fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   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 ([email protected]), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static const char *ext4_decode_error(struct super_block *sb, int errno,
  73                                      char nbuf[16]);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static void ext4_write_super(struct super_block *sb);
  78 static int ext4_freeze(struct super_block *sb);
  79 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  80                        const char *dev_name, void *data);
  81 static inline int ext2_feature_set_ok(struct super_block *sb);
  82 static inline int ext3_feature_set_ok(struct super_block *sb);
  83 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  84 static void ext4_destroy_lazyinit_thread(void);
  85 static void ext4_unregister_li_request(struct super_block *sb);
  86 static void ext4_clear_request_list(void);
  87
  88 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  89 static struct file_system_type ext2_fs_type = {
  90         .owner          = THIS_MODULE,
  91         .name           = "ext2",
  92         .mount          = ext4_mount,
  93         .kill_sb        = kill_block_super,
  94         .fs_flags       = FS_REQUIRES_DEV,
  95 };
  96 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  97 #else
  98 #define IS_EXT2_SB(sb) (0)
  99 #endif
 100
 101
 102 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 103 static struct file_system_type ext3_fs_type = {
 104         .owner          = THIS_MODULE,
 105         .name           = "ext3",
 106         .mount          = ext4_mount,
 107         .kill_sb        = kill_block_super,
 108         .fs_flags       = FS_REQUIRES_DEV,
 109 };
 110 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 111 #else
 112 #define IS_EXT3_SB(sb) (0)
 113 #endif
 114
 115 static int ext4_verify_csum_type(struct super_block *sb,
 116                                  struct ext4_super_block *es)
 117 {
 118         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 119                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 120                 return 1;
 121
 122         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 123 }
 124
 125 static __le32 ext4_superblock_csum(struct super_block *sb,
 126                                    struct ext4_super_block *es)
 127 {
 128         struct ext4_sb_info *sbi = EXT4_SB(sb);
 129         int offset = offsetof(struct ext4_super_block, s_checksum);
 130         __u32 csum;
 131
 132         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 133
 134         return cpu_to_le32(csum);
 135 }
 136
 137 int ext4_superblock_csum_verify(struct super_block *sb,
 138                                 struct ext4_super_block *es)
 139 {
 140         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 141                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 142                 return 1;
 143
 144         return es->s_checksum == ext4_superblock_csum(sb, es);
 145 }
 146
 147 void ext4_superblock_csum_set(struct super_block *sb,
 148                               struct ext4_super_block *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  * To avoid j_barrier hold in userspace when a user calls freeze(),
 337  * ext4 prevents a new handle from being started by s_frozen, which
 338  * is in an upper layer.
 339  */
 340 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 341 {
 342         journal_t *journal;
 343         handle_t  *handle;
 344
 345         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 346         if (sb->s_flags & MS_RDONLY)
 347                 return ERR_PTR(-EROFS);
 348
 349         journal = EXT4_SB(sb)->s_journal;
 350         handle = ext4_journal_current_handle();
 351
 352         /*
 353          * If a handle has been started, it should be allowed to
 354          * finish, otherwise deadlock could happen between freeze
 355          * and others(e.g. truncate) due to the restart of the
 356          * journal handle if the filesystem is forzen and active
 357          * handles are not stopped.
 358          */
 359         if (!handle)
 360                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
 361
 362         if (!journal)
 363                 return ext4_get_nojournal();
 364         /*
 365          * Special case here: if the journal has aborted behind our
 366          * backs (eg. EIO in the commit thread), then we still need to
 367          * take the FS itself readonly cleanly.
 368          */
 369         if (is_journal_aborted(journal)) {
 370                 ext4_abort(sb, "Detected aborted journal");
 371                 return ERR_PTR(-EROFS);
 372         }
 373         return jbd2_journal_start(journal, nblocks);
 374 }
 375
 376 /*
 377  * The only special thing we need to do here is to make sure that all
 378  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 379  * that sync() will call the filesystem's write_super callback if
 380  * appropriate.
 381  */
 382 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 383 {
 384         struct super_block *sb;
 385         int err;
 386         int rc;
 387
 388         if (!ext4_handle_valid(handle)) {
 389                 ext4_put_nojournal(handle);
 390                 return 0;
 391         }
 392         sb = handle->h_transaction->t_journal->j_private;
 393         err = handle->h_err;
 394         rc = jbd2_journal_stop(handle);
 395
 396         if (!err)
 397                 err = rc;
 398         if (err)
 399                 __ext4_std_error(sb, where, line, err);
 400         return err;
 401 }
 402
 403 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 404                                const char *err_fn, struct buffer_head *bh,
 405                                handle_t *handle, int err)
 406 {
 407         char nbuf[16];
 408         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 409
 410         BUG_ON(!ext4_handle_valid(handle));
 411
 412         if (bh)
 413                 BUFFER_TRACE(bh, "abort");
 414
 415         if (!handle->h_err)
 416                 handle->h_err = err;
 417
 418         if (is_handle_aborted(handle))
 419                 return;
 420
 421         printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
 422                caller, line, errstr, err_fn);
 423
 424         jbd2_journal_abort_handle(handle);
 425 }
 426
 427 static void __save_error_info(struct super_block *sb, const char *func,
 428                             unsigned int line)
 429 {
 430         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 431
 432         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 433         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 434         es->s_last_error_time = cpu_to_le32(get_seconds());
 435         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 436         es->s_last_error_line = cpu_to_le32(line);
 437         if (!es->s_first_error_time) {
 438                 es->s_first_error_time = es->s_last_error_time;
 439                 strncpy(es->s_first_error_func, func,
 440                         sizeof(es->s_first_error_func));
 441                 es->s_first_error_line = cpu_to_le32(line);
 442                 es->s_first_error_ino = es->s_last_error_ino;
 443                 es->s_first_error_block = es->s_last_error_block;
 444         }
 445         /*
 446          * Start the daily error reporting function if it hasn't been
 447          * started already
 448          */
 449         if (!es->s_error_count)
 450                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 451         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 452 }
 453
 454 static void save_error_info(struct super_block *sb, const char *func,
 455                             unsigned int line)
 456 {
 457         __save_error_info(sb, func, line);
 458         ext4_commit_super(sb, 1);
 459 }
 460
 461 /*
 462  * The del_gendisk() function uninitializes the disk-specific data
 463  * structures, including the bdi structure, without telling anyone
 464  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 465  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 466  * This is a kludge to prevent these oops until we can put in a proper
 467  * hook in del_gendisk() to inform the VFS and file system layers.
 468  */
 469 static int block_device_ejected(struct super_block *sb)
 470 {
 471         struct inode *bd_inode = sb->s_bdev->bd_inode;
 472         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 473
 474         return bdi->dev == NULL;
 475 }
 476
 477 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 478 {
 479         struct super_block              *sb = journal->j_private;
 480         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 481         int                             error = is_journal_aborted(journal);
 482         struct ext4_journal_cb_entry    *jce, *tmp;
 483
 484         spin_lock(&sbi->s_md_lock);
 485         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 486                 list_del_init(&jce->jce_list);
 487                 spin_unlock(&sbi->s_md_lock);
 488                 jce->jce_func(sb, jce, error);
 489                 spin_lock(&sbi->s_md_lock);
 490         }
 491         spin_unlock(&sbi->s_md_lock);
 492 }
 493
 494 /* Deal with the reporting of failure conditions on a filesystem such as
 495  * inconsistencies detected or read IO failures.
 496  *
 497  * On ext2, we can store the error state of the filesystem in the
 498  * superblock.  That is not possible on ext4, because we may have other
 499  * write ordering constraints on the superblock which prevent us from
 500  * writing it out straight away; and given that the journal is about to
 501  * be aborted, we can't rely on the current, or future, transactions to
 502  * write out the superblock safely.
 503  *
 504  * We'll just use the jbd2_journal_abort() error code to record an error in
 505  * the journal instead.  On recovery, the journal will complain about
 506  * that error until we've noted it down and cleared it.
 507  */
 508
 509 static void ext4_handle_error(struct super_block *sb)
 510 {
 511         if (sb->s_flags & MS_RDONLY)
 512                 return;
 513
 514         if (!test_opt(sb, ERRORS_CONT)) {
 515                 journal_t *journal = EXT4_SB(sb)->s_journal;
 516
 517                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 518                 if (journal)
 519                         jbd2_journal_abort(journal, -EIO);
 520         }
 521         if (test_opt(sb, ERRORS_RO)) {
 522                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 523                 sb->s_flags |= MS_RDONLY;
 524         }
 525         if (test_opt(sb, ERRORS_PANIC))
 526                 panic("EXT4-fs (device %s): panic forced after error\n",
 527                         sb->s_id);
 528 }
 529
 530 void __ext4_error(struct super_block *sb, const char *function,
 531                   unsigned int line, const char *fmt, ...)
 532 {
 533         struct va_format vaf;
 534         va_list args;
 535
 536         va_start(args, fmt);
 537         vaf.fmt = fmt;
 538         vaf.va = &args;
 539         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 540                sb->s_id, function, line, current->comm, &vaf);
 541         va_end(args);
 542         save_error_info(sb, function, line);
 543
 544         ext4_handle_error(sb);
 545 }
 546
 547 void ext4_error_inode(struct inode *inode, const char *function,
 548                       unsigned int line, ext4_fsblk_t block,
 549                       const char *fmt, ...)
 550 {
 551         va_list args;
 552         struct va_format vaf;
 553         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 554
 555         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 556         es->s_last_error_block = cpu_to_le64(block);
 557         save_error_info(inode->i_sb, function, line);
 558         va_start(args, fmt);
 559         vaf.fmt = fmt;
 560         vaf.va = &args;
 561         if (block)
 562                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 563                        "inode #%lu: block %llu: comm %s: %pV\n",
 564                        inode->i_sb->s_id, function, line, inode->i_ino,
 565                        block, current->comm, &vaf);
 566         else
 567                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 568                        "inode #%lu: comm %s: %pV\n",
 569                        inode->i_sb->s_id, function, line, inode->i_ino,
 570                        current->comm, &vaf);
 571         va_end(args);
 572
 573         ext4_handle_error(inode->i_sb);
 574 }
 575
 576 void ext4_error_file(struct file *file, const char *function,
 577                      unsigned int line, ext4_fsblk_t block,
 578                      const char *fmt, ...)
 579 {
 580         va_list args;
 581         struct va_format vaf;
 582         struct ext4_super_block *es;
 583         struct inode *inode = file->f_dentry->d_inode;
 584         char pathname[80], *path;
 585
 586         es = EXT4_SB(inode->i_sb)->s_es;
 587         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 588         save_error_info(inode->i_sb, function, line);
 589         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 590         if (IS_ERR(path))
 591                 path = "(unknown)";
 592         va_start(args, fmt);
 593         vaf.fmt = fmt;
 594         vaf.va = &args;
 595         if (block)
 596                 printk(KERN_CRIT
 597                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 598                        "block %llu: comm %s: path %s: %pV\n",
 599                        inode->i_sb->s_id, function, line, inode->i_ino,
 600                        block, current->comm, path, &vaf);
 601         else
 602                 printk(KERN_CRIT
 603                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 604                        "comm %s: path %s: %pV\n",
 605                        inode->i_sb->s_id, function, line, inode->i_ino,
 606                        current->comm, path, &vaf);
 607         va_end(args);
 608
 609         ext4_handle_error(inode->i_sb);
 610 }
 611
 612 static const char *ext4_decode_error(struct super_block *sb, int errno,
 613                                      char nbuf[16])
 614 {
 615         char *errstr = NULL;
 616
 617         switch (errno) {
 618         case -EIO:
 619                 errstr = "IO failure";
 620                 break;
 621         case -ENOMEM:
 622                 errstr = "Out of memory";
 623                 break;
 624         case -EROFS:
 625                 if (!sb || (EXT4_SB(sb)->s_journal &&
 626                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 627                         errstr = "Journal has aborted";
 628                 else
 629                         errstr = "Readonly filesystem";
 630                 break;
 631         default:
 632                 /* If the caller passed in an extra buffer for unknown
 633                  * errors, textualise them now.  Else we just return
 634                  * NULL. */
 635                 if (nbuf) {
 636                         /* Check for truncated error codes... */
 637                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 638                                 errstr = nbuf;
 639                 }
 640                 break;
 641         }
 642
 643         return errstr;
 644 }
 645
 646 /* __ext4_std_error decodes expected errors from journaling functions
 647  * automatically and invokes the appropriate error response.  */
 648
 649 void __ext4_std_error(struct super_block *sb, const char *function,
 650                       unsigned int line, int errno)
 651 {
 652         char nbuf[16];
 653         const char *errstr;
 654
 655         /* Special case: if the error is EROFS, and we're not already
 656          * inside a transaction, then there's really no point in logging
 657          * an error. */
 658         if (errno == -EROFS && journal_current_handle() == NULL &&
 659             (sb->s_flags & MS_RDONLY))
 660                 return;
 661
 662         errstr = ext4_decode_error(sb, errno, nbuf);
 663         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 664                sb->s_id, function, line, errstr);
 665         save_error_info(sb, function, line);
 666
 667         ext4_handle_error(sb);
 668 }
 669
 670 /*
 671  * ext4_abort is a much stronger failure handler than ext4_error.  The
 672  * abort function may be used to deal with unrecoverable failures such
 673  * as journal IO errors or ENOMEM at a critical moment in log management.
 674  *
 675  * We unconditionally force the filesystem into an ABORT|READONLY state,
 676  * unless the error response on the fs has been set to panic in which
 677  * case we take the easy way out and panic immediately.
 678  */
 679
 680 void __ext4_abort(struct super_block *sb, const char *function,
 681                 unsigned int line, const char *fmt, ...)
 682 {
 683         va_list args;
 684
 685         save_error_info(sb, function, line);
 686         va_start(args, fmt);
 687         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 688                function, line);
 689         vprintk(fmt, args);
 690         printk("\n");
 691         va_end(args);
 692
 693         if ((sb->s_flags & MS_RDONLY) == 0) {
 694                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 695                 sb->s_flags |= MS_RDONLY;
 696                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 697                 if (EXT4_SB(sb)->s_journal)
 698                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 699                 save_error_info(sb, function, line);
 700         }
 701         if (test_opt(sb, ERRORS_PANIC))
 702                 panic("EXT4-fs panic from previous error\n");
 703 }
 704
 705 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 706 {
 707         struct va_format vaf;
 708         va_list args;
 709
 710         va_start(args, fmt);
 711         vaf.fmt = fmt;
 712         vaf.va = &args;
 713         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 714         va_end(args);
 715 }
 716
 717 void __ext4_warning(struct super_block *sb, const char *function,
 718                     unsigned int line, const char *fmt, ...)
 719 {
 720         struct va_format vaf;
 721         va_list args;
 722
 723         va_start(args, fmt);
 724         vaf.fmt = fmt;
 725         vaf.va = &args;
 726         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 727                sb->s_id, function, line, &vaf);
 728         va_end(args);
 729 }
 730
 731 void __ext4_grp_locked_error(const char *function, unsigned int line,
 732                              struct super_block *sb, ext4_group_t grp,
 733                              unsigned long ino, ext4_fsblk_t block,
 734                              const char *fmt, ...)
 735 __releases(bitlock)
 736 __acquires(bitlock)
 737 {
 738         struct va_format vaf;
 739         va_list args;
 740         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 741
 742         es->s_last_error_ino = cpu_to_le32(ino);
 743         es->s_last_error_block = cpu_to_le64(block);
 744         __save_error_info(sb, function, line);
 745
 746         va_start(args, fmt);
 747
 748         vaf.fmt = fmt;
 749         vaf.va = &args;
 750         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 751                sb->s_id, function, line, grp);
 752         if (ino)
 753                 printk(KERN_CONT "inode %lu: ", ino);
 754         if (block)
 755                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 756         printk(KERN_CONT "%pV\n", &vaf);
 757         va_end(args);
 758
 759         if (test_opt(sb, ERRORS_CONT)) {
 760                 ext4_commit_super(sb, 0);
 761                 return;
 762         }
 763
 764         ext4_unlock_group(sb, grp);
 765         ext4_handle_error(sb);
 766         /*
 767          * We only get here in the ERRORS_RO case; relocking the group
 768          * may be dangerous, but nothing bad will happen since the
 769          * filesystem will have already been marked read/only and the
 770          * journal has been aborted.  We return 1 as a hint to callers
 771          * who might what to use the return value from
 772          * ext4_grp_locked_error() to distinguish between the
 773          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 774          * aggressively from the ext4 function in question, with a
 775          * more appropriate error code.
