /*
3858  * Setup any per-fs journal parameters now.  We'll do this both on
3859  * initial mount, once the journal has been initialised but before we've
3860  * done any recovery; and again on any subsequent remount.
3861  */
3862 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3863 {
3864         struct ext4_sb_info *sbi = EXT4_SB(sb);
3865
3866         journal->j_commit_interval = sbi->s_commit_interval;
3867         journal->j_min_batch_time = sbi->s_min_batch_time;
3868         journal->j_max_batch_time = sbi->s_max_batch_time;
3869
3870         write_lock(&journal->j_state_lock);
3871         if (test_opt(sb, BARRIER))
3872                 journal->j_flags |= JBD2_BARRIER;
3873         else
3874                 journal->j_flags &= ~JBD2_BARRIER;
3875         if (test_opt(sb, DATA_ERR_ABORT))
3876                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3877         else
3878                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3879         write_unlock(&journal->j_state_lock);
3880 }
3881
3882 static journal_t *ext4_get_journal(struct super_block *sb,
3883                                    unsigned int journal_inum)
3884 {
3885         struct inode *journal_inode;
3886         journal_t *journal;
3887
3888         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3889
3890         /* First, test for the existence of a valid inode on disk.  Bad
3891          * things happen if we iget() an unused inode, as the subsequent
3892          * iput() will try to delete it. */
3893
3894         journal_inode = ext4_iget(sb, journal_inum);
3895         if (IS_ERR(journal_inode)) {
3896                 ext4_msg(sb, KERN_ERR, "no journal found");
3897                 return NULL;
3898         }
3899         if (!journal_inode->i_nlink) {
3900                 make_bad_inode(journal_inode);
3901                 iput(journal_inode);
3902                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3903                 return NULL;
3904         }
3905
3906         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3907                   journal_inode, journal_inode->i_size);
3908         if (!S_ISREG(journal_inode->i_mode)) {
3909                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3910                 iput(journal_inode);
3911                 return NULL;
3912         }
3913
3914         journal = jbd2_journal_init_inode(journal_inode);
3915         if (!journal) {
3916                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3917                 iput(journal_inode);
3918                 return NULL;
3919         }
3920         journal->j_private = sb;
3921         ext4_init_journal_params(sb, journal);
3922         return journal;
3923 }
3924
3925 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3926                                        dev_t j_dev)
3927 {
3928         struct buffer_head *bh;
3929         journal_t *journal;
3930         ext4_fsblk_t start;
3931         ext4_fsblk_t len;
3932         int hblock, blocksize;
3933         ext4_fsblk_t sb_block;
3934         unsigned long offset;
3935         struct ext4_super_block *es;
3936         struct block_device *bdev;
3937
3938         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3939
3940         bdev = ext4_blkdev_get(j_dev, sb);
3941         if (bdev == NULL)
3942                 return NULL;
3943
3944         blocksize = sb->s_blocksize;
3945         hblock = bdev_logical_block_size(bdev);
3946         if (blocksize < hblock) {
3947                 ext4_msg(sb, KERN_ERR,
3948                         "blocksize too small for journal device");
3949                 goto out_bdev;
3950         }
3951
3952         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3953         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3954         set_blocksize(bdev, blocksize);
3955         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3956                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3957                        "external journal");
3958                 goto out_bdev;
3959         }
3960
3961         es = (struct ext4_super_block *) (((char *)bh->b_data) + offset);
3962         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
3963             !(le32_to_cpu(es->s_feature_incompat) &
3964               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
3965                 ext4_msg(sb, KERN_ERR, "external journal has "
3966                                         "bad superblock");
3967                 brelse(bh);
3968                 goto out_bdev;
3969         }
3970
3971         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
3972                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
3973                 brelse(bh);
3974                 goto out_bdev;
3975         }
3976
3977         len = ext4_blocks_count(es);
3978         start = sb_block + 1;
3979         brelse(bh);     /* we're done with the superblock */
3980
3981         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
3982                                         start, len, blocksize);
3983         if (!journal) {
3984                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
3985                 goto out_bdev;
3986         }
3987         journal->j_private = sb;
3988         ll_rw_block(READ, 1, &journal->j_sb_buffer);
3989         wait_on_buffer(journal->j_sb_buffer);
3990         if (!buffer_uptodate(journal->j_sb_buffer)) {
3991                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
3992                 goto out_journal;
3993         }
3994         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
3995                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
3996                                         "user (unsupported) - %d",
3997                         be32_to_cpu(journal->j_superblock->s_nr_users));
3998                 goto out_journal;
3999         }
4000         EXT4_SB(sb)->journal_bdev = bdev;
4001         ext4_init_journal_params(sb, journal);
4002         return journal;
4003
4004 out_journal:
4005         jbd2_journal_destroy(journal);
4006 out_bdev:
4007         ext4_blkdev_put(bdev);
4008         return NULL;
4009 }
4010
4011 static int ext4_load_journal(struct super_block *sb,
4012                              struct ext4_super_block *es,
4013                              unsigned long journal_devnum)
4014 {
4015         journal_t *journal;
4016         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4017         dev_t journal_dev;
4018         int err = 0;
4019         int really_read_only;
4020
4021         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4022
4023         if (journal_devnum &&
4024             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4025                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4026                         "numbers have changed");
4027                 journal_dev = new_decode_dev(journal_devnum);
4028         } else
4029                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4030
4031         really_read_only = bdev_read_only(sb->s_bdev);
4032
4033         /*
4034          * Are we loading a blank journal or performing recovery after a
4035          * crash?  For recovery, we need to check in advance whether we
4036          * can get read-write access to the device.
