fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static const char *ext4_decode_error(struct super_block *sb, int errno,
  73                                      char nbuf[16]);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static 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 static int ext4_verify_csum_type(struct super_block *sb,
 115                                  struct ext4_super_block *es)
 116 {
 117         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 118                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 119                 return 1;
 120
 121         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 122 }
 123
 124 static __le32 ext4_superblock_csum(struct super_block *sb,
 125                                    struct ext4_super_block *es)
 126 {
 127         struct ext4_sb_info *sbi = EXT4_SB(sb);
 128         int offset = offsetof(struct ext4_super_block, s_checksum);
 129         __u32 csum;
 130
 131         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 132
 133         return cpu_to_le32(csum);
 134 }
 135
 136 int ext4_superblock_csum_verify(struct super_block *sb,
 137                                 struct ext4_super_block *es)
 138 {
 139         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 140                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 141                 return 1;
 142
 143         return es->s_checksum == ext4_superblock_csum(sb, es);
 144 }
 145
 146 void ext4_superblock_csum_set(struct super_block *sb)
 147 {
 148         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 149
 150         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 151                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 152                 return;
 153
 154         es->s_checksum = ext4_superblock_csum(sb, es);
 155 }
 156
 157 void *ext4_kvmalloc(size_t size, gfp_t flags)
 158 {
 159         void *ret;
 160
 161         ret = kmalloc(size, flags);
 162         if (!ret)
 163                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 164         return ret;
 165 }
 166
 167 void *ext4_kvzalloc(size_t size, gfp_t flags)
 168 {
 169         void *ret;
 170
 171         ret = kzalloc(size, flags);
 172         if (!ret)
 173                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 174         return ret;
 175 }
 176
 177 void ext4_kvfree(void *ptr)
 178 {
 179         if (is_vmalloc_addr(ptr))
 180                 vfree(ptr);
 181         else
 182                 kfree(ptr);
 183
 184 }
 185
 186 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 187                                struct ext4_group_desc *bg)
 188 {
 189         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 190                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 191                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 192 }
 193
 194 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 195                                struct ext4_group_desc *bg)
 196 {
 197         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 198                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 199                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 200 }
 201
 202 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 203                               struct ext4_group_desc *bg)
 204 {
 205         return le32_to_cpu(bg->bg_inode_table_lo) |
 206                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 207                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 208 }
 209
 210 __u32 ext4_free_group_clusters(struct super_block *sb,
 211                                struct ext4_group_desc *bg)
 212 {
 213         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 214                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 215                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 216 }
 217
 218 __u32 ext4_free_inodes_count(struct super_block *sb,
 219                               struct ext4_group_desc *bg)
 220 {
 221         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 222                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 223                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 224 }
 225
 226 __u32 ext4_used_dirs_count(struct super_block *sb,
 227                               struct ext4_group_desc *bg)
 228 {
 229         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 230                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 231                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 232 }
 233
 234 __u32 ext4_itable_unused_count(struct super_block *sb,
 235                               struct ext4_group_desc *bg)
 236 {
 237         return le16_to_cpu(bg->bg_itable_unused_lo) |
 238                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 239                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 240 }
 241
 242 void ext4_block_bitmap_set(struct super_block *sb,
 243                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 244 {
 245         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 246         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 247                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 248 }
 249
 250 void ext4_inode_bitmap_set(struct super_block *sb,
 251                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 252 {
 253         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 254         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 255                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 256 }
 257
 258 void ext4_inode_table_set(struct super_block *sb,
 259                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 260 {
 261         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 262         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 263                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 264 }
 265
 266 void ext4_free_group_clusters_set(struct super_block *sb,
 267                                   struct ext4_group_desc *bg, __u32 count)
 268 {
 269         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 270         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 271                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 272 }
 273
 274 void ext4_free_inodes_set(struct super_block *sb,
 275                           struct ext4_group_desc *bg, __u32 count)
 276 {
 277         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 278         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 279                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 280 }
 281
 282 void ext4_used_dirs_set(struct super_block *sb,
 283                           struct ext4_group_desc *bg, __u32 count)
 284 {
 285         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 286         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 287                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 288 }
 289
 290 void ext4_itable_unused_set(struct super_block *sb,
 291                           struct ext4_group_desc *bg, __u32 count)
 292 {
 293         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 294         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 295                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 296 }
 297
 298
 299 /* Just increment the non-pointer handle value */
 300 static handle_t *ext4_get_nojournal(void)
 301 {
 302         handle_t *handle = current->journal_info;
 303         unsigned long ref_cnt = (unsigned long)handle;
 304
 305         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 306
 307         ref_cnt++;
 308         handle = (handle_t *)ref_cnt;
 309
 310         current->journal_info = handle;
 311         return handle;
 312 }
 313
 314
 315 /* Decrement the non-pointer handle value */
 316 static void ext4_put_nojournal(handle_t *handle)
 317 {
 318         unsigned long ref_cnt = (unsigned long)handle;
 319
 320         BUG_ON(ref_cnt == 0);
 321
 322         ref_cnt--;
 323         handle = (handle_t *)ref_cnt;
 324
 325         current->journal_info = handle;
 326 }
 327
 328 /*
 329  * Wrappers for jbd2_journal_start/end.
 330  *
 331  * The only special thing we need to do here is to make sure that all
 332  * journal_end calls result in the superblock being marked dirty, so
 333  * that sync() will call the filesystem's write_super callback if
 334  * appropriate.
 335  */
 336 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 337 {
 338         journal_t *journal;
 339
 340         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 341         if (sb->s_flags & MS_RDONLY)
 342                 return ERR_PTR(-EROFS);
 343
 344         WARN_ON(sb->s_writers.frozen == SB_FREEZE_COMPLETE);
 345         journal = EXT4_SB(sb)->s_journal;
 346         if (!journal)
 347                 return ext4_get_nojournal();
 348         /*
 349          * Special case here: if the journal has aborted behind our
 350          * backs (eg. EIO in the commit thread), then we still need to
 351          * take the FS itself readonly cleanly.
 352          */
 353         if (is_journal_aborted(journal)) {
 354                 ext4_abort(sb, "Detected aborted journal");
 355                 return ERR_PTR(-EROFS);
 356         }
 357         return jbd2_journal_start(journal, nblocks);
 358 }
 359
 360 /*
 361  * The only special thing we need to do here is to make sure that all
 362  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 363  * that sync() will call the filesystem's write_super callback if
 364  * appropriate.
 365  */
 366 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 367 {
 368         struct super_block *sb;
 369         int err;
 370         int rc;
 371
 372         if (!ext4_handle_valid(handle)) {
 373                 ext4_put_nojournal(handle);
 374                 return 0;
 375         }
 376         sb = handle->h_transaction->t_journal->j_private;
 377         err = handle->h_err;
 378         rc = jbd2_journal_stop(handle);
 379
 380         if (!err)
 381                 err = rc;
 382         if (err)
 383                 __ext4_std_error(sb, where, line, err);
 384         return err;
 385 }
 386
 387 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 388                                const char *err_fn, struct buffer_head *bh,
 389                                handle_t *handle, int err)
 390 {
 391         char nbuf[16];
 392         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 393
 394         BUG_ON(!ext4_handle_valid(handle));
 395
 396         if (bh)
 397                 BUFFER_TRACE(bh, "abort");
 398
 399         if (!handle->h_err)
 400                 handle->h_err = err;
 401
 402         if (is_handle_aborted(handle))
 403                 return;
 404
 405         printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
 406                caller, line, errstr, err_fn);
 407
 408         jbd2_journal_abort_handle(handle);
 409 }
 410
 411 static void __save_error_info(struct super_block *sb, const char *func,
 412                             unsigned int line)
 413 {
 414         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 415
 416         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 417         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 418         es->s_last_error_time = cpu_to_le32(get_seconds());
 419         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 420         es->s_last_error_line = cpu_to_le32(line);
 421         if (!es->s_first_error_time) {
 422                 es->s_first_error_time = es->s_last_error_time;
 423                 strncpy(es->s_first_error_func, func,
 424                         sizeof(es->s_first_error_func));
 425                 es->s_first_error_line = cpu_to_le32(line);
 426                 es->s_first_error_ino = es->s_last_error_ino;
 427                 es->s_first_error_block = es->s_last_error_block;
 428         }
 429         /*
 430          * Start the daily error reporting function if it hasn't been
 431          * started already
 432          */
 433         if (!es->s_error_count)
 434                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 435         le32_add_cpu(&es->s_error_count, 1);
 436 }
 437
 438 static void save_error_info(struct super_block *sb, const char *func,
 439                             unsigned int line)
 440 {
 441         __save_error_info(sb, func, line);
 442         ext4_commit_super(sb, 1);
 443 }
 444
 445 /*
 446  * The del_gendisk() function uninitializes the disk-specific data
 447  * structures, including the bdi structure, without telling anyone
 448  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 449  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 450  * This is a kludge to prevent these oops until we can put in a proper
 451  * hook in del_gendisk() to inform the VFS and file system layers.
 452  */
 453 static int block_device_ejected(struct super_block *sb)
 454 {
 455         struct inode *bd_inode = sb->s_bdev->bd_inode;
 456         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 457
 458         return bdi->dev == NULL;
 459 }
 460
 461 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 462 {
 463         struct super_block              *sb = journal->j_private;
 464         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 465         int                             error = is_journal_aborted(journal);
 466         struct ext4_journal_cb_entry    *jce, *tmp;
 467
 468         spin_lock(&sbi->s_md_lock);
 469         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 470                 list_del_init(&jce->jce_list);
 471                 spin_unlock(&sbi->s_md_lock);
 472                 jce->jce_func(sb, jce, error);
 473                 spin_lock(&sbi->s_md_lock);
 474         }
 475         spin_unlock(&sbi->s_md_lock);
 476 }
 477
 478 /* Deal with the reporting of failure conditions on a filesystem such as
 479  * inconsistencies detected or read IO failures.
 480  *
 481  * On ext2, we can store the error state of the filesystem in the
 482  * superblock.  That is not possible on ext4, because we may have other
 483  * write ordering constraints on the superblock which prevent us from
 484  * writing it out straight away; and given that the journal is about to
 485  * be aborted, we can't rely on the current, or future, transactions to
 486  * write out the superblock safely.
 487  *
 488  * We'll just use the jbd2_journal_abort() error code to record an error in
 489  * the journal instead.  On recovery, the journal will complain about
 490  * that error until we've noted it down and cleared it.
 491  */
 492
 493 static void ext4_handle_error(struct super_block *sb)
 494 {
 495         if (sb->s_flags & MS_RDONLY)
 496                 return;
 497
 498         if (!test_opt(sb, ERRORS_CONT)) {
 499                 journal_t *journal = EXT4_SB(sb)->s_journal;
 500
 501                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 502                 if (journal)
 503                         jbd2_journal_abort(journal, -EIO);
 504         }
 505         if (test_opt(sb, ERRORS_RO)) {
 506                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 507                 sb->s_flags |= MS_RDONLY;
 508         }
 509         if (test_opt(sb, ERRORS_PANIC))
 510                 panic("EXT4-fs (device %s): panic forced after error\n",
 511                         sb->s_id);
 512 }
 513
 514 void __ext4_error(struct super_block *sb, const char *function,
 515                   unsigned int line, const char *fmt, ...)
 516 {
 517         struct va_format vaf;
 518         va_list args;
 519
 520         va_start(args, fmt);
 521         vaf.fmt = fmt;
 522         vaf.va = &args;
 523         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 524                sb->s_id, function, line, current->comm, &vaf);
 525         va_end(args);
 526         save_error_info(sb, function, line);
 527
 528         ext4_handle_error(sb);
 529 }
 530
 531 void ext4_error_inode(struct inode *inode, const char *function,
 532                       unsigned int line, ext4_fsblk_t block,
 533                       const char *fmt, ...)
 534 {
 535         va_list args;
 536         struct va_format vaf;
 537         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 538
 539         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 540         es->s_last_error_block = cpu_to_le64(block);
 541         save_error_info(inode->i_sb, function, line);
 542         va_start(args, fmt);
 543         vaf.fmt = fmt;
 544         vaf.va = &args;
 545         if (block)
 546                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 547                        "inode #%lu: block %llu: comm %s: %pV\n",
 548                        inode->i_sb->s_id, function, line, inode->i_ino,
 549                        block, current->comm, &vaf);
 550         else
 551                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 552                        "inode #%lu: comm %s: %pV\n",
 553                        inode->i_sb->s_id, function, line, inode->i_ino,
 554                        current->comm, &vaf);
 555         va_end(args);
 556
 557         ext4_handle_error(inode->i_sb);
 558 }
 559
 560 void ext4_error_file(struct file *file, const char *function,
 561                      unsigned int line, ext4_fsblk_t block,
 562                      const char *fmt, ...)
 563 {
 564         va_list args;
 565         struct va_format vaf;
 566         struct ext4_super_block *es;
 567         struct inode *inode = file->f_dentry->d_inode;
 568         char pathname[80], *path;
 569
 570         es = EXT4_SB(inode->i_sb)->s_es;
 571         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 572         save_error_info(inode->i_sb, function, line);
 573         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 574         if (IS_ERR(path))
 575                 path = "(unknown)";
 576         va_start(args, fmt);
 577         vaf.fmt = fmt;
 578         vaf.va = &args;
 579         if (block)
 580                 printk(KERN_CRIT
 581                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 582                        "block %llu: comm %s: path %s: %pV\n",
 583                        inode->i_sb->s_id, function, line, inode->i_ino,
 584                        block, current->comm, path, &vaf);
 585         else
 586                 printk(KERN_CRIT
 587                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 588                        "comm %s: path %s: %pV\n",
 589                        inode->i_sb->s_id, function, line, inode->i_ino,
 590                        current->comm, path, &vaf);
 591         va_end(args);
 592
 593         ext4_handle_error(inode->i_sb);
 594 }
 595
 596 static const char *ext4_decode_error(struct super_block *sb, int errno,
 597                                      char nbuf[16])
 598 {
 599         char *errstr = NULL;
 600
 601         switch (errno) {
 602         case -EIO:
 603                 errstr = "IO failure";
 604                 break;
 605         case -ENOMEM:
 606                 errstr = "Out of memory";
 607                 break;
 608         case -EROFS:
 609                 if (!sb || (EXT4_SB(sb)->s_journal &&
 610                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 611                         errstr = "Journal has aborted";
 612                 else
 613                         errstr = "Readonly filesystem";
 614                 break;
 615         default:
 616                 /* If the caller passed in an extra buffer for unknown
 617                  * errors, textualise them now.  Else we just return
 618                  * NULL. */
 619                 if (nbuf) {
 620                         /* Check for truncated error codes... */
 621                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 622                                 errstr = nbuf;
 623                 }
 624                 break;
 625         }
 626
 627         return errstr;
 628 }
 629
 630 /* __ext4_std_error decodes expected errors from journaling functions
 631  * automatically and invokes the appropriate error response.  */
 632
 633 void __ext4_std_error(struct super_block *sb, const char *function,
 634                       unsigned int line, int errno)
 635 {
 636         char nbuf[16];
 637         const char *errstr;
 638
 639         /* Special case: if the error is EROFS, and we're not already
 640          * inside a transaction, then there's really no point in logging
 641          * an error. */
 642         if (errno == -EROFS && journal_current_handle() == NULL &&
 643             (sb->s_flags & MS_RDONLY))
 644                 return;
 645
 646         errstr = ext4_decode_error(sb, errno, nbuf);
 647         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 648                sb->s_id, function, line, errstr);
 649         save_error_info(sb, function, line);
 650
 651         ext4_handle_error(sb);
 652 }
 653
 654 /*
 655  * ext4_abort is a much stronger failure handler than ext4_error.  The
 656  * abort function may be used to deal with unrecoverable failures such
 657  * as journal IO errors or ENOMEM at a critical moment in log management.
 658  *
 659  * We unconditionally force the filesystem into an ABORT|READONLY state,
 660  * unless the error response on the fs has been set to panic in which
 661  * case we take the easy way out and panic immediately.
 662  */
 663
 664 void __ext4_abort(struct super_block *sb, const char *function,
 665                 unsigned int line, const char *fmt, ...)
 666 {
 667         va_list args;
 668
 669         save_error_info(sb, function, line);
 670         va_start(args, fmt);
 671         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 672                function, line);
 673         vprintk(fmt, args);
 674         printk("\n");
 675         va_end(args);
 676
 677         if ((sb->s_flags & MS_RDONLY) == 0) {
 678                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 679                 sb->s_flags |= MS_RDONLY;
 680                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 681                 if (EXT4_SB(sb)->s_journal)
 682                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 683                 save_error_info(sb, function, line);
 684         }
 685         if (test_opt(sb, ERRORS_PANIC))
 686                 panic("EXT4-fs panic from previous error\n");
 687 }
 688
 689 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 690 {
 691         struct va_format vaf;
 692         va_list args;
 693
 694         va_start(args, fmt);
 695         vaf.fmt = fmt;
 696         vaf.va = &args;
 697         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 698         va_end(args);
 699 }
 700
 701 void __ext4_warning(struct super_block *sb, const char *function,
 702                     unsigned int line, const char *fmt, ...)
 703 {
 704         struct va_format vaf;
 705         va_list args;
 706
 707         va_start(args, fmt);
 708         vaf.fmt = fmt;
 709         vaf.va = &args;
 710         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 711                sb->s_id, function, line, &vaf);
 712         va_end(args);
 713 }
 714
 715 void __ext4_grp_locked_error(const char *function, unsigned int line,
 716                              struct super_block *sb, ext4_group_t grp,
 717                              unsigned long ino, ext4_fsblk_t block,
 718                              const char *fmt, ...)
 719 __releases(bitlock)
 720 __acquires(bitlock)
 721 {
 722         struct va_format vaf;
 723         va_list args;
 724         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 725
 726         es->s_last_error_ino = cpu_to_le32(ino);
 727         es->s_last_error_block = cpu_to_le64(block);
 728         __save_error_info(sb, function, line);
 729
 730         va_start(args, fmt);
 731
 732         vaf.fmt = fmt;
 733         vaf.va = &args;
 734         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 735                sb->s_id, function, line, grp);
 736         if (ino)
 737                 printk(KERN_CONT "inode %lu: ", ino);
 738         if (block)
 739                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 740         printk(KERN_CONT "%pV\n", &vaf);
 741         va_end(args);
 742
 743         if (test_opt(sb, ERRORS_CONT)) {
 744                 ext4_commit_super(sb, 0);
 745                 return;
 746         }
 747
 748         ext4_unlock_group(sb, grp);
 749         ext4_handle_error(sb);
 750         /*
 751          * We only get here in the ERRORS_RO case; relocking the group
 752          * may be dangerous, but nothing bad will happen since the
 753          * filesystem will have already been marked read/only and the
 754          * journal has been aborted.  We return 1 as a hint to callers
 755          * who might what to use the return value from
 756          * ext4_grp_locked_error() to distinguish between the
 757          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 758          * aggressively from the ext4 function in question, with a
 759          * more appropriate error code.