 776          */
 777         ext4_lock_group(sb, grp);
 778         return;
 779 }
 780
 781 void ext4_update_dynamic_rev(struct super_block *sb)
 782 {
 783         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 784
 785         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 786                 return;
 787
 788         ext4_warning(sb,
 789                      "updating to rev %d because of new feature flag, "
 790                      "running e2fsck is recommended",
 791                      EXT4_DYNAMIC_REV);
 792
 793         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 794         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 795         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 796         /* leave es->s_feature_*compat flags alone */
 797         /* es->s_uuid will be set by e2fsck if empty */
 798
 799         /*
 800          * The rest of the superblock fields should be zero, and if not it
 801          * means they are likely already in use, so leave them alone.  We
 802          * can leave it up to e2fsck to clean up any inconsistencies there.
 803          */
 804 }
 805
 806 /*
 807  * Open the external journal device
 808  */
 809 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 810 {
 811         struct block_device *bdev;
 812         char b[BDEVNAME_SIZE];
 813
 814         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 815         if (IS_ERR(bdev))
 816                 goto fail;
 817         return bdev;
 818
 819 fail:
 820         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 821                         __bdevname(dev, b), PTR_ERR(bdev));
 822         return NULL;
 823 }
 824
 825 /*
 826  * Release the journal device
 827  */
 828 static int ext4_blkdev_put(struct block_device *bdev)
 829 {
 830         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 831 }
 832
 833 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 834 {
 835         struct block_device *bdev;
 836         int ret = -ENODEV;
 837
 838         bdev = sbi->journal_bdev;
 839         if (bdev) {
 840                 ret = ext4_blkdev_put(bdev);
 841                 sbi->journal_bdev = NULL;
 842         }
 843         return ret;
 844 }
 845
 846 static inline struct inode *orphan_list_entry(struct list_head *l)
 847 {
 848         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 849 }
 850
 851 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 852 {
 853         struct list_head *l;
 854
 855         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 856                  le32_to_cpu(sbi->s_es->s_last_orphan));
 857
 858         printk(KERN_ERR "sb_info orphan list:\n");
 859         list_for_each(l, &sbi->s_orphan) {
 860                 struct inode *inode = orphan_list_entry(l);
 861                 printk(KERN_ERR "  "
 862                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 863                        inode->i_sb->s_id, inode->i_ino, inode,
 864                        inode->i_mode, inode->i_nlink,
 865                        NEXT_ORPHAN(inode));
 866         }
 867 }
 868
 869 static void ext4_put_super(struct super_block *sb)
 870 {
 871         struct ext4_sb_info *sbi = EXT4_SB(sb);
 872         struct ext4_super_block *es = sbi->s_es;
 873         int i, err;
 874
 875         ext4_unregister_li_request(sb);
 876         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 877
 878         flush_workqueue(sbi->dio_unwritten_wq);
 879         destroy_workqueue(sbi->dio_unwritten_wq);
 880
 881         lock_super(sb);
 882         if (sbi->s_journal) {
 883                 err = jbd2_journal_destroy(sbi->s_journal);
 884                 sbi->s_journal = NULL;
 885                 if (err < 0)
 886                         ext4_abort(sb, "Couldn't clean up the journal");
 887         }
 888
 889         del_timer(&sbi->s_err_report);
 890         ext4_release_system_zone(sb);
 891         ext4_mb_release(sb);
 892         ext4_ext_release(sb);
 893         ext4_xattr_put_super(sb);
 894
 895         if (!(sb->s_flags & MS_RDONLY)) {
 896                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 897                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 898         }
 899         if (sb->s_dirt || !(sb->s_flags & MS_RDONLY))
 900                 ext4_commit_super(sb, 1);
 901
 902         if (sbi->s_proc) {
 903                 remove_proc_entry("options", sbi->s_proc);
 904                 remove_proc_entry(sb->s_id, ext4_proc_root);
 905         }
 906         kobject_del(&sbi->s_kobj);
 907
 908         for (i = 0; i < sbi->s_gdb_count; i++)
 909                 brelse(sbi->s_group_desc[i]);
 910         ext4_kvfree(sbi->s_group_desc);
 911         ext4_kvfree(sbi->s_flex_groups);
 912         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 913         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 914         percpu_counter_destroy(&sbi->s_dirs_counter);
 915         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 916         brelse(sbi->s_sbh);
 917 #ifdef CONFIG_QUOTA
 918         for (i = 0; i < MAXQUOTAS; i++)
 919                 kfree(sbi->s_qf_names[i]);
 920 #endif
 921
 922         /* Debugging code just in case the in-memory inode orphan list
 923          * isn't empty.  The on-disk one can be non-empty if we've
 924          * detected an error and taken the fs readonly, but the
 925          * in-memory list had better be clean by this point. */
 926         if (!list_empty(&sbi->s_orphan))
 927                 dump_orphan_list(sb, sbi);
 928         J_ASSERT(list_empty(&sbi->s_orphan));
 929
 930         invalidate_bdev(sb->s_bdev);
 931         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 932                 /*
 933                  * Invalidate the journal device's buffers.  We don't want them
 934                  * floating about in memory - the physical journal device may
 935                  * hotswapped, and it breaks the `ro-after' testing code.
 936                  */
 937                 sync_blockdev(sbi->journal_bdev);
 938                 invalidate_bdev(sbi->journal_bdev);
 939                 ext4_blkdev_remove(sbi);
 940         }
 941         if (sbi->s_mmp_tsk)
 942                 kthread_stop(sbi->s_mmp_tsk);
 943         sb->s_fs_info = NULL;
 944         /*
 945          * Now that we are completely done shutting down the
 946          * superblock, we need to actually destroy the kobject.
 947          */
 948         unlock_super(sb);
 949         kobject_put(&sbi->s_kobj);
 950         wait_for_completion(&sbi->s_kobj_unregister);
 951         if (sbi->s_chksum_driver)
 952                 crypto_free_shash(sbi->s_chksum_driver);
 953         kfree(sbi->s_blockgroup_lock);
 954         kfree(sbi);
 955 }
 956
 957 static struct kmem_cache *ext4_inode_cachep;
 958
 959 /*
 960  * Called inside transaction, so use GFP_NOFS
 961  */
 962 static struct inode *ext4_alloc_inode(struct super_block *sb)
 963 {
 964         struct ext4_inode_info *ei;
 965
 966         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 967         if (!ei)
 968                 return NULL;
 969
 970         ei->vfs_inode.i_version = 1;
 971         ei->vfs_inode.i_data.writeback_index = 0;
 972         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 973         INIT_LIST_HEAD(&ei->i_prealloc_list);
 974         spin_lock_init(&ei->i_prealloc_lock);
 975         ei->i_reserved_data_blocks = 0;
 976         ei->i_reserved_meta_blocks = 0;
 977         ei->i_allocated_meta_blocks = 0;
 978         ei->i_da_metadata_calc_len = 0;
 979         spin_lock_init(&(ei->i_block_reservation_lock));
 980 #ifdef CONFIG_QUOTA
 981         ei->i_reserved_quota = 0;
 982 #endif
 983         ei->jinode = NULL;
 984         INIT_LIST_HEAD(&ei->i_completed_io_list);
 985         spin_lock_init(&ei->i_completed_io_lock);
 986         ei->cur_aio_dio = NULL;
 987         ei->i_sync_tid = 0;
 988         ei->i_datasync_tid = 0;
 989         atomic_set(&ei->i_ioend_count, 0);
 990         atomic_set(&ei->i_aiodio_unwritten, 0);
 991
 992         return &ei->vfs_inode;
 993 }
 994
 995 static int ext4_drop_inode(struct inode *inode)
 996 {
 997         int drop = generic_drop_inode(inode);
 998
 999         trace_ext4_drop_inode(inode, drop);
1000         return drop;
1001 }
1002
1003 static void ext4_i_callback(struct rcu_head *head)
1004 {
1005         struct inode *inode = container_of(head, struct inode, i_rcu);
1006         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
1007 }
1008
1009 static void ext4_destroy_inode(struct inode *inode)
1010 {
1011         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
1012                 ext4_msg(inode->i_sb, KERN_ERR,
1013                          "Inode %lu (%p): orphan list check failed!",
1014                          inode->i_ino, EXT4_I(inode));
1015                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1016                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1017                                 true);
1018                 dump_stack();
1019         }
1020         call_rcu(&inode->i_rcu, ext4_i_callback);
1021 }
1022
1023 static void init_once(void *foo)
1024 {
1025         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1026
1027         INIT_LIST_HEAD(&ei->i_orphan);
1028 #ifdef CONFIG_EXT4_FS_XATTR
1029         init_rwsem(&ei->xattr_sem);
1030 #endif
1031         init_rwsem(&ei->i_data_sem);
1032         inode_init_once(&ei->vfs_inode);
1033 }
1034
1035 static int init_inodecache(void)
1036 {
1037         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1038                                              sizeof(struct ext4_inode_info),
1039                                              0, (SLAB_RECLAIM_ACCOUNT|
1040                                                 SLAB_MEM_SPREAD),
1041                                              init_once);
1042         if (ext4_inode_cachep == NULL)
1043                 return -ENOMEM;
1044         return 0;
1045 }
1046
1047 static void destroy_inodecache(void)
1048 {
1049         kmem_cache_destroy(ext4_inode_cachep);
1050 }
1051
1052 void ext4_clear_inode(struct inode *inode)
1053 {
1054         invalidate_inode_buffers(inode);
1055         clear_inode(inode);
1056         dquot_drop(inode);
1057         ext4_discard_preallocations(inode);
1058         if (EXT4_I(inode)->jinode) {
1059                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1060                                                EXT4_I(inode)->jinode);
1061                 jbd2_free_inode(EXT4_I(inode)->jinode);
1062                 EXT4_I(inode)->jinode = NULL;
1063         }
1064 }
1065
1066 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1067                                         u64 ino, u32 generation)
1068 {
1069         struct inode *inode;
1070
1071         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1072                 return ERR_PTR(-ESTALE);
1073         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1074                 return ERR_PTR(-ESTALE);
1075
1076         /* iget isn't really right if the inode is currently unallocated!!
1077          *
1078          * ext4_read_inode will return a bad_inode if the inode had been
1079          * deleted, so we should be safe.
1080          *
1081          * Currently we don't know the generation for parent directory, so
1082          * a generation of 0 means "accept any"
1083          */
1084         inode = ext4_iget(sb, ino);
1085         if (IS_ERR(inode))
1086                 return ERR_CAST(inode);
1087         if (generation && inode->i_generation != generation) {
1088                 iput(inode);
1089                 return ERR_PTR(-ESTALE);
1090         }
1091
1092         return inode;
1093 }
1094
1095 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1096                                         int fh_len, int fh_type)
1097 {
1098         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1099                                     ext4_nfs_get_inode);
1100 }
1101
1102 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1103                                         int fh_len, int fh_type)
1104 {
1105         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1106                                     ext4_nfs_get_inode);
1107 }
1108
1109 /*
1110  * Try to release metadata pages (indirect blocks, directories) which are
1111  * mapped via the block device.  Since these pages could have journal heads
1112  * which would prevent try_to_free_buffers() from freeing them, we must use
1113  * jbd2 layer's try_to_free_buffers() function to release them.