4037          */
4038         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4039                 if (sb->s_flags & MS_RDONLY) {
4040                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4041                                         "required on readonly filesystem");
4042                         if (really_read_only) {
4043                                 ext4_msg(sb, KERN_ERR, "write access "
4044                                         "unavailable, cannot proceed");
4045                                 return -EROFS;
4046                         }
4047                         ext4_msg(sb, KERN_INFO, "write access will "
4048                                "be enabled during recovery");
4049                 }
4050         }
4051
4052         if (journal_inum && journal_dev) {
4053                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4054                        "and inode journals!");
4055                 return -EINVAL;
4056         }
4057
4058         if (journal_inum) {
4059                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4060                         return -EINVAL;
4061         } else {
4062                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4063                         return -EINVAL;
4064         }
4065
4066         if (!(journal->j_flags & JBD2_BARRIER))
4067                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4068
4069         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4070                 err = jbd2_journal_wipe(journal, !really_read_only);
4071         if (!err) {
4072                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4073                 if (save)
4074                         memcpy(save, ((char *) es) +
4075                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4076                 err = jbd2_journal_load(journal);
4077                 if (save)
4078                         memcpy(((char *) es) + EXT4_S_ERR_START,
4079                                save, EXT4_S_ERR_LEN);
4080                 kfree(save);
4081         }
4082
4083         if (err) {
4084                 ext4_msg(sb, KERN_ERR, "error loading journal");
4085                 jbd2_journal_destroy(journal);
4086                 return err;
4087         }
4088
4089         EXT4_SB(sb)->s_journal = journal;
4090         ext4_clear_journal_err(sb, es);
4091
4092         if (!really_read_only && journal_devnum &&
4093             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4094                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4095
4096                 /* Make sure we flush the recovery flag to disk. */
4097                 ext4_commit_super(sb, 1);
4098         }
4099
4100         return 0;
4101 }
4102
4103 static int ext4_commit_super(struct super_block *sb, int sync)
4104 {
4105         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4106         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4107         int error = 0;
4108
4109         if (!sbh || block_device_ejected(sb))
4110                 return error;
4111         if (buffer_write_io_error(sbh)) {
4112                 /*
4113                  * Oh, dear.  A previous attempt to write the
4114                  * superblock failed.  This could happen because the
4115                  * USB device was yanked out.  Or it could happen to
4116                  * be a transient write error and maybe the block will
4117                  * be remapped.  Nothing we can do but to retry the
4118                  * write and hope for the best.
4119                  */
4120                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4121                        "superblock detected");
4122                 clear_buffer_write_io_error(sbh);
4123                 set_buffer_uptodate(sbh);
4124         }
4125         /*
4126          * If the file system is mounted read-only, don't update the
4127          * superblock write time.  This avoids updating the superblock
4128          * write time when we are mounting the root file system
4129          * read/only but we need to replay the journal; at that point,
4130          * for people who are east of GMT and who make their clock
4131          * tick in localtime for Windows bug-for-bug compatibility,
4132          * the clock is set in the future, and this will cause e2fsck
4133          * to complain and force a full file system check.
4134          */
4135         if (!(sb->s_flags & MS_RDONLY))
4136                 es->s_wtime = cpu_to_le32(get_seconds());
4137         if (sb->s_bdev->bd_part)
4138                 es->s_kbytes_written =
4139                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4140                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4141                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4142         else
4143                 es->s_kbytes_written =
4144                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4145         ext4_free_blocks_count_set(es,
4146                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4147                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4148         es->s_free_inodes_count =
4149                 cpu_to_le32(percpu_counter_sum_positive(
4150                                 &EXT4_SB(sb)->s_freeinodes_counter));
4151         sb->s_dirt = 0;
4152         BUFFER_TRACE(sbh, "marking dirty");
4153         mark_buffer_dirty(sbh);
4154         if (sync) {
4155                 error = sync_dirty_buffer(sbh);
4156                 if (error)
4157                         return error;
4158
4159                 error = buffer_write_io_error(sbh);
4160                 if (error) {
4161                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4162                                "superblock");
4163                         clear_buffer_write_io_error(sbh);
4164                         set_buffer_uptodate(sbh);
4165                 }
4166         }
4167         return error;
4168 }
4169
4170 /*
4171  * Have we just finished recovery?  If so, and if we are mounting (or
4172  * remounting) the filesystem readonly, then we will end up with a
4173  * consistent fs on disk.  Record that fact.