 760          */
 761         ext4_lock_group(sb, grp);
 762         return;
 763 }
 764
 765 void ext4_update_dynamic_rev(struct super_block *sb)
 766 {
 767         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 768
 769         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 770                 return;
 771
 772         ext4_warning(sb,
 773                      "updating to rev %d because of new feature flag, "
 774                      "running e2fsck is recommended",
 775                      EXT4_DYNAMIC_REV);
 776
 777         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 778         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 779         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 780         /* leave es->s_feature_*compat flags alone */
 781         /* es->s_uuid will be set by e2fsck if empty */
 782
 783         /*
 784          * The rest of the superblock fields should be zero, and if not it
 785          * means they are likely already in use, so leave them alone.  We
 786          * can leave it up to e2fsck to clean up any inconsistencies there.
 787          */
 788 }
 789
 790 /*
 791  * Open the external journal device
 792  */
 793 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 794 {
 795         struct block_device *bdev;
 796         char b[BDEVNAME_SIZE];
 797
 798         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 799         if (IS_ERR(bdev))
 800                 goto fail;
 801         return bdev;
 802
 803 fail:
 804         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 805                         __bdevname(dev, b), PTR_ERR(bdev));
 806         return NULL;
 807 }
 808
 809 /*
 810  * Release the journal device
 811  */
 812 static int ext4_blkdev_put(struct block_device *bdev)
 813 {
 814         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 815 }
 816
 817 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 818 {
 819         struct block_device *bdev;
 820         int ret = -ENODEV;
 821
 822         bdev = sbi->journal_bdev;
 823         if (bdev) {
 824                 ret = ext4_blkdev_put(bdev);
 825                 sbi->journal_bdev = NULL;
 826         }
 827         return ret;
 828 }
 829
 830 static inline struct inode *orphan_list_entry(struct list_head *l)
 831 {
 832         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 833 }
 834
 835 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 836 {
 837         struct list_head *l;
 838
 839         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 840                  le32_to_cpu(sbi->s_es->s_last_orphan));
 841
 842         printk(KERN_ERR "sb_info orphan list:\n");
 843         list_for_each(l, &sbi->s_orphan) {
 844                 struct inode *inode = orphan_list_entry(l);
 845                 printk(KERN_ERR "  "
 846                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 847                        inode->i_sb->s_id, inode->i_ino, inode,
 848                        inode->i_mode, inode->i_nlink,
 849                        NEXT_ORPHAN(inode));
 850         }
 851 }
 852
 853 static void ext4_put_super(struct super_block *sb)
 854 {
 855         struct ext4_sb_info *sbi = EXT4_SB(sb);
 856         struct ext4_super_block *es = sbi->s_es;
 857         int i, err;
 858
 859         ext4_unregister_li_request(sb);
 860         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 861
 862         flush_workqueue(sbi->dio_unwritten_wq);
 863         destroy_workqueue(sbi->dio_unwritten_wq);
 864
 865         if (sbi->s_journal) {
 866                 err = jbd2_journal_destroy(sbi->s_journal);
 867                 sbi->s_journal = NULL;
 868                 if (err < 0)
 869                         ext4_abort(sb, "Couldn't clean up the journal");
 870         }
 871
 872         del_timer(&sbi->s_err_report);
 873         ext4_release_system_zone(sb);
 874         ext4_mb_release(sb);
 875         ext4_ext_release(sb);
 876         ext4_xattr_put_super(sb);
 877
 878         if (!(sb->s_flags & MS_RDONLY)) {
 879                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 880                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 881         }
 882         if (!(sb->s_flags & MS_RDONLY))
 883                 ext4_commit_super(sb, 1);
 884
 885         if (sbi->s_proc) {
 886                 remove_proc_entry("options", sbi->s_proc);
 887                 remove_proc_entry(sb->s_id, ext4_proc_root);
 888         }
 889         kobject_del(&sbi->s_kobj);
 890
 891         for (i = 0; i < sbi->s_gdb_count; i++)
 892                 brelse(sbi->s_group_desc[i]);
 893         ext4_kvfree(sbi->s_group_desc);
 894         ext4_kvfree(sbi->s_flex_groups);
 895         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 896         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 897         percpu_counter_destroy(&sbi->s_dirs_counter);
 898         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 899         brelse(sbi->s_sbh);
 900 #ifdef CONFIG_QUOTA
 901         for (i = 0; i < MAXQUOTAS; i++)
 902                 kfree(sbi->s_qf_names[i]);
 903 #endif
 904
 905         /* Debugging code just in case the in-memory inode orphan list
 906          * isn't empty.  The on-disk one can be non-empty if we've
 907          * detected an error and taken the fs readonly, but the
 908          * in-memory list had better be clean by this point. */
 909         if (!list_empty(&sbi->s_orphan))
 910                 dump_orphan_list(sb, sbi);
 911         J_ASSERT(list_empty(&sbi->s_orphan));
 912
 913         invalidate_bdev(sb->s_bdev);
 914         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 915                 /*
 916                  * Invalidate the journal device's buffers.  We don't want them
 917                  * floating about in memory - the physical journal device may
 918                  * hotswapped, and it breaks the `ro-after' testing code.
 919                  */
 920                 sync_blockdev(sbi->journal_bdev);
 921                 invalidate_bdev(sbi->journal_bdev);
 922                 ext4_blkdev_remove(sbi);
 923         }
 924         if (sbi->s_mmp_tsk)
 925                 kthread_stop(sbi->s_mmp_tsk);
 926         sb->s_fs_info = NULL;
 927         /*
 928          * Now that we are completely done shutting down the
 929          * superblock, we need to actually destroy the kobject.
 930          */
 931         kobject_put(&sbi->s_kobj);
 932         wait_for_completion(&sbi->s_kobj_unregister);
 933         if (sbi->s_chksum_driver)
 934                 crypto_free_shash(sbi->s_chksum_driver);
 935         kfree(sbi->s_blockgroup_lock);
 936         kfree(sbi);
 937 }
 938
 939 static struct kmem_cache *ext4_inode_cachep;
 940
 941 /*
 942  * Called inside transaction, so use GFP_NOFS
 943  */
 944 static struct inode *ext4_alloc_inode(struct super_block *sb)
 945 {
 946         struct ext4_inode_info *ei;
 947
 948         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 949         if (!ei)
 950                 return NULL;
 951
 952         ei->vfs_inode.i_version = 1;
 953         ei->vfs_inode.i_data.writeback_index = 0;
 954         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 955         INIT_LIST_HEAD(&ei->i_prealloc_list);
 956         spin_lock_init(&ei->i_prealloc_lock);
 957         ei->i_reserved_data_blocks = 0;
 958         ei->i_reserved_meta_blocks = 0;
 959         ei->i_allocated_meta_blocks = 0;
 960         ei->i_da_metadata_calc_len = 0;
 961         ei->i_da_metadata_calc_last_lblock = 0;
 962         spin_lock_init(&(ei->i_block_reservation_lock));
 963 #ifdef CONFIG_QUOTA
 964         ei->i_reserved_quota = 0;
 965 #endif
 966         ei->jinode = NULL;
 967         INIT_LIST_HEAD(&ei->i_completed_io_list);
 968         spin_lock_init(&ei->i_completed_io_lock);
 969         ei->i_sync_tid = 0;
 970         ei->i_datasync_tid = 0;
 971         atomic_set(&ei->i_ioend_count, 0);
 972         atomic_set(&ei->i_unwritten, 0);
 973
 974         return &ei->vfs_inode;
 975 }
 976
 977 static int ext4_drop_inode(struct inode *inode)
 978 {
 979         int drop = generic_drop_inode(inode);
 980
 981         trace_ext4_drop_inode(inode, drop);
 982         return drop;
 983 }
 984
 985 static void ext4_i_callback(struct rcu_head *head)
 986 {
 987         struct inode *inode = container_of(head, struct inode, i_rcu);
 988         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 989 }
 990
 991 static void ext4_destroy_inode(struct inode *inode)
 992 {
 993         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 994                 ext4_msg(inode->i_sb, KERN_ERR,
 995                          "Inode %lu (%p): orphan list check failed!",
 996                          inode->i_ino, EXT4_I(inode));
 997                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 998                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 999                                 true);
1000                 dump_stack();
1001         }
1002         call_rcu(&inode->i_rcu, ext4_i_callback);
1003 }
1004
1005 static void init_once(void *foo)
1006 {
1007         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1008
1009         INIT_LIST_HEAD(&ei->i_orphan);
1010 #ifdef CONFIG_EXT4_FS_XATTR
1011         init_rwsem(&ei->xattr_sem);
1012 #endif
1013         init_rwsem(&ei->i_data_sem);
1014         inode_init_once(&ei->vfs_inode);
1015 }
1016
1017 static int init_inodecache(void)
1018 {
1019         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1020                                              sizeof(struct ext4_inode_info),
1021                                              0, (SLAB_RECLAIM_ACCOUNT|
1022                                                 SLAB_MEM_SPREAD),
1023                                              init_once);
1024         if (ext4_inode_cachep == NULL)
1025                 return -ENOMEM;
1026         return 0;
1027 }
1028
1029 static void destroy_inodecache(void)
1030 {
1031         kmem_cache_destroy(ext4_inode_cachep);
1032 }
1033
1034 void ext4_clear_inode(struct inode *inode)
1035 {
1036         invalidate_inode_buffers(inode);
1037         clear_inode(inode);
1038         dquot_drop(inode);
1039         ext4_discard_preallocations(inode);
1040         if (EXT4_I(inode)->jinode) {
1041                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1042                                                EXT4_I(inode)->jinode);
1043                 jbd2_free_inode(EXT4_I(inode)->jinode);
1044                 EXT4_I(inode)->jinode = NULL;
1045         }
1046 }
1047
1048 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1049                                         u64 ino, u32 generation)
1050 {
1051         struct inode *inode;
1052
1053         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1054                 return ERR_PTR(-ESTALE);
1055         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1056                 return ERR_PTR(-ESTALE);
1057
1058         /* iget isn't really right if the inode is currently unallocated!!
1059          *
1060          * ext4_read_inode will return a bad_inode if the inode had been
1061          * deleted, so we should be safe.
1062          *
1063          * Currently we don't know the generation for parent directory, so
1064          * a generation of 0 means "accept any"
1065          */
1066         inode = ext4_iget(sb, ino);
1067         if (IS_ERR(inode))
1068                 return ERR_CAST(inode);
1069         if (generation && inode->i_generation != generation) {
1070                 iput(inode);
1071                 return ERR_PTR(-ESTALE);
1072         }
1073
1074         return inode;
1075 }
1076
1077 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1078                                         int fh_len, int fh_type)
1079 {
1080         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1081                                     ext4_nfs_get_inode);
1082 }
1083
1084 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1085                                         int fh_len, int fh_type)
1086 {
1087         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1088                                     ext4_nfs_get_inode);
1089 }
1090
1091 /*
1092  * Try to release metadata pages (indirect blocks, directories) which are
1093  * mapped via the block device.  Since these pages could have journal heads
1094  * which would prevent try_to_free_buffers() from freeing them, we must use
1095  * jbd2 layer's try_to_free_buffers() function to release them.
1096  */
1097 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1098                                  gfp_t wait)
1099 {
1100         journal_t *journal = EXT4_SB(sb)->s_journal;
1101
1102         WARN_ON(PageChecked(page));
1103         if (!page_has_buffers(page))
1104                 return 0;
1105         if (journal)
1106                 return jbd2_journal_try_to_free_buffers(journal, page,
1107                                                         wait & ~__GFP_WAIT);
1108         return try_to_free_buffers(page);
1109 }
1110
1111 #ifdef CONFIG_QUOTA
1112 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1113 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1114
1115 static int ext4_write_dquot(struct dquot *dquot);
1116 static int ext4_acquire_dquot(struct dquot *dquot);
1117 static int ext4_release_dquot(struct dquot *dquot);
1118 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1119 static int ext4_write_info(struct super_block *sb, int type);
1120 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1121                          struct path *path);
1122 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1123                                  int format_id);
1124 static int ext4_quota_off(struct super_block *sb, int type);
1125 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1126 static int ext4_quota_on_mount(struct super_block *sb, int type);
1127 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1128                                size_t len, loff_t off);
1129 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1130                                 const char *data, size_t len, loff_t off);
1131 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1132                              unsigned int flags);
1133 static int ext4_enable_quotas(struct super_block *sb);
1134
1135 static const struct dquot_operations ext4_quota_operations = {
1136         .get_reserved_space = ext4_get_reserved_space,
1137         .write_dquot    = ext4_write_dquot,
1138         .acquire_dquot  = ext4_acquire_dquot,
1139         .release_dquot  = ext4_release_dquot,
1140         .mark_dirty     = ext4_mark_dquot_dirty,
1141         .write_info     = ext4_write_info,
1142         .alloc_dquot    = dquot_alloc,
1143         .destroy_dquot  = dquot_destroy,
1144 };
1145
1146 static const struct quotactl_ops ext4_qctl_operations = {
1147         .quota_on       = ext4_quota_on,
1148         .quota_off      = ext4_quota_off,
1149         .quota_sync     = dquot_quota_sync,
1150         .get_info       = dquot_get_dqinfo,
1151         .set_info       = dquot_set_dqinfo,
1152         .get_dqblk      = dquot_get_dqblk,
1153         .set_dqblk      = dquot_set_dqblk
1154 };
1155
1156 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1157         .quota_on_meta  = ext4_quota_on_sysfile,
1158         .quota_off      = ext4_quota_off_sysfile,
1159         .quota_sync     = dquot_quota_sync,
1160         .get_info       = dquot_get_dqinfo,
1161         .set_info       = dquot_set_dqinfo,
1162         .get_dqblk      = dquot_get_dqblk,
1163         .set_dqblk      = dquot_set_dqblk
1164 };
1165 #endif
1166
1167 static const struct super_operations ext4_sops = {
1168         .alloc_inode    = ext4_alloc_inode,
1169         .destroy_inode  = ext4_destroy_inode,
1170         .write_inode    = ext4_write_inode,
1171         .dirty_inode    = ext4_dirty_inode,
1172         .drop_inode     = ext4_drop_inode,
1173         .evict_inode    = ext4_evict_inode,
1174         .put_super      = ext4_put_super,
1175         .sync_fs        = ext4_sync_fs,
1176         .freeze_fs      = ext4_freeze,
1177         .unfreeze_fs    = ext4_unfreeze,
1178         .statfs         = ext4_statfs,
1179         .remount_fs     = ext4_remount,
1180         .show_options   = ext4_show_options,
1181 #ifdef CONFIG_QUOTA
1182         .quota_read     = ext4_quota_read,
1183         .quota_write    = ext4_quota_write,
1184 #endif
1185         .bdev_try_to_free_page = bdev_try_to_free_page,
1186 };
1187
1188 static const struct super_operations ext4_nojournal_sops = {
1189         .alloc_inode    = ext4_alloc_inode,
1190         .destroy_inode  = ext4_destroy_inode,
1191         .write_inode    = ext4_write_inode,
1192         .dirty_inode    = ext4_dirty_inode,
1193         .drop_inode     = ext4_drop_inode,
1194         .evict_inode    = ext4_evict_inode,
1195         .put_super      = ext4_put_super,
1196         .statfs         = ext4_statfs,
1197         .remount_fs     = ext4_remount,
1198         .show_options   = ext4_show_options,
1199 #ifdef CONFIG_QUOTA
1200         .quota_read     = ext4_quota_read,
1201         .quota_write    = ext4_quota_write,
1202 #endif
1203         .bdev_try_to_free_page = bdev_try_to_free_page,
1204 };
1205
1206 static const struct export_operations ext4_export_ops = {
1207         .fh_to_dentry = ext4_fh_to_dentry,
1208         .fh_to_parent = ext4_fh_to_parent,
1209         .get_parent = ext4_get_parent,
1210 };
1211
1212 enum {
1213         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1214         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1215         Opt_nouid32, Opt_debug, Opt_removed,
1216         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1217         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1218         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1219         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1220         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1221         Opt_data_err_abort, Opt_data_err_ignore,
1222         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1223         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1224         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1225         Opt_usrquota, Opt_grpquota, Opt_i_version,
1226         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1227         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1228         Opt_inode_readahead_blks, Opt_journal_ioprio,
1229         Opt_dioread_nolock, Opt_dioread_lock,
1230         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1231         Opt_max_dir_size_kb,
1232 };
1233
1234 static const match_table_t tokens = {
1235         {Opt_bsd_df, "bsddf"},
1236         {Opt_minix_df, "minixdf"},
1237         {Opt_grpid, "grpid"},
1238         {Opt_grpid, "bsdgroups"},
1239         {Opt_nogrpid, "nogrpid"},
1240         {Opt_nogrpid, "sysvgroups"},
1241         {Opt_resgid, "resgid=%u"},
1242         {Opt_resuid, "resuid=%u"},
1243         {Opt_sb, "sb=%u"},
1244         {Opt_err_cont, "errors=continue"},
1245         {Opt_err_panic, "errors=panic"},
1246         {Opt_err_ro, "errors=remount-ro"},
1247         {Opt_nouid32, "nouid32"},
1248         {Opt_debug, "debug"},
1249         {Opt_removed, "oldalloc"},
1250         {Opt_removed, "orlov"},
1251         {Opt_user_xattr, "user_xattr"},
1252         {Opt_nouser_xattr, "nouser_xattr"},
1253         {Opt_acl, "acl"},
1254         {Opt_noacl, "noacl"},
1255         {Opt_noload, "norecovery"},
1256         {Opt_noload, "noload"},
1257         {Opt_removed, "nobh"},
1258         {Opt_removed, "bh"},
1259         {Opt_commit, "commit=%u"},
1260         {Opt_min_batch_time, "min_batch_time=%u"},
1261         {Opt_max_batch_time, "max_batch_time=%u"},
1262         {Opt_journal_dev, "journal_dev=%u"},
1263         {Opt_journal_checksum, "journal_checksum"},
1264         {Opt_journal_async_commit, "journal_async_commit"},
1265         {Opt_abort, "abort"},
1266         {Opt_data_journal, "data=journal"},
1267         {Opt_data_ordered, "data=ordered"},
1268         {Opt_data_writeback, "data=writeback"},
1269         {Opt_data_err_abort, "data_err=abort"},
1270         {Opt_data_err_ignore, "data_err=ignore"},
1271         {Opt_offusrjquota, "usrjquota="},
1272         {Opt_usrjquota, "usrjquota=%s"},
1273         {Opt_offgrpjquota, "grpjquota="},
1274         {Opt_grpjquota, "grpjquota=%s"},
1275         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1276         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1277         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1278         {Opt_grpquota, "grpquota"},
1279         {Opt_noquota, "noquota"},
1280         {Opt_quota, "quota"},
1281         {Opt_usrquota, "usrquota"},
1282         {Opt_barrier, "barrier=%u"},
1283         {Opt_barrier, "barrier"},
1284         {Opt_nobarrier, "nobarrier"},
1285         {Opt_i_version, "i_version"},
1286         {Opt_stripe, "stripe=%u"},
1287         {Opt_delalloc, "delalloc"},
1288         {Opt_nodelalloc, "nodelalloc"},
1289         {Opt_mblk_io_submit, "mblk_io_submit"},
1290         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1291         {Opt_block_validity, "block_validity"},
1292         {Opt_noblock_validity, "noblock_validity"},
1293         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1294         {Opt_journal_ioprio, "journal_ioprio=%u"},
1295         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1296         {Opt_auto_da_alloc, "auto_da_alloc"},
1297         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1298         {Opt_dioread_nolock, "dioread_nolock"},
1299         {Opt_dioread_lock, "dioread_lock"},
1300         {Opt_discard, "discard"},
1301         {Opt_nodiscard, "nodiscard"},
1302         {Opt_init_itable, "init_itable=%u"},
1303         {Opt_init_itable, "init_itable"},
1304         {Opt_noinit_itable, "noinit_itable"},
1305         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1306         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1307         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1308         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1309         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1310         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1311         {Opt_err, NULL},
1312 };
1313
1314 static ext4_fsblk_t get_sb_block(void **data)
1315 {
1316         ext4_fsblk_t    sb_block;
1317         char            *options = (char *) *data;
1318
1319         if (!options || strncmp(options, "sb=", 3) != 0)
1320                 return 1;       /* Default location */
1321
1322         options += 3;
1323         /* TODO: use simple_strtoll with >32bit ext4 */
1324         sb_block = simple_strtoul(options, &options, 0);
1325         if (*options && *options != ',') {
1326                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1327                        (char *) *data);
1328                 return 1;
1329         }
1330         if (*options == ',')
1331                 options++;
1332         *data = (void *) options;
1333
1334         return sb_block;
1335 }
1336
1337 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1338 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1339         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1340
1341 #ifdef CONFIG_QUOTA
1342 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1343 {
1344         struct ext4_sb_info *sbi = EXT4_SB(sb);
1345         char *qname;
1346
1347         if (sb_any_quota_loaded(sb) &&
1348                 !sbi->s_qf_names[qtype]) {
1349                 ext4_msg(sb, KERN_ERR,
1350                         "Cannot change journaled "
1351                         "quota options when quota turned on");
1352                 return -1;
1353         }
1354         qname = match_strdup(args);
1355         if (!qname) {
1356                 ext4_msg(sb, KERN_ERR,
1357                         "Not enough memory for storing quotafile name");
1358                 return -1;
1359         }
1360         if (sbi->s_qf_names[qtype] &&
1361                 strcmp(sbi->s_qf_names[qtype], qname)) {
1362                 ext4_msg(sb, KERN_ERR,
1363                         "%s quota file already specified", QTYPE2NAME(qtype));
1364                 kfree(qname);
1365                 return -1;
1366         }
1367         sbi->s_qf_names[qtype] = qname;
1368         if (strchr(sbi->s_qf_names[qtype], '/')) {
1369                 ext4_msg(sb, KERN_ERR,
1370                         "quotafile must be on filesystem root");
1371                 kfree(sbi->s_qf_names[qtype]);
1372                 sbi->s_qf_names[qtype] = NULL;
1373                 return -1;
1374         }
1375         set_opt(sb, QUOTA);
1376         return 1;
1377 }
1378
1379 static int clear_qf_name(struct super_block *sb, int qtype)
1380 {
1381
1382         struct ext4_sb_info *sbi = EXT4_SB(sb);
1383
1384         if (sb_any_quota_loaded(sb) &&
1385                 sbi->s_qf_names[qtype]) {