1114  */
1115 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1116                                  gfp_t wait)
1117 {
1118         journal_t *journal = EXT4_SB(sb)->s_journal;
1119
1120         WARN_ON(PageChecked(page));
1121         if (!page_has_buffers(page))
1122                 return 0;
1123         if (journal)
1124                 return jbd2_journal_try_to_free_buffers(journal, page,
1125                                                         wait & ~__GFP_WAIT);
1126         return try_to_free_buffers(page);
1127 }
1128
1129 #ifdef CONFIG_QUOTA
1130 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1131 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1132
1133 static int ext4_write_dquot(struct dquot *dquot);
1134 static int ext4_acquire_dquot(struct dquot *dquot);
1135 static int ext4_release_dquot(struct dquot *dquot);
1136 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1137 static int ext4_write_info(struct super_block *sb, int type);
1138 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1139                          struct path *path);
1140 static int ext4_quota_off(struct super_block *sb, int type);
1141 static int ext4_quota_on_mount(struct super_block *sb, int type);
1142 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1143                                size_t len, loff_t off);
1144 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1145                                 const char *data, size_t len, loff_t off);
1146
1147 static const struct dquot_operations ext4_quota_operations = {
1148         .get_reserved_space = ext4_get_reserved_space,
1149         .write_dquot    = ext4_write_dquot,
1150         .acquire_dquot  = ext4_acquire_dquot,
1151         .release_dquot  = ext4_release_dquot,
1152         .mark_dirty     = ext4_mark_dquot_dirty,
1153         .write_info     = ext4_write_info,
1154         .alloc_dquot    = dquot_alloc,
1155         .destroy_dquot  = dquot_destroy,
1156 };
1157
1158 static const struct quotactl_ops ext4_qctl_operations = {
1159         .quota_on       = ext4_quota_on,
1160         .quota_off      = ext4_quota_off,
1161         .quota_sync     = dquot_quota_sync,
1162         .get_info       = dquot_get_dqinfo,
1163         .set_info       = dquot_set_dqinfo,
1164         .get_dqblk      = dquot_get_dqblk,
1165         .set_dqblk      = dquot_set_dqblk
1166 };
1167 #endif
1168
1169 static const struct super_operations ext4_sops = {
1170         .alloc_inode    = ext4_alloc_inode,
1171         .destroy_inode  = ext4_destroy_inode,
1172         .write_inode    = ext4_write_inode,
1173         .dirty_inode    = ext4_dirty_inode,
1174         .drop_inode     = ext4_drop_inode,
1175         .evict_inode    = ext4_evict_inode,
1176         .put_super      = ext4_put_super,
1177         .sync_fs        = ext4_sync_fs,
1178         .freeze_fs      = ext4_freeze,
1179         .unfreeze_fs    = ext4_unfreeze,
1180         .statfs         = ext4_statfs,
1181         .remount_fs     = ext4_remount,
1182         .show_options   = ext4_show_options,
1183 #ifdef CONFIG_QUOTA
1184         .quota_read     = ext4_quota_read,
1185         .quota_write    = ext4_quota_write,
1186 #endif
1187         .bdev_try_to_free_page = bdev_try_to_free_page,
1188 };
1189
1190 static const struct super_operations ext4_nojournal_sops = {
1191         .alloc_inode    = ext4_alloc_inode,
1192         .destroy_inode  = ext4_destroy_inode,
1193         .write_inode    = ext4_write_inode,
1194         .dirty_inode    = ext4_dirty_inode,
1195         .drop_inode     = ext4_drop_inode,
1196         .evict_inode    = ext4_evict_inode,
1197         .write_super    = ext4_write_super,
1198         .put_super      = ext4_put_super,
1199         .statfs         = ext4_statfs,
1200         .remount_fs     = ext4_remount,
1201         .show_options   = ext4_show_options,
1202 #ifdef CONFIG_QUOTA
1203         .quota_read     = ext4_quota_read,
1204         .quota_write    = ext4_quota_write,
1205 #endif
1206         .bdev_try_to_free_page = bdev_try_to_free_page,
1207 };
1208
1209 static const struct export_operations ext4_export_ops = {
1210         .fh_to_dentry = ext4_fh_to_dentry,
1211         .fh_to_parent = ext4_fh_to_parent,
1212         .get_parent = ext4_get_parent,
1213 };
1214
1215 enum {
1216         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1217         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1218         Opt_nouid32, Opt_debug, Opt_removed,
1219         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1220         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1221         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1222         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1223         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1224         Opt_data_err_abort, Opt_data_err_ignore,
1225         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1226         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1227         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1228         Opt_usrquota, Opt_grpquota, Opt_i_version,
1229         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1230         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1231         Opt_inode_readahead_blks, Opt_journal_ioprio,
1232         Opt_dioread_nolock, Opt_dioread_lock,
1233         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1234 };
1235
1236 static const match_table_t tokens = {
1237         {Opt_bsd_df, "bsddf"},
1238         {Opt_minix_df, "minixdf"},
1239         {Opt_grpid, "grpid"},
1240         {Opt_grpid, "bsdgroups"},
1241         {Opt_nogrpid, "nogrpid"},
1242         {Opt_nogrpid, "sysvgroups"},
1243         {Opt_resgid, "resgid=%u"},
1244         {Opt_resuid, "resuid=%u"},
1245         {Opt_sb, "sb=%u"},
1246         {Opt_err_cont, "errors=continue"},
1247         {Opt_err_panic, "errors=panic"},
1248         {Opt_err_ro, "errors=remount-ro"},
1249         {Opt_nouid32, "nouid32"},
1250         {Opt_debug, "debug"},
1251         {Opt_removed, "oldalloc"},
1252         {Opt_removed, "orlov"},
1253         {Opt_user_xattr, "user_xattr"},
1254         {Opt_nouser_xattr, "nouser_xattr"},
1255         {Opt_acl, "acl"},
1256         {Opt_noacl, "noacl"},
1257         {Opt_noload, "norecovery"},
1258         {Opt_noload, "noload"},
1259         {Opt_removed, "nobh"},
1260         {Opt_removed, "bh"},
1261         {Opt_commit, "commit=%u"},
1262         {Opt_min_batch_time, "min_batch_time=%u"},
1263         {Opt_max_batch_time, "max_batch_time=%u"},
1264         {Opt_journal_dev, "journal_dev=%u"},
1265         {Opt_journal_checksum, "journal_checksum"},
1266         {Opt_journal_async_commit, "journal_async_commit"},
1267         {Opt_abort, "abort"},
1268         {Opt_data_journal, "data=journal"},
1269         {Opt_data_ordered, "data=ordered"},
1270         {Opt_data_writeback, "data=writeback"},
1271         {Opt_data_err_abort, "data_err=abort"},
1272         {Opt_data_err_ignore, "data_err=ignore"},
1273         {Opt_offusrjquota, "usrjquota="},
1274         {Opt_usrjquota, "usrjquota=%s"},
1275         {Opt_offgrpjquota, "grpjquota="},
1276         {Opt_grpjquota, "grpjquota=%s"},
1277         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1278         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1279         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1280         {Opt_grpquota, "grpquota"},
1281         {Opt_noquota, "noquota"},
1282         {Opt_quota, "quota"},
1283         {Opt_usrquota, "usrquota"},
1284         {Opt_barrier, "barrier=%u"},
1285         {Opt_barrier, "barrier"},
1286         {Opt_nobarrier, "nobarrier"},
1287         {Opt_i_version, "i_version"},
1288         {Opt_stripe, "stripe=%u"},
1289         {Opt_delalloc, "delalloc"},
1290         {Opt_nodelalloc, "nodelalloc"},
1291         {Opt_mblk_io_submit, "mblk_io_submit"},
1292         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1293         {Opt_block_validity, "block_validity"},
1294         {Opt_noblock_validity, "noblock_validity"},
1295         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1296         {Opt_journal_ioprio, "journal_ioprio=%u"},
1297         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1298         {Opt_auto_da_alloc, "auto_da_alloc"},
1299         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1300         {Opt_dioread_nolock, "dioread_nolock"},
1301         {Opt_dioread_lock, "dioread_lock"},
1302         {Opt_discard, "discard"},
1303         {Opt_nodiscard, "nodiscard"},
1304         {Opt_init_itable, "init_itable=%u"},
1305         {Opt_init_itable, "init_itable"},
1306         {Opt_noinit_itable, "noinit_itable"},
1307         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1308         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1309         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1310         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1311         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1312         {Opt_err, NULL},
1313 };
1314
1315 static ext4_fsblk_t get_sb_block(void **data)
1316 {
1317         ext4_fsblk_t    sb_block;
1318         char            *options = (char *) *data;
1319
1320         if (!options || strncmp(options, "sb=", 3) != 0)
1321                 return 1;       /* Default location */
1322
1323         options += 3;
1324         /* TODO: use simple_strtoll with >32bit ext4 */
1325         sb_block = simple_strtoul(options, &options, 0);
1326         if (*options && *options != ',') {
1327                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1328                        (char *) *data);
1329                 return 1;
1330         }
1331         if (*options == ',')
1332                 options++;
1333         *data = (void *) options;
1334
1335         return sb_block;
1336 }
1337
1338 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1339 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1340         "Contact [email protected] if you think we should keep it.\n";
1341
1342 #ifdef CONFIG_QUOTA
1343 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1344 {
1345         struct ext4_sb_info *sbi = EXT4_SB(sb);
1346         char *qname;
1347
1348         if (sb_any_quota_loaded(sb) &&
1349                 !sbi->s_qf_names[qtype]) {
1350                 ext4_msg(sb, KERN_ERR,
1351                         "Cannot change journaled "
1352                         "quota options when quota turned on");
1353                 return -1;
1354         }
1355         qname = match_strdup(args);
1356         if (!qname) {
1357                 ext4_msg(sb, KERN_ERR,
1358                         "Not enough memory for storing quotafile name");
1359                 return -1;
1360         }
1361         if (sbi->s_qf_names[qtype] &&
1362                 strcmp(sbi->s_qf_names[qtype], qname)) {
1363                 ext4_msg(sb, KERN_ERR,
1364                         "%s quota file already specified", QTYPE2NAME(qtype));
1365                 kfree(qname);
1366                 return -1;
1367         }
1368         sbi->s_qf_names[qtype] = qname;
1369         if (strchr(sbi->s_qf_names[qtype], '/')) {
1370                 ext4_msg(sb, KERN_ERR,
1371                         "quotafile must be on filesystem root");
1372                 kfree(sbi->s_qf_names[qtype]);
1373                 sbi->s_qf_names[qtype] = NULL;
1374                 return -1;
1375         }
1376         set_opt(sb, QUOTA);
1377         return 1;
1378 }
1379
1380 static int clear_qf_name(struct super_block *sb, int qtype)
1381 {
1382
1383         struct ext4_sb_info *sbi = EXT4_SB(sb);
1384
1385         if (sb_any_quota_loaded(sb) &&
1386                 sbi->s_qf_names[qtype]) {
1387                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1388                         " when quota turned on");
1389                 return -1;
1390         }
1391         /*
1392          * The space will be released later when all options are confirmed
1393          * to be correct
1394          */
1395         sbi->s_qf_names[qtype] = NULL;
1396         return 1;
1397 }
1398 #endif
1399
1400 #define MOPT_SET        0x0001
1401 #define MOPT_CLEAR      0x0002
1402 #define MOPT_NOSUPPORT  0x0004
1403 #define MOPT_EXPLICIT   0x0008
1404 #define MOPT_CLEAR_ERR  0x0010
1405 #define MOPT_GTE0       0x0020
1406 #ifdef CONFIG_QUOTA
1407 #define MOPT_Q          0
1408 #define MOPT_QFMT       0x0040
1409 #else
1410 #define MOPT_Q          MOPT_NOSUPPORT
1411 #define MOPT_QFMT       MOPT_NOSUPPORT
1412 #endif
1413 #define MOPT_DATAJ      0x0080
1414
1415 static const struct mount_opts {
1416         int     token;
1417         int     mount_opt;
1418         int     flags;
1419 } ext4_mount_opts[] = {
1420         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1421         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1422         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1423         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1424         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1425         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1426         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1427         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1428         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1429         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1430         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1431         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1432         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1433         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1434         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1435         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1436                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1437         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1438         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1439         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1440         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1441         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1442         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1443         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1444         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1445         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1446         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1447         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1448         {Opt_commit, 0, MOPT_GTE0},
1449         {Opt_max_batch_time, 0, MOPT_GTE0},
1450         {Opt_min_batch_time, 0, MOPT_GTE0},
1451         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1452         {Opt_init_itable, 0, MOPT_GTE0},
1453         {Opt_stripe, 0, MOPT_GTE0},
1454         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1455         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1456         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1457 #ifdef CONFIG_EXT4_FS_XATTR
1458         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1459         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1460 #else
1461         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1462         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1463 #endif
1464 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1465         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1466         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1467 #else
1468         {Opt_acl, 0, MOPT_NOSUPPORT},
1469         {Opt_noacl, 0, MOPT_NOSUPPORT},
1470 #endif
1471         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1472         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1473         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1474         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1475                                                         MOPT_SET | MOPT_Q},
1476         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1477                                                         MOPT_SET | MOPT_Q},
1478         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1479                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1480         {Opt_usrjquota, 0, MOPT_Q},
1481         {Opt_grpjquota, 0, MOPT_Q},
1482         {Opt_offusrjquota, 0, MOPT_Q},
1483         {Opt_offgrpjquota, 0, MOPT_Q},
1484         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1485         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1486         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
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_stripe) {
1603                         sbi->s_stripe = arg;
1604                 } else if (m->flags & MOPT_DATAJ) {
1605                         if (is_remount) {
1606                                 if (!sbi->s_journal)
1607                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1608                                 else if (test_opt(sb, DATA_FLAGS) !=
1609                                          m->mount_opt) {
1610                                         ext4_msg(sb, KERN_ERR,
1611                                          "Cannot change data mode on remount");
1612                                         return -1;
1613                                 }
1614                         } else {
1615                                 clear_opt(sb, DATA_FLAGS);
1616                                 sbi->s_mount_opt |= m->mount_opt;
1617                         }
1618 #ifdef CONFIG_QUOTA
1619                 } else if (m->flags & MOPT_QFMT) {
1620                         if (sb_any_quota_loaded(sb) &&
1621                             sbi->s_jquota_fmt != m->mount_opt) {
1622                                 ext4_msg(sb, KERN_ERR, "Cannot "
1623                                          "change journaled quota options "
1624                                          "when quota turned on");
1625                                 return -1;
1626                         }
1627                         sbi->s_jquota_fmt = m->mount_opt;
1628 #endif
1629                 } else {
1630                         if (!args->from)
1631                                 arg = 1;
1632                         if (m->flags & MOPT_CLEAR)
1633                                 arg = !arg;
1634                         else if (unlikely(!(m->flags & MOPT_SET))) {
1635                                 ext4_msg(sb, KERN_WARNING,
1636                                          "buggy handling of option %s", opt);
1637                                 WARN_ON(1);
1638                                 return -1;
1639                         }
1640                         if (arg != 0)
1641                                 sbi->s_mount_opt |= m->mount_opt;
1642                         else
1643                                 sbi->s_mount_opt &= ~m->mount_opt;
1644                 }
1645                 return 1;
1646         }
1647         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1648                  "or missing value", opt);
1649         return -1;
1650 }
1651
1652 static int parse_options(char *options, struct super_block *sb,
1653                          unsigned long *journal_devnum,
1654                          unsigned int *journal_ioprio,
1655                          int is_remount)
1656 {
1657 #ifdef CONFIG_QUOTA
1658         struct ext4_sb_info *sbi = EXT4_SB(sb);
1659 #endif
1660         char *p;
1661         substring_t args[MAX_OPT_ARGS];
1662         int token;
1663
1664         if (!options)
1665                 return 1;
1666
1667         while ((p = strsep(&options, ",")) != NULL) {
1668                 if (!*p)
1669                         continue;
1670                 /*
1671                  * Initialize args struct so we know whether arg was
1672                  * found; some options take optional arguments.