4174  */
4175 static void ext4_mark_recovery_complete(struct super_block *sb,
4176                                         struct ext4_super_block *es)
4177 {
4178         journal_t *journal = EXT4_SB(sb)->s_journal;
4179
4180         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4181                 BUG_ON(journal != NULL);
4182                 return;
4183         }
4184         jbd2_journal_lock_updates(journal);
4185         if (jbd2_journal_flush(journal) < 0)
4186                 goto out;
4187
4188         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4189             sb->s_flags & MS_RDONLY) {
4190                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4191                 ext4_commit_super(sb, 1);
4192         }
4193
4194 out:
4195         jbd2_journal_unlock_updates(journal);
4196 }
4197
4198 /*
4199  * If we are mounting (or read-write remounting) a filesystem whose journal
4200  * has recorded an error from a previous lifetime, move that error to the
4201  * main filesystem now.
4202  */
4203 static void ext4_clear_journal_err(struct super_block *sb,
4204                                    struct ext4_super_block *es)
4205 {
4206         journal_t *journal;
4207         int j_errno;
4208         const char *errstr;
4209
4210         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4211
4212         journal = EXT4_SB(sb)->s_journal;
4213
4214         /*
4215          * Now check for any error status which may have been recorded in the
4216          * journal by a prior ext4_error() or ext4_abort()
4217          */
4218
4219         j_errno = jbd2_journal_errno(journal);
4220         if (j_errno) {
4221                 char nbuf[16];
4222
4223                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4224                 ext4_warning(sb, "Filesystem error recorded "
4225                              "from previous mount: %s", errstr);
4226                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4227
4228                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4229                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4230                 ext4_commit_super(sb, 1);
4231
4232                 jbd2_journal_clear_err(journal);
4233         }
4234 }
4235
4236 /*
4237  * Force the running and committing transactions to commit,
4238  * and wait on the commit.
4239  */
4240 int ext4_force_commit(struct super_block *sb)
4241 {
4242         journal_t *journal;
4243         int ret = 0;
4244
4245         if (sb->s_flags & MS_RDONLY)
4246                 return 0;
4247
4248         journal = EXT4_SB(sb)->s_journal;
4249         if (journal) {
4250                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4251                 ret = ext4_journal_force_commit(journal);
4252         }
4253
4254         return ret;
4255 }
4256
4257 static void ext4_write_super(struct super_block *sb)
4258 {
4259         lock_super(sb);
4260         ext4_commit_super(sb, 1);
4261         unlock_super(sb);
4262 }
4263
4264 static int ext4_sync_fs(struct super_block *sb, int wait)
4265 {
4266         int ret = 0;
4267         tid_t target;
4268         struct ext4_sb_info *sbi = EXT4_SB(sb);
4269
4270         trace_ext4_sync_fs(sb, wait);
4271         flush_workqueue(sbi->dio_unwritten_wq);
4272         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4273                 if (wait)
4274                         jbd2_log_wait_commit(sbi->s_journal, target);
4275         }
4276         return ret;
4277 }
4278
4279 /*
4280  * LVM calls this function before a (read-only) snapshot is created.  This
4281  * gives us a chance to flush the journal completely and mark the fs clean.
4282  *
4283  * Note that only this function cannot bring a filesystem to be in a clean
4284  * state independently, because ext4 prevents a new handle from being started
4285  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4286  * the upper layer.
4287  */
4288 static int ext4_freeze(struct super_block *sb)
4289 {
4290         int error = 0;
4291         journal_t *journal;
4292
4293         if (sb->s_flags & MS_RDONLY)
4294                 return 0;
4295
4296         journal = EXT4_SB(sb)->s_journal;
4297
4298         /* Now we set up the journal barrier. */
4299         jbd2_journal_lock_updates(journal);
4300
4301         /*
4302          * Don't clear the needs_recovery flag if we failed to flush
4303          * the journal.
4304          */
4305         error = jbd2_journal_flush(journal);
4306         if (error < 0)
4307                 goto out;
4308
4309         /* Journal blocked and flushed, clear needs_recovery flag. */
4310         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4311         error = ext4_commit_super(sb, 1);
4312 out:
4313         /* we rely on s_frozen to stop further updates */
4314         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4315         return error;
4316 }
4317
4318 /*
4319  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4320  * flag here, even though the filesystem is not technically dirty yet.