1386                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1387                         " when quota turned on");
1388                 return -1;
1389         }
1390         /*
1391          * The space will be released later when all options are confirmed
1392          * to be correct
1393          */
1394         sbi->s_qf_names[qtype] = NULL;
1395         return 1;
1396 }
1397 #endif
1398
1399 #define MOPT_SET        0x0001
1400 #define MOPT_CLEAR      0x0002
1401 #define MOPT_NOSUPPORT  0x0004
1402 #define MOPT_EXPLICIT   0x0008
1403 #define MOPT_CLEAR_ERR  0x0010
1404 #define MOPT_GTE0       0x0020
1405 #ifdef CONFIG_QUOTA
1406 #define MOPT_Q          0
1407 #define MOPT_QFMT       0x0040
1408 #else
1409 #define MOPT_Q          MOPT_NOSUPPORT
1410 #define MOPT_QFMT       MOPT_NOSUPPORT
1411 #endif
1412 #define MOPT_DATAJ      0x0080
1413
1414 static const struct mount_opts {
1415         int     token;
1416         int     mount_opt;
1417         int     flags;
1418 } ext4_mount_opts[] = {
1419         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1420         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1421         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1422         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1423         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1424         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1425         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1426         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1427         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1428         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1429         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1430         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1431         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1432         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1433         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1434         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1435                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1436         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1437         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1438         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1439         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1440         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1441         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1442         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1443         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1444         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1445         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1446         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1447         {Opt_commit, 0, MOPT_GTE0},
1448         {Opt_max_batch_time, 0, MOPT_GTE0},
1449         {Opt_min_batch_time, 0, MOPT_GTE0},
1450         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1451         {Opt_init_itable, 0, MOPT_GTE0},
1452         {Opt_stripe, 0, MOPT_GTE0},
1453         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1454         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1455         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1456 #ifdef CONFIG_EXT4_FS_XATTR
1457         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1458         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1459 #else
1460         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1461         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1462 #endif
1463 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1464         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1465         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1466 #else
1467         {Opt_acl, 0, MOPT_NOSUPPORT},
1468         {Opt_noacl, 0, MOPT_NOSUPPORT},
1469 #endif
1470         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1471         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1472         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1473         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1474                                                         MOPT_SET | MOPT_Q},
1475         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1476                                                         MOPT_SET | MOPT_Q},
1477         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1478                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1479         {Opt_usrjquota, 0, MOPT_Q},
1480         {Opt_grpjquota, 0, MOPT_Q},
1481         {Opt_offusrjquota, 0, MOPT_Q},
1482         {Opt_offgrpjquota, 0, MOPT_Q},
1483         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1484         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1485         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1486         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1487         {Opt_err, 0, 0}
1488 };
1489
1490 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1491                             substring_t *args, unsigned long *journal_devnum,
1492                             unsigned int *journal_ioprio, int is_remount)
1493 {
1494         struct ext4_sb_info *sbi = EXT4_SB(sb);
1495         const struct mount_opts *m;
1496         kuid_t uid;
1497         kgid_t gid;
1498         int arg = 0;
1499
1500 #ifdef CONFIG_QUOTA
1501         if (token == Opt_usrjquota)
1502                 return set_qf_name(sb, USRQUOTA, &args[0]);
1503         else if (token == Opt_grpjquota)
1504                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1505         else if (token == Opt_offusrjquota)
1506                 return clear_qf_name(sb, USRQUOTA);
1507         else if (token == Opt_offgrpjquota)
1508                 return clear_qf_name(sb, GRPQUOTA);
1509 #endif
1510         if (args->from && match_int(args, &arg))
1511                 return -1;
1512         switch (token) {
1513         case Opt_noacl:
1514         case Opt_nouser_xattr:
1515                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1516                 break;
1517         case Opt_sb:
1518                 return 1;       /* handled by get_sb_block() */
1519         case Opt_removed:
1520                 ext4_msg(sb, KERN_WARNING,
1521                          "Ignoring removed %s option", opt);
1522                 return 1;
1523         case Opt_resuid:
1524                 uid = make_kuid(current_user_ns(), arg);
1525                 if (!uid_valid(uid)) {
1526                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1527                         return -1;
1528                 }
1529                 sbi->s_resuid = uid;
1530                 return 1;
1531         case Opt_resgid:
1532                 gid = make_kgid(current_user_ns(), arg);
1533                 if (!gid_valid(gid)) {
1534                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1535                         return -1;
1536                 }
1537                 sbi->s_resgid = gid;
1538                 return 1;
1539         case Opt_abort:
1540                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1541                 return 1;
1542         case Opt_i_version:
1543                 sb->s_flags |= MS_I_VERSION;
1544                 return 1;
1545         case Opt_journal_dev:
1546                 if (is_remount) {
1547                         ext4_msg(sb, KERN_ERR,
1548                                  "Cannot specify journal on remount");
1549                         return -1;
1550                 }
1551                 *journal_devnum = arg;
1552                 return 1;
1553         case Opt_journal_ioprio:
1554                 if (arg < 0 || arg > 7)
1555                         return -1;
1556                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1557                 return 1;
1558         }
1559
1560         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1561                 if (token != m->token)
1562                         continue;
1563                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1564                         return -1;
1565                 if (m->flags & MOPT_EXPLICIT)
1566                         set_opt2(sb, EXPLICIT_DELALLOC);
1567                 if (m->flags & MOPT_CLEAR_ERR)
1568                         clear_opt(sb, ERRORS_MASK);
1569                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1570                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1571                                  "options when quota turned on");
1572                         return -1;
1573                 }
1574
1575                 if (m->flags & MOPT_NOSUPPORT) {
1576                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1577                 } else if (token == Opt_commit) {
1578                         if (arg == 0)
1579                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1580                         sbi->s_commit_interval = HZ * arg;
1581                 } else if (token == Opt_max_batch_time) {
1582                         if (arg == 0)
1583                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1584                         sbi->s_max_batch_time = arg;
1585                 } else if (token == Opt_min_batch_time) {
1586                         sbi->s_min_batch_time = arg;
1587                 } else if (token == Opt_inode_readahead_blks) {
1588                         if (arg > (1 << 30))
1589                                 return -1;
1590                         if (arg && !is_power_of_2(arg)) {
1591                                 ext4_msg(sb, KERN_ERR,
1592                                          "EXT4-fs: inode_readahead_blks"
1593                                          " must be a power of 2");
1594                                 return -1;
1595                         }
1596                         sbi->s_inode_readahead_blks = arg;
1597                 } else if (token == Opt_init_itable) {
1598                         set_opt(sb, INIT_INODE_TABLE);
1599                         if (!args->from)
1600                                 arg = EXT4_DEF_LI_WAIT_MULT;
1601                         sbi->s_li_wait_mult = arg;
1602                 } else if (token == Opt_max_dir_size_kb) {
1603                         sbi->s_max_dir_size_kb = arg;
1604                 } else if (token == Opt_stripe) {
1605                         sbi->s_stripe = arg;
1606                 } else if (m->flags & MOPT_DATAJ) {
1607                         if (is_remount) {
1608                                 if (!sbi->s_journal)
1609                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1610                                 else if (test_opt(sb, DATA_FLAGS) !=
1611                                          m->mount_opt) {
1612                                         ext4_msg(sb, KERN_ERR,
1613                                          "Cannot change data mode on remount");
1614                                         return -1;
1615                                 }
1616                         } else {
1617                                 clear_opt(sb, DATA_FLAGS);
1618                                 sbi->s_mount_opt |= m->mount_opt;
1619                         }
1620 #ifdef CONFIG_QUOTA
1621                 } else if (m->flags & MOPT_QFMT) {
1622                         if (sb_any_quota_loaded(sb) &&
1623                             sbi->s_jquota_fmt != m->mount_opt) {
1624                                 ext4_msg(sb, KERN_ERR, "Cannot "
1625                                          "change journaled quota options "
1626                                          "when quota turned on");
1627                                 return -1;
1628                         }
1629                         sbi->s_jquota_fmt = m->mount_opt;
1630 #endif
1631                 } else {
1632                         if (!args->from)
1633                                 arg = 1;
1634                         if (m->flags & MOPT_CLEAR)
1635                                 arg = !arg;
1636                         else if (unlikely(!(m->flags & MOPT_SET))) {
1637                                 ext4_msg(sb, KERN_WARNING,
1638                                          "buggy handling of option %s", opt);
1639                                 WARN_ON(1);
1640                                 return -1;
1641                         }
1642                         if (arg != 0)
1643                                 sbi->s_mount_opt |= m->mount_opt;
1644                         else
1645                                 sbi->s_mount_opt &= ~m->mount_opt;
1646                 }
1647                 return 1;
1648         }
1649         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1650                  "or missing value", opt);
1651         return -1;
1652 }
1653
1654 static int parse_options(char *options, struct super_block *sb,
1655                          unsigned long *journal_devnum,
1656                          unsigned int *journal_ioprio,
1657                          int is_remount)
1658 {
1659 #ifdef CONFIG_QUOTA
1660         struct ext4_sb_info *sbi = EXT4_SB(sb);
1661 #endif
1662         char *p;
1663         substring_t args[MAX_OPT_ARGS];
1664         int token;
1665
1666         if (!options)
1667                 return 1;
1668
1669         while ((p = strsep(&options, ",")) != NULL) {
1670                 if (!*p)
1671                         continue;
1672                 /*
1673                  * Initialize args struct so we know whether arg was
1674                  * found; some options take optional arguments.
1675                  */
1676                 args[0].to = args[0].from = NULL;
1677                 token = match_token(p, tokens, args);
1678                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1679                                      journal_ioprio, is_remount) < 0)
1680                         return 0;
1681         }
1682 #ifdef CONFIG_QUOTA
1683         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1684                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1685                         clear_opt(sb, USRQUOTA);
1686
1687                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1688                         clear_opt(sb, GRPQUOTA);
1689
1690                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1691                         ext4_msg(sb, KERN_ERR, "old and new quota "
1692                                         "format mixing");
1693                         return 0;
1694                 }
1695
1696                 if (!sbi->s_jquota_fmt) {
1697                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1698                                         "not specified");
1699                         return 0;
1700                 }
1701         } else {
1702                 if (sbi->s_jquota_fmt) {
1703                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1704                                         "specified with no journaling "
1705                                         "enabled");
1706                         return 0;
1707                 }
1708         }
1709 #endif
1710         return 1;
1711 }
1712
1713 static inline void ext4_show_quota_options(struct seq_file *seq,
1714                                            struct super_block *sb)
1715 {
1716 #if defined(CONFIG_QUOTA)
1717         struct ext4_sb_info *sbi = EXT4_SB(sb);
1718
1719         if (sbi->s_jquota_fmt) {
1720                 char *fmtname = "";
1721
1722                 switch (sbi->s_jquota_fmt) {
1723                 case QFMT_VFS_OLD:
1724                         fmtname = "vfsold";
1725                         break;
1726                 case QFMT_VFS_V0:
1727                         fmtname = "vfsv0";
1728                         break;
1729                 case QFMT_VFS_V1:
1730                         fmtname = "vfsv1";
1731                         break;
1732                 }
1733                 seq_printf(seq, ",jqfmt=%s", fmtname);
1734         }
1735
1736         if (sbi->s_qf_names[USRQUOTA])
1737                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1738
1739         if (sbi->s_qf_names[GRPQUOTA])
1740                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1741
1742         if (test_opt(sb, USRQUOTA))
1743                 seq_puts(seq, ",usrquota");
1744
1745         if (test_opt(sb, GRPQUOTA))
1746                 seq_puts(seq, ",grpquota");
1747 #endif
1748 }
1749
1750 static const char *token2str(int token)
1751 {
1752         const struct match_token *t;
1753
1754         for (t = tokens; t->token != Opt_err; t++)
1755                 if (t->token == token && !strchr(t->pattern, '='))
1756                         break;
1757         return t->pattern;
1758 }
1759
1760 /*
1761  * Show an option if
1762  *  - it's set to a non-default value OR
1763  *  - if the per-sb default is different from the global default
1764  */
1765 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1766                               int nodefs)
1767 {
1768         struct ext4_sb_info *sbi = EXT4_SB(sb);
1769         struct ext4_super_block *es = sbi->s_es;
1770         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1771         const struct mount_opts *m;
1772         char sep = nodefs ? '\n' : ',';
1773
1774 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1775 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1776
1777         if (sbi->s_sb_block != 1)
1778                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1779
1780         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1781                 int want_set = m->flags & MOPT_SET;
1782                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1783                     (m->flags & MOPT_CLEAR_ERR))
1784                         continue;
1785                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1786                         continue; /* skip if same as the default */
1787                 if ((want_set &&
1788                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1789                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1790                         continue; /* select Opt_noFoo vs Opt_Foo */
1791                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1792         }
1793
1794         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1795             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1796                 SEQ_OPTS_PRINT("resuid=%u",
1797                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1798         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1799             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1800                 SEQ_OPTS_PRINT("resgid=%u",
1801                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1802         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1803         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1804                 SEQ_OPTS_PUTS("errors=remount-ro");
1805         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1806                 SEQ_OPTS_PUTS("errors=continue");
1807         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1808                 SEQ_OPTS_PUTS("errors=panic");
1809         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1810                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1811         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1812                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1813         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1814                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1815         if (sb->s_flags & MS_I_VERSION)
1816                 SEQ_OPTS_PUTS("i_version");
1817         if (nodefs || sbi->s_stripe)
1818                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1819         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1820                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1821                         SEQ_OPTS_PUTS("data=journal");
1822                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1823                         SEQ_OPTS_PUTS("data=ordered");
1824                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1825                         SEQ_OPTS_PUTS("data=writeback");
1826         }
1827         if (nodefs ||
1828             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1829                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1830                                sbi->s_inode_readahead_blks);
1831
1832         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1833                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1834                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1835         if (nodefs || sbi->s_max_dir_size_kb)
1836                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1837
1838         ext4_show_quota_options(seq, sb);
1839         return 0;
1840 }
1841
1842 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1843 {
1844         return _ext4_show_options(seq, root->d_sb, 0);
1845 }
1846
1847 static int options_seq_show(struct seq_file *seq, void *offset)
1848 {
1849         struct super_block *sb = seq->private;
1850         int rc;
1851
1852         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1853         rc = _ext4_show_options(seq, sb, 1);
1854         seq_puts(seq, "\n");
1855         return rc;
1856 }
1857
1858 static int options_open_fs(struct inode *inode, struct file *file)
1859 {
1860         return single_open(file, options_seq_show, PDE(inode)->data);
1861 }
1862
1863 static const struct file_operations ext4_seq_options_fops = {
1864         .owner = THIS_MODULE,
1865         .open = options_open_fs,
1866         .read = seq_read,
1867         .llseek = seq_lseek,
1868         .release = single_release,
1869 };
1870
1871 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1872                             int read_only)
1873 {
1874         struct ext4_sb_info *sbi = EXT4_SB(sb);
1875         int res = 0;
1876
1877         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1878                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1879                          "forcing read-only mode");
1880                 res = MS_RDONLY;
1881         }
1882         if (read_only)
1883                 goto done;
1884         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1885                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1886                          "running e2fsck is recommended");
1887         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1888                 ext4_msg(sb, KERN_WARNING,
1889                          "warning: mounting fs with errors, "
1890                          "running e2fsck is recommended");
1891         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1892                  le16_to_cpu(es->s_mnt_count) >=
1893                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1894                 ext4_msg(sb, KERN_WARNING,
1895                          "warning: maximal mount count reached, "
1896                          "running e2fsck is recommended");
1897         else if (le32_to_cpu(es->s_checkinterval) &&
1898                 (le32_to_cpu(es->s_lastcheck) +
1899                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1900                 ext4_msg(sb, KERN_WARNING,
1901                          "warning: checktime reached, "
1902                          "running e2fsck is recommended");
1903         if (!sbi->s_journal)
1904                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1905         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1906                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1907         le16_add_cpu(&es->s_mnt_count, 1);
1908         es->s_mtime = cpu_to_le32(get_seconds());
1909         ext4_update_dynamic_rev(sb);
1910         if (sbi->s_journal)
1911                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1912
1913         ext4_commit_super(sb, 1);
1914 done:
1915         if (test_opt(sb, DEBUG))
1916                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1917                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1918                         sb->s_blocksize,
1919                         sbi->s_groups_count,
1920                         EXT4_BLOCKS_PER_GROUP(sb),
1921                         EXT4_INODES_PER_GROUP(sb),
1922                         sbi->s_mount_opt, sbi->s_mount_opt2);
1923
1924         cleancache_init_fs(sb);
1925         return res;
1926 }
1927
1928 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1929 {
1930         struct ext4_sb_info *sbi = EXT4_SB(sb);
1931         struct flex_groups *new_groups;
1932         int size;
1933
1934         if (!sbi->s_log_groups_per_flex)