1673                  */
1674                 args[0].to = args[0].from = 0;
1675                 token = match_token(p, tokens, args);
1676                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1677                                      journal_ioprio, is_remount) < 0)
1678                         return 0;
1679         }
1680 #ifdef CONFIG_QUOTA
1681         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1682                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1683                         clear_opt(sb, USRQUOTA);
1684
1685                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1686                         clear_opt(sb, GRPQUOTA);
1687
1688                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1689                         ext4_msg(sb, KERN_ERR, "old and new quota "
1690                                         "format mixing");
1691                         return 0;
1692                 }
1693
1694                 if (!sbi->s_jquota_fmt) {
1695                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1696                                         "not specified");
1697                         return 0;
1698                 }
1699         } else {
1700                 if (sbi->s_jquota_fmt) {
1701                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1702                                         "specified with no journaling "
1703                                         "enabled");
1704                         return 0;
1705                 }
1706         }
1707 #endif
1708         return 1;
1709 }
1710
1711 static inline void ext4_show_quota_options(struct seq_file *seq,
1712                                            struct super_block *sb)
1713 {
1714 #if defined(CONFIG_QUOTA)
1715         struct ext4_sb_info *sbi = EXT4_SB(sb);
1716
1717         if (sbi->s_jquota_fmt) {
1718                 char *fmtname = "";
1719
1720                 switch (sbi->s_jquota_fmt) {
1721                 case QFMT_VFS_OLD:
1722                         fmtname = "vfsold";
1723                         break;
1724                 case QFMT_VFS_V0:
1725                         fmtname = "vfsv0";
1726                         break;
1727                 case QFMT_VFS_V1:
1728                         fmtname = "vfsv1";
1729                         break;
1730                 }
1731                 seq_printf(seq, ",jqfmt=%s", fmtname);
1732         }
1733
1734         if (sbi->s_qf_names[USRQUOTA])
1735                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1736
1737         if (sbi->s_qf_names[GRPQUOTA])
1738                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1739
1740         if (test_opt(sb, USRQUOTA))
1741                 seq_puts(seq, ",usrquota");
1742
1743         if (test_opt(sb, GRPQUOTA))
1744                 seq_puts(seq, ",grpquota");
1745 #endif
1746 }
1747
1748 static const char *token2str(int token)
1749 {
1750         static const struct match_token *t;
1751
1752         for (t = tokens; t->token != Opt_err; t++)
1753                 if (t->token == token && !strchr(t->pattern, '='))
1754                         break;
1755         return t->pattern;
1756 }
1757
1758 /*
1759  * Show an option if
1760  *  - it's set to a non-default value OR
1761  *  - if the per-sb default is different from the global default
1762  */
1763 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1764                               int nodefs)
1765 {
1766         struct ext4_sb_info *sbi = EXT4_SB(sb);
1767         struct ext4_super_block *es = sbi->s_es;
1768         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1769         const struct mount_opts *m;
1770         char sep = nodefs ? '\n' : ',';
1771
1772 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1773 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1774
1775         if (sbi->s_sb_block != 1)
1776                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1777
1778         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1779                 int want_set = m->flags & MOPT_SET;
1780                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1781                     (m->flags & MOPT_CLEAR_ERR))
1782                         continue;
1783                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1784                         continue; /* skip if same as the default */
1785                 if ((want_set &&
1786                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1787                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1788                         continue; /* select Opt_noFoo vs Opt_Foo */
1789                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1790         }
1791
1792         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1793             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1794                 SEQ_OPTS_PRINT("resuid=%u",
1795                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1796         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1797             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1798                 SEQ_OPTS_PRINT("resgid=%u",
1799                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1800         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1801         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1802                 SEQ_OPTS_PUTS("errors=remount-ro");
1803         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1804                 SEQ_OPTS_PUTS("errors=continue");
1805         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1806                 SEQ_OPTS_PUTS("errors=panic");
1807         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1808                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1809         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1810                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1811         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1812                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1813         if (sb->s_flags & MS_I_VERSION)
1814                 SEQ_OPTS_PUTS("i_version");
1815         if (nodefs || sbi->s_stripe)
1816                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1817         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1818                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1819                         SEQ_OPTS_PUTS("data=journal");
1820                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1821                         SEQ_OPTS_PUTS("data=ordered");
1822                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1823                         SEQ_OPTS_PUTS("data=writeback");
1824         }
1825         if (nodefs ||
1826             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1827                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1828                                sbi->s_inode_readahead_blks);
1829
1830         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1831                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1832                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1833
1834         ext4_show_quota_options(seq, sb);
1835         return 0;
1836 }
1837
1838 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1839 {
1840         return _ext4_show_options(seq, root->d_sb, 0);
1841 }
1842
1843 static int options_seq_show(struct seq_file *seq, void *offset)
1844 {
1845         struct super_block *sb = seq->private;
1846         int rc;
1847
1848         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1849         rc = _ext4_show_options(seq, sb, 1);
1850         seq_puts(seq, "\n");
1851         return rc;
1852 }
1853
1854 static int options_open_fs(struct inode *inode, struct file *file)
1855 {
1856         return single_open(file, options_seq_show, PDE(inode)->data);
1857 }
1858
1859 static const struct file_operations ext4_seq_options_fops = {
1860         .owner = THIS_MODULE,
1861         .open = options_open_fs,
1862         .read = seq_read,
1863         .llseek = seq_lseek,
1864         .release = single_release,
1865 };
1866
1867 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1868                             int read_only)
1869 {
1870         struct ext4_sb_info *sbi = EXT4_SB(sb);
1871         int res = 0;
1872
1873         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1874                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1875                          "forcing read-only mode");
1876                 res = MS_RDONLY;
1877         }
1878         if (read_only)
1879                 goto done;
1880         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1881                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1882                          "running e2fsck is recommended");
1883         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1884                 ext4_msg(sb, KERN_WARNING,
1885                          "warning: mounting fs with errors, "
1886                          "running e2fsck is recommended");
1887         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1888                  le16_to_cpu(es->s_mnt_count) >=
1889                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1890                 ext4_msg(sb, KERN_WARNING,
1891                          "warning: maximal mount count reached, "
1892                          "running e2fsck is recommended");
1893         else if (le32_to_cpu(es->s_checkinterval) &&
1894                 (le32_to_cpu(es->s_lastcheck) +
1895                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1896                 ext4_msg(sb, KERN_WARNING,
1897                          "warning: checktime reached, "
1898                          "running e2fsck is recommended");
1899         if (!sbi->s_journal)
1900                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1901         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1902                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1903         le16_add_cpu(&es->s_mnt_count, 1);
1904         es->s_mtime = cpu_to_le32(get_seconds());
1905         ext4_update_dynamic_rev(sb);
1906         if (sbi->s_journal)
1907                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1908
1909         ext4_commit_super(sb, 1);
1910 done:
1911         if (test_opt(sb, DEBUG))
1912                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1913                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1914                         sb->s_blocksize,
1915                         sbi->s_groups_count,
1916                         EXT4_BLOCKS_PER_GROUP(sb),
1917                         EXT4_INODES_PER_GROUP(sb),
1918                         sbi->s_mount_opt, sbi->s_mount_opt2);
1919
1920         cleancache_init_fs(sb);
1921         return res;
1922 }
1923
1924 static int ext4_fill_flex_info(struct super_block *sb)
1925 {
1926         struct ext4_sb_info *sbi = EXT4_SB(sb);
1927         struct ext4_group_desc *gdp = NULL;
1928         ext4_group_t flex_group_count;
1929         ext4_group_t flex_group;
1930         unsigned int groups_per_flex = 0;
1931         size_t size;
1932         int i;
1933
1934         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1935         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1936                 sbi->s_log_groups_per_flex = 0;
1937                 return 1;
1938         }
1939         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1940
1941         /* We allocate both existing and potentially added groups */
1942         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1943                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1944                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1945         size = flex_group_count * sizeof(struct flex_groups);
1946         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1947         if (sbi->s_flex_groups == NULL) {
1948                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1949                          flex_group_count);
1950                 goto failed;
1951         }
1952
1953         for (i = 0; i < sbi->s_groups_count; i++) {
1954                 gdp = ext4_get_group_desc(sb, i, NULL);
1955
1956                 flex_group = ext4_flex_group(sbi, i);
1957                 atomic_add(ext4_free_inodes_count(sb, gdp),
1958                            &sbi->s_flex_groups[flex_group].free_inodes);
1959                 atomic_add(ext4_free_group_clusters(sb, gdp),
1960                            &sbi->s_flex_groups[flex_group].free_clusters);
1961                 atomic_add(ext4_used_dirs_count(sb, gdp),
1962                            &sbi->s_flex_groups[flex_group].used_dirs);
1963         }
1964
1965         return 1;
1966 failed:
1967         return 0;
1968 }
1969
1970 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1971                                    struct ext4_group_desc *gdp)
1972 {
1973         int offset;
1974         __u16 crc = 0;
1975         __le32 le_group = cpu_to_le32(block_group);
1976
1977         if ((sbi->s_es->s_feature_ro_compat &
1978              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1979                 /* Use new metadata_csum algorithm */
1980                 __u16 old_csum;
1981                 __u32 csum32;
1982
1983                 old_csum = gdp->bg_checksum;
1984                 gdp->bg_checksum = 0;
1985                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1986                                      sizeof(le_group));
1987                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1988                                      sbi->s_desc_size);
1989                 gdp->bg_checksum = old_csum;
1990
1991                 crc = csum32 & 0xFFFF;
1992                 goto out;
1993         }
1994
1995         /* old crc16 code */
1996         offset = offsetof(struct ext4_group_desc, bg_checksum);
1997
1998         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1999         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2000         crc = crc16(crc, (__u8 *)gdp, offset);
2001         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2002         /* for checksum of struct ext4_group_desc do the rest...*/
2003         if ((sbi->s_es->s_feature_incompat &
2004              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2005             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2006                 crc = crc16(crc, (__u8 *)gdp + offset,
2007                             le16_to_cpu(sbi->s_es->s_desc_size) -
2008                                 offset);
2009
2010 out:
2011         return cpu_to_le16(crc);
2012 }
2013
2014 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2015                                 struct ext4_group_desc *gdp)
2016 {
2017         if (ext4_has_group_desc_csum(sb) &&
2018             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2019                                                       block_group, gdp)))
2020                 return 0;
2021
2022         return 1;
2023 }
2024
2025 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2026                               struct ext4_group_desc *gdp)
2027 {
2028         if (!ext4_has_group_desc_csum(sb))
2029                 return;
2030         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2031 }
2032
2033 /* Called at mount-time, super-block is locked */
2034 static int ext4_check_descriptors(struct super_block *sb,
2035                                   ext4_group_t *first_not_zeroed)
2036 {
2037         struct ext4_sb_info *sbi = EXT4_SB(sb);
2038         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2039         ext4_fsblk_t last_block;
2040         ext4_fsblk_t block_bitmap;
2041         ext4_fsblk_t inode_bitmap;
2042         ext4_fsblk_t inode_table;
2043         int flexbg_flag = 0;
2044         ext4_group_t i, grp = sbi->s_groups_count;
2045
2046         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2047                 flexbg_flag = 1;
2048
2049         ext4_debug("Checking group descriptors");
2050
2051         for (i = 0; i < sbi->s_groups_count; i++) {
2052                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2053
2054                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2055                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2056                 else
2057                         last_block = first_block +
2058                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2059
2060                 if ((grp == sbi->s_groups_count) &&
2061                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2062                         grp = i;
2063
2064                 block_bitmap = ext4_block_bitmap(sb, gdp);
2065                 if (block_bitmap < first_block || block_bitmap > last_block) {
2066                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2067                                "Block bitmap for group %u not in group "
2068                                "(block %llu)!", i, block_bitmap);
2069                         return 0;
2070                 }
2071                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2072                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2073                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2074                                "Inode bitmap for group %u not in group "
2075                                "(block %llu)!", i, inode_bitmap);
2076                         return 0;
2077                 }
2078                 inode_table = ext4_inode_table(sb, gdp);
2079                 if (inode_table < first_block ||
2080                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2081                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2082                                "Inode table for group %u not in group "
2083                                "(block %llu)!", i, inode_table);
2084                         return 0;
2085                 }
2086                 ext4_lock_group(sb, i);
2087                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2088                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2089                                  "Checksum for group %u failed (%u!=%u)",
2090                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2091                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2092                         if (!(sb->s_flags & MS_RDONLY)) {
2093                                 ext4_unlock_group(sb, i);
2094                                 return 0;
2095                         }
2096                 }
2097                 ext4_unlock_group(sb, i);
2098                 if (!flexbg_flag)
2099                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2100         }
2101         if (NULL != first_not_zeroed)
2102                 *first_not_zeroed = grp;
2103
2104         ext4_free_blocks_count_set(sbi->s_es,
2105                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2106         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2107         return 1;
2108 }
2109
2110 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2111  * the superblock) which were deleted from all directories, but held open by
2112  * a process at the time of a crash.  We walk the list and try to delete these
2113  * inodes at recovery time (only with a read-write filesystem).
2114  *
2115  * In order to keep the orphan inode chain consistent during traversal (in
2116  * case of crash during recovery), we link each inode into the superblock
2117  * orphan list_head and handle it the same way as an inode deletion during
2118  * normal operation (which journals the operations for us).
2119  *
2120  * We only do an iget() and an iput() on each inode, which is very safe if we
2121  * accidentally point at an in-use or already deleted inode.  The worst that
2122  * can happen in this case is that we get a "bit already cleared" message from
2123  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2124  * e2fsck was run on this filesystem, and it must have already done the orphan
2125  * inode cleanup for us, so we can safely abort without any further action.
2126  */
2127 static void ext4_orphan_cleanup(struct super_block *sb,
2128                                 struct ext4_super_block *es)
2129 {
2130         unsigned int s_flags = sb->s_flags;
2131         int nr_orphans = 0, nr_truncates = 0;
2132 #ifdef CONFIG_QUOTA
2133         int i;
2134 #endif
2135         if (!es->s_last_orphan) {
2136                 jbd_debug(4, "no orphan inodes to clean up\n");
2137                 return;
2138         }
2139
2140         if (bdev_read_only(sb->s_bdev)) {
2141                 ext4_msg(sb, KERN_ERR, "write access "
2142                         "unavailable, skipping orphan cleanup");
2143                 return;
2144         }
2145
2146         /* Check if feature set would not allow a r/w mount */
2147         if (!ext4_feature_set_ok(sb, 0)) {
2148                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2149                          "unknown ROCOMPAT features");
2150                 return;
2151         }
2152
2153         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2154                 if (es->s_last_orphan)
2155                         jbd_debug(1, "Errors on filesystem, "
2156                                   "clearing orphan list.\n");
2157                 es->s_last_orphan = 0;
2158                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2159                 return;
2160         }
2161
2162         if (s_flags & MS_RDONLY) {
2163                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2164                 sb->s_flags &= ~MS_RDONLY;
2165         }
2166 #ifdef CONFIG_QUOTA
2167         /* Needed for iput() to work correctly and not trash data */
2168         sb->s_flags |= MS_ACTIVE;
2169         /* Turn on quotas so that they are updated correctly */
2170         for (i = 0; i < MAXQUOTAS; i++) {
2171                 if (EXT4_SB(sb)->s_qf_names[i]) {
2172                         int ret = ext4_quota_on_mount(sb, i);
2173                         if (ret < 0)
2174                                 ext4_msg(sb, KERN_ERR,
2175                                         "Cannot turn on journaled "
2176                                         "quota: error %d", ret);
2177                 }
2178         }
2179 #endif
2180
2181         while (es->s_last_orphan) {
2182                 struct inode *inode;
2183
2184                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2185                 if (IS_ERR(inode)) {
2186                         es->s_last_orphan = 0;
2187                         break;
2188                 }
2189
2190                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2191                 dquot_initialize(inode);
2192                 if (inode->i_nlink) {
2193                         ext4_msg(sb, KERN_DEBUG,
2194                                 "%s: truncating inode %lu to %lld bytes",
2195                                 __func__, inode->i_ino, inode->i_size);
2196                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2197                                   inode->i_ino, inode->i_size);
2198                         ext4_truncate(inode);
2199                         nr_truncates++;
2200                 } else {
2201                         ext4_msg(sb, KERN_DEBUG,
2202                                 "%s: deleting unreferenced inode %lu",
2203                                 __func__, inode->i_ino);
2204                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2205                                   inode->i_ino);
2206                         nr_orphans++;
2207                 }
2208                 iput(inode);  /* The delete magic happens here! */
2209         }
2210
2211 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2212
2213         if (nr_orphans)
2214                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2215                        PLURAL(nr_orphans));
2216         if (nr_truncates)
2217                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2218                        PLURAL(nr_truncates));
2219 #ifdef CONFIG_QUOTA
2220         /* Turn quotas off */
2221         for (i = 0; i < MAXQUOTAS; i++) {
2222                 if (sb_dqopt(sb)->files[i])
2223                         dquot_quota_off(sb, i);
2224         }
2225 #endif
2226         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2227 }
2228
2229 /*
2230  * Maximal extent format file size.
2231  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2232  * extent format containers, within a sector_t, and within i_blocks
2233  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2234  * so that won't be a limiting factor.
2235  *
2236  * However there is other limiting factor. We do store extents in the form
2237  * of starting block and length, hence the resulting length of the extent
2238  * covering maximum file size must fit into on-disk format containers as
2239  * well. Given that length is always by 1 unit bigger than max unit (because
2240  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2241  *
2242  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2243  */
2244 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2245 {
2246         loff_t res;
2247         loff_t upper_limit = MAX_LFS_FILESIZE;
2248
2249         /* small i_blocks in vfs inode? */
2250         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2251                 /*
2252                  * CONFIG_LBDAF is not enabled implies the inode
2253                  * i_block represent total blocks in 512 bytes
2254                  * 32 == size of vfs inode i_blocks * 8
2255                  */
2256                 upper_limit = (1LL << 32) - 1;
2257
2258                 /* total blocks in file system block size */
2259                 upper_limit >>= (blkbits - 9);
2260                 upper_limit <<= blkbits;
2261         }
2262
2263         /*
2264          * 32-bit extent-start container, ee_block. We lower the maxbytes
2265          * by one fs block, so ee_len can cover the extent of maximum file
2266          * size
2267          */
2268         res = (1LL << 32) - 1;
2269         res <<= blkbits;
2270
2271         /* Sanity check against vm- & vfs- imposed limits */
2272         if (res > upper_limit)
2273                 res = upper_limit;
2274
2275         return res;
2276 }
2277
2278 /*
2279  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2280  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2281  * We need to be 1 filesystem block less than the 2^48 sector limit.
2282  */
2283 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2284 {
2285         loff_t res = EXT4_NDIR_BLOCKS;
2286         int meta_blocks;
2287         loff_t upper_limit;
2288         /* This is calculated to be the largest file size for a dense, block
2289          * mapped file such that the file's total number of 512-byte sectors,
2290          * including data and all indirect blocks, does not exceed (2^48 - 1).
2291          *
2292          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2293          * number of 512-byte sectors of the file.
2294          */
2295
2296         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2297                 /*
2298                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2299                  * the inode i_block field represents total file blocks in
2300                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2301                  */
2302                 upper_limit = (1LL << 32) - 1;
2303
2304                 /* total blocks in file system block size */
2305                 upper_limit >>= (bits - 9);
2306
2307         } else {
2308                 /*
2309                  * We use 48 bit ext4_inode i_blocks
2310                  * With EXT4_HUGE_FILE_FL set the i_blocks
2311                  * represent total number of blocks in
2312                  * file system block size
2313                  */
2314                 upper_limit = (1LL << 48) - 1;
2315
2316         }
2317
2318         /* indirect blocks */
2319         meta_blocks = 1;
2320         /* double indirect blocks */
2321         meta_blocks += 1 + (1LL << (bits-2));
2322         /* tripple indirect blocks */
2323         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2324
2325         upper_limit -= meta_blocks;
2326         upper_limit <<= bits;
2327
2328         res += 1LL << (bits-2);
2329         res += 1LL << (2*(bits-2));
2330         res += 1LL << (3*(bits-2));
2331         res <<= bits;
2332         if (res > upper_limit)
2333                 res = upper_limit;
2334
2335         if (res > MAX_LFS_FILESIZE)
2336                 res = MAX_LFS_FILESIZE;
2337
2338         return res;
2339 }
2340
2341 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2342                                    ext4_fsblk_t logical_sb_block, int nr)
2343 {
2344         struct ext4_sb_info *sbi = EXT4_SB(sb);
2345         ext4_group_t bg, first_meta_bg;
2346         int has_super = 0;
2347
2348         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2349
2350         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2351             nr < first_meta_bg)
2352                 return logical_sb_block + nr + 1;
2353         bg = sbi->s_desc_per_block * nr;
2354         if (ext4_bg_has_super(sb, bg))
2355                 has_super = 1;
2356
2357         return (has_super + ext4_group_first_block_no(sb, bg));
2358 }
2359
2360 /**
2361  * ext4_get_stripe_size: Get the stripe size.