4321  */
4322 static int ext4_unfreeze(struct super_block *sb)
4323 {
4324         if (sb->s_flags & MS_RDONLY)
4325                 return 0;
4326
4327         lock_super(sb);
4328         /* Reset the needs_recovery flag before the fs is unlocked. */
4329         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4330         ext4_commit_super(sb, 1);
4331         unlock_super(sb);
4332         return 0;
4333 }
4334
4335 /*
4336  * Structure to save mount options for ext4_remount's benefit
4337  */
4338 struct ext4_mount_options {
4339         unsigned long s_mount_opt;
4340         unsigned long s_mount_opt2;
4341         uid_t s_resuid;
4342         gid_t s_resgid;
4343         unsigned long s_commit_interval;
4344         u32 s_min_batch_time, s_max_batch_time;
4345 #ifdef CONFIG_QUOTA
4346         int s_jquota_fmt;
4347         char *s_qf_names[MAXQUOTAS];
4348 #endif
4349 };
4350
4351 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4352 {
4353         struct ext4_super_block *es;
4354         struct ext4_sb_info *sbi = EXT4_SB(sb);
4355         unsigned long old_sb_flags;
4356         struct ext4_mount_options old_opts;
4357         int enable_quota = 0;
4358         ext4_group_t g;
4359         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4360         int err = 0;
4361 #ifdef CONFIG_QUOTA
4362         int i;
4363 #endif
4364         char *orig_data = kstrdup(data, GFP_KERNEL);
4365
4366         /* Store the original options */
4367         lock_super(sb);
4368         old_sb_flags = sb->s_flags;
4369         old_opts.s_mount_opt = sbi->s_mount_opt;
4370         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4371         old_opts.s_resuid = sbi->s_resuid;
4372         old_opts.s_resgid = sbi->s_resgid;
4373         old_opts.s_commit_interval = sbi->s_commit_interval;
4374         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4375         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4376 #ifdef CONFIG_QUOTA
4377         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4378         for (i = 0; i < MAXQUOTAS; i++)
4379                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4380 #endif
4381         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4382                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4383
4384         /*
4385          * Allow the "check" option to be passed as a remount option.
4386          */
4387         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4388                 err = -EINVAL;
4389                 goto restore_opts;
4390         }
4391
4392         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4393                 ext4_abort(sb, "Abort forced by user");
4394
4395         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4396                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4397
4398         es = sbi->s_es;
4399
4400         if (sbi->s_journal) {
4401                 ext4_init_journal_params(sb, sbi->s_journal);
4402                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4403         }
4404
4405         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4406                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4407                         err = -EROFS;
4408                         goto restore_opts;
4409                 }
4410
4411                 if (*flags & MS_RDONLY) {
4412                         err = dquot_suspend(sb, -1);
4413                         if (err < 0)
4414                                 goto restore_opts;
4415
4416                         /*
4417                          * First of all, the unconditional stuff we have to do
4418                          * to disable replay of the journal when we next remount
4419                          */
4420                         sb->s_flags |= MS_RDONLY;
4421
4422                         /*
4423                          * OK, test if we are remounting a valid rw partition
4424                          * readonly, and if so set the rdonly flag and then
4425                          * mark the partition as valid again.
4426                          */
4427                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4428                             (sbi->s_mount_state & EXT4_VALID_FS))
4429                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4430
4431                         if (sbi->s_journal)
4432                                 ext4_mark_recovery_complete(sb, es);
4433                 } else {
4434                         /* Make sure we can mount this feature set readwrite */
4435                         if (!ext4_feature_set_ok(sb, 0)) {
4436                                 err = -EROFS;
4437                                 goto restore_opts;
4438                         }
4439                         /*
4440                          * Make sure the group descriptor checksums
4441                          * are sane.  If they aren't, refuse to remount r/w.
4442                          */
4443                         for (g = 0; g < sbi->s_groups_count; g++) {
4444                                 struct ext4_group_desc *gdp =
4445                                         ext4_get_group_desc(sb, g, NULL);
4446
4447                                 if (!ext4_group_desc_csum_verify(sbi, g, gdp)) {
4448                                         ext4_msg(sb, KERN_ERR,
4449                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4450                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4451                                                le16_to_cpu(gdp->bg_checksum));
4452                                         err = -EINVAL;
4453                                         goto restore_opts;
4454                                 }
4455                         }
4456
4457                         /*
4458                          * If we have an unprocessed orphan list hanging
4459                          * around from a previously readonly bdev mount,
4460                          * require a full umount/remount for now.
4461                          */
4462                         if (es->s_last_orphan) {
4463                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4464                                        "remount RDWR because of unprocessed "
4465                                        "orphan inode list.  Please "
4466                                        "umount/remount instead");
4467                                 err = -EINVAL;
4468                                 goto restore_opts;
4469                         }
4470
4471                         /*
4472                          * Mounting a RDONLY partition read-write, so reread
4473                          * and store the current valid flag.  (It may have
4474                          * been changed by e2fsck since we originally mounted
4475                          * the partition.)