1935                 return 0;
1936
1937         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1938         if (size <= sbi->s_flex_groups_allocated)
1939                 return 0;
1940
1941         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1942         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1943         if (!new_groups) {
1944                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1945                          size / (int) sizeof(struct flex_groups));
1946                 return -ENOMEM;
1947         }
1948
1949         if (sbi->s_flex_groups) {
1950                 memcpy(new_groups, sbi->s_flex_groups,
1951                        (sbi->s_flex_groups_allocated *
1952                         sizeof(struct flex_groups)));
1953                 ext4_kvfree(sbi->s_flex_groups);
1954         }
1955         sbi->s_flex_groups = new_groups;
1956         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1957         return 0;
1958 }
1959
1960 static int ext4_fill_flex_info(struct super_block *sb)
1961 {
1962         struct ext4_sb_info *sbi = EXT4_SB(sb);
1963         struct ext4_group_desc *gdp = NULL;
1964         ext4_group_t flex_group;
1965         unsigned int groups_per_flex = 0;
1966         int i, err;
1967
1968         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1969         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1970                 sbi->s_log_groups_per_flex = 0;
1971                 return 1;
1972         }
1973         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1974
1975         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1976         if (err)
1977                 goto failed;
1978
1979         for (i = 0; i < sbi->s_groups_count; i++) {
1980                 gdp = ext4_get_group_desc(sb, i, NULL);
1981
1982                 flex_group = ext4_flex_group(sbi, i);
1983                 atomic_add(ext4_free_inodes_count(sb, gdp),
1984                            &sbi->s_flex_groups[flex_group].free_inodes);
1985                 atomic_add(ext4_free_group_clusters(sb, gdp),
1986                            &sbi->s_flex_groups[flex_group].free_clusters);
1987                 atomic_add(ext4_used_dirs_count(sb, gdp),
1988                            &sbi->s_flex_groups[flex_group].used_dirs);
1989         }
1990
1991         return 1;
1992 failed:
1993         return 0;
1994 }
1995
1996 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1997                                    struct ext4_group_desc *gdp)
1998 {
1999         int offset;
2000         __u16 crc = 0;
2001         __le32 le_group = cpu_to_le32(block_group);
2002
2003         if ((sbi->s_es->s_feature_ro_compat &
2004              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
2005                 /* Use new metadata_csum algorithm */
2006                 __u16 old_csum;
2007                 __u32 csum32;
2008
2009                 old_csum = gdp->bg_checksum;
2010                 gdp->bg_checksum = 0;
2011                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2012                                      sizeof(le_group));
2013                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2014                                      sbi->s_desc_size);
2015                 gdp->bg_checksum = old_csum;
2016
2017                 crc = csum32 & 0xFFFF;
2018                 goto out;
2019         }
2020
2021         /* old crc16 code */
2022         offset = offsetof(struct ext4_group_desc, bg_checksum);
2023
2024         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2025         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2026         crc = crc16(crc, (__u8 *)gdp, offset);
2027         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2028         /* for checksum of struct ext4_group_desc do the rest...*/
2029         if ((sbi->s_es->s_feature_incompat &
2030              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2031             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2032                 crc = crc16(crc, (__u8 *)gdp + offset,
2033                             le16_to_cpu(sbi->s_es->s_desc_size) -
2034                                 offset);
2035
2036 out:
2037         return cpu_to_le16(crc);
2038 }
2039
2040 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2041                                 struct ext4_group_desc *gdp)
2042 {
2043         if (ext4_has_group_desc_csum(sb) &&
2044             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2045                                                       block_group, gdp)))
2046                 return 0;
2047
2048         return 1;
2049 }
2050
2051 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2052                               struct ext4_group_desc *gdp)
2053 {
2054         if (!ext4_has_group_desc_csum(sb))
2055                 return;
2056         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2057 }
2058
2059 /* Called at mount-time, super-block is locked */
2060 static int ext4_check_descriptors(struct super_block *sb,
2061                                   ext4_group_t *first_not_zeroed)
2062 {
2063         struct ext4_sb_info *sbi = EXT4_SB(sb);
2064         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2065         ext4_fsblk_t last_block;
2066         ext4_fsblk_t block_bitmap;
2067         ext4_fsblk_t inode_bitmap;
2068         ext4_fsblk_t inode_table;
2069         int flexbg_flag = 0;
2070         ext4_group_t i, grp = sbi->s_groups_count;
2071
2072         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2073                 flexbg_flag = 1;
2074
2075         ext4_debug("Checking group descriptors");
2076
2077         for (i = 0; i < sbi->s_groups_count; i++) {
2078                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2079
2080                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2081                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2082                 else
2083                         last_block = first_block +
2084                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2085
2086                 if ((grp == sbi->s_groups_count) &&
2087                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2088                         grp = i;
2089
2090                 block_bitmap = ext4_block_bitmap(sb, gdp);
2091                 if (block_bitmap < first_block || block_bitmap > last_block) {
2092                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2093                                "Block bitmap for group %u not in group "
2094                                "(block %llu)!", i, block_bitmap);
2095                         return 0;
2096                 }
2097                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2098                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2099                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2100                                "Inode bitmap for group %u not in group "
2101                                "(block %llu)!", i, inode_bitmap);
2102                         return 0;
2103                 }
2104                 inode_table = ext4_inode_table(sb, gdp);
2105                 if (inode_table < first_block ||
2106                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2107                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2108                                "Inode table for group %u not in group "
2109                                "(block %llu)!", i, inode_table);
2110                         return 0;
2111                 }
2112                 ext4_lock_group(sb, i);
2113                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2114                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2115                                  "Checksum for group %u failed (%u!=%u)",
2116                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2117                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2118                         if (!(sb->s_flags & MS_RDONLY)) {
2119                                 ext4_unlock_group(sb, i);
2120                                 return 0;
2121                         }
2122                 }
2123                 ext4_unlock_group(sb, i);
2124                 if (!flexbg_flag)
2125                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2126         }
2127         if (NULL != first_not_zeroed)
2128                 *first_not_zeroed = grp;
2129
2130         ext4_free_blocks_count_set(sbi->s_es,
2131                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2132         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2133         return 1;
2134 }
2135
2136 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2137  * the superblock) which were deleted from all directories, but held open by
2138  * a process at the time of a crash.  We walk the list and try to delete these
2139  * inodes at recovery time (only with a read-write filesystem).
2140  *
2141  * In order to keep the orphan inode chain consistent during traversal (in
2142  * case of crash during recovery), we link each inode into the superblock
2143  * orphan list_head and handle it the same way as an inode deletion during
2144  * normal operation (which journals the operations for us).
2145  *
2146  * We only do an iget() and an iput() on each inode, which is very safe if we
2147  * accidentally point at an in-use or already deleted inode.  The worst that
2148  * can happen in this case is that we get a "bit already cleared" message from
2149  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2150  * e2fsck was run on this filesystem, and it must have already done the orphan
2151  * inode cleanup for us, so we can safely abort without any further action.
2152  */
2153 static void ext4_orphan_cleanup(struct super_block *sb,
2154                                 struct ext4_super_block *es)
2155 {
2156         unsigned int s_flags = sb->s_flags;
2157         int nr_orphans = 0, nr_truncates = 0;
2158 #ifdef CONFIG_QUOTA
2159         int i;
2160 #endif
2161         if (!es->s_last_orphan) {
2162                 jbd_debug(4, "no orphan inodes to clean up\n");
2163                 return;
2164         }
2165
2166         if (bdev_read_only(sb->s_bdev)) {
2167                 ext4_msg(sb, KERN_ERR, "write access "
2168                         "unavailable, skipping orphan cleanup");
2169                 return;
2170         }
2171
2172         /* Check if feature set would not allow a r/w mount */
2173         if (!ext4_feature_set_ok(sb, 0)) {
2174                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2175                          "unknown ROCOMPAT features");
2176                 return;
2177         }
2178
2179         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2180                 /* don't clear list on RO mount w/ errors */
2181                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2182                         jbd_debug(1, "Errors on filesystem, "
2183                                   "clearing orphan list.\n");
2184                         es->s_last_orphan = 0;
2185                 }
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 inode_readahead_blks_store(struct ext4_attr *a,
2483                                           struct ext4_sb_info *sbi,
2484                                           const char *buf, size_t count)
2485 {
2486         unsigned long t;
2487
2488         if (parse_strtoul(buf, 0x40000000, &t))
2489                 return -EINVAL;
2490
2491         if (t && !is_power_of_2(t))
2492                 return -EINVAL;
2493
2494         sbi->s_inode_readahead_blks = t;
2495         return count;
2496 }
2497
2498 static ssize_t sbi_ui_show(struct ext4_attr *a,
2499                            struct ext4_sb_info *sbi, char *buf)
2500 {
2501         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2502
2503         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2504 }
2505
2506 static ssize_t sbi_ui_store(struct ext4_attr *a,
2507                             struct ext4_sb_info *sbi,
2508                             const char *buf, size_t count)
2509 {
2510         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2511         unsigned long t;
2512
2513         if (parse_strtoul(buf, 0xffffffff, &t))
2514                 return -EINVAL;
2515         *ui = t;
2516         return count;
2517 }
2518
2519 static ssize_t trigger_test_error(struct ext4_attr *a,
2520                                   struct ext4_sb_info *sbi,
2521                                   const char *buf, size_t count)
2522 {
2523         int len = count;
2524
2525         if (!capable(CAP_SYS_ADMIN))
2526                 return -EPERM;
2527
2528         if (len && buf[len-1] == '\n')
2529                 len--;
2530
2531         if (len)
2532                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2533         return count;
2534 }
2535
2536 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2537 static struct ext4_attr ext4_attr_##_name = {                   \
2538         .attr = {.name = __stringify(_name), .mode = _mode },   \
2539         .show   = _show,                                        \
2540         .store  = _store,                                       \
2541         .offset = offsetof(struct ext4_sb_info, _elname),       \
2542 }
2543 #define EXT4_ATTR(name, mode, show, store) \
2544 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2545
2546 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2547 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2548 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2549 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2550         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2551 #define ATTR_LIST(name) &ext4_attr_##name.attr
2552
2553 EXT4_RO_ATTR(delayed_allocation_blocks);
2554 EXT4_RO_ATTR(session_write_kbytes);
2555 EXT4_RO_ATTR(lifetime_write_kbytes);
2556 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2557                  inode_readahead_blks_store, s_inode_readahead_blks);
2558 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2559 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2560 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2561 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2562 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2563 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2564 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2565 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2566 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2567 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2568
2569 static struct attribute *ext4_attrs[] = {
2570         ATTR_LIST(delayed_allocation_blocks),
2571         ATTR_LIST(session_write_kbytes),
2572         ATTR_LIST(lifetime_write_kbytes),
2573         ATTR_LIST(inode_readahead_blks),
2574         ATTR_LIST(inode_goal),
2575         ATTR_LIST(mb_stats),
2576         ATTR_LIST(mb_max_to_scan),
2577         ATTR_LIST(mb_min_to_scan),
2578         ATTR_LIST(mb_order2_req),
2579         ATTR_LIST(mb_stream_req),
2580         ATTR_LIST(mb_group_prealloc),
2581         ATTR_LIST(max_writeback_mb_bump),
2582         ATTR_LIST(extent_max_zeroout_kb),
2583         ATTR_LIST(trigger_fs_error),
2584         NULL,
2585 };
2586
2587 /* Features this copy of ext4 supports */
2588 EXT4_INFO_ATTR(lazy_itable_init);
2589 EXT4_INFO_ATTR(batched_discard);
2590 EXT4_INFO_ATTR(meta_bg_resize);
2591
2592 static struct attribute *ext4_feat_attrs[] = {
2593         ATTR_LIST(lazy_itable_init),
2594         ATTR_LIST(batched_discard),
2595         ATTR_LIST(meta_bg_resize),
2596         NULL,
2597 };
2598
2599 static ssize_t ext4_attr_show(struct kobject *kobj,
2600                               struct attribute *attr, char *buf)
2601 {
2602         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2603                                                 s_kobj);
2604         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2605
2606         return a->show ? a->show(a, sbi, buf) : 0;
2607 }
2608
2609 static ssize_t ext4_attr_store(struct kobject *kobj,
2610                                struct attribute *attr,
2611                                const char *buf, size_t len)
2612 {
2613         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2614                                                 s_kobj);
2615         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2616
2617         return a->store ? a->store(a, sbi, buf, len) : 0;
2618 }
2619
2620 static void ext4_sb_release(struct kobject *kobj)
2621 {
2622         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2623                                                 s_kobj);
2624         complete(&sbi->s_kobj_unregister);
2625 }
2626
2627 static const struct sysfs_ops ext4_attr_ops = {
2628         .show   = ext4_attr_show,
2629         .store  = ext4_attr_store,
2630 };
2631
2632 static struct kobj_type ext4_ktype = {
2633         .default_attrs  = ext4_attrs,
2634         .sysfs_ops      = &ext4_attr_ops,
2635         .release        = ext4_sb_release,
2636 };
2637
2638 static void ext4_feat_release(struct kobject *kobj)
2639 {
2640         complete(&ext4_feat->f_kobj_unregister);
2641 }
2642
2643 static struct kobj_type ext4_feat_ktype = {
2644         .default_attrs  = ext4_feat_attrs,
2645         .sysfs_ops      = &ext4_attr_ops,
2646         .release        = ext4_feat_release,
2647 };
2648
2649 /*
2650  * Check whether this filesystem can be mounted based on
2651  * the features present and the RDONLY/RDWR mount requested.
2652  * Returns 1 if this filesystem can be mounted as requested,
2653  * 0 if it cannot be.
2654  */
2655 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2656 {
2657         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2658                 ext4_msg(sb, KERN_ERR,
2659                         "Couldn't mount because of "
2660                         "unsupported optional features (%x)",
2661                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2662                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2663                 return 0;
2664         }
2665
2666         if (readonly)
2667                 return 1;
2668
2669         /* Check that feature set is OK for a read-write mount */
2670         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2671                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2672                          "unsupported optional features (%x)",
2673                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2674                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2675                 return 0;
2676         }
2677         /*
2678          * Large file size enabled file system can only be mounted
2679          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2680          */
2681         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2682                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2683                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2684                                  "cannot be mounted RDWR without "
2685                                  "CONFIG_LBDAF");
2686                         return 0;
2687                 }
2688         }
2689         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2690             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2691                 ext4_msg(sb, KERN_ERR,
2692                          "Can't support bigalloc feature without "
2693                          "extents feature\n");
2694                 return 0;
2695         }
2696
2697 #ifndef CONFIG_QUOTA
2698         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2699             !readonly) {
2700                 ext4_msg(sb, KERN_ERR,
2701                          "Filesystem with quota feature cannot be mounted RDWR "
2702                          "without CONFIG_QUOTA");
2703                 return 0;
2704         }
2705 #endif  /* CONFIG_QUOTA */
2706         return 1;
2707 }
2708
2709 /*
2710  * This function is called once a day if we have errors logged
2711  * on the file system
2712  */
2713 static void print_daily_error_info(unsigned long arg)
2714 {
2715         struct super_block *sb = (struct super_block *) arg;
2716         struct ext4_sb_info *sbi;
2717         struct ext4_super_block *es;
2718
2719         sbi = EXT4_SB(sb);
2720         es = sbi->s_es;
2721
2722         if (es->s_error_count)
2723                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2724                          le32_to_cpu(es->s_error_count));
2725         if (es->s_first_error_time) {
2726                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2727                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2728                        (int) sizeof(es->s_first_error_func),
2729                        es->s_first_error_func,
2730                        le32_to_cpu(es->s_first_error_line));
2731                 if (es->s_first_error_ino)
2732                         printk(": inode %u",
2733                                le32_to_cpu(es->s_first_error_ino));
2734                 if (es->s_first_error_block)
2735                         printk(": block %llu", (unsigned long long)
2736                                le64_to_cpu(es->s_first_error_block));
2737                 printk("\n");
2738         }
2739         if (es->s_last_error_time) {
2740                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2741                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2742                        (int) sizeof(es->s_last_error_func),
2743                        es->s_last_error_func,
2744                        le32_to_cpu(es->s_last_error_line));
2745                 if (es->s_last_error_ino)
2746                         printk(": inode %u",
2747                                le32_to_cpu(es->s_last_error_ino));
2748                 if (es->s_last_error_block)
2749                         printk(": block %llu", (unsigned long long)
2750                                le64_to_cpu(es->s_last_error_block));
2751                 printk("\n");
2752         }
2753         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2754 }
2755
2756 /* Find next suitable group and run ext4_init_inode_table */
2757 static int ext4_run_li_request(struct ext4_li_request *elr)
2758 {
2759         struct ext4_group_desc *gdp = NULL;
2760         ext4_group_t group, ngroups;
2761         struct super_block *sb;
2762         unsigned long timeout = 0;
2763         int ret = 0;
2764
2765         sb = elr->lr_super;
2766         ngroups = EXT4_SB(sb)->s_groups_count;
2767
2768         sb_start_write(sb);
2769         for (group = elr->lr_next_group; group < ngroups; group++) {
2770                 gdp = ext4_get_group_desc(sb, group, NULL);
2771                 if (!gdp) {
2772                         ret = 1;
2773                         break;
2774                 }
2775
2776                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2777                         break;
2778         }
2779
2780         if (group == ngroups)
2781                 ret = 1;
2782
2783         if (!ret) {
2784                 timeout = jiffies;
2785                 ret = ext4_init_inode_table(sb, group,
2786                                             elr->lr_timeout ? 0 : 1);
2787                 if (elr->lr_timeout == 0) {
2788                         timeout = (jiffies - timeout) *
2789                                   elr->lr_sbi->s_li_wait_mult;
2790                         elr->lr_timeout = timeout;
2791                 }
2792                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2793                 elr->lr_next_group = group + 1;
2794         }
2795         sb_end_write(sb);
2796
2797         return ret;
2798 }
2799
2800 /*
2801  * Remove lr_request from the list_request and free the
2802  * request structure. Should be called with li_list_mtx held
2803  */
2804 static void ext4_remove_li_request(struct ext4_li_request *elr)
2805 {
2806         struct ext4_sb_info *sbi;
2807
2808         if (!elr)
2809                 return;
2810
2811         sbi = elr->lr_sbi;
2812
2813         list_del(&elr->lr_request);
2814         sbi->s_li_request = NULL;
2815         kfree(elr);
2816 }
2817
2818 static void ext4_unregister_li_request(struct super_block *sb)
2819 {
2820         mutex_lock(&ext4_li_mtx);
2821         if (!ext4_li_info) {
2822                 mutex_unlock(&ext4_li_mtx);
2823                 return;
2824         }
2825
2826         mutex_lock(&ext4_li_info->li_list_mtx);
2827         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2828         mutex_unlock(&ext4_li_info->li_list_mtx);
2829         mutex_unlock(&ext4_li_mtx);
2830 }
2831
2832 static struct task_struct *ext4_lazyinit_task;
2833
2834 /*
2835  * This is the function where ext4lazyinit thread lives. It walks
2836  * through the request list searching for next scheduled filesystem.