2362  * @sbi: In memory super block info
2363  *
2364  * If we have specified it via mount option, then
2365  * use the mount option value. If the value specified at mount time is
2366  * greater than the blocks per group use the super block value.
2367  * If the super block value is greater than blocks per group return 0.
2368  * Allocator needs it be less than blocks per group.
2369  *
2370  */
2371 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2372 {
2373         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2374         unsigned long stripe_width =
2375                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2376         int ret;
2377
2378         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2379                 ret = sbi->s_stripe;
2380         else if (stripe_width <= sbi->s_blocks_per_group)
2381                 ret = stripe_width;
2382         else if (stride <= sbi->s_blocks_per_group)
2383                 ret = stride;
2384         else
2385                 ret = 0;
2386
2387         /*
2388          * If the stripe width is 1, this makes no sense and
2389          * we set it to 0 to turn off stripe handling code.
2390          */
2391         if (ret <= 1)
2392                 ret = 0;
2393
2394         return ret;
2395 }
2396
2397 /* sysfs supprt */
2398
2399 struct ext4_attr {
2400         struct attribute attr;
2401         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2402         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2403                          const char *, size_t);
2404         int offset;
2405 };
2406
2407 static int parse_strtoul(const char *buf,
2408                 unsigned long max, unsigned long *value)
2409 {
2410         char *endp;
2411
2412         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2413         endp = skip_spaces(endp);
2414         if (*endp || *value > max)
2415                 return -EINVAL;
2416
2417         return 0;
2418 }
2419
2420 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2421                                               struct ext4_sb_info *sbi,
2422                                               char *buf)
2423 {
2424         return snprintf(buf, PAGE_SIZE, "%llu\n",
2425                 (s64) EXT4_C2B(sbi,
2426                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2427 }
2428
2429 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2430                                          struct ext4_sb_info *sbi, char *buf)
2431 {
2432         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2433
2434         if (!sb->s_bdev->bd_part)
2435                 return snprintf(buf, PAGE_SIZE, "0\n");
2436         return snprintf(buf, PAGE_SIZE, "%lu\n",
2437                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2438                          sbi->s_sectors_written_start) >> 1);
2439 }
2440
2441 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2442                                           struct ext4_sb_info *sbi, char *buf)
2443 {
2444         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2445
2446         if (!sb->s_bdev->bd_part)
2447                 return snprintf(buf, PAGE_SIZE, "0\n");
2448         return snprintf(buf, PAGE_SIZE, "%llu\n",
2449                         (unsigned long long)(sbi->s_kbytes_written +
2450                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2451                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2452 }
2453
2454 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2455                                           struct ext4_sb_info *sbi,
2456                                           const char *buf, size_t count)
2457 {
2458         unsigned long t;
2459
2460         if (parse_strtoul(buf, 0x40000000, &t))
2461                 return -EINVAL;
2462
2463         if (t && !is_power_of_2(t))
2464                 return -EINVAL;
2465
2466         sbi->s_inode_readahead_blks = t;
2467         return count;
2468 }
2469
2470 static ssize_t sbi_ui_show(struct ext4_attr *a,
2471                            struct ext4_sb_info *sbi, char *buf)
2472 {
2473         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2474
2475         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2476 }
2477
2478 static ssize_t sbi_ui_store(struct ext4_attr *a,
2479                             struct ext4_sb_info *sbi,
2480                             const char *buf, size_t count)
2481 {
2482         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2483         unsigned long t;
2484
2485         if (parse_strtoul(buf, 0xffffffff, &t))
2486                 return -EINVAL;
2487         *ui = t;
2488         return count;
2489 }
2490
2491 static ssize_t trigger_test_error(struct ext4_attr *a,
2492                                   struct ext4_sb_info *sbi,
2493                                   const char *buf, size_t count)
2494 {
2495         int len = count;
2496
2497         if (!capable(CAP_SYS_ADMIN))
2498                 return -EPERM;
2499
2500         if (len && buf[len-1] == '\n')
2501                 len--;
2502
2503         if (len)
2504                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2505         return count;
2506 }
2507
2508 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2509 static struct ext4_attr ext4_attr_##_name = {                   \
2510         .attr = {.name = __stringify(_name), .mode = _mode },   \
2511         .show   = _show,                                        \
2512         .store  = _store,                                       \
2513         .offset = offsetof(struct ext4_sb_info, _elname),       \
2514 }
2515 #define EXT4_ATTR(name, mode, show, store) \
2516 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2517
2518 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2519 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2520 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2521 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2522         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2523 #define ATTR_LIST(name) &ext4_attr_##name.attr
2524
2525 EXT4_RO_ATTR(delayed_allocation_blocks);
2526 EXT4_RO_ATTR(session_write_kbytes);
2527 EXT4_RO_ATTR(lifetime_write_kbytes);
2528 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2529                  inode_readahead_blks_store, s_inode_readahead_blks);
2530 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2531 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2532 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2533 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2534 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2535 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2536 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2537 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2538 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2539
2540 static struct attribute *ext4_attrs[] = {
2541         ATTR_LIST(delayed_allocation_blocks),
2542         ATTR_LIST(session_write_kbytes),
2543         ATTR_LIST(lifetime_write_kbytes),
2544         ATTR_LIST(inode_readahead_blks),
2545         ATTR_LIST(inode_goal),
2546         ATTR_LIST(mb_stats),
2547         ATTR_LIST(mb_max_to_scan),
2548         ATTR_LIST(mb_min_to_scan),
2549         ATTR_LIST(mb_order2_req),
2550         ATTR_LIST(mb_stream_req),
2551         ATTR_LIST(mb_group_prealloc),
2552         ATTR_LIST(max_writeback_mb_bump),
2553         ATTR_LIST(trigger_fs_error),
2554         NULL,
2555 };
2556
2557 /* Features this copy of ext4 supports */
2558 EXT4_INFO_ATTR(lazy_itable_init);
2559 EXT4_INFO_ATTR(batched_discard);
2560
2561 static struct attribute *ext4_feat_attrs[] = {
2562         ATTR_LIST(lazy_itable_init),
2563         ATTR_LIST(batched_discard),
2564         NULL,
2565 };
2566
2567 static ssize_t ext4_attr_show(struct kobject *kobj,
2568                               struct attribute *attr, char *buf)
2569 {
2570         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2571                                                 s_kobj);
2572         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2573
2574         return a->show ? a->show(a, sbi, buf) : 0;
2575 }
2576
2577 static ssize_t ext4_attr_store(struct kobject *kobj,
2578                                struct attribute *attr,
2579                                const char *buf, size_t len)
2580 {
2581         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2582                                                 s_kobj);
2583         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2584
2585         return a->store ? a->store(a, sbi, buf, len) : 0;
2586 }
2587
2588 static void ext4_sb_release(struct kobject *kobj)
2589 {
2590         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2591                                                 s_kobj);
2592         complete(&sbi->s_kobj_unregister);
2593 }
2594
2595 static const struct sysfs_ops ext4_attr_ops = {
2596         .show   = ext4_attr_show,
2597         .store  = ext4_attr_store,
2598 };
2599
2600 static struct kobj_type ext4_ktype = {
2601         .default_attrs  = ext4_attrs,
2602         .sysfs_ops      = &ext4_attr_ops,
2603         .release        = ext4_sb_release,
2604 };
2605
2606 static void ext4_feat_release(struct kobject *kobj)
2607 {
2608         complete(&ext4_feat->f_kobj_unregister);
2609 }
2610
2611 static struct kobj_type ext4_feat_ktype = {
2612         .default_attrs  = ext4_feat_attrs,
2613         .sysfs_ops      = &ext4_attr_ops,
2614         .release        = ext4_feat_release,
2615 };
2616
2617 /*
2618  * Check whether this filesystem can be mounted based on
2619  * the features present and the RDONLY/RDWR mount requested.
2620  * Returns 1 if this filesystem can be mounted as requested,
2621  * 0 if it cannot be.
2622  */
2623 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2624 {
2625         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2626                 ext4_msg(sb, KERN_ERR,
2627                         "Couldn't mount because of "
2628                         "unsupported optional features (%x)",
2629                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2630                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2631                 return 0;
2632         }
2633
2634         if (readonly)
2635                 return 1;
2636
2637         /* Check that feature set is OK for a read-write mount */
2638         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2639                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2640                          "unsupported optional features (%x)",
2641                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2642                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2643                 return 0;
2644         }
2645         /*
2646          * Large file size enabled file system can only be mounted
2647          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2648          */
2649         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2650                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2651                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2652                                  "cannot be mounted RDWR without "
2653                                  "CONFIG_LBDAF");
2654                         return 0;
2655                 }
2656         }
2657         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2658             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2659                 ext4_msg(sb, KERN_ERR,
2660                          "Can't support bigalloc feature without "
2661                          "extents feature\n");
2662                 return 0;
2663         }
2664         return 1;
2665 }
2666
2667 /*
2668  * This function is called once a day if we have errors logged
2669  * on the file system
2670  */
2671 static void print_daily_error_info(unsigned long arg)
2672 {
2673         struct super_block *sb = (struct super_block *) arg;
2674         struct ext4_sb_info *sbi;
2675         struct ext4_super_block *es;
2676
2677         sbi = EXT4_SB(sb);
2678         es = sbi->s_es;
2679
2680         if (es->s_error_count)
2681                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2682                          le32_to_cpu(es->s_error_count));
2683         if (es->s_first_error_time) {
2684                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2685                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2686                        (int) sizeof(es->s_first_error_func),
2687                        es->s_first_error_func,
2688                        le32_to_cpu(es->s_first_error_line));
2689                 if (es->s_first_error_ino)
2690                         printk(": inode %u",
2691                                le32_to_cpu(es->s_first_error_ino));
2692                 if (es->s_first_error_block)
2693                         printk(": block %llu", (unsigned long long)
2694                                le64_to_cpu(es->s_first_error_block));
2695                 printk("\n");
2696         }
2697         if (es->s_last_error_time) {
2698                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2699                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2700                        (int) sizeof(es->s_last_error_func),
2701                        es->s_last_error_func,
2702                        le32_to_cpu(es->s_last_error_line));
2703                 if (es->s_last_error_ino)
2704                         printk(": inode %u",
2705                                le32_to_cpu(es->s_last_error_ino));
2706                 if (es->s_last_error_block)
2707                         printk(": block %llu", (unsigned long long)
2708                                le64_to_cpu(es->s_last_error_block));
2709                 printk("\n");
2710         }
2711         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2712 }
2713
2714 /* Find next suitable group and run ext4_init_inode_table */
2715 static int ext4_run_li_request(struct ext4_li_request *elr)
2716 {
2717         struct ext4_group_desc *gdp = NULL;
2718         ext4_group_t group, ngroups;
2719         struct super_block *sb;
2720         unsigned long timeout = 0;
2721         int ret = 0;
2722
2723         sb = elr->lr_super;
2724         ngroups = EXT4_SB(sb)->s_groups_count;
2725
2726         for (group = elr->lr_next_group; group < ngroups; group++) {
2727                 gdp = ext4_get_group_desc(sb, group, NULL);
2728                 if (!gdp) {
2729                         ret = 1;
2730                         break;
2731                 }
2732
2733                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2734                         break;
2735         }
2736
2737         if (group == ngroups)
2738                 ret = 1;
2739
2740         if (!ret) {
2741                 timeout = jiffies;
2742                 ret = ext4_init_inode_table(sb, group,
2743                                             elr->lr_timeout ? 0 : 1);
2744                 if (elr->lr_timeout == 0) {
2745                         timeout = (jiffies - timeout) *
2746                                   elr->lr_sbi->s_li_wait_mult;
2747                         elr->lr_timeout = timeout;
2748                 }
2749                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2750                 elr->lr_next_group = group + 1;
2751         }
2752
2753         return ret;
2754 }
2755
2756 /*
2757  * Remove lr_request from the list_request and free the
2758  * request structure. Should be called with li_list_mtx held
2759  */
2760 static void ext4_remove_li_request(struct ext4_li_request *elr)
2761 {
2762         struct ext4_sb_info *sbi;
2763
2764         if (!elr)
2765                 return;
2766
2767         sbi = elr->lr_sbi;
2768
2769         list_del(&elr->lr_request);
2770         sbi->s_li_request = NULL;
2771         kfree(elr);
2772 }
2773
2774 static void ext4_unregister_li_request(struct super_block *sb)
2775 {
2776         mutex_lock(&ext4_li_mtx);
2777         if (!ext4_li_info) {
2778                 mutex_unlock(&ext4_li_mtx);
2779                 return;
2780         }
2781
2782         mutex_lock(&ext4_li_info->li_list_mtx);
2783         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2784         mutex_unlock(&ext4_li_info->li_list_mtx);
2785         mutex_unlock(&ext4_li_mtx);
2786 }
2787
2788 static struct task_struct *ext4_lazyinit_task;
2789
2790 /*
2791  * This is the function where ext4lazyinit thread lives. It walks
2792  * through the request list searching for next scheduled filesystem.
2793  * When such a fs is found, run the lazy initialization request
2794  * (ext4_rn_li_request) and keep track of the time spend in this
2795  * function. Based on that time we compute next schedule time of
2796  * the request. When walking through the list is complete, compute
2797  * next waking time and put itself into sleep.
2798  */
2799 static int ext4_lazyinit_thread(void *arg)
2800 {
2801         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2802         struct list_head *pos, *n;
2803         struct ext4_li_request *elr;
2804         unsigned long next_wakeup, cur;
2805
2806         BUG_ON(NULL == eli);
2807
2808 cont_thread:
2809         while (true) {
2810                 next_wakeup = MAX_JIFFY_OFFSET;
2811
2812                 mutex_lock(&eli->li_list_mtx);
2813                 if (list_empty(&eli->li_request_list)) {
2814                         mutex_unlock(&eli->li_list_mtx);
2815                         goto exit_thread;
2816                 }
2817
2818                 list_for_each_safe(pos, n, &eli->li_request_list) {
2819                         elr = list_entry(pos, struct ext4_li_request,
2820                                          lr_request);
2821
2822                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2823                                 if (ext4_run_li_request(elr) != 0) {
2824                                         /* error, remove the lazy_init job */
2825                                         ext4_remove_li_request(elr);
2826                                         continue;
2827                                 }
2828                         }
2829
2830                         if (time_before(elr->lr_next_sched, next_wakeup))
2831                                 next_wakeup = elr->lr_next_sched;
2832                 }
2833                 mutex_unlock(&eli->li_list_mtx);
2834
2835                 try_to_freeze();
2836
2837                 cur = jiffies;
2838                 if ((time_after_eq(cur, next_wakeup)) ||
2839                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2840                         cond_resched();
2841                         continue;
2842                 }
2843
2844                 schedule_timeout_interruptible(next_wakeup - cur);
2845
2846                 if (kthread_should_stop()) {
2847                         ext4_clear_request_list();
2848                         goto exit_thread;
2849                 }
2850         }
2851
2852 exit_thread:
2853         /*
2854          * It looks like the request list is empty, but we need
2855          * to check it under the li_list_mtx lock, to prevent any
2856          * additions into it, and of course we should lock ext4_li_mtx
2857          * to atomically free the list and ext4_li_info, because at
2858          * this point another ext4 filesystem could be registering
2859          * new one.