4476                          */
4477                         if (sbi->s_journal)
4478                                 ext4_clear_journal_err(sb, es);
4479                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4480                         if (!ext4_setup_super(sb, es, 0))
4481                                 sb->s_flags &= ~MS_RDONLY;
4482                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4483                                                      EXT4_FEATURE_INCOMPAT_MMP))
4484                                 if (ext4_multi_mount_protect(sb,
4485                                                 le64_to_cpu(es->s_mmp_block))) {
4486                                         err = -EROFS;
4487                                         goto restore_opts;
4488                                 }
4489                         enable_quota = 1;
4490                 }
4491         }
4492
4493         /*
4494          * Reinitialize lazy itable initialization thread based on
4495          * current settings
4496          */
4497         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4498                 ext4_unregister_li_request(sb);
4499         else {
4500                 ext4_group_t first_not_zeroed;
4501                 first_not_zeroed = ext4_has_uninit_itable(sb);
4502                 ext4_register_li_request(sb, first_not_zeroed);
4503         }
4504
4505         ext4_setup_system_zone(sb);
4506         if (sbi->s_journal == NULL)
4507                 ext4_commit_super(sb, 1);
4508
4509 #ifdef CONFIG_QUOTA
4510         /* Release old quota file names */
4511         for (i = 0; i < MAXQUOTAS; i++)
4512                 if (old_opts.s_qf_names[i] &&
4513                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4514                         kfree(old_opts.s_qf_names[i]);
4515 #endif
4516         unlock_super(sb);
4517         if (enable_quota)
4518                 dquot_resume(sb, -1);
4519
4520         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4521         kfree(orig_data);
4522         return 0;
4523
4524 restore_opts:
4525         sb->s_flags = old_sb_flags;
4526         sbi->s_mount_opt = old_opts.s_mount_opt;
4527         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4528         sbi->s_resuid = old_opts.s_resuid;
4529         sbi->s_resgid = old_opts.s_resgid;
4530         sbi->s_commit_interval = old_opts.s_commit_interval;
4531         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4532         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4533 #ifdef CONFIG_QUOTA
4534         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4535         for (i = 0; i < MAXQUOTAS; i++) {
4536                 if (sbi->s_qf_names[i] &&
4537                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4538                         kfree(sbi->s_qf_names[i]);
4539                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4540         }
4541 #endif
4542         unlock_super(sb);
4543         kfree(orig_data);
4544         return err;
4545 }
4546
4547 /*
4548  * Note: calculating the overhead so we can be compatible with
4549  * historical BSD practice is quite difficult in the face of
4550  * clusters/bigalloc.  This is because multiple metadata blocks from
4551  * different block group can end up in the same allocation cluster.
4552  * Calculating the exact overhead in the face of clustered allocation
4553  * requires either O(all block bitmaps) in memory or O(number of block
4554  * groups**2) in time.  We will still calculate the superblock for
4555  * older file systems --- and if we come across with a bigalloc file
4556  * system with zero in s_overhead_clusters the estimate will be close to
4557  * correct especially for very large cluster sizes --- but for newer
4558  * file systems, it's better to calculate this figure once at mkfs
4559  * time, and store it in the superblock.  If the superblock value is
4560  * present (even for non-bigalloc file systems), we will use it.
4561  */
4562 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4563 {
4564         struct super_block *sb = dentry->d_sb;
4565         struct ext4_sb_info *sbi = EXT4_SB(sb);
4566         struct ext4_super_block *es = sbi->s_es;
4567         struct ext4_group_desc *gdp;
4568         u64 fsid;
4569         s64 bfree;
4570
4571         if (test_opt(sb, MINIX_DF)) {
4572                 sbi->s_overhead_last = 0;
4573         } else if (es->s_overhead_clusters) {
4574                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4575         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4576                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4577                 ext4_fsblk_t overhead = 0;
4578
4579                 /*
4580                  * Compute the overhead (FS structures).  This is constant
4581                  * for a given filesystem unless the number of block groups
4582                  * changes so we cache the previous value until it does.