2837  * When such a fs is found, run the lazy initialization request
2838  * (ext4_rn_li_request) and keep track of the time spend in this
2839  * function. Based on that time we compute next schedule time of
2840  * the request. When walking through the list is complete, compute
2841  * next waking time and put itself into sleep.
2842  */
2843 static int ext4_lazyinit_thread(void *arg)
2844 {
2845         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2846         struct list_head *pos, *n;
2847         struct ext4_li_request *elr;
2848         unsigned long next_wakeup, cur;
2849
2850         BUG_ON(NULL == eli);
2851
2852 cont_thread:
2853         while (true) {
2854                 next_wakeup = MAX_JIFFY_OFFSET;
2855
2856                 mutex_lock(&eli->li_list_mtx);
2857                 if (list_empty(&eli->li_request_list)) {
2858                         mutex_unlock(&eli->li_list_mtx);
2859                         goto exit_thread;
2860                 }
2861
2862                 list_for_each_safe(pos, n, &eli->li_request_list) {
2863                         elr = list_entry(pos, struct ext4_li_request,
2864                                          lr_request);
2865
2866                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2867                                 if (ext4_run_li_request(elr) != 0) {
2868                                         /* error, remove the lazy_init job */
2869                                         ext4_remove_li_request(elr);
2870                                         continue;
2871                                 }
2872                         }
2873
2874                         if (time_before(elr->lr_next_sched, next_wakeup))
2875                                 next_wakeup = elr->lr_next_sched;
2876                 }
2877                 mutex_unlock(&eli->li_list_mtx);
2878
2879                 try_to_freeze();
2880
2881                 cur = jiffies;
2882                 if ((time_after_eq(cur, next_wakeup)) ||
2883                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2884                         cond_resched();
2885                         continue;
2886                 }
2887
2888                 schedule_timeout_interruptible(next_wakeup - cur);
2889
2890                 if (kthread_should_stop()) {
2891                         ext4_clear_request_list();
2892                         goto exit_thread;
2893                 }
2894         }
2895
2896 exit_thread:
2897         /*
2898          * It looks like the request list is empty, but we need
2899          * to check it under the li_list_mtx lock, to prevent any
2900          * additions into it, and of course we should lock ext4_li_mtx
2901          * to atomically free the list and ext4_li_info, because at
2902          * this point another ext4 filesystem could be registering
2903          * new one.
2904          */
2905         mutex_lock(&ext4_li_mtx);
2906         mutex_lock(&eli->li_list_mtx);
2907         if (!list_empty(&eli->li_request_list)) {
2908                 mutex_unlock(&eli->li_list_mtx);
2909                 mutex_unlock(&ext4_li_mtx);
2910                 goto cont_thread;
2911         }
2912         mutex_unlock(&eli->li_list_mtx);
2913         kfree(ext4_li_info);
2914         ext4_li_info = NULL;
2915         mutex_unlock(&ext4_li_mtx);
2916
2917         return 0;
2918 }
2919
2920 static void ext4_clear_request_list(void)
2921 {
2922         struct list_head *pos, *n;
2923         struct ext4_li_request *elr;
2924
2925         mutex_lock(&ext4_li_info->li_list_mtx);
2926         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2927                 elr = list_entry(pos, struct ext4_li_request,
2928                                  lr_request);
2929                 ext4_remove_li_request(elr);
2930         }
2931         mutex_unlock(&ext4_li_info->li_list_mtx);
2932 }
2933
2934 static int ext4_run_lazyinit_thread(void)
2935 {
2936         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2937                                          ext4_li_info, "ext4lazyinit");
2938         if (IS_ERR(ext4_lazyinit_task)) {
2939                 int err = PTR_ERR(ext4_lazyinit_task);
2940                 ext4_clear_request_list();
2941                 kfree(ext4_li_info);
2942                 ext4_li_info = NULL;
2943                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2944                                  "initialization thread\n",
2945                                  err);
2946                 return err;
2947         }
2948         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2949         return 0;
2950 }
2951
2952 /*
2953  * Check whether it make sense to run itable init. thread or not.
2954  * If there is at least one uninitialized inode table, return
2955  * corresponding group number, else the loop goes through all
2956  * groups and return total number of groups.
2957  */
2958 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2959 {
2960         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2961         struct ext4_group_desc *gdp = NULL;
2962
2963         for (group = 0; group < ngroups; group++) {
2964                 gdp = ext4_get_group_desc(sb, group, NULL);
2965                 if (!gdp)
2966                         continue;
2967
2968                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2969                         break;
2970         }
2971
2972         return group;
2973 }
2974
2975 static int ext4_li_info_new(void)
2976 {
2977         struct ext4_lazy_init *eli = NULL;
2978
2979         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2980         if (!eli)
2981                 return -ENOMEM;
2982
2983         INIT_LIST_HEAD(&eli->li_request_list);
2984         mutex_init(&eli->li_list_mtx);
2985
2986         eli->li_state |= EXT4_LAZYINIT_QUIT;
2987
2988         ext4_li_info = eli;
2989
2990         return 0;
2991 }
2992
2993 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2994                                             ext4_group_t start)
2995 {
2996         struct ext4_sb_info *sbi = EXT4_SB(sb);
2997         struct ext4_li_request *elr;
2998         unsigned long rnd;
2999
3000         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3001         if (!elr)
3002                 return NULL;
3003
3004         elr->lr_super = sb;
3005         elr->lr_sbi = sbi;
3006         elr->lr_next_group = start;
3007
3008         /*
3009          * Randomize first schedule time of the request to
3010          * spread the inode table initialization requests
3011          * better.
3012          */
3013         get_random_bytes(&rnd, sizeof(rnd));
3014         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3015                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3016
3017         return elr;
3018 }
3019
3020 static int ext4_register_li_request(struct super_block *sb,
3021                                     ext4_group_t first_not_zeroed)
3022 {
3023         struct ext4_sb_info *sbi = EXT4_SB(sb);
3024         struct ext4_li_request *elr;
3025         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3026         int ret = 0;
3027
3028         if (sbi->s_li_request != NULL) {
3029                 /*
3030                  * Reset timeout so it can be computed again, because
3031                  * s_li_wait_mult might have changed.
3032                  */
3033                 sbi->s_li_request->lr_timeout = 0;
3034                 return 0;
3035         }
3036
3037         if (first_not_zeroed == ngroups ||
3038             (sb->s_flags & MS_RDONLY) ||
3039             !test_opt(sb, INIT_INODE_TABLE))
3040                 return 0;
3041
3042         elr = ext4_li_request_new(sb, first_not_zeroed);
3043         if (!elr)
3044                 return -ENOMEM;
3045
3046         mutex_lock(&ext4_li_mtx);
3047
3048         if (NULL == ext4_li_info) {
3049                 ret = ext4_li_info_new();
3050                 if (ret)
3051                         goto out;
3052         }
3053
3054         mutex_lock(&ext4_li_info->li_list_mtx);
3055         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3056         mutex_unlock(&ext4_li_info->li_list_mtx);
3057
3058         sbi->s_li_request = elr;
3059         /*
3060          * set elr to NULL here since it has been inserted to
3061          * the request_list and the removal and free of it is
3062          * handled by ext4_clear_request_list from now on.
3063          */
3064         elr = NULL;
3065
3066         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3067                 ret = ext4_run_lazyinit_thread();
3068                 if (ret)
3069                         goto out;
3070         }
3071 out:
3072         mutex_unlock(&ext4_li_mtx);
3073         if (ret)
3074                 kfree(elr);
3075         return ret;
3076 }
3077
3078 /*
3079  * We do not need to lock anything since this is called on
3080  * module unload.
3081  */
3082 static void ext4_destroy_lazyinit_thread(void)
3083 {
3084         /*
3085          * If thread exited earlier
3086          * there's nothing to be done.
3087          */
3088         if (!ext4_li_info || !ext4_lazyinit_task)
3089                 return;
3090
3091         kthread_stop(ext4_lazyinit_task);
3092 }
3093
3094 static int set_journal_csum_feature_set(struct super_block *sb)
3095 {
3096         int ret = 1;
3097         int compat, incompat;
3098         struct ext4_sb_info *sbi = EXT4_SB(sb);
3099
3100         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3101                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3102                 /* journal checksum v2 */
3103                 compat = 0;
3104                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3105         } else {
3106                 /* journal checksum v1 */
3107                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3108                 incompat = 0;
3109         }
3110
3111         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3112                 ret = jbd2_journal_set_features(sbi->s_journal,
3113                                 compat, 0,
3114                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3115                                 incompat);
3116         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3117                 ret = jbd2_journal_set_features(sbi->s_journal,
3118                                 compat, 0,
3119                                 incompat);
3120                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3121                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3122         } else {
3123                 jbd2_journal_clear_features(sbi->s_journal,
3124                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3125                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3126                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3127         }
3128
3129         return ret;
3130 }
3131
3132 /*
3133  * Note: calculating the overhead so we can be compatible with
3134  * historical BSD practice is quite difficult in the face of
3135  * clusters/bigalloc.  This is because multiple metadata blocks from
3136  * different block group can end up in the same allocation cluster.
3137  * Calculating the exact overhead in the face of clustered allocation
3138  * requires either O(all block bitmaps) in memory or O(number of block
3139  * groups**2) in time.  We will still calculate the superblock for
3140  * older file systems --- and if we come across with a bigalloc file
3141  * system with zero in s_overhead_clusters the estimate will be close to
3142  * correct especially for very large cluster sizes --- but for newer
3143  * file systems, it's better to calculate this figure once at mkfs
3144  * time, and store it in the superblock.  If the superblock value is
3145  * present (even for non-bigalloc file systems), we will use it.
3146  */
3147 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3148                           char *buf)
3149 {
3150         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3151         struct ext4_group_desc  *gdp;
3152         ext4_fsblk_t            first_block, last_block, b;
3153         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3154         int                     s, j, count = 0;
3155
3156         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3157                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3158                         sbi->s_itb_per_group + 2);
3159
3160         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3161                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3162         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3163         for (i = 0; i < ngroups; i++) {
3164                 gdp = ext4_get_group_desc(sb, i, NULL);
3165                 b = ext4_block_bitmap(sb, gdp);
3166                 if (b >= first_block && b <= last_block) {
3167                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3168                         count++;
3169                 }
3170                 b = ext4_inode_bitmap(sb, gdp);
3171                 if (b >= first_block && b <= last_block) {
3172                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3173                         count++;
3174                 }
3175                 b = ext4_inode_table(sb, gdp);
3176                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3177                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3178                                 int c = EXT4_B2C(sbi, b - first_block);
3179                                 ext4_set_bit(c, buf);
3180                                 count++;
3181                         }
3182                 if (i != grp)
3183                         continue;
3184                 s = 0;
3185                 if (ext4_bg_has_super(sb, grp)) {
3186                         ext4_set_bit(s++, buf);
3187                         count++;
3188                 }
3189                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3190                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3191                         count++;
3192                 }
3193         }
3194         if (!count)
3195                 return 0;
3196         return EXT4_CLUSTERS_PER_GROUP(sb) -
3197                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3198 }
3199
3200 /*
3201  * Compute the overhead and stash it in sbi->s_overhead
3202  */
3203 int ext4_calculate_overhead(struct super_block *sb)
3204 {
3205         struct ext4_sb_info *sbi = EXT4_SB(sb);
3206         struct ext4_super_block *es = sbi->s_es;
3207         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3208         ext4_fsblk_t overhead = 0;
3209         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3210
3211         memset(buf, 0, PAGE_SIZE);
3212         if (!buf)
3213                 return -ENOMEM;
3214
3215         /*
3216          * Compute the overhead (FS structures).  This is constant
3217          * for a given filesystem unless the number of block groups
3218          * changes so we cache the previous value until it does.
3219          */
3220
3221         /*
3222          * All of the blocks before first_data_block are overhead
3223          */
3224         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3225
3226         /*
3227          * Add the overhead found in each block group
3228          */
3229         for (i = 0; i < ngroups; i++) {
3230                 int blks;
3231
3232                 blks = count_overhead(sb, i, buf);
3233                 overhead += blks;
3234                 if (blks)
3235                         memset(buf, 0, PAGE_SIZE);
3236                 cond_resched();
3237         }
3238         sbi->s_overhead = overhead;
3239         smp_wmb();
3240         free_page((unsigned long) buf);
3241         return 0;
3242 }
3243
3244 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3245 {
3246         char *orig_data = kstrdup(data, GFP_KERNEL);
3247         struct buffer_head *bh;
3248         struct ext4_super_block *es = NULL;
3249         struct ext4_sb_info *sbi;
3250         ext4_fsblk_t block;
3251         ext4_fsblk_t sb_block = get_sb_block(&data);
3252         ext4_fsblk_t logical_sb_block;
3253         unsigned long offset = 0;
3254         unsigned long journal_devnum = 0;
3255         unsigned long def_mount_opts;
3256         struct inode *root;
3257         char *cp;
3258         const char *descr;
3259         int ret = -ENOMEM;
3260         int blocksize, clustersize;
3261         unsigned int db_count;
3262         unsigned int i;
3263         int needs_recovery, has_huge_files, has_bigalloc;
3264         __u64 blocks_count;
3265         int err;
3266         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3267         ext4_group_t first_not_zeroed;
3268
3269         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3270         if (!sbi)
3271                 goto out_free_orig;
3272
3273         sbi->s_blockgroup_lock =
3274                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3275         if (!sbi->s_blockgroup_lock) {
3276                 kfree(sbi);
3277                 goto out_free_orig;
3278         }
3279         sb->s_fs_info = sbi;
3280         sbi->s_sb = sb;
3281         sbi->s_mount_opt = 0;
3282         sbi->s_resuid = make_kuid(&init_user_ns, EXT4_DEF_RESUID);
3283         sbi->s_resgid = make_kgid(&init_user_ns, EXT4_DEF_RESGID);
3284         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3285         sbi->s_sb_block = sb_block;
3286         if (sb->s_bdev->bd_part)
3287                 sbi->s_sectors_written_start =
3288                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3289
3290         /* Cleanup superblock name */
3291         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3292                 *cp = '!';
3293
3294         ret = -EINVAL;
3295         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3296         if (!blocksize) {
3297                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3298                 goto out_fail;
3299         }
3300
3301         /*
3302          * The ext4 superblock will not be buffer aligned for other than 1kB
3303          * block sizes.  We need to calculate the offset from buffer start.
3304          */
3305         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3306                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3307                 offset = do_div(logical_sb_block, blocksize);
3308         } else {
3309                 logical_sb_block = sb_block;
3310         }
3311
3312         if (!(bh = sb_bread(sb, logical_sb_block))) {
3313                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3314                 goto out_fail;
3315         }
3316         /*
3317          * Note: s_es must be initialized as soon as possible because
3318          *       some ext4 macro-instructions depend on its value
3319          */
3320         es = (struct ext4_super_block *) (bh->b_data + offset);
3321         sbi->s_es = es;
3322         sb->s_magic = le16_to_cpu(es->s_magic);
3323         if (sb->s_magic != EXT4_SUPER_MAGIC)
3324                 goto cantfind_ext4;
3325         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3326
3327         /* Warn if metadata_csum and gdt_csum are both set. */
3328         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3329                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3330             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3331                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3332                              "redundant flags; please run fsck.");
3333
3334         /* Check for a known checksum algorithm */
3335         if (!ext4_verify_csum_type(sb, es)) {
3336                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3337                          "unknown checksum algorithm.");
3338                 silent = 1;
3339                 goto cantfind_ext4;
3340         }
3341
3342         /* Load the checksum driver */
3343         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3344                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3345                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3346                 if (IS_ERR(sbi->s_chksum_driver)) {
3347                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3348                         ret = PTR_ERR(sbi->s_chksum_driver);
3349                         sbi->s_chksum_driver = NULL;
3350                         goto failed_mount;
3351                 }
3352         }
3353
3354         /* Check superblock checksum */
3355         if (!ext4_superblock_csum_verify(sb, es)) {
3356                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3357                          "invalid superblock checksum.  Run e2fsck?");
3358                 silent = 1;
3359                 goto cantfind_ext4;
3360         }
3361
3362         /* Precompute checksum seed for all metadata */
3363         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3364                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3365                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3366                                                sizeof(es->s_uuid));
3367
3368         /* Set defaults before we parse the mount options */
3369         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3370         set_opt(sb, INIT_INODE_TABLE);
3371         if (def_mount_opts & EXT4_DEFM_DEBUG)
3372                 set_opt(sb, DEBUG);
3373         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3374                 set_opt(sb, GRPID);
3375         if (def_mount_opts & EXT4_DEFM_UID16)
3376                 set_opt(sb, NO_UID32);
3377         /* xattr user namespace & acls are now defaulted on */
3378 #ifdef CONFIG_EXT4_FS_XATTR
3379         set_opt(sb, XATTR_USER);
3380 #endif
3381 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3382         set_opt(sb, POSIX_ACL);
3383 #endif
3384         set_opt(sb, MBLK_IO_SUBMIT);
3385         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3386                 set_opt(sb, JOURNAL_DATA);
3387         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3388                 set_opt(sb, ORDERED_DATA);
3389         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3390                 set_opt(sb, WRITEBACK_DATA);
3391
3392         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3393                 set_opt(sb, ERRORS_PANIC);
3394         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3395                 set_opt(sb, ERRORS_CONT);
3396         else
3397                 set_opt(sb, ERRORS_RO);
3398         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3399                 set_opt(sb, BLOCK_VALIDITY);
3400         if (def_mount_opts & EXT4_DEFM_DISCARD)
3401                 set_opt(sb, DISCARD);
3402
3403         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3404         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3405         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3406         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3407         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3408
3409         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3410                 set_opt(sb, BARRIER);
3411
3412         /*
3413          * enable delayed allocation by default
3414          * Use -o nodelalloc to turn it off
3415          */
3416         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3417             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3418                 set_opt(sb, DELALLOC);
3419
3420         /*
3421          * set default s_li_wait_mult for lazyinit, for the case there is
3422          * no mount option specified.