2860          */
2861         mutex_lock(&ext4_li_mtx);
2862         mutex_lock(&eli->li_list_mtx);
2863         if (!list_empty(&eli->li_request_list)) {
2864                 mutex_unlock(&eli->li_list_mtx);
2865                 mutex_unlock(&ext4_li_mtx);
2866                 goto cont_thread;
2867         }
2868         mutex_unlock(&eli->li_list_mtx);
2869         kfree(ext4_li_info);
2870         ext4_li_info = NULL;
2871         mutex_unlock(&ext4_li_mtx);
2872
2873         return 0;
2874 }
2875
2876 static void ext4_clear_request_list(void)
2877 {
2878         struct list_head *pos, *n;
2879         struct ext4_li_request *elr;
2880
2881         mutex_lock(&ext4_li_info->li_list_mtx);
2882         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2883                 elr = list_entry(pos, struct ext4_li_request,
2884                                  lr_request);
2885                 ext4_remove_li_request(elr);
2886         }
2887         mutex_unlock(&ext4_li_info->li_list_mtx);
2888 }
2889
2890 static int ext4_run_lazyinit_thread(void)
2891 {
2892         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2893                                          ext4_li_info, "ext4lazyinit");
2894         if (IS_ERR(ext4_lazyinit_task)) {
2895                 int err = PTR_ERR(ext4_lazyinit_task);
2896                 ext4_clear_request_list();
2897                 kfree(ext4_li_info);
2898                 ext4_li_info = NULL;
2899                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2900                                  "initialization thread\n",
2901                                  err);
2902                 return err;
2903         }
2904         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2905         return 0;
2906 }
2907
2908 /*
2909  * Check whether it make sense to run itable init. thread or not.
2910  * If there is at least one uninitialized inode table, return
2911  * corresponding group number, else the loop goes through all
2912  * groups and return total number of groups.
2913  */
2914 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2915 {
2916         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2917         struct ext4_group_desc *gdp = NULL;
2918
2919         for (group = 0; group < ngroups; group++) {
2920                 gdp = ext4_get_group_desc(sb, group, NULL);
2921                 if (!gdp)
2922                         continue;
2923
2924                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2925                         break;
2926         }
2927
2928         return group;
2929 }
2930
2931 static int ext4_li_info_new(void)
2932 {
2933         struct ext4_lazy_init *eli = NULL;
2934
2935         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2936         if (!eli)
2937                 return -ENOMEM;
2938
2939         INIT_LIST_HEAD(&eli->li_request_list);
2940         mutex_init(&eli->li_list_mtx);
2941
2942         eli->li_state |= EXT4_LAZYINIT_QUIT;
2943
2944         ext4_li_info = eli;
2945
2946         return 0;
2947 }
2948
2949 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2950                                             ext4_group_t start)
2951 {
2952         struct ext4_sb_info *sbi = EXT4_SB(sb);
2953         struct ext4_li_request *elr;
2954         unsigned long rnd;
2955
2956         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2957         if (!elr)
2958                 return NULL;
2959
2960         elr->lr_super = sb;
2961         elr->lr_sbi = sbi;
2962         elr->lr_next_group = start;
2963
2964         /*
2965          * Randomize first schedule time of the request to
2966          * spread the inode table initialization requests
2967          * better.
2968          */
2969         get_random_bytes(&rnd, sizeof(rnd));
2970         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2971                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2972
2973         return elr;
2974 }
2975
2976 static int ext4_register_li_request(struct super_block *sb,
2977                                     ext4_group_t first_not_zeroed)
2978 {
2979         struct ext4_sb_info *sbi = EXT4_SB(sb);
2980         struct ext4_li_request *elr;
2981         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2982         int ret = 0;
2983
2984         if (sbi->s_li_request != NULL) {
2985                 /*
2986                  * Reset timeout so it can be computed again, because
2987                  * s_li_wait_mult might have changed.
2988                  */
2989                 sbi->s_li_request->lr_timeout = 0;
2990                 return 0;
2991         }
2992
2993         if (first_not_zeroed == ngroups ||
2994             (sb->s_flags & MS_RDONLY) ||
2995             !test_opt(sb, INIT_INODE_TABLE))
2996                 return 0;
2997
2998         elr = ext4_li_request_new(sb, first_not_zeroed);
2999         if (!elr)
3000                 return -ENOMEM;
3001
3002         mutex_lock(&ext4_li_mtx);
3003
3004         if (NULL == ext4_li_info) {
3005                 ret = ext4_li_info_new();
3006                 if (ret)
3007                         goto out;
3008         }
3009
3010         mutex_lock(&ext4_li_info->li_list_mtx);
3011         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3012         mutex_unlock(&ext4_li_info->li_list_mtx);
3013
3014         sbi->s_li_request = elr;
3015         /*
3016          * set elr to NULL here since it has been inserted to
3017          * the request_list and the removal and free of it is
3018          * handled by ext4_clear_request_list from now on.
3019          */
3020         elr = NULL;
3021
3022         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3023                 ret = ext4_run_lazyinit_thread();
3024                 if (ret)
3025                         goto out;
3026         }
3027 out:
3028         mutex_unlock(&ext4_li_mtx);
3029         if (ret)
3030                 kfree(elr);
3031         return ret;
3032 }
3033
3034 /*
3035  * We do not need to lock anything since this is called on
3036  * module unload.
3037  */
3038 static void ext4_destroy_lazyinit_thread(void)
3039 {
3040         /*
3041          * If thread exited earlier
3042          * there's nothing to be done.
3043          */
3044         if (!ext4_li_info || !ext4_lazyinit_task)
3045                 return;
3046
3047         kthread_stop(ext4_lazyinit_task);
3048 }
3049
3050 static int set_journal_csum_feature_set(struct super_block *sb)
3051 {
3052         int ret = 1;
3053         int compat, incompat;
3054         struct ext4_sb_info *sbi = EXT4_SB(sb);
3055
3056         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3057                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3058                 /* journal checksum v2 */
3059                 compat = 0;
3060                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3061         } else {
3062                 /* journal checksum v1 */
3063                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3064                 incompat = 0;
3065         }
3066
3067         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3068                 ret = jbd2_journal_set_features(sbi->s_journal,
3069                                 compat, 0,
3070                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3071                                 incompat);
3072         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3073                 ret = jbd2_journal_set_features(sbi->s_journal,
3074                                 compat, 0,
3075                                 incompat);
3076                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3077                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3078         } else {
3079                 jbd2_journal_clear_features(sbi->s_journal,
3080                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3081                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3082                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3083         }
3084
3085         return ret;
3086 }
3087
3088 /*
3089  * Note: calculating the overhead so we can be compatible with
3090  * historical BSD practice is quite difficult in the face of
3091  * clusters/bigalloc.  This is because multiple metadata blocks from
3092  * different block group can end up in the same allocation cluster.
3093  * Calculating the exact overhead in the face of clustered allocation
3094  * requires either O(all block bitmaps) in memory or O(number of block
3095  * groups**2) in time.  We will still calculate the superblock for
3096  * older file systems --- and if we come across with a bigalloc file
3097  * system with zero in s_overhead_clusters the estimate will be close to
3098  * correct especially for very large cluster sizes --- but for newer
3099  * file systems, it's better to calculate this figure once at mkfs
3100  * time, and store it in the superblock.  If the superblock value is
3101  * present (even for non-bigalloc file systems), we will use it.
3102  */
3103 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3104                           char *buf)
3105 {
3106         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3107         struct ext4_group_desc  *gdp;
3108         ext4_fsblk_t            first_block, last_block, b;
3109         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3110         int                     s, j, count = 0;
3111
3112         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3113                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3114         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3115         for (i = 0; i < ngroups; i++) {
3116                 gdp = ext4_get_group_desc(sb, i, NULL);
3117                 b = ext4_block_bitmap(sb, gdp);
3118                 if (b >= first_block && b <= last_block) {
3119                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3120                         count++;
3121                 }
3122                 b = ext4_inode_bitmap(sb, gdp);
3123                 if (b >= first_block && b <= last_block) {
3124                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3125                         count++;
3126                 }
3127                 b = ext4_inode_table(sb, gdp);
3128                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3129                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3130                                 int c = EXT4_B2C(sbi, b - first_block);
3131                                 ext4_set_bit(c, buf);
3132                                 count++;
3133                         }
3134                 if (i != grp)
3135                         continue;
3136                 s = 0;
3137                 if (ext4_bg_has_super(sb, grp)) {
3138                         ext4_set_bit(s++, buf);
3139                         count++;
3140                 }
3141                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3142                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3143                         count++;
3144                 }
3145         }
3146         if (!count)
3147                 return 0;
3148         return EXT4_CLUSTERS_PER_GROUP(sb) -
3149                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3150 }
3151
3152 /*
3153  * Compute the overhead and stash it in sbi->s_overhead
3154  */
3155 int ext4_calculate_overhead(struct super_block *sb)
3156 {
3157         struct ext4_sb_info *sbi = EXT4_SB(sb);
3158         struct ext4_super_block *es = sbi->s_es;
3159         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3160         ext4_fsblk_t overhead = 0;
3161         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3162
3163         memset(buf, 0, PAGE_SIZE);
3164         if (!buf)
3165                 return -ENOMEM;
3166
3167         /*
3168          * Compute the overhead (FS structures).  This is constant
3169          * for a given filesystem unless the number of block groups
3170          * changes so we cache the previous value until it does.
3171          */
3172
3173         /*
3174          * All of the blocks before first_data_block are overhead
3175          */
3176         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3177
3178         /*
3179          * Add the overhead found in each block group
3180          */
3181         for (i = 0; i < ngroups; i++) {
3182                 int blks;
3183
3184                 blks = count_overhead(sb, i, buf);
3185                 overhead += blks;
3186                 if (blks)
3187                         memset(buf, 0, PAGE_SIZE);
3188                 cond_resched();
3189         }
3190         sbi->s_overhead = overhead;
3191         smp_wmb();
3192         free_page((unsigned long) buf);
3193         return 0;
3194 }
3195
3196 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3197 {
3198         char *orig_data = kstrdup(data, GFP_KERNEL);
3199         struct buffer_head *bh;
3200         struct ext4_super_block *es = NULL;
3201         struct ext4_sb_info *sbi;
3202         ext4_fsblk_t block;
3203         ext4_fsblk_t sb_block = get_sb_block(&data);
3204         ext4_fsblk_t logical_sb_block;
3205         unsigned long offset = 0;
3206         unsigned long journal_devnum = 0;
3207         unsigned long def_mount_opts;
3208         struct inode *root;
3209         char *cp;
3210         const char *descr;
3211         int ret = -ENOMEM;
3212         int blocksize, clustersize;
3213         unsigned int db_count;
3214         unsigned int i;
3215         int needs_recovery, has_huge_files, has_bigalloc;
3216         __u64 blocks_count;
3217         int err;
3218         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3219         ext4_group_t first_not_zeroed;
3220
3221         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3222         if (!sbi)
3223                 goto out_free_orig;
3224
3225         sbi->s_blockgroup_lock =
3226                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3227         if (!sbi->s_blockgroup_lock) {
3228                 kfree(sbi);
3229                 goto out_free_orig;
3230         }
3231         sb->s_fs_info = sbi;
3232         sbi->s_sb = sb;
3233         sbi->s_mount_opt = 0;
3234         sbi->s_resuid = make_kuid(&init_user_ns, EXT4_DEF_RESUID);
3235         sbi->s_resgid = make_kgid(&init_user_ns, EXT4_DEF_RESGID);
3236         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3237         sbi->s_sb_block = sb_block;
3238         if (sb->s_bdev->bd_part)
3239                 sbi->s_sectors_written_start =
3240                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3241
3242         /* Cleanup superblock name */
3243         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3244                 *cp = '!';
3245
3246         ret = -EINVAL;
3247         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3248         if (!blocksize) {
3249                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3250                 goto out_fail;
3251         }
3252
3253         /*
3254          * The ext4 superblock will not be buffer aligned for other than 1kB
3255          * block sizes.  We need to calculate the offset from buffer start.
3256          */
3257         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3258                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3259                 offset = do_div(logical_sb_block, blocksize);
3260         } else {
3261                 logical_sb_block = sb_block;
3262         }
3263
3264         if (!(bh = sb_bread(sb, logical_sb_block))) {
3265                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3266                 goto out_fail;
3267         }
3268         /*
3269          * Note: s_es must be initialized as soon as possible because
3270          *       some ext4 macro-instructions depend on its value
3271          */
3272         es = (struct ext4_super_block *) (bh->b_data + offset);
3273         sbi->s_es = es;
3274         sb->s_magic = le16_to_cpu(es->s_magic);
3275         if (sb->s_magic != EXT4_SUPER_MAGIC)
3276                 goto cantfind_ext4;
3277         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3278
3279         /* Warn if metadata_csum and gdt_csum are both set. */
3280         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3281                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3282             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3283                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3284                              "redundant flags; please run fsck.");
3285
3286         /* Check for a known checksum algorithm */
3287         if (!ext4_verify_csum_type(sb, es)) {
3288                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3289                          "unknown checksum algorithm.");
3290                 silent = 1;
3291                 goto cantfind_ext4;
3292         }
3293
3294         /* Load the checksum driver */
3295         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3296                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3297                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3298                 if (IS_ERR(sbi->s_chksum_driver)) {
3299                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3300                         ret = PTR_ERR(sbi->s_chksum_driver);
3301                         sbi->s_chksum_driver = NULL;
3302                         goto failed_mount;
3303                 }
3304         }
3305
3306         /* Check superblock checksum */
3307         if (!ext4_superblock_csum_verify(sb, es)) {
3308                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3309                          "invalid superblock checksum.  Run e2fsck?");
3310                 silent = 1;
3311                 goto cantfind_ext4;
3312         }
3313
3314         /* Precompute checksum seed for all metadata */
3315         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3316                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3317                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3318                                                sizeof(es->s_uuid));
3319
3320         /* Set defaults before we parse the mount options */
3321         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3322         set_opt(sb, INIT_INODE_TABLE);
3323         if (def_mount_opts & EXT4_DEFM_DEBUG)
3324                 set_opt(sb, DEBUG);
3325         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3326                 set_opt(sb, GRPID);
3327         if (def_mount_opts & EXT4_DEFM_UID16)
3328                 set_opt(sb, NO_UID32);
3329         /* xattr user namespace & acls are now defaulted on */
3330 #ifdef CONFIG_EXT4_FS_XATTR
3331         set_opt(sb, XATTR_USER);
3332 #endif
3333 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3334         set_opt(sb, POSIX_ACL);
3335 #endif
3336         set_opt(sb, MBLK_IO_SUBMIT);
3337         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3338                 set_opt(sb, JOURNAL_DATA);
3339         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3340                 set_opt(sb, ORDERED_DATA);
3341         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3342                 set_opt(sb, WRITEBACK_DATA);
3343
3344         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3345                 set_opt(sb, ERRORS_PANIC);
3346         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3347                 set_opt(sb, ERRORS_CONT);
3348         else
3349                 set_opt(sb, ERRORS_RO);
3350         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3351                 set_opt(sb, BLOCK_VALIDITY);
3352         if (def_mount_opts & EXT4_DEFM_DISCARD)
3353                 set_opt(sb, DISCARD);
3354
3355         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3356         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3357         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3358         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3359         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3360
3361         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3362                 set_opt(sb, BARRIER);
3363
3364         /*
3365          * enable delayed allocation by default
3366          * Use -o nodelalloc to turn it off
3367          */
3368         if (!IS_EXT3_SB(sb) &&
3369             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3370                 set_opt(sb, DELALLOC);
3371
3372         /*
3373          * set default s_li_wait_mult for lazyinit, for the case there is
3374          * no mount option specified.