4583                  */
4584
4585                 /*
4586                  * All of the blocks before first_data_block are
4587                  * overhead
4588                  */
4589                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4590
4591                 /*
4592                  * Add the overhead found in each block group
4593                  */
4594                 for (i = 0; i < ngroups; i++) {
4595                         gdp = ext4_get_group_desc(sb, i, NULL);
4596                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4597                         cond_resched();
4598                 }
4599                 sbi->s_overhead_last = overhead;
4600                 smp_wmb();
4601                 sbi->s_blocks_last = ext4_blocks_count(es);
4602         }
4603
4604         buf->f_type = EXT4_SUPER_MAGIC;
4605         buf->f_bsize = sb->s_blocksize;
4606         buf->f_blocks = (ext4_blocks_count(es) -
4607                          EXT4_C2B(sbi, sbi->s_overhead_last));
4608         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4609                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4610         /* prevent underflow in case that few free space is available */
4611         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4612         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4613         if (buf->f_bfree < ext4_r_blocks_count(es))
4614                 buf->f_bavail = 0;
4615         buf->f_files = le32_to_cpu(es->s_inodes_count);
4616         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4617         buf->f_namelen = EXT4_NAME_LEN;
4618         fsid = le64_to_cpup((void *)es->s_uuid) ^
4619                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4620         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4621         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4622
4623         return 0;
4624 }
4625
4626 /* Helper function for writing quotas on sync - we need to start transaction
4627  * before quota file is locked for write. Otherwise the are possible deadlocks:
4628  * Process 1                         Process 2
4629  * ext4_create()                     quota_sync()
4630  *   jbd2_journal_start()                  write_dquot()
4631  *   dquot_initialize()                         down(dqio_mutex)
4632  *     down(dqio_mutex)                    jbd2_journal_start()
4633  *
4634  */
4635
4636 #ifdef CONFIG_QUOTA
4637
4638 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4639 {
4640         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4641 }
4642
4643 static int ext4_write_dquot(struct dquot *dquot)
4644 {
4645         int ret, err;
4646         handle_t *handle;
4647         struct inode *inode;
4648
4649         inode = dquot_to_inode(dquot);
4650         handle = ext4_journal_start(inode,
4651                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4652         if (IS_ERR(handle))
4653                 return PTR_ERR(handle);
4654         ret = dquot_commit(dquot);
4655         err = ext4_journal_stop(handle);
4656         if (!ret)
4657                 ret = err;
4658         return ret;
4659 }
4660
4661 static int ext4_acquire_dquot(struct dquot *dquot)
4662 {
4663         int ret, err;
4664         handle_t *handle;
4665
4666         handle = ext4_journal_start(dquot_to_inode(dquot),
4667                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4668         if (IS_ERR(handle))
4669                 return PTR_ERR(handle);
4670         ret = dquot_acquire(dquot);
4671         err = ext4_journal_stop(handle);
4672         if (!ret)
4673                 ret = err;
4674         return ret;
4675 }
4676
4677 static int ext4_release_dquot(struct dquot *dquot)
4678 {
4679         int ret, err;
4680         handle_t *handle;
4681
4682         handle = ext4_journal_start(dquot_to_inode(dquot),
4683                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4684         if (IS_ERR(handle)) {
4685                 /* Release dquot anyway to avoid endless cycle in dqput() */
4686                 dquot_release(dquot);
4687                 return PTR_ERR(handle);
4688         }
4689         ret = dquot_release(dquot);
4690         err = ext4_journal_stop(handle);
4691         if (!ret)
4692                 ret = err;
4693         return ret;
4694 }
4695
4696 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4697 {
4698         /* Are we journaling quotas? */
4699         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4700             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4701                 dquot_mark_dquot_dirty(dquot);
4702                 return ext4_write_dquot(dquot);
4703         } else {
4704                 return dquot_mark_dquot_dirty(dquot);
4705         }
4706 }
4707
4708 static int ext4_write_info(struct super_block *sb, int type)
4709 {
4710         int ret, err;
4711         handle_t *handle;
4712
4713         /* Data block + inode block */
4714         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4715         if (IS_ERR(handle))
4716                 return PTR_ERR(handle);
4717         ret = dquot_commit_info(sb, type);
4718         err = ext4_journal_stop(handle);
4719         if (!ret)
4720                 ret = err;
4721         return ret;
4722 }
4723
4724 /*
4725  * Turn on quotas during mount time - we need to find
4726  * the quota file and such...
4727  */
4728 static int ext4_quota_on_mount(struct super_block *sb, int type)
4729 {
4730         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4731                                         EXT4_SB(sb)->s_jquota_fmt, type);
4732 }
4733
4734 /*
4735  * Standard function to be called on quota_on
4736  */
4737 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4738                          struct path *path)
4739 {
4740         int err;
4741
4742         if (!test_opt(sb, QUOTA))
4743                 return -EINVAL;
4744
4745         /* Quotafile not on the same filesystem? */
4746         if (path->dentry->d_sb != sb)
4747                 return -EXDEV;
4748         /* Journaling quota? */
4749         if (EXT4_SB(sb)->s_qf_names[type]) {
4750                 /* Quotafile not in fs root? */
4751                 if (path->dentry->d_parent != sb->s_root)
4752                         ext4_msg(sb, KERN_WARNING,
4753                                 "Quota file not on filesystem root. "
4754                                 "Journaled quota will not work");
4755         }
4756
4757         /*
4758          * When we journal data on quota file, we have to flush journal to see
4759          * all updates to the file when we bypass pagecache...
4760          */
4761         if (EXT4_SB(sb)->s_journal &&
4762             ext4_should_journal_data(path->dentry->d_inode)) {
4763                 /*
4764                  * We don't need to lock updates but journal_flush() could
4765                  * otherwise be livelocked...