3423          */
3424         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3425
3426         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3427                            &journal_devnum, &journal_ioprio, 0)) {
3428                 ext4_msg(sb, KERN_WARNING,
3429                          "failed to parse options in superblock: %s",
3430                          sbi->s_es->s_mount_opts);
3431         }
3432         sbi->s_def_mount_opt = sbi->s_mount_opt;
3433         if (!parse_options((char *) data, sb, &journal_devnum,
3434                            &journal_ioprio, 0))
3435                 goto failed_mount;
3436
3437         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3438                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3439                             "with data=journal disables delayed "
3440                             "allocation and O_DIRECT support!\n");
3441                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3442                         ext4_msg(sb, KERN_ERR, "can't mount with "
3443                                  "both data=journal and delalloc");
3444                         goto failed_mount;
3445                 }
3446                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3447                         ext4_msg(sb, KERN_ERR, "can't mount with "
3448                                  "both data=journal and delalloc");
3449                         goto failed_mount;
3450                 }
3451                 if (test_opt(sb, DELALLOC))
3452                         clear_opt(sb, DELALLOC);
3453         }
3454
3455         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3456         if (test_opt(sb, DIOREAD_NOLOCK)) {
3457                 if (blocksize < PAGE_SIZE) {
3458                         ext4_msg(sb, KERN_ERR, "can't mount with "
3459                                  "dioread_nolock if block size != PAGE_SIZE");
3460                         goto failed_mount;
3461                 }
3462         }
3463
3464         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3465                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3466
3467         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3468             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3469              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3470              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3471                 ext4_msg(sb, KERN_WARNING,
3472                        "feature flags set on rev 0 fs, "
3473                        "running e2fsck is recommended");
3474
3475         if (IS_EXT2_SB(sb)) {
3476                 if (ext2_feature_set_ok(sb))
3477                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3478                                  "using the ext4 subsystem");
3479                 else {
3480                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3481                                  "to feature incompatibilities");
3482                         goto failed_mount;
3483                 }
3484         }
3485
3486         if (IS_EXT3_SB(sb)) {
3487                 if (ext3_feature_set_ok(sb))
3488                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3489                                  "using the ext4 subsystem");
3490                 else {
3491                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3492                                  "to feature incompatibilities");
3493                         goto failed_mount;
3494                 }
3495         }
3496
3497         /*
3498          * Check feature flags regardless of the revision level, since we
3499          * previously didn't change the revision level when setting the flags,
3500          * so there is a chance incompat flags are set on a rev 0 filesystem.
3501          */
3502         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3503                 goto failed_mount;
3504
3505         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3506             blocksize > EXT4_MAX_BLOCK_SIZE) {
3507                 ext4_msg(sb, KERN_ERR,
3508                        "Unsupported filesystem blocksize %d", blocksize);
3509                 goto failed_mount;
3510         }
3511
3512         if (sb->s_blocksize != blocksize) {
3513                 /* Validate the filesystem blocksize */
3514                 if (!sb_set_blocksize(sb, blocksize)) {
3515                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3516                                         blocksize);
3517                         goto failed_mount;
3518                 }
3519
3520                 brelse(bh);
3521                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3522                 offset = do_div(logical_sb_block, blocksize);
3523                 bh = sb_bread(sb, logical_sb_block);
3524                 if (!bh) {
3525                         ext4_msg(sb, KERN_ERR,
3526                                "Can't read superblock on 2nd try");
3527                         goto failed_mount;
3528                 }
3529                 es = (struct ext4_super_block *)(bh->b_data + offset);
3530                 sbi->s_es = es;
3531                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3532                         ext4_msg(sb, KERN_ERR,
3533                                "Magic mismatch, very weird!");
3534                         goto failed_mount;
3535                 }
3536         }
3537
3538         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3539                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3540         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3541                                                       has_huge_files);
3542         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3543
3544         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3545                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3546                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3547         } else {
3548                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3549                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3550                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3551                     (!is_power_of_2(sbi->s_inode_size)) ||
3552                     (sbi->s_inode_size > blocksize)) {
3553                         ext4_msg(sb, KERN_ERR,
3554                                "unsupported inode size: %d",
3555                                sbi->s_inode_size);
3556                         goto failed_mount;
3557                 }
3558                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3559                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3560         }
3561
3562         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3563         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3564                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3565                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3566                     !is_power_of_2(sbi->s_desc_size)) {
3567                         ext4_msg(sb, KERN_ERR,
3568                                "unsupported descriptor size %lu",
3569                                sbi->s_desc_size);
3570                         goto failed_mount;
3571                 }
3572         } else
3573                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3574
3575         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3576         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3577         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3578                 goto cantfind_ext4;
3579
3580         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3581         if (sbi->s_inodes_per_block == 0)
3582                 goto cantfind_ext4;
3583         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3584                                         sbi->s_inodes_per_block;
3585         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3586         sbi->s_sbh = bh;
3587         sbi->s_mount_state = le16_to_cpu(es->s_state);
3588         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3589         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3590
3591         for (i = 0; i < 4; i++)
3592                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3593         sbi->s_def_hash_version = es->s_def_hash_version;
3594         i = le32_to_cpu(es->s_flags);
3595         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3596                 sbi->s_hash_unsigned = 3;
3597         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3598 #ifdef __CHAR_UNSIGNED__
3599                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3600                 sbi->s_hash_unsigned = 3;
3601 #else
3602                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3603 #endif
3604         }
3605
3606         /* Handle clustersize */
3607         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3608         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3609                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3610         if (has_bigalloc) {
3611                 if (clustersize < blocksize) {
3612                         ext4_msg(sb, KERN_ERR,
3613                                  "cluster size (%d) smaller than "
3614                                  "block size (%d)", clustersize, blocksize);
3615                         goto failed_mount;
3616                 }
3617                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3618                         le32_to_cpu(es->s_log_block_size);
3619                 sbi->s_clusters_per_group =
3620                         le32_to_cpu(es->s_clusters_per_group);
3621                 if (sbi->s_clusters_per_group > blocksize * 8) {
3622                         ext4_msg(sb, KERN_ERR,
3623                                  "#clusters per group too big: %lu",
3624                                  sbi->s_clusters_per_group);
3625                         goto failed_mount;
3626                 }
3627                 if (sbi->s_blocks_per_group !=
3628                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3629                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3630                                  "clusters per group (%lu) inconsistent",
3631                                  sbi->s_blocks_per_group,
3632                                  sbi->s_clusters_per_group);
3633                         goto failed_mount;
3634                 }
3635         } else {
3636                 if (clustersize != blocksize) {
3637                         ext4_warning(sb, "fragment/cluster size (%d) != "
3638                                      "block size (%d)", clustersize,
3639                                      blocksize);
3640                         clustersize = blocksize;
3641                 }
3642                 if (sbi->s_blocks_per_group > blocksize * 8) {
3643                         ext4_msg(sb, KERN_ERR,
3644                                  "#blocks per group too big: %lu",
3645                                  sbi->s_blocks_per_group);
3646                         goto failed_mount;
3647                 }
3648                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3649                 sbi->s_cluster_bits = 0;
3650         }
3651         sbi->s_cluster_ratio = clustersize / blocksize;
3652
3653         if (sbi->s_inodes_per_group > blocksize * 8) {
3654                 ext4_msg(sb, KERN_ERR,
3655                        "#inodes per group too big: %lu",
3656                        sbi->s_inodes_per_group);
3657                 goto failed_mount;
3658         }
3659
3660         /*
3661          * Test whether we have more sectors than will fit in sector_t,
3662          * and whether the max offset is addressable by the page cache.
3663          */
3664         err = generic_check_addressable(sb->s_blocksize_bits,
3665                                         ext4_blocks_count(es));
3666         if (err) {
3667                 ext4_msg(sb, KERN_ERR, "filesystem"
3668                          " too large to mount safely on this system");
3669                 if (sizeof(sector_t) < 8)
3670                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3671                 ret = err;
3672                 goto failed_mount;
3673         }
3674
3675         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3676                 goto cantfind_ext4;
3677
3678         /* check blocks count against device size */
3679         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3680         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3681                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3682                        "exceeds size of device (%llu blocks)",
3683                        ext4_blocks_count(es), blocks_count);
3684                 goto failed_mount;
3685         }
3686
3687         /*
3688          * It makes no sense for the first data block to be beyond the end
3689          * of the filesystem.
3690          */
3691         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3692                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3693                          "block %u is beyond end of filesystem (%llu)",
3694                          le32_to_cpu(es->s_first_data_block),
3695                          ext4_blocks_count(es));
3696                 goto failed_mount;
3697         }
3698         blocks_count = (ext4_blocks_count(es) -
3699                         le32_to_cpu(es->s_first_data_block) +
3700                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3701         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3702         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3703                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3704                        "(block count %llu, first data block %u, "
3705                        "blocks per group %lu)", sbi->s_groups_count,
3706                        ext4_blocks_count(es),
3707                        le32_to_cpu(es->s_first_data_block),
3708                        EXT4_BLOCKS_PER_GROUP(sb));
3709                 goto failed_mount;
3710         }
3711         sbi->s_groups_count = blocks_count;
3712         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3713                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3714         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3715                    EXT4_DESC_PER_BLOCK(sb);
3716         sbi->s_group_desc = ext4_kvmalloc(db_count *
3717                                           sizeof(struct buffer_head *),
3718                                           GFP_KERNEL);
3719         if (sbi->s_group_desc == NULL) {
3720                 ext4_msg(sb, KERN_ERR, "not enough memory");
3721                 ret = -ENOMEM;
3722                 goto failed_mount;
3723         }
3724
3725         if (ext4_proc_root)
3726                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3727
3728         if (sbi->s_proc)
3729                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3730                                  &ext4_seq_options_fops, sb);
3731
3732         bgl_lock_init(sbi->s_blockgroup_lock);
3733
3734         for (i = 0; i < db_count; i++) {
3735                 block = descriptor_loc(sb, logical_sb_block, i);
3736                 sbi->s_group_desc[i] = sb_bread(sb, block);
3737                 if (!sbi->s_group_desc[i]) {
3738                         ext4_msg(sb, KERN_ERR,
3739                                "can't read group descriptor %d", i);
3740                         db_count = i;
3741                         goto failed_mount2;
3742                 }
3743         }
3744         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3745                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3746                 goto failed_mount2;
3747         }
3748         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3749                 if (!ext4_fill_flex_info(sb)) {
3750                         ext4_msg(sb, KERN_ERR,
3751                                "unable to initialize "
3752                                "flex_bg meta info!");
3753                         goto failed_mount2;
3754                 }
3755
3756         sbi->s_gdb_count = db_count;
3757         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3758         spin_lock_init(&sbi->s_next_gen_lock);
3759
3760         init_timer(&sbi->s_err_report);
3761         sbi->s_err_report.function = print_daily_error_info;
3762         sbi->s_err_report.data = (unsigned long) sb;
3763
3764         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3765                         ext4_count_free_clusters(sb));
3766         if (!err) {
3767                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3768                                 ext4_count_free_inodes(sb));
3769         }
3770         if (!err) {
3771                 err = percpu_counter_init(&sbi->s_dirs_counter,
3772                                 ext4_count_dirs(sb));
3773         }
3774         if (!err) {
3775                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3776         }
3777         if (err) {
3778                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3779                 ret = err;
3780                 goto failed_mount3;
3781         }
3782
3783         sbi->s_stripe = ext4_get_stripe_size(sbi);
3784         sbi->s_max_writeback_mb_bump = 128;
3785         sbi->s_extent_max_zeroout_kb = 32;
3786
3787         /*
3788          * set up enough so that it can read an inode
3789          */
3790         if (!test_opt(sb, NOLOAD) &&
3791             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3792                 sb->s_op = &ext4_sops;
3793         else
3794                 sb->s_op = &ext4_nojournal_sops;
3795         sb->s_export_op = &ext4_export_ops;
3796         sb->s_xattr = ext4_xattr_handlers;
3797 #ifdef CONFIG_QUOTA
3798         sb->s_qcop = &ext4_qctl_operations;
3799         sb->dq_op = &ext4_quota_operations;
3800
3801         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
3802                 /* Use qctl operations for hidden quota files. */
3803                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3804         }
3805 #endif
3806         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3807
3808         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3809         mutex_init(&sbi->s_orphan_lock);
3810         sbi->s_resize_flags = 0;
3811
3812         sb->s_root = NULL;
3813
3814         needs_recovery = (es->s_last_orphan != 0 ||
3815                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3816                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3817
3818         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3819             !(sb->s_flags & MS_RDONLY))
3820                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3821                         goto failed_mount3;
3822
3823         /*
3824          * The first inode we look at is the journal inode.  Don't try
3825          * root first: it may be modified in the journal!
3826          */
3827         if (!test_opt(sb, NOLOAD) &&
3828             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3829                 if (ext4_load_journal(sb, es, journal_devnum))
3830                         goto failed_mount3;
3831         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3832               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3833                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3834                        "suppressed and not mounted read-only");
3835                 goto failed_mount_wq;
3836         } else {
3837                 clear_opt(sb, DATA_FLAGS);
3838                 sbi->s_journal = NULL;
3839                 needs_recovery = 0;
3840                 goto no_journal;
3841         }
3842
3843         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3844             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3845                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3846                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3847                 goto failed_mount_wq;
3848         }
3849
3850         if (!set_journal_csum_feature_set(sb)) {
3851                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3852                          "feature set");
3853                 goto failed_mount_wq;
3854         }
3855
3856         /* We have now updated the journal if required, so we can
3857          * validate the data journaling mode. */
3858         switch (test_opt(sb, DATA_FLAGS)) {
3859         case 0:
3860                 /* No mode set, assume a default based on the journal
3861                  * capabilities: ORDERED_DATA if the journal can
3862                  * cope, else JOURNAL_DATA
3863                  */
3864                 if (jbd2_journal_check_available_features
3865                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3866                         set_opt(sb, ORDERED_DATA);
3867                 else
3868                         set_opt(sb, JOURNAL_DATA);
3869                 break;
3870
3871         case EXT4_MOUNT_ORDERED_DATA:
3872         case EXT4_MOUNT_WRITEBACK_DATA:
3873                 if (!jbd2_journal_check_available_features
3874                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3875                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3876                                "requested data journaling mode");
3877                         goto failed_mount_wq;
3878                 }
3879         default:
3880                 break;
3881         }
3882         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3883
3884         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3885
3886         /*
3887          * The journal may have updated the bg summary counts, so we
3888          * need to update the global counters.
3889          */
3890         percpu_counter_set(&sbi->s_freeclusters_counter,
3891                            ext4_count_free_clusters(sb));
3892         percpu_counter_set(&sbi->s_freeinodes_counter,
3893                            ext4_count_free_inodes(sb));
3894         percpu_counter_set(&sbi->s_dirs_counter,
3895                            ext4_count_dirs(sb));
3896         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3897
3898 no_journal:
3899         /*
3900          * Get the # of file system overhead blocks from the
3901          * superblock if present.
3902          */
3903         if (es->s_overhead_clusters)
3904                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3905         else {
3906                 ret = ext4_calculate_overhead(sb);
3907                 if (ret)
3908                         goto failed_mount_wq;
3909         }
3910
3911         /*
3912          * The maximum number of concurrent works can be high and
3913          * concurrency isn't really necessary.  Limit it to 1.
3914          */
3915         EXT4_SB(sb)->dio_unwritten_wq =
3916                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3917         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3918                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3919                 goto failed_mount_wq;
3920         }
3921
3922         /*
3923          * The jbd2_journal_load will have done any necessary log recovery,
3924          * so we can safely mount the rest of the filesystem now.
3925          */
3926
3927         root = ext4_iget(sb, EXT4_ROOT_INO);
3928         if (IS_ERR(root)) {
3929                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3930                 ret = PTR_ERR(root);
3931                 root = NULL;
3932                 goto failed_mount4;
3933         }
3934         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3935                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3936                 iput(root);
3937                 goto failed_mount4;
3938         }
3939         sb->s_root = d_make_root(root);
3940         if (!sb->s_root) {
3941                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3942                 ret = -ENOMEM;
3943                 goto failed_mount4;
3944         }
3945
3946         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3947                 sb->s_flags |= MS_RDONLY;
3948
3949         /* determine the minimum size of new large inodes, if present */
3950         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3951                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3952                                                      EXT4_GOOD_OLD_INODE_SIZE;
3953                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3954                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3955                         if (sbi->s_want_extra_isize <
3956                             le16_to_cpu(es->s_want_extra_isize))
3957                                 sbi->s_want_extra_isize =
3958                                         le16_to_cpu(es->s_want_extra_isize);
3959                         if (sbi->s_want_extra_isize <
3960                             le16_to_cpu(es->s_min_extra_isize))
3961                                 sbi->s_want_extra_isize =
3962                                         le16_to_cpu(es->s_min_extra_isize);
3963                 }
3964         }
3965         /* Check if enough inode space is available */
3966         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3967                                                         sbi->s_inode_size) {
3968                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3969                                                        EXT4_GOOD_OLD_INODE_SIZE;
3970                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3971                          "available");
3972         }
3973
3974         err = ext4_setup_system_zone(sb);
3975         if (err) {
3976                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3977                          "zone (%d)", err);
3978                 goto failed_mount4a;
3979         }
3980
3981         ext4_ext_init(sb);
3982         err = ext4_mb_init(sb);
3983         if (err) {
3984                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3985                          err);
3986                 goto failed_mount5;
3987         }
3988
3989         err = ext4_register_li_request(sb, first_not_zeroed);
3990         if (err)
3991                 goto failed_mount6;
3992
3993         sbi->s_kobj.kset = ext4_kset;
3994         init_completion(&sbi->s_kobj_unregister);
3995         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3996                                    "%s", sb->s_id);
3997         if (err)
3998                 goto failed_mount7;
3999
4000         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4001         ext4_orphan_cleanup(sb, es);
4002         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4003         if (needs_recovery) {
4004                 ext4_msg(sb, KERN_INFO, "recovery complete");
4005                 ext4_mark_recovery_complete(sb, es);
4006         }
4007         if (EXT4_SB(sb)->s_journal) {
4008                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4009                         descr = " journalled data mode";
4010                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4011                         descr = " ordered data mode";
4012                 else
4013                         descr = " writeback data mode";
4014         } else
4015                 descr = "out journal";
4016
4017 #ifdef CONFIG_QUOTA
4018         /* Enable quota usage during mount. */
4019         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4020             !(sb->s_flags & MS_RDONLY)) {
4021                 ret = ext4_enable_quotas(sb);
4022                 if (ret)
4023                         goto failed_mount7;
4024         }
4025 #endif  /* CONFIG_QUOTA */
4026
4027         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4028                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4029                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4030
4031         if (es->s_error_count)
4032                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4033
4034         kfree(orig_data);
4035         return 0;
4036
4037 cantfind_ext4:
4038         if (!silent)
4039                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4040         goto failed_mount;
4041
4042 failed_mount7:
4043         ext4_unregister_li_request(sb);
4044 failed_mount6:
4045         ext4_mb_release(sb);
4046 failed_mount5:
4047         ext4_ext_release(sb);
4048         ext4_release_system_zone(sb);
4049 failed_mount4a:
4050         dput(sb->s_root);
4051         sb->s_root = NULL;
4052 failed_mount4:
4053         ext4_msg(sb, KERN_ERR, "mount failed");
4054         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4055 failed_mount_wq:
4056         if (sbi->s_journal) {
4057                 jbd2_journal_destroy(sbi->s_journal);
4058                 sbi->s_journal = NULL;
4059         }
4060 failed_mount3:
4061         del_timer(&sbi->s_err_report);
4062         if (sbi->s_flex_groups)
4063                 ext4_kvfree(sbi->s_flex_groups);
4064         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4065         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4066         percpu_counter_destroy(&sbi->s_dirs_counter);
4067         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4068         if (sbi->s_mmp_tsk)
4069                 kthread_stop(sbi->s_mmp_tsk);
4070 failed_mount2:
4071         for (i = 0; i < db_count; i++)
4072                 brelse(sbi->s_group_desc[i]);
4073         ext4_kvfree(sbi->s_group_desc);
4074 failed_mount:
4075         if (sbi->s_chksum_driver)
4076                 crypto_free_shash(sbi->s_chksum_driver);
4077         if (sbi->s_proc) {
4078                 remove_proc_entry("options", sbi->s_proc);
4079                 remove_proc_entry(sb->s_id, ext4_proc_root);
4080         }
4081 #ifdef CONFIG_QUOTA
4082         for (i = 0; i < MAXQUOTAS; i++)
4083                 kfree(sbi->s_qf_names[i]);
4084 #endif
4085         ext4_blkdev_remove(sbi);
4086         brelse(bh);
4087 out_fail:
4088         sb->s_fs_info = NULL;
4089         kfree(sbi->s_blockgroup_lock);
4090         kfree(sbi);
4091 out_free_orig:
4092         kfree(orig_data);
4093         return ret;
4094 }
4095
4096 /*
4097  * Setup any per-fs journal parameters now.  We'll do this both on
4098  * initial mount, once the journal has been initialised but before we've
4099  * done any recovery; and again on any subsequent remount.