3375          */
3376         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3377
3378         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3379                            &journal_devnum, &journal_ioprio, 0)) {
3380                 ext4_msg(sb, KERN_WARNING,
3381                          "failed to parse options in superblock: %s",
3382                          sbi->s_es->s_mount_opts);
3383         }
3384         sbi->s_def_mount_opt = sbi->s_mount_opt;
3385         if (!parse_options((char *) data, sb, &journal_devnum,
3386                            &journal_ioprio, 0))
3387                 goto failed_mount;
3388
3389         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3390                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3391                             "with data=journal disables delayed "
3392                             "allocation and O_DIRECT support!\n");
3393                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3394                         ext4_msg(sb, KERN_ERR, "can't mount with "
3395                                  "both data=journal and delalloc");
3396                         goto failed_mount;
3397                 }
3398                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3399                         ext4_msg(sb, KERN_ERR, "can't mount with "
3400                                  "both data=journal and delalloc");
3401                         goto failed_mount;
3402                 }
3403                 if (test_opt(sb, DELALLOC))
3404                         clear_opt(sb, DELALLOC);
3405         }
3406
3407         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3408         if (test_opt(sb, DIOREAD_NOLOCK)) {
3409                 if (blocksize < PAGE_SIZE) {
3410                         ext4_msg(sb, KERN_ERR, "can't mount with "
3411                                  "dioread_nolock if block size != PAGE_SIZE");
3412                         goto failed_mount;
3413                 }
3414         }
3415
3416         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3417                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3418
3419         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3420             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3421              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3422              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3423                 ext4_msg(sb, KERN_WARNING,
3424                        "feature flags set on rev 0 fs, "
3425                        "running e2fsck is recommended");
3426
3427         if (IS_EXT2_SB(sb)) {
3428                 if (ext2_feature_set_ok(sb))
3429                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3430                                  "using the ext4 subsystem");
3431                 else {
3432                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3433                                  "to feature incompatibilities");
3434                         goto failed_mount;
3435                 }
3436         }
3437
3438         if (IS_EXT3_SB(sb)) {
3439                 if (ext3_feature_set_ok(sb))
3440                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3441                                  "using the ext4 subsystem");
3442                 else {
3443                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3444                                  "to feature incompatibilities");
3445                         goto failed_mount;
3446                 }
3447         }
3448
3449         /*
3450          * Check feature flags regardless of the revision level, since we
3451          * previously didn't change the revision level when setting the flags,
3452          * so there is a chance incompat flags are set on a rev 0 filesystem.
3453          */
3454         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3455                 goto failed_mount;
3456
3457         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3458             blocksize > EXT4_MAX_BLOCK_SIZE) {
3459                 ext4_msg(sb, KERN_ERR,
3460                        "Unsupported filesystem blocksize %d", blocksize);
3461                 goto failed_mount;
3462         }
3463
3464         if (sb->s_blocksize != blocksize) {
3465                 /* Validate the filesystem blocksize */
3466                 if (!sb_set_blocksize(sb, blocksize)) {
3467                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3468                                         blocksize);
3469                         goto failed_mount;
3470                 }
3471
3472                 brelse(bh);
3473                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3474                 offset = do_div(logical_sb_block, blocksize);
3475                 bh = sb_bread(sb, logical_sb_block);
3476                 if (!bh) {
3477                         ext4_msg(sb, KERN_ERR,
3478                                "Can't read superblock on 2nd try");
3479                         goto failed_mount;
3480                 }
3481                 es = (struct ext4_super_block *)(bh->b_data + offset);
3482                 sbi->s_es = es;
3483                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3484                         ext4_msg(sb, KERN_ERR,
3485                                "Magic mismatch, very weird!");
3486                         goto failed_mount;
3487                 }
3488         }
3489
3490         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3491                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3492         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3493                                                       has_huge_files);
3494         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3495
3496         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3497                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3498                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3499         } else {
3500                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3501                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3502                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3503                     (!is_power_of_2(sbi->s_inode_size)) ||
3504                     (sbi->s_inode_size > blocksize)) {
3505                         ext4_msg(sb, KERN_ERR,
3506                                "unsupported inode size: %d",
3507                                sbi->s_inode_size);
3508                         goto failed_mount;
3509                 }
3510                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3511                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3512         }
3513
3514         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3515         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3516                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3517                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3518                     !is_power_of_2(sbi->s_desc_size)) {
3519                         ext4_msg(sb, KERN_ERR,
3520                                "unsupported descriptor size %lu",
3521                                sbi->s_desc_size);
3522                         goto failed_mount;
3523                 }
3524         } else
3525                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3526
3527         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3528         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3529         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3530                 goto cantfind_ext4;
3531
3532         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3533         if (sbi->s_inodes_per_block == 0)
3534                 goto cantfind_ext4;
3535         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3536                                         sbi->s_inodes_per_block;
3537         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3538         sbi->s_sbh = bh;
3539         sbi->s_mount_state = le16_to_cpu(es->s_state);
3540         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3541         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3542
3543         for (i = 0; i < 4; i++)
3544                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3545         sbi->s_def_hash_version = es->s_def_hash_version;
3546         i = le32_to_cpu(es->s_flags);
3547         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3548                 sbi->s_hash_unsigned = 3;
3549         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3550 #ifdef __CHAR_UNSIGNED__
3551                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3552                 sbi->s_hash_unsigned = 3;
3553 #else
3554                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3555 #endif
3556         }
3557
3558         /* Handle clustersize */
3559         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3560         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3561                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3562         if (has_bigalloc) {
3563                 if (clustersize < blocksize) {
3564                         ext4_msg(sb, KERN_ERR,
3565                                  "cluster size (%d) smaller than "
3566                                  "block size (%d)", clustersize, blocksize);
3567                         goto failed_mount;
3568                 }
3569                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3570                         le32_to_cpu(es->s_log_block_size);
3571                 sbi->s_clusters_per_group =
3572                         le32_to_cpu(es->s_clusters_per_group);
3573                 if (sbi->s_clusters_per_group > blocksize * 8) {
3574                         ext4_msg(sb, KERN_ERR,
3575                                  "#clusters per group too big: %lu",
3576                                  sbi->s_clusters_per_group);
3577                         goto failed_mount;
3578                 }
3579                 if (sbi->s_blocks_per_group !=
3580                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3581                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3582                                  "clusters per group (%lu) inconsistent",
3583                                  sbi->s_blocks_per_group,
3584                                  sbi->s_clusters_per_group);
3585                         goto failed_mount;
3586                 }
3587         } else {
3588                 if (clustersize != blocksize) {
3589                         ext4_warning(sb, "fragment/cluster size (%d) != "
3590                                      "block size (%d)", clustersize,
3591                                      blocksize);
3592                         clustersize = blocksize;
3593                 }
3594                 if (sbi->s_blocks_per_group > blocksize * 8) {
3595                         ext4_msg(sb, KERN_ERR,
3596                                  "#blocks per group too big: %lu",
3597                                  sbi->s_blocks_per_group);
3598                         goto failed_mount;
3599                 }
3600                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3601                 sbi->s_cluster_bits = 0;
3602         }
3603         sbi->s_cluster_ratio = clustersize / blocksize;
3604
3605         if (sbi->s_inodes_per_group > blocksize * 8) {
3606                 ext4_msg(sb, KERN_ERR,
3607                        "#inodes per group too big: %lu",
3608                        sbi->s_inodes_per_group);
3609                 goto failed_mount;
3610         }
3611
3612         /*
3613          * Test whether we have more sectors than will fit in sector_t,
3614          * and whether the max offset is addressable by the page cache.
3615          */
3616         err = generic_check_addressable(sb->s_blocksize_bits,
3617                                         ext4_blocks_count(es));
3618         if (err) {
3619                 ext4_msg(sb, KERN_ERR, "filesystem"
3620                          " too large to mount safely on this system");
3621                 if (sizeof(sector_t) < 8)
3622                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3623                 ret = err;
3624                 goto failed_mount;
3625         }
3626
3627         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3628                 goto cantfind_ext4;
3629
3630         /* check blocks count against device size */
3631         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3632         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3633                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3634                        "exceeds size of device (%llu blocks)",
3635                        ext4_blocks_count(es), blocks_count);
3636                 goto failed_mount;
3637         }
3638
3639         /*
3640          * It makes no sense for the first data block to be beyond the end
3641          * of the filesystem.
3642          */
3643         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3644                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3645                          "block %u is beyond end of filesystem (%llu)",
3646                          le32_to_cpu(es->s_first_data_block),
3647                          ext4_blocks_count(es));
3648                 goto failed_mount;
3649         }
3650         blocks_count = (ext4_blocks_count(es) -
3651                         le32_to_cpu(es->s_first_data_block) +
3652                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3653         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3654         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3655                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3656                        "(block count %llu, first data block %u, "
3657                        "blocks per group %lu)", sbi->s_groups_count,
3658                        ext4_blocks_count(es),
3659                        le32_to_cpu(es->s_first_data_block),
3660                        EXT4_BLOCKS_PER_GROUP(sb));
3661                 goto failed_mount;
3662         }
3663         sbi->s_groups_count = blocks_count;
3664         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3665                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3666         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3667                    EXT4_DESC_PER_BLOCK(sb);
3668         sbi->s_group_desc = ext4_kvmalloc(db_count *
3669                                           sizeof(struct buffer_head *),
3670                                           GFP_KERNEL);
3671         if (sbi->s_group_desc == NULL) {
3672                 ext4_msg(sb, KERN_ERR, "not enough memory");
3673                 ret = -ENOMEM;
3674                 goto failed_mount;
3675         }
3676
3677         if (ext4_proc_root)
3678                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3679
3680         if (sbi->s_proc)
3681                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3682                                  &ext4_seq_options_fops, sb);
3683
3684         bgl_lock_init(sbi->s_blockgroup_lock);
3685
3686         for (i = 0; i < db_count; i++) {
3687                 block = descriptor_loc(sb, logical_sb_block, i);
3688                 sbi->s_group_desc[i] = sb_bread(sb, block);
3689                 if (!sbi->s_group_desc[i]) {
3690                         ext4_msg(sb, KERN_ERR,
3691                                "can't read group descriptor %d", i);
3692                         db_count = i;
3693                         goto failed_mount2;
3694                 }
3695         }
3696         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3697                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3698                 goto failed_mount2;
3699         }
3700         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3701                 if (!ext4_fill_flex_info(sb)) {
3702                         ext4_msg(sb, KERN_ERR,
3703                                "unable to initialize "
3704                                "flex_bg meta info!");
3705                         goto failed_mount2;
3706                 }
3707
3708         sbi->s_gdb_count = db_count;
3709         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3710         spin_lock_init(&sbi->s_next_gen_lock);
3711
3712         init_timer(&sbi->s_err_report);
3713         sbi->s_err_report.function = print_daily_error_info;
3714         sbi->s_err_report.data = (unsigned long) sb;
3715
3716         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3717                         ext4_count_free_clusters(sb));
3718         if (!err) {
3719                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3720                                 ext4_count_free_inodes(sb));
3721         }
3722         if (!err) {
3723                 err = percpu_counter_init(&sbi->s_dirs_counter,
3724                                 ext4_count_dirs(sb));
3725         }
3726         if (!err) {
3727                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3728         }
3729         if (err) {
3730                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3731                 ret = err;
3732                 goto failed_mount3;
3733         }
3734
3735         sbi->s_stripe = ext4_get_stripe_size(sbi);
3736         sbi->s_max_writeback_mb_bump = 128;
3737
3738         /*
3739          * set up enough so that it can read an inode
3740          */
3741         if (!test_opt(sb, NOLOAD) &&
3742             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3743                 sb->s_op = &ext4_sops;
3744         else
3745                 sb->s_op = &ext4_nojournal_sops;
3746         sb->s_export_op = &ext4_export_ops;
3747         sb->s_xattr = ext4_xattr_handlers;
3748 #ifdef CONFIG_QUOTA
3749         sb->s_qcop = &ext4_qctl_operations;
3750         sb->dq_op = &ext4_quota_operations;
3751 #endif
3752         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3753
3754         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3755         mutex_init(&sbi->s_orphan_lock);
3756         sbi->s_resize_flags = 0;
3757
3758         sb->s_root = NULL;
3759
3760         needs_recovery = (es->s_last_orphan != 0 ||
3761                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3762                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3763
3764         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3765             !(sb->s_flags & MS_RDONLY))
3766                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3767                         goto failed_mount3;
3768
3769         /*
3770          * The first inode we look at is the journal inode.  Don't try
3771          * root first: it may be modified in the journal!
3772          */
3773         if (!test_opt(sb, NOLOAD) &&
3774             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3775                 if (ext4_load_journal(sb, es, journal_devnum))
3776                         goto failed_mount3;
3777         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3778               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3779                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3780                        "suppressed and not mounted read-only");
3781                 goto failed_mount_wq;
3782         } else {
3783                 clear_opt(sb, DATA_FLAGS);
3784                 sbi->s_journal = NULL;
3785                 needs_recovery = 0;
3786                 goto no_journal;
3787         }
3788
3789         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3790             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3791                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3792                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3793                 goto failed_mount_wq;
3794         }
3795
3796         if (!set_journal_csum_feature_set(sb)) {
3797                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3798                          "feature set");
3799                 goto failed_mount_wq;
3800         }
3801
3802         /* We have now updated the journal if required, so we can
3803          * validate the data journaling mode. */
3804         switch (test_opt(sb, DATA_FLAGS)) {
3805         case 0:
3806                 /* No mode set, assume a default based on the journal
3807                  * capabilities: ORDERED_DATA if the journal can
3808                  * cope, else JOURNAL_DATA
3809                  */
3810                 if (jbd2_journal_check_available_features
3811                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3812                         set_opt(sb, ORDERED_DATA);
3813                 else
3814                         set_opt(sb, JOURNAL_DATA);
3815                 break;
3816
3817         case EXT4_MOUNT_ORDERED_DATA:
3818         case EXT4_MOUNT_WRITEBACK_DATA:
3819                 if (!jbd2_journal_check_available_features
3820                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3821                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3822                                "requested data journaling mode");
3823                         goto failed_mount_wq;
3824                 }
3825         default:
3826                 break;
3827         }
3828         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3829
3830         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3831
3832         /*
3833          * The journal may have updated the bg summary counts, so we
3834          * need to update the global counters.
3835          */
3836         percpu_counter_set(&sbi->s_freeclusters_counter,
3837                            ext4_count_free_clusters(sb));
3838         percpu_counter_set(&sbi->s_freeinodes_counter,
3839                            ext4_count_free_inodes(sb));
3840         percpu_counter_set(&sbi->s_dirs_counter,
3841                            ext4_count_dirs(sb));
3842         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3843
3844 no_journal:
3845         /*
3846          * Get the # of file system overhead blocks from the
3847          * superblock if present.
3848          */
3849         if (es->s_overhead_clusters)
3850                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3851         else {
3852                 ret = ext4_calculate_overhead(sb);
3853                 if (ret)
3854                         goto failed_mount_wq;
3855         }
3856
3857         /*
3858          * The maximum number of concurrent works can be high and
3859          * concurrency isn't really necessary.  Limit it to 1.
3860          */
3861         EXT4_SB(sb)->dio_unwritten_wq =
3862                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3863         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3864                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3865                 goto failed_mount_wq;
3866         }
3867
3868         /*
3869          * The jbd2_journal_load will have done any necessary log recovery,
3870          * so we can safely mount the rest of the filesystem now.