4766                  */
4767                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4768                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4769                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4770                 if (err)
4771                         return err;
4772         }
4773
4774         return dquot_quota_on(sb, type, format_id, path);
4775 }
4776
4777 static int ext4_quota_off(struct super_block *sb, int type)
4778 {
4779         struct inode *inode = sb_dqopt(sb)->files[type];
4780         handle_t *handle;
4781
4782         /* Force all delayed allocation blocks to be allocated.
4783          * Caller already holds s_umount sem */
4784         if (test_opt(sb, DELALLOC))
4785                 sync_filesystem(sb);
4786
4787         if (!inode)
4788                 goto out;
4789
4790         /* Update modification times of quota files when userspace can
4791          * start looking at them */
4792         handle = ext4_journal_start(inode, 1);
4793         if (IS_ERR(handle))
4794                 goto out;
4795         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4796         ext4_mark_inode_dirty(handle, inode);
4797         ext4_journal_stop(handle);
4798
4799 out:
4800         return dquot_quota_off(sb, type);
4801 }
4802
4803 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4804  * acquiring the locks... As quota files are never truncated and quota code
4805  * itself serializes the operations (and no one else should touch the files)
4806  * we don't have to be afraid of races */
4807 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4808                                size_t len, loff_t off)
4809 {
4810         struct inode *inode = sb_dqopt(sb)->files[type];
4811         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4812         int err = 0;
4813         int offset = off & (sb->s_blocksize - 1);
4814         int tocopy;
4815         size_t toread;
4816         struct buffer_head *bh;
4817         loff_t i_size = i_size_read(inode);
4818
4819         if (off > i_size)
4820                 return 0;
4821         if (off+len > i_size)
4822                 len = i_size-off;
4823         toread = len;
4824         while (toread > 0) {
4825                 tocopy = sb->s_blocksize - offset < toread ?
4826                                 sb->s_blocksize - offset : toread;
4827                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4828                 if (err)
4829                         return err;
4830                 if (!bh)        /* A hole? */
4831                         memset(data, 0, tocopy);
4832                 else
4833                         memcpy(data, bh->b_data+offset, tocopy);
4834                 brelse(bh);
4835                 offset = 0;
4836                 toread -= tocopy;
4837                 data += tocopy;
4838                 blk++;
4839         }
4840         return len;
4841 }
4842
4843 /* Write to quotafile (we know the transaction is already started and has
4844  * enough credits) */
4845 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4846                                 const char *data, size_t len, loff_t off)
4847 {
4848         struct inode *inode = sb_dqopt(sb)->files[type];
4849         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4850         int err = 0;
4851         int offset = off & (sb->s_blocksize - 1);
4852         struct buffer_head *bh;
4853         handle_t *handle = journal_current_handle();
4854
4855         if (EXT4_SB(sb)->s_journal && !handle) {
4856                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4857                         " cancelled because transaction is not started",
4858                         (unsigned long long)off, (unsigned long long)len);
4859                 return -EIO;
4860         }
4861         /*
4862          * Since we account only one data block in transaction credits,
4863          * then it is impossible to cross a block boundary.
4864          */
4865         if (sb->s_blocksize - offset < len) {
4866                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4867                         " cancelled because not block aligned",
4868                         (unsigned long long)off, (unsigned long long)len);
4869                 return -EIO;
4870         }
4871
4872         mutex_lock_nested(&inode->i_mutex, I_MUTEX_QUOTA);
4873         bh = ext4_bread(handle, inode, blk, 1, &err);
4874         if (!bh)
4875                 goto out;
4876         err = ext4_journal_get_write_access(handle, bh);
4877         if (err) {
4878                 brelse(bh);
4879                 goto out;
4880         }
4881         lock_buffer(bh);
4882         memcpy(bh->b_data+offset, data, len);
4883         flush_dcache_page(bh->b_page);
4884         unlock_buffer(bh);
4885         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4886         brelse(bh);
4887 out:
4888         if (err) {
4889                 mutex_unlock(&inode->i_mutex);
4890                 return err;
4891         }
4892         if (inode->i_size < off + len) {
4893                 i_size_write(inode, off + len);
4894                 EXT4_I(inode)->i_disksize = inode->i_size;
4895                 ext4_mark_inode_dirty(handle, inode);
4896         }
4897         mutex_unlock(&inode->i_mutex);
4898         return len;
4899 }
4900
4901 #endif
4902
4903 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4904                        const char *dev_name, void *data)
4905 {
4906         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4907 }
4908
4909 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4910 static inline void register_as_ext2(void)
4911 {
4912         int err = register_filesystem(&ext2_fs_type);
4913         if (err)
4914                 printk(KERN_WARNING