4100  */
4101 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4102 {
4103         struct ext4_sb_info *sbi = EXT4_SB(sb);
4104
4105         journal->j_commit_interval = sbi->s_commit_interval;
4106         journal->j_min_batch_time = sbi->s_min_batch_time;
4107         journal->j_max_batch_time = sbi->s_max_batch_time;
4108
4109         write_lock(&journal->j_state_lock);
4110         if (test_opt(sb, BARRIER))
4111                 journal->j_flags |= JBD2_BARRIER;
4112         else
4113                 journal->j_flags &= ~JBD2_BARRIER;
4114         if (test_opt(sb, DATA_ERR_ABORT))
4115                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4116         else
4117                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4118         write_unlock(&journal->j_state_lock);
4119 }
4120
4121 static journal_t *ext4_get_journal(struct super_block *sb,
4122                                    unsigned int journal_inum)
4123 {
4124         struct inode *journal_inode;
4125         journal_t *journal;
4126
4127         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4128
4129         /* First, test for the existence of a valid inode on disk.  Bad
4130          * things happen if we iget() an unused inode, as the subsequent
4131          * iput() will try to delete it. */
4132
4133         journal_inode = ext4_iget(sb, journal_inum);
4134         if (IS_ERR(journal_inode)) {
4135                 ext4_msg(sb, KERN_ERR, "no journal found");
4136                 return NULL;
4137         }
4138         if (!journal_inode->i_nlink) {
4139                 make_bad_inode(journal_inode);
4140                 iput(journal_inode);
4141                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4142                 return NULL;
4143         }
4144
4145         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4146                   journal_inode, journal_inode->i_size);
4147         if (!S_ISREG(journal_inode->i_mode)) {
4148                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4149                 iput(journal_inode);
4150                 return NULL;
4151         }
4152
4153         journal = jbd2_journal_init_inode(journal_inode);
4154         if (!journal) {
4155                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4156                 iput(journal_inode);
4157                 return NULL;
4158         }
4159         journal->j_private = sb;
4160         ext4_init_journal_params(sb, journal);
4161         return journal;
4162 }
4163
4164 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4165                                        dev_t j_dev)
4166 {
4167         struct buffer_head *bh;
4168         journal_t *journal;
4169         ext4_fsblk_t start;
4170         ext4_fsblk_t len;
4171         int hblock, blocksize;
4172         ext4_fsblk_t sb_block;
4173         unsigned long offset;
4174         struct ext4_super_block *es;
4175         struct block_device *bdev;
4176
4177         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4178
4179         bdev = ext4_blkdev_get(j_dev, sb);
4180         if (bdev == NULL)
4181                 return NULL;
4182
4183         blocksize = sb->s_blocksize;
4184         hblock = bdev_logical_block_size(bdev);
4185         if (blocksize < hblock) {
4186                 ext4_msg(sb, KERN_ERR,
4187                         "blocksize too small for journal device");
4188                 goto out_bdev;
4189         }
4190
4191         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4192         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4193         set_blocksize(bdev, blocksize);
4194         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4195                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4196                        "external journal");
4197                 goto out_bdev;
4198         }
4199
4200         es = (struct ext4_super_block *) (bh->b_data + offset);
4201         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4202             !(le32_to_cpu(es->s_feature_incompat) &
4203               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4204                 ext4_msg(sb, KERN_ERR, "external journal has "
4205                                         "bad superblock");
4206                 brelse(bh);
4207                 goto out_bdev;
4208         }
4209
4210         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4211                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4212                 brelse(bh);
4213                 goto out_bdev;
4214         }
4215
4216         len = ext4_blocks_count(es);
4217         start = sb_block + 1;
4218         brelse(bh);     /* we're done with the superblock */
4219
4220         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4221                                         start, len, blocksize);
4222         if (!journal) {
4223                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4224                 goto out_bdev;
4225         }
4226         journal->j_private = sb;
4227         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4228         wait_on_buffer(journal->j_sb_buffer);
4229         if (!buffer_uptodate(journal->j_sb_buffer)) {
4230                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4231                 goto out_journal;
4232         }
4233         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4234                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4235                                         "user (unsupported) - %d",
4236                         be32_to_cpu(journal->j_superblock->s_nr_users));
4237                 goto out_journal;
4238         }
4239         EXT4_SB(sb)->journal_bdev = bdev;
4240         ext4_init_journal_params(sb, journal);
4241         return journal;
4242
4243 out_journal:
4244         jbd2_journal_destroy(journal);
4245 out_bdev:
4246         ext4_blkdev_put(bdev);
4247         return NULL;
4248 }
4249
4250 static int ext4_load_journal(struct super_block *sb,
4251                              struct ext4_super_block *es,
4252                              unsigned long journal_devnum)
4253 {
4254         journal_t *journal;
4255         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4256         dev_t journal_dev;
4257         int err = 0;
4258         int really_read_only;
4259
4260         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4261
4262         if (journal_devnum &&
4263             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4264                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4265                         "numbers have changed");
4266                 journal_dev = new_decode_dev(journal_devnum);
4267         } else
4268                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4269
4270         really_read_only = bdev_read_only(sb->s_bdev);
4271
4272         /*
4273          * Are we loading a blank journal or performing recovery after a
4274          * crash?  For recovery, we need to check in advance whether we
4275          * can get read-write access to the device.
4276          */
4277         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4278                 if (sb->s_flags & MS_RDONLY) {
4279                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4280                                         "required on readonly filesystem");
4281                         if (really_read_only) {
4282                                 ext4_msg(sb, KERN_ERR, "write access "
4283                                         "unavailable, cannot proceed");
4284                                 return -EROFS;
4285                         }
4286                         ext4_msg(sb, KERN_INFO, "write access will "
4287                                "be enabled during recovery");
4288                 }
4289         }
4290
4291         if (journal_inum && journal_dev) {
4292                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4293                        "and inode journals!");
4294                 return -EINVAL;
4295         }
4296
4297         if (journal_inum) {
4298                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4299                         return -EINVAL;
4300         } else {
4301                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4302                         return -EINVAL;
4303         }
4304
4305         if (!(journal->j_flags & JBD2_BARRIER))
4306                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4307
4308         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4309                 err = jbd2_journal_wipe(journal, !really_read_only);
4310         if (!err) {
4311                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4312                 if (save)
4313                         memcpy(save, ((char *) es) +
4314                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4315                 err = jbd2_journal_load(journal);
4316                 if (save)
4317                         memcpy(((char *) es) + EXT4_S_ERR_START,
4318                                save, EXT4_S_ERR_LEN);
4319                 kfree(save);
4320         }
4321
4322         if (err) {
4323                 ext4_msg(sb, KERN_ERR, "error loading journal");
4324                 jbd2_journal_destroy(journal);
4325                 return err;
4326         }
4327
4328         EXT4_SB(sb)->s_journal = journal;
4329         ext4_clear_journal_err(sb, es);
4330
4331         if (!really_read_only && journal_devnum &&
4332             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4333                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4334
4335                 /* Make sure we flush the recovery flag to disk. */
4336                 ext4_commit_super(sb, 1);
4337         }
4338
4339         return 0;
4340 }
4341
4342 static int ext4_commit_super(struct super_block *sb, int sync)
4343 {
4344         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4345         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4346         int error = 0;
4347
4348         if (!sbh || block_device_ejected(sb))
4349                 return error;
4350         if (buffer_write_io_error(sbh)) {
4351                 /*
4352                  * Oh, dear.  A previous attempt to write the
4353                  * superblock failed.  This could happen because the
4354                  * USB device was yanked out.  Or it could happen to
4355                  * be a transient write error and maybe the block will
4356                  * be remapped.  Nothing we can do but to retry the
4357                  * write and hope for the best.
4358                  */
4359                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4360                        "superblock detected");
4361                 clear_buffer_write_io_error(sbh);
4362                 set_buffer_uptodate(sbh);
4363         }
4364         /*
4365          * If the file system is mounted read-only, don't update the
4366          * superblock write time.  This avoids updating the superblock
4367          * write time when we are mounting the root file system
4368          * read/only but we need to replay the journal; at that point,
4369          * for people who are east of GMT and who make their clock
4370          * tick in localtime for Windows bug-for-bug compatibility,
4371          * the clock is set in the future, and this will cause e2fsck
4372          * to complain and force a full file system check.
4373          */
4374         if (!(sb->s_flags & MS_RDONLY))
4375                 es->s_wtime = cpu_to_le32(get_seconds());
4376         if (sb->s_bdev->bd_part)
4377                 es->s_kbytes_written =
4378                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4379                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4380                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4381         else
4382                 es->s_kbytes_written =
4383                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4384         ext4_free_blocks_count_set(es,
4385                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4386                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4387         es->s_free_inodes_count =
4388                 cpu_to_le32(percpu_counter_sum_positive(
4389                                 &EXT4_SB(sb)->s_freeinodes_counter));
4390         BUFFER_TRACE(sbh, "marking dirty");
4391         ext4_superblock_csum_set(sb);
4392         mark_buffer_dirty(sbh);
4393         if (sync) {
4394                 error = sync_dirty_buffer(sbh);
4395                 if (error)
4396                         return error;
4397
4398                 error = buffer_write_io_error(sbh);
4399                 if (error) {
4400                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4401                                "superblock");
4402                         clear_buffer_write_io_error(sbh);
4403                         set_buffer_uptodate(sbh);
4404                 }
4405         }
4406         return error;
4407 }
4408
4409 /*
4410  * Have we just finished recovery?  If so, and if we are mounting (or
4411  * remounting) the filesystem readonly, then we will end up with a
4412  * consistent fs on disk.  Record that fact.
4413  */
4414 static void ext4_mark_recovery_complete(struct super_block *sb,
4415                                         struct ext4_super_block *es)
4416 {
4417         journal_t *journal = EXT4_SB(sb)->s_journal;
4418
4419         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4420                 BUG_ON(journal != NULL);
4421                 return;
4422         }
4423         jbd2_journal_lock_updates(journal);
4424         if (jbd2_journal_flush(journal) < 0)
4425                 goto out;
4426
4427         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4428             sb->s_flags & MS_RDONLY) {
4429                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4430                 ext4_commit_super(sb, 1);
4431         }
4432
4433 out:
4434         jbd2_journal_unlock_updates(journal);
4435 }
4436
4437 /*
4438  * If we are mounting (or read-write remounting) a filesystem whose journal
4439  * has recorded an error from a previous lifetime, move that error to the
4440  * main filesystem now.
4441  */
4442 static void ext4_clear_journal_err(struct super_block *sb,
4443                                    struct ext4_super_block *es)
4444 {
4445         journal_t *journal;
4446         int j_errno;
4447         const char *errstr;
4448
4449         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4450
4451         journal = EXT4_SB(sb)->s_journal;
4452
4453         /*
4454          * Now check for any error status which may have been recorded in the
4455          * journal by a prior ext4_error() or ext4_abort()
4456          */
4457
4458         j_errno = jbd2_journal_errno(journal);
4459         if (j_errno) {
4460                 char nbuf[16];
4461
4462                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4463                 ext4_warning(sb, "Filesystem error recorded "
4464                              "from previous mount: %s", errstr);
4465                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4466
4467                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4468                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4469                 ext4_commit_super(sb, 1);
4470
4471                 jbd2_journal_clear_err(journal);
4472                 jbd2_journal_update_sb_errno(journal);
4473         }
4474 }
4475
4476 /*
4477  * Force the running and committing transactions to commit,
4478  * and wait on the commit.
4479  */
4480 int ext4_force_commit(struct super_block *sb)
4481 {
4482         journal_t *journal;
4483         int ret = 0;
4484
4485         if (sb->s_flags & MS_RDONLY)
4486                 return 0;
4487
4488         journal = EXT4_SB(sb)->s_journal;
4489         if (journal)
4490                 ret = ext4_journal_force_commit(journal);
4491
4492         return ret;
4493 }
4494
4495 static int ext4_sync_fs(struct super_block *sb, int wait)
4496 {
4497         int ret = 0;
4498         tid_t target;
4499         struct ext4_sb_info *sbi = EXT4_SB(sb);
4500
4501         trace_ext4_sync_fs(sb, wait);
4502         flush_workqueue(sbi->dio_unwritten_wq);
4503         /*
4504          * Writeback quota in non-journalled quota case - journalled quota has
4505          * no dirty dquots
4506          */
4507         dquot_writeback_dquots(sb, -1);
4508         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4509                 if (wait)
4510                         jbd2_log_wait_commit(sbi->s_journal, target);
4511         }
4512         return ret;
4513 }
4514
4515 /*
4516  * LVM calls this function before a (read-only) snapshot is created.  This
4517  * gives us a chance to flush the journal completely and mark the fs clean.
4518  *
4519  * Note that only this function cannot bring a filesystem to be in a clean
4520  * state independently. It relies on upper layer to stop all data & metadata
4521  * modifications.
4522  */
4523 static int ext4_freeze(struct super_block *sb)
4524 {
4525         int error = 0;
4526         journal_t *journal;
4527
4528         if (sb->s_flags & MS_RDONLY)
4529                 return 0;
4530
4531         journal = EXT4_SB(sb)->s_journal;
4532
4533         /* Now we set up the journal barrier. */
4534         jbd2_journal_lock_updates(journal);
4535
4536         /*
4537          * Don't clear the needs_recovery flag if we failed to flush
4538          * the journal.
4539          */
4540         error = jbd2_journal_flush(journal);
4541         if (error < 0)
4542                 goto out;
4543
4544         /* Journal blocked and flushed, clear needs_recovery flag. */
4545         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4546         error = ext4_commit_super(sb, 1);
4547 out:
4548         /* we rely on upper layer to stop further updates */
4549         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4550         return error;
4551 }
4552
4553 /*
4554  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4555  * flag here, even though the filesystem is not technically dirty yet.
4556  */
4557 static int ext4_unfreeze(struct super_block *sb)
4558 {
4559         if (sb->s_flags & MS_RDONLY)
4560                 return 0;
4561
4562         /* Reset the needs_recovery flag before the fs is unlocked. */
4563         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4564         ext4_commit_super(sb, 1);
4565         return 0;
4566 }
4567
4568 /*
4569  * Structure to save mount options for ext4_remount's benefit
4570  */
4571 struct ext4_mount_options {
4572         unsigned long s_mount_opt;
4573         unsigned long s_mount_opt2;
4574         kuid_t s_resuid;
4575         kgid_t s_resgid;
4576         unsigned long s_commit_interval;
4577         u32 s_min_batch_time, s_max_batch_time;
4578 #ifdef CONFIG_QUOTA
4579         int s_jquota_fmt;
4580         char *s_qf_names[MAXQUOTAS];
4581 #endif
4582 };
4583
4584 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4585 {
4586         struct ext4_super_block *es;
4587         struct ext4_sb_info *sbi = EXT4_SB(sb);
4588         unsigned long old_sb_flags;
4589         struct ext4_mount_options old_opts;
4590         int enable_quota = 0;
4591         ext4_group_t g;
4592         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4593         int err = 0;
4594 #ifdef CONFIG_QUOTA
4595         int i;
4596 #endif
4597         char *orig_data = kstrdup(data, GFP_KERNEL);
4598
4599         /* Store the original options */
4600         old_sb_flags = sb->s_flags;
4601         old_opts.s_mount_opt = sbi->s_mount_opt;
4602         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4603         old_opts.s_resuid = sbi->s_resuid;
4604         old_opts.s_resgid = sbi->s_resgid;
4605         old_opts.s_commit_interval = sbi->s_commit_interval;
4606         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4607         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4608 #ifdef CONFIG_QUOTA
4609         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4610         for (i = 0; i < MAXQUOTAS; i++)
4611                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4612 #endif
4613         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4614                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4615
4616         /*
4617          * Allow the "check" option to be passed as a remount option.
4618          */
4619         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4620                 err = -EINVAL;
4621                 goto restore_opts;
4622         }
4623
4624         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4625                 ext4_abort(sb, "Abort forced by user");
4626
4627         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4628                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4629
4630         es = sbi->s_es;
4631
4632         if (sbi->s_journal) {
4633                 ext4_init_journal_params(sb, sbi->s_journal);
4634                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4635         }
4636
4637         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4638                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4639                         err = -EROFS;
4640                         goto restore_opts;
4641                 }
4642
4643                 if (*flags & MS_RDONLY) {
4644                         err = dquot_suspend(sb, -1);
4645                         if (err < 0)
4646                                 goto restore_opts;
4647
4648                         /*
4649                          * First of all, the unconditional stuff we have to do
4650                          * to disable replay of the journal when we next remount
4651                          */
4652                         sb->s_flags |= MS_RDONLY;
4653
4654                         /*
4655                          * OK, test if we are remounting a valid rw partition
4656                          * readonly, and if so set the rdonly flag and then
4657                          * mark the partition as valid again.
4658                          */
4659                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4660                             (sbi->s_mount_state & EXT4_VALID_FS))
4661                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4662
4663                         if (sbi->s_journal)
4664                                 ext4_mark_recovery_complete(sb, es);
4665                 } else {
4666                         /* Make sure we can mount this feature set readwrite */
4667                         if (!ext4_feature_set_ok(sb, 0)) {
4668                                 err = -EROFS;
4669                                 goto restore_opts;
4670                         }
4671                         /*
4672                          * Make sure the group descriptor checksums
4673                          * are sane.  If they aren't, refuse to remount r/w.
4674                          */
4675                         for (g = 0; g < sbi->s_groups_count; g++) {
4676                                 struct ext4_group_desc *gdp =
4677                                         ext4_get_group_desc(sb, g, NULL);
4678
4679                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4680                                         ext4_msg(sb, KERN_ERR,
4681                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4682                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4683                                                le16_to_cpu(gdp->bg_checksum));
4684                                         err = -EINVAL;
4685                                         goto restore_opts;
4686                                 }
4687                         }
4688
4689                         /*
4690                          * If we have an unprocessed orphan list hanging
4691                          * around from a previously readonly bdev mount,
4692                          * require a full umount/remount for now.
4693                          */
4694                         if (es->s_last_orphan) {
4695                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4696                                        "remount RDWR because of unprocessed "
4697                                        "orphan inode list.  Please "
4698                                        "umount/remount instead");
4699                                 err = -EINVAL;
4700                                 goto restore_opts;
4701                         }
4702
4703                         /*
4704                          * Mounting a RDONLY partition read-write, so reread
4705                          * and store the current valid flag.  (It may have
4706                          * been changed by e2fsck since we originally mounted
4707                          * the partition.)