4915                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4916 }
4917
4918 static inline void unregister_as_ext2(void)
4919 {
4920         unregister_filesystem(&ext2_fs_type);
4921 }
4922
4923 static inline int ext2_feature_set_ok(struct super_block *sb)
4924 {
4925         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4926                 return 0;
4927         if (sb->s_flags & MS_RDONLY)
4928                 return 1;
4929         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4930                 return 0;
4931         return 1;
4932 }
4933 MODULE_ALIAS("ext2");
4934 #else
4935 static inline void register_as_ext2(void) { }
4936 static inline void unregister_as_ext2(void) { }
4937 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4938 #endif
4939
4940 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4941 static inline void register_as_ext3(void)
4942 {
4943         int err = register_filesystem(&ext3_fs_type);
4944         if (err)
4945                 printk(KERN_WARNING
4946                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4947 }
4948
4949 static inline void unregister_as_ext3(void)
4950 {
4951         unregister_filesystem(&ext3_fs_type);
4952 }
4953
4954 static inline int ext3_feature_set_ok(struct super_block *sb)
4955 {
4956         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4957                 return 0;
4958         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4959                 return 0;
4960         if (sb->s_flags & MS_RDONLY)
4961                 return 1;
4962         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
4963                 return 0;
4964         return 1;
4965 }
4966 MODULE_ALIAS("ext3");
4967 #else
4968 static inline void register_as_ext3(void) { }
4969 static inline void unregister_as_ext3(void) { }
4970 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
4971 #endif
4972
4973 static struct file_system_type ext4_fs_type = {
4974         .owner          = THIS_MODULE,
4975         .name           = "ext4",
4976         .mount          = ext4_mount,
4977         .kill_sb        = kill_block_super,
4978         .fs_flags       = FS_REQUIRES_DEV,
4979 };
4980
4981 static int __init ext4_init_feat_adverts(void)
4982 {
4983         struct ext4_features *ef;
4984         int ret = -ENOMEM;
4985
4986         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
4987         if (!ef)
4988                 goto out;
4989
4990         ef->f_kobj.kset = ext4_kset;
4991         init_completion(&ef->f_kobj_unregister);
4992         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
4993                                    "features");
4994         if (ret) {
4995                 kfree(ef);
4996                 goto out;
4997         }
4998
4999         ext4_feat = ef;
5000         ret = 0;
5001 out:
5002         return ret;
5003 }
5004
5005 static void ext4_exit_feat_adverts(void)
5006 {
5007         kobject_put(&ext4_feat->f_kobj);
5008         wait_for_completion(&ext4_feat->f_kobj_unregister);
5009         kfree(ext4_feat);
5010 }
5011
5012 /* Shared across all ext4 file systems */
5013 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5014 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5015
5016 static int __init ext4_init_fs(void)
5017 {
5018         int i, err;
5019
5020         ext4_check_flag_values();
5021
5022         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5023                 mutex_init(&ext4__aio_mutex[i]);
5024                 init_waitqueue_head(&ext4__ioend_wq[i]);
5025         }
5026
5027         err = ext4_init_pageio();
5028         if (err)
5029                 return err;
5030         err = ext4_init_system_zone();
5031         if (err)
5032                 goto out6;
5033         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5034         if (!ext4_kset)
5035                 goto out5;
5036         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5037
5038         err = ext4_init_feat_adverts();
5039         if (err)
5040                 goto out4;
5041
5042         err = ext4_init_mballoc();
5043         if (err)
5044                 goto out3;
5045
5046         err = ext4_init_xattr();
5047         if (err)
5048                 goto out2;
5049         err = init_inodecache();
5050         if (err)
5051                 goto out1;
5052         register_as_ext3();
5053         register_as_ext2();
5054         err = register_filesystem(&ext4_fs_type);
5055         if (err)
5056                 goto out;
5057
5058         ext4_li_info = NULL;
5059         mutex_init(&ext4_li_mtx);
5060         return 0;
5061 out:
5062         unregister_as_ext2();
5063         unregister_as_ext3();
5064         destroy_inodecache();
5065 out1:
5066         ext4_exit_xattr();
5067 out2:
5068         ext4_exit_mballoc();
5069 out3:
5070         ext4_exit_feat_adverts();
5071 out4:
5072         if (ext4_proc_root)
5073                 remove_proc_entry("fs/ext4", NULL);
5074         kset_unregister(ext4_kset);
5075 out5:
5076         ext4_exit_system_zone();
5077 out6:
5078         ext4_exit_pageio();
5079         return err;
5080 }
5081
5082 static void __exit ext4_exit_fs(void)
5083 {
5084         ext4_destroy_lazyinit_thread();
5085         unregister_as_ext2();
5086         unregister_as_ext3();
5087         unregister_filesystem(&ext4_fs_type);
5088         destroy_inodecache();
5089         ext4_exit_xattr();
5090         ext4_exit_mballoc();
5091         ext4_exit_feat_adverts();
5092         remove_proc_entry("fs/ext4", NULL);
5093         kset_unregister(ext4_kset);
5094         ext4_exit_system_zone();
5095         ext4_exit_pageio();
5096 }
5097
5098 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5099 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5100 MODULE_LICENSE("GPL");
5101 module_init(ext4_init_fs)
5102 module_exit(ext4_exit_fs)