4708                          */
4709                         if (sbi->s_journal)
4710                                 ext4_clear_journal_err(sb, es);
4711                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4712                         if (!ext4_setup_super(sb, es, 0))
4713                                 sb->s_flags &= ~MS_RDONLY;
4714                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4715                                                      EXT4_FEATURE_INCOMPAT_MMP))
4716                                 if (ext4_multi_mount_protect(sb,
4717                                                 le64_to_cpu(es->s_mmp_block))) {
4718                                         err = -EROFS;
4719                                         goto restore_opts;
4720                                 }
4721                         enable_quota = 1;
4722                 }
4723         }
4724
4725         /*
4726          * Reinitialize lazy itable initialization thread based on
4727          * current settings
4728          */
4729         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4730                 ext4_unregister_li_request(sb);
4731         else {
4732                 ext4_group_t first_not_zeroed;
4733                 first_not_zeroed = ext4_has_uninit_itable(sb);
4734                 ext4_register_li_request(sb, first_not_zeroed);
4735         }
4736
4737         ext4_setup_system_zone(sb);
4738         if (sbi->s_journal == NULL)
4739                 ext4_commit_super(sb, 1);
4740
4741 #ifdef CONFIG_QUOTA
4742         /* Release old quota file names */
4743         for (i = 0; i < MAXQUOTAS; i++)
4744                 if (old_opts.s_qf_names[i] &&
4745                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4746                         kfree(old_opts.s_qf_names[i]);
4747         if (enable_quota) {
4748                 if (sb_any_quota_suspended(sb))
4749                         dquot_resume(sb, -1);
4750                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4751                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4752                         err = ext4_enable_quotas(sb);
4753                         if (err)
4754                                 goto restore_opts;
4755                 }
4756         }
4757 #endif
4758
4759         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4760         kfree(orig_data);
4761         return 0;
4762
4763 restore_opts:
4764         sb->s_flags = old_sb_flags;
4765         sbi->s_mount_opt = old_opts.s_mount_opt;
4766         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4767         sbi->s_resuid = old_opts.s_resuid;
4768         sbi->s_resgid = old_opts.s_resgid;
4769         sbi->s_commit_interval = old_opts.s_commit_interval;
4770         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4771         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4772 #ifdef CONFIG_QUOTA
4773         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4774         for (i = 0; i < MAXQUOTAS; i++) {
4775                 if (sbi->s_qf_names[i] &&
4776                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4777                         kfree(sbi->s_qf_names[i]);
4778                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4779         }
4780 #endif
4781         kfree(orig_data);
4782         return err;
4783 }
4784
4785 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4786 {
4787         struct super_block *sb = dentry->d_sb;
4788         struct ext4_sb_info *sbi = EXT4_SB(sb);
4789         struct ext4_super_block *es = sbi->s_es;
4790         ext4_fsblk_t overhead = 0;
4791         u64 fsid;
4792         s64 bfree;
4793
4794         if (!test_opt(sb, MINIX_DF))
4795                 overhead = sbi->s_overhead;
4796
4797         buf->f_type = EXT4_SUPER_MAGIC;
4798         buf->f_bsize = sb->s_blocksize;
4799         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, sbi->s_overhead);
4800         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4801                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4802         /* prevent underflow in case that few free space is available */
4803         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4804         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4805         if (buf->f_bfree < ext4_r_blocks_count(es))
4806                 buf->f_bavail = 0;
4807         buf->f_files = le32_to_cpu(es->s_inodes_count);
4808         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4809         buf->f_namelen = EXT4_NAME_LEN;
4810         fsid = le64_to_cpup((void *)es->s_uuid) ^
4811                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4812         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4813         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4814
4815         return 0;
4816 }
4817
4818 /* Helper function for writing quotas on sync - we need to start transaction
4819  * before quota file is locked for write. Otherwise the are possible deadlocks:
4820  * Process 1                         Process 2
4821  * ext4_create()                     quota_sync()
4822  *   jbd2_journal_start()                  write_dquot()
4823  *   dquot_initialize()                         down(dqio_mutex)
4824  *     down(dqio_mutex)                    jbd2_journal_start()
4825  *
4826  */
4827
4828 #ifdef CONFIG_QUOTA
4829
4830 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4831 {
4832         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4833 }
4834
4835 static int ext4_write_dquot(struct dquot *dquot)
4836 {
4837         int ret, err;
4838         handle_t *handle;
4839         struct inode *inode;
4840
4841         inode = dquot_to_inode(dquot);
4842         handle = ext4_journal_start(inode,
4843                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4844         if (IS_ERR(handle))
4845                 return PTR_ERR(handle);
4846         ret = dquot_commit(dquot);
4847         err = ext4_journal_stop(handle);
4848         if (!ret)
4849                 ret = err;
4850         return ret;
4851 }
4852
4853 static int ext4_acquire_dquot(struct dquot *dquot)
4854 {
4855         int ret, err;
4856         handle_t *handle;
4857
4858         handle = ext4_journal_start(dquot_to_inode(dquot),
4859                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4860         if (IS_ERR(handle))
4861                 return PTR_ERR(handle);
4862         ret = dquot_acquire(dquot);
4863         err = ext4_journal_stop(handle);
4864         if (!ret)
4865                 ret = err;
4866         return ret;
4867 }
4868
4869 static int ext4_release_dquot(struct dquot *dquot)
4870 {
4871         int ret, err;
4872         handle_t *handle;
4873
4874         handle = ext4_journal_start(dquot_to_inode(dquot),
4875                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4876         if (IS_ERR(handle)) {
4877                 /* Release dquot anyway to avoid endless cycle in dqput() */
4878                 dquot_release(dquot);
4879                 return PTR_ERR(handle);
4880         }
4881         ret = dquot_release(dquot);
4882         err = ext4_journal_stop(handle);
4883         if (!ret)
4884                 ret = err;
4885         return ret;
4886 }
4887
4888 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4889 {
4890         /* Are we journaling quotas? */
4891         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4892             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4893                 dquot_mark_dquot_dirty(dquot);
4894                 return ext4_write_dquot(dquot);
4895         } else {
4896                 return dquot_mark_dquot_dirty(dquot);
4897         }
4898 }
4899
4900 static int ext4_write_info(struct super_block *sb, int type)
4901 {
4902         int ret, err;
4903         handle_t *handle;
4904
4905         /* Data block + inode block */
4906         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4907         if (IS_ERR(handle))
4908                 return PTR_ERR(handle);
4909         ret = dquot_commit_info(sb, type);
4910         err = ext4_journal_stop(handle);
4911         if (!ret)
4912                 ret = err;
4913         return ret;
4914 }
4915
4916 /*
4917  * Turn on quotas during mount time - we need to find
4918  * the quota file and such...
4919  */
4920 static int ext4_quota_on_mount(struct super_block *sb, int type)
4921 {
4922         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4923                                         EXT4_SB(sb)->s_jquota_fmt, type);
4924 }
4925
4926 /*
4927  * Standard function to be called on quota_on
4928  */
4929 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4930                          struct path *path)
4931 {
4932         int err;
4933
4934         if (!test_opt(sb, QUOTA))
4935                 return -EINVAL;
4936
4937         /* Quotafile not on the same filesystem? */
4938         if (path->dentry->d_sb != sb)
4939                 return -EXDEV;
4940         /* Journaling quota? */
4941         if (EXT4_SB(sb)->s_qf_names[type]) {
4942                 /* Quotafile not in fs root? */
4943                 if (path->dentry->d_parent != sb->s_root)
4944                         ext4_msg(sb, KERN_WARNING,
4945                                 "Quota file not on filesystem root. "
4946                                 "Journaled quota will not work");
4947         }
4948
4949         /*
4950          * When we journal data on quota file, we have to flush journal to see
4951          * all updates to the file when we bypass pagecache...
4952          */
4953         if (EXT4_SB(sb)->s_journal &&
4954             ext4_should_journal_data(path->dentry->d_inode)) {
4955                 /*
4956                  * We don't need to lock updates but journal_flush() could
4957                  * otherwise be livelocked...
4958                  */
4959                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4960                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4961                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4962                 if (err)
4963                         return err;
4964         }
4965
4966         return dquot_quota_on(sb, type, format_id, path);
4967 }
4968
4969 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
4970                              unsigned int flags)
4971 {
4972         int err;
4973         struct inode *qf_inode;
4974         unsigned long qf_inums[MAXQUOTAS] = {
4975                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4976                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4977         };
4978
4979         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
4980
4981         if (!qf_inums[type])
4982                 return -EPERM;
4983
4984         qf_inode = ext4_iget(sb, qf_inums[type]);
4985         if (IS_ERR(qf_inode)) {
4986                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
4987                 return PTR_ERR(qf_inode);
4988         }
4989
4990         err = dquot_enable(qf_inode, type, format_id, flags);
4991         iput(qf_inode);
4992
4993         return err;
4994 }
4995
4996 /* Enable usage tracking for all quota types. */
4997 static int ext4_enable_quotas(struct super_block *sb)
4998 {
4999         int type, err = 0;
5000         unsigned long qf_inums[MAXQUOTAS] = {
5001                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5002                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5003         };
5004
5005         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5006         for (type = 0; type < MAXQUOTAS; type++) {
5007                 if (qf_inums[type]) {
5008                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5009                                                 DQUOT_USAGE_ENABLED);
5010                         if (err) {
5011                                 ext4_warning(sb,
5012                                         "Failed to enable quota (type=%d) "
5013                                         "tracking. Please run e2fsck to fix.",
5014                                         type);
5015                                 return err;
5016                         }
5017                 }
5018         }
5019         return 0;
5020 }
5021
5022 /*
5023  * quota_on function that is used when QUOTA feature is set.
5024  */
5025 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5026                                  int format_id)
5027 {
5028         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5029                 return -EINVAL;
5030
5031         /*
5032          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5033          */
5034         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5035 }
5036
5037 static int ext4_quota_off(struct super_block *sb, int type)
5038 {
5039         struct inode *inode = sb_dqopt(sb)->files[type];
5040         handle_t *handle;
5041
5042         /* Force all delayed allocation blocks to be allocated.
5043          * Caller already holds s_umount sem */
5044         if (test_opt(sb, DELALLOC))
5045                 sync_filesystem(sb);
5046
5047         if (!inode)
5048                 goto out;
5049
5050         /* Update modification times of quota files when userspace can
5051          * start looking at them */
5052         handle = ext4_journal_start(inode, 1);
5053         if (IS_ERR(handle))
5054                 goto out;
5055         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5056         ext4_mark_inode_dirty(handle, inode);
5057         ext4_journal_stop(handle);
5058
5059 out:
5060         return dquot_quota_off(sb, type);
5061 }
5062
5063 /*
5064  * quota_off function that is used when QUOTA feature is set.
5065  */
5066 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5067 {
5068         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5069                 return -EINVAL;
5070
5071         /* Disable only the limits. */
5072         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5073 }
5074
5075 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5076  * acquiring the locks... As quota files are never truncated and quota code
5077  * itself serializes the operations (and no one else should touch the files)
5078  * we don't have to be afraid of races */
5079 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5080                                size_t len, loff_t off)
5081 {
5082         struct inode *inode = sb_dqopt(sb)->files[type];
5083         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5084         int err = 0;
5085         int offset = off & (sb->s_blocksize - 1);
5086         int tocopy;
5087         size_t toread;
5088         struct buffer_head *bh;
5089         loff_t i_size = i_size_read(inode);
5090
5091         if (off > i_size)
5092                 return 0;
5093         if (off+len > i_size)
5094                 len = i_size-off;
5095         toread = len;
5096         while (toread > 0) {
5097                 tocopy = sb->s_blocksize - offset < toread ?
5098                                 sb->s_blocksize - offset : toread;
5099                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5100                 if (err)
5101                         return err;
5102                 if (!bh)        /* A hole? */
5103                         memset(data, 0, tocopy);
5104                 else
5105                         memcpy(data, bh->b_data+offset, tocopy);
5106                 brelse(bh);
5107                 offset = 0;
5108                 toread -= tocopy;
5109                 data += tocopy;
5110                 blk++;
5111         }
5112         return len;
5113 }
5114
5115 /* Write to quotafile (we know the transaction is already started and has
5116  * enough credits) */
5117 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5118                                 const char *data, size_t len, loff_t off)
5119 {
5120         struct inode *inode = sb_dqopt(sb)->files[type];
5121         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5122         int err = 0;
5123         int offset = off & (sb->s_blocksize - 1);
5124         struct buffer_head *bh;
5125         handle_t *handle = journal_current_handle();
5126
5127         if (EXT4_SB(sb)->s_journal && !handle) {
5128                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5129                         " cancelled because transaction is not started",
5130                         (unsigned long long)off, (unsigned long long)len);
5131                 return -EIO;
5132         }
5133         /*
5134          * Since we account only one data block in transaction credits,
5135          * then it is impossible to cross a block boundary.
5136          */
5137         if (sb->s_blocksize - offset < len) {
5138                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5139                         " cancelled because not block aligned",
5140                         (unsigned long long)off, (unsigned long long)len);
5141                 return -EIO;
5142         }
5143
5144         bh = ext4_bread(handle, inode, blk, 1, &err);
5145         if (!bh)
5146                 goto out;
5147         err = ext4_journal_get_write_access(handle, bh);
5148         if (err) {
5149                 brelse(bh);
5150                 goto out;
5151         }
5152         lock_buffer(bh);
5153         memcpy(bh->b_data+offset, data, len);
5154         flush_dcache_page(bh->b_page);
5155         unlock_buffer(bh);
5156         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5157         brelse(bh);
5158 out:
5159         if (err)
5160                 return err;
5161         if (inode->i_size < off + len) {
5162                 i_size_write(inode, off + len);
5163                 EXT4_I(inode)->i_disksize = inode->i_size;
5164                 ext4_mark_inode_dirty(handle, inode);
5165         }
5166         return len;
5167 }
5168
5169 #endif
5170
5171 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5172                        const char *dev_name, void *data)
5173 {
5174         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5175 }
5176
5177 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5178 static inline void register_as_ext2(void)
5179 {
5180         int err = register_filesystem(&ext2_fs_type);
5181         if (err)
5182                 printk(KERN_WARNING
5183                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5184 }
5185
5186 static inline void unregister_as_ext2(void)
5187 {
5188         unregister_filesystem(&ext2_fs_type);
5189 }
5190
5191 static inline int ext2_feature_set_ok(struct super_block *sb)
5192 {
5193         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5194                 return 0;
5195         if (sb->s_flags & MS_RDONLY)
5196                 return 1;
5197         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5198                 return 0;
5199         return 1;
5200 }
5201 MODULE_ALIAS("ext2");
5202 #else
5203 static inline void register_as_ext2(void) { }
5204 static inline void unregister_as_ext2(void) { }
5205 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5206 #endif
5207
5208 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5209 static inline void register_as_ext3(void)
5210 {
5211         int err = register_filesystem(&ext3_fs_type);
5212         if (err)
5213                 printk(KERN_WARNING
5214                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5215 }
5216
5217 static inline void unregister_as_ext3(void)
5218 {
5219         unregister_filesystem(&ext3_fs_type);
5220 }
5221
5222 static inline int ext3_feature_set_ok(struct super_block *sb)
5223 {
5224         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5225                 return 0;
5226         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5227                 return 0;
5228         if (sb->s_flags & MS_RDONLY)
5229                 return 1;
5230         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5231                 return 0;
5232         return 1;
5233 }
5234 MODULE_ALIAS("ext3");
5235 #else
5236 static inline void register_as_ext3(void) { }
5237 static inline void unregister_as_ext3(void) { }
5238 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5239 #endif
5240
5241 static struct file_system_type ext4_fs_type = {
5242         .owner          = THIS_MODULE,
5243         .name           = "ext4",
5244         .mount          = ext4_mount,
5245         .kill_sb        = kill_block_super,
5246         .fs_flags       = FS_REQUIRES_DEV,
5247 };
5248
5249 static int __init ext4_init_feat_adverts(void)
5250 {
5251         struct ext4_features *ef;
5252         int ret = -ENOMEM;
5253
5254         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5255         if (!ef)
5256                 goto out;
5257
5258         ef->f_kobj.kset = ext4_kset;
5259         init_completion(&ef->f_kobj_unregister);
5260         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5261                                    "features");
5262         if (ret) {
5263                 kfree(ef);
5264                 goto out;
5265         }
5266
5267         ext4_feat = ef;
5268         ret = 0;
5269 out:
5270         return ret;
5271 }
5272
5273 static void ext4_exit_feat_adverts(void)
5274 {
5275         kobject_put(&ext4_feat->f_kobj);
5276         wait_for_completion(&ext4_feat->f_kobj_unregister);
5277         kfree(ext4_feat);
5278 }
5279
5280 /* Shared across all ext4 file systems */
5281 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5282 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5283
5284 static int __init ext4_init_fs(void)
5285 {
5286         int i, err;
5287
5288         ext4_li_info = NULL;
5289         mutex_init(&ext4_li_mtx);
5290
5291         ext4_check_flag_values();
5292
5293         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5294                 mutex_init(&ext4__aio_mutex[i]);
5295                 init_waitqueue_head(&ext4__ioend_wq[i]);
5296         }
5297
5298         err = ext4_init_pageio();
5299         if (err)
5300                 return err;
5301         err = ext4_init_system_zone();
5302         if (err)
5303                 goto out6;
5304         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5305         if (!ext4_kset) {
5306                 err = -ENOMEM;
5307                 goto out5;
5308         }
5309         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5310
5311         err = ext4_init_feat_adverts();
5312         if (err)
5313                 goto out4;
5314
5315         err = ext4_init_mballoc();
5316         if (err)
5317                 goto out3;
5318
5319         err = ext4_init_xattr();
5320         if (err)
5321                 goto out2;
5322         err = init_inodecache();
5323         if (err)
5324                 goto out1;
5325         register_as_ext3();
5326         register_as_ext2();
5327         err = register_filesystem(&ext4_fs_type);
5328         if (err)
5329                 goto out;
5330
5331         return 0;
5332 out:
5333         unregister_as_ext2();
5334         unregister_as_ext3();
5335         destroy_inodecache();
5336 out1:
5337         ext4_exit_xattr();
5338 out2:
5339         ext4_exit_mballoc();
5340 out3:
5341         ext4_exit_feat_adverts();
5342 out4:
5343         if (ext4_proc_root)
5344                 remove_proc_entry("fs/ext4", NULL);
5345         kset_unregister(ext4_kset);
5346 out5:
5347         ext4_exit_system_zone();
5348 out6:
5349         ext4_exit_pageio();
5350         return err;
5351 }
5352
5353 static void __exit ext4_exit_fs(void)
5354 {
5355         ext4_destroy_lazyinit_thread();
5356         unregister_as_ext2();
5357         unregister_as_ext3();
5358         unregister_filesystem(&ext4_fs_type);
5359         destroy_inodecache();
5360         ext4_exit_xattr();
5361         ext4_exit_mballoc();
5362         ext4_exit_feat_adverts();
5363         remove_proc_entry("fs/ext4", NULL);
5364         kset_unregister(ext4_kset);
5365         ext4_exit_system_zone();
5366         ext4_exit_pageio();
5367 }
5368
5369 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5370 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5371 MODULE_LICENSE("GPL");
5372 module_init(ext4_init_fs)
5373 module_exit(ext4_exit_fs)