3871          */
3872
3873         root = ext4_iget(sb, EXT4_ROOT_INO);
3874         if (IS_ERR(root)) {
3875                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3876                 ret = PTR_ERR(root);
3877                 root = NULL;
3878                 goto failed_mount4;
3879         }
3880         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3881                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3882                 iput(root);
3883                 goto failed_mount4;
3884         }
3885         sb->s_root = d_make_root(root);
3886         if (!sb->s_root) {
3887                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3888                 ret = -ENOMEM;
3889                 goto failed_mount4;
3890         }
3891
3892         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3893                 sb->s_flags |= MS_RDONLY;
3894
3895         /* determine the minimum size of new large inodes, if present */
3896         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3897                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3898                                                      EXT4_GOOD_OLD_INODE_SIZE;
3899                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3900                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3901                         if (sbi->s_want_extra_isize <
3902                             le16_to_cpu(es->s_want_extra_isize))
3903                                 sbi->s_want_extra_isize =
3904                                         le16_to_cpu(es->s_want_extra_isize);
3905                         if (sbi->s_want_extra_isize <
3906                             le16_to_cpu(es->s_min_extra_isize))
3907                                 sbi->s_want_extra_isize =
3908                                         le16_to_cpu(es->s_min_extra_isize);
3909                 }
3910         }
3911         /* Check if enough inode space is available */
3912         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3913                                                         sbi->s_inode_size) {
3914                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3915                                                        EXT4_GOOD_OLD_INODE_SIZE;
3916                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3917                          "available");
3918         }
3919
3920         err = ext4_setup_system_zone(sb);
3921         if (err) {
3922                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3923                          "zone (%d)", err);
3924                 goto failed_mount4a;
3925         }
3926
3927         ext4_ext_init(sb);
3928         err = ext4_mb_init(sb);
3929         if (err) {
3930                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3931                          err);
3932                 goto failed_mount5;
3933         }
3934
3935         err = ext4_register_li_request(sb, first_not_zeroed);
3936         if (err)
3937                 goto failed_mount6;
3938
3939         sbi->s_kobj.kset = ext4_kset;
3940         init_completion(&sbi->s_kobj_unregister);
3941         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3942                                    "%s", sb->s_id);
3943         if (err)
3944                 goto failed_mount7;
3945
3946         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3947         ext4_orphan_cleanup(sb, es);
3948         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3949         if (needs_recovery) {
3950                 ext4_msg(sb, KERN_INFO, "recovery complete");
3951                 ext4_mark_recovery_complete(sb, es);
3952         }
3953         if (EXT4_SB(sb)->s_journal) {
3954                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3955                         descr = " journalled data mode";
3956                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3957                         descr = " ordered data mode";
3958                 else
3959                         descr = " writeback data mode";
3960         } else
3961                 descr = "out journal";
3962
3963         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3964                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3965                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3966
3967         if (es->s_error_count)
3968                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3969
3970         kfree(orig_data);
3971         return 0;
3972
3973 cantfind_ext4:
3974         if (!silent)
3975                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3976         goto failed_mount;
3977
3978 failed_mount7:
3979         ext4_unregister_li_request(sb);
3980 failed_mount6:
3981         ext4_mb_release(sb);
3982 failed_mount5:
3983         ext4_ext_release(sb);
3984         ext4_release_system_zone(sb);
3985 failed_mount4a:
3986         dput(sb->s_root);
3987         sb->s_root = NULL;
3988 failed_mount4:
3989         ext4_msg(sb, KERN_ERR, "mount failed");
3990         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3991 failed_mount_wq:
3992         if (sbi->s_journal) {
3993                 jbd2_journal_destroy(sbi->s_journal);
3994                 sbi->s_journal = NULL;
3995         }
3996 failed_mount3:
3997         del_timer(&sbi->s_err_report);
3998         if (sbi->s_flex_groups)
3999                 ext4_kvfree(sbi->s_flex_groups);
4000         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4001         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4002         percpu_counter_destroy(&sbi->s_dirs_counter);
4003         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4004         if (sbi->s_mmp_tsk)
4005                 kthread_stop(sbi->s_mmp_tsk);
4006 failed_mount2:
4007         for (i = 0; i < db_count; i++)
4008                 brelse(sbi->s_group_desc[i]);
4009         ext4_kvfree(sbi->s_group_desc);
4010 failed_mount:
4011         if (sbi->s_chksum_driver)
4012                 crypto_free_shash(sbi->s_chksum_driver);
4013         if (sbi->s_proc) {
4014                 remove_proc_entry("options", sbi->s_proc);
4015                 remove_proc_entry(sb->s_id, ext4_proc_root);
4016         }
4017 #ifdef CONFIG_QUOTA
4018         for (i = 0; i < MAXQUOTAS; i++)
4019                 kfree(sbi->s_qf_names[i]);
4020 #endif
4021         ext4_blkdev_remove(sbi);
4022         brelse(bh);
4023 out_fail:
4024         sb->s_fs_info = NULL;
4025         kfree(sbi->s_blockgroup_lock);
4026         kfree(sbi);
4027 out_free_orig:
4028         kfree(orig_data);
4029         return ret;
4030 }
4031
4032 /*
4033  * Setup any per-fs journal parameters now.  We'll do this both on
4034  * initial mount, once the journal has been initialised but before we've
4035  * done any recovery; and again on any subsequent remount.
4036  */
4037 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4038 {
4039         struct ext4_sb_info *sbi = EXT4_SB(sb);
4040
4041         journal->j_commit_interval = sbi->s_commit_interval;
4042         journal->j_min_batch_time = sbi->s_min_batch_time;
4043         journal->j_max_batch_time = sbi->s_max_batch_time;
4044
4045         write_lock(&journal->j_state_lock);
4046         if (test_opt(sb, BARRIER))
4047                 journal->j_flags |= JBD2_BARRIER;
4048         else
4049                 journal->j_flags &= ~JBD2_BARRIER;
4050         if (test_opt(sb, DATA_ERR_ABORT))
4051                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4052         else
4053                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4054         write_unlock(&journal->j_state_lock);
4055 }
4056
4057 static journal_t *ext4_get_journal(struct super_block *sb,
4058                                    unsigned int journal_inum)
4059 {
4060         struct inode *journal_inode;
4061         journal_t *journal;
4062
4063         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4064
4065         /* First, test for the existence of a valid inode on disk.  Bad
4066          * things happen if we iget() an unused inode, as the subsequent
4067          * iput() will try to delete it. */
4068
4069         journal_inode = ext4_iget(sb, journal_inum);
4070         if (IS_ERR(journal_inode)) {
4071                 ext4_msg(sb, KERN_ERR, "no journal found");
4072                 return NULL;
4073         }
4074         if (!journal_inode->i_nlink) {
4075                 make_bad_inode(journal_inode);
4076                 iput(journal_inode);
4077                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4078                 return NULL;
4079         }
4080
4081         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4082                   journal_inode, journal_inode->i_size);
4083         if (!S_ISREG(journal_inode->i_mode)) {
4084                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4085                 iput(journal_inode);
4086                 return NULL;
4087         }
4088
4089         journal = jbd2_journal_init_inode(journal_inode);
4090         if (!journal) {
4091                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4092                 iput(journal_inode);
4093                 return NULL;
4094         }
4095         journal->j_private = sb;
4096         ext4_init_journal_params(sb, journal);
4097         return journal;
4098 }
4099
4100 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4101                                        dev_t j_dev)
4102 {
4103         struct buffer_head *bh;
4104         journal_t *journal;
4105         ext4_fsblk_t start;
4106         ext4_fsblk_t len;
4107         int hblock, blocksize;
4108         ext4_fsblk_t sb_block;
4109         unsigned long offset;
4110         struct ext4_super_block *es;
4111         struct block_device *bdev;
4112
4113         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4114
4115         bdev = ext4_blkdev_get(j_dev, sb);
4116         if (bdev == NULL)
4117                 return NULL;
4118
4119         blocksize = sb->s_blocksize;
4120         hblock = bdev_logical_block_size(bdev);
4121         if (blocksize < hblock) {
4122                 ext4_msg(sb, KERN_ERR,
4123                         "blocksize too small for journal device");
4124                 goto out_bdev;
4125         }
4126
4127         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4128         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4129         set_blocksize(bdev, blocksize);
4130         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4131                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4132                        "external journal");
4133                 goto out_bdev;
4134         }
4135
4136         es = (struct ext4_super_block *) (bh->b_data + offset);
4137         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4138             !(le32_to_cpu(es->s_feature_incompat) &
4139               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4140                 ext4_msg(sb, KERN_ERR, "external journal has "
4141                                         "bad superblock");
4142                 brelse(bh);
4143                 goto out_bdev;
4144         }
4145
4146         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4147                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4148                 brelse(bh);
4149                 goto out_bdev;
4150         }
4151
4152         len = ext4_blocks_count(es);
4153         start = sb_block + 1;
4154         brelse(bh);     /* we're done with the superblock */
4155
4156         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4157                                         start, len, blocksize);
4158         if (!journal) {
4159                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4160                 goto out_bdev;
4161         }
4162         journal->j_private = sb;
4163         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4164         wait_on_buffer(journal->j_sb_buffer);
4165         if (!buffer_uptodate(journal->j_sb_buffer)) {
4166                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4167                 goto out_journal;
4168         }
4169         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4170                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4171                                         "user (unsupported) - %d",
4172                         be32_to_cpu(journal->j_superblock->s_nr_users));
4173                 goto out_journal;
4174         }
4175         EXT4_SB(sb)->journal_bdev = bdev;
4176         ext4_init_journal_params(sb, journal);
4177         return journal;
4178
4179 out_journal:
4180         jbd2_journal_destroy(journal);
4181 out_bdev:
4182         ext4_blkdev_put(bdev);
4183         return NULL;
4184 }
4185
4186 static int ext4_load_journal(struct super_block *sb,
4187                              struct ext4_super_block *es,
4188                              unsigned long journal_devnum)
4189 {
4190         journal_t *journal;
4191         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4192         dev_t journal_dev;
4193         int err = 0;
4194         int really_read_only;
4195
4196         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4197
4198         if (journal_devnum &&
4199             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4200                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4201                         "numbers have changed");
4202                 journal_dev = new_decode_dev(journal_devnum);
4203         } else
4204                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4205
4206         really_read_only = bdev_read_only(sb->s_bdev);
4207
4208         /*
4209          * Are we loading a blank journal or performing recovery after a
4210          * crash?  For recovery, we need to check in advance whether we
4211          * can get read-write access to the device.
4212          */
4213         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4214                 if (sb->s_flags & MS_RDONLY) {
4215                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4216                                         "required on readonly filesystem");
4217                         if (really_read_only) {
4218                                 ext4_msg(sb, KERN_ERR, "write access "
4219                                         "unavailable, cannot proceed");
4220                                 return -EROFS;
4221                         }
4222                         ext4_msg(sb, KERN_INFO, "write access will "
4223                                "be enabled during recovery");
4224                 }
4225         }
4226
4227         if (journal_inum && journal_dev) {
4228                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4229                        "and inode journals!");
4230                 return -EINVAL;
4231         }
4232
4233         if (journal_inum) {
4234                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4235                         return -EINVAL;
4236         } else {
4237                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4238                         return -EINVAL;
4239         }
4240
4241         if (!(journal->j_flags & JBD2_BARRIER))
4242                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4243
4244         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4245                 err = jbd2_journal_wipe(journal, !really_read_only);
4246         if (!err) {
4247                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4248                 if (save)
4249                         memcpy(save, ((char *) es) +
4250                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4251                 err = jbd2_journal_load(journal);
4252                 if (save)
4253                         memcpy(((char *) es) + EXT4_S_ERR_START,
4254                                save, EXT4_S_ERR_LEN);
4255                 kfree(save);
4256         }
4257
4258         if (err) {
4259                 ext4_msg(sb, KERN_ERR, "error loading journal");
4260                 jbd2_journal_destroy(journal);
4261                 return err;
4262         }
4263
4264         EXT4_SB(sb)->s_journal = journal;
4265         ext4_clear_journal_err(sb, es);
4266
4267         if (!really_read_only && journal_devnum &&
4268             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4269                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4270
4271                 /* Make sure we flush the recovery flag to disk. */
4272                 ext4_commit_super(sb, 1);
4273         }
4274
4275         return 0;
4276 }
4277
4278 static int ext4_commit_super(struct super_block *sb, int sync)
4279 {
4280         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4281         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4282         int error = 0;
4283
4284         if (!sbh || block_device_ejected(sb))
4285                 return error;
4286         if (buffer_write_io_error(sbh)) {
4287                 /*
4288                  * Oh, dear.  A previous attempt to write the
4289                  * superblock failed.  This could happen because the
4290                  * USB device was yanked out.  Or it could happen to
4291                  * be a transient write error and maybe the block will
4292                  * be remapped.  Nothing we can do but to retry the
4293                  * write and hope for the best.
4294                  */
4295                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4296                        "superblock detected");
4297                 clear_buffer_write_io_error(sbh);
4298                 set_buffer_uptodate(sbh);
4299         }
4300         /*
4301          * If the file system is mounted read-only, don't update the
4302          * superblock write time.  This avoids updating the superblock
4303          * write time when we are mounting the root file system
4304          * read/only but we need to replay the journal; at that point,
4305          * for people who are east of GMT and who make their clock
4306          * tick in localtime for Windows bug-for-bug compatibility,
4307          * the clock is set in the future, and this will cause e2fsck
4308          * to complain and force a full file system check.
4309          */
4310         if (!(sb->s_flags & MS_RDONLY))
4311                 es->s_wtime = cpu_to_le32(get_seconds());
4312         if (sb->s_bdev->bd_part)
4313                 es->s_kbytes_written =
4314                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4315                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4316                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4317         else
4318                 es->s_kbytes_written =
4319                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4320         ext4_free_blocks_count_set(es,
4321                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4322                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4323         es->s_free_inodes_count =
4324                 cpu_to_le32(percpu_counter_sum_positive(
4325                                 &EXT4_SB(sb)->s_freeinodes_counter));
4326         sb->s_dirt = 0;
4327         BUFFER_TRACE(sbh, "marking dirty");
4328         ext4_superblock_csum_set(sb, es);
4329         mark_buffer_dirty(sbh);
4330         if (sync) {
4331                 error = sync_dirty_buffer(sbh);
4332                 if (error)
4333                         return error;
4334
4335                 error = buffer_write_io_error(sbh);
4336                 if (error) {
4337                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4338                                "superblock");
4339                         clear_buffer_write_io_error(sbh);
4340                         set_buffer_uptodate(sbh);
4341                 }
4342         }
4343         return error;
4344 }
4345
4346 /*
4347  * Have we just finished recovery?  If so, and if we are mounting (or
4348  * remounting) the filesystem readonly, then we will end up with a
4349  * consistent fs on disk.  Record that fact.
4350  */
4351 static void ext4_mark_recovery_complete(struct super_block *sb,
4352                                         struct ext4_super_block *es)
4353 {
4354         journal_t *journal = EXT4_SB(sb)->s_journal;
4355
4356         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4357                 BUG_ON(journal != NULL);
4358                 return;
4359         }
4360         jbd2_journal_lock_updates(journal);
4361         if (jbd2_journal_flush(journal) < 0)
4362                 goto out;
4363
4364         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4365             sb->s_flags & MS_RDONLY) {
4366                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4367                 ext4_commit_super(sb, 1);
4368         }
4369
4370 out:
4371         jbd2_journal_unlock_updates(journal);
4372 }
4373
4374 /*
4375  * If we are mounting (or read-write remounting) a filesystem whose journal
4376  * has recorded an error from a previous lifetime, move that error to the
4377  * main filesystem now.
4378  */
4379 static void ext4_clear_journal_err(struct super_block *sb,
4380                                    struct ext4_super_block *es)
4381 {
4382         journal_t *journal;
4383         int j_errno;
4384         const char *errstr;
4385
4386         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4387
4388         journal = EXT4_SB(sb)->s_journal;
4389
4390         /*
4391          * Now check for any error status which may have been recorded in the
4392          * journal by a prior ext4_error() or ext4_abort()
4393          */
4394
4395         j_errno = jbd2_journal_errno(journal);
4396         if (j_errno) {
4397                 char nbuf[16];
4398
4399                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4400                 ext4_warning(sb, "Filesystem error recorded "
4401                              "from previous mount: %s", errstr);
4402                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4403
4404                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4405                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4406                 ext4_commit_super(sb, 1);
4407
4408                 jbd2_journal_clear_err(journal);
4409         }
4410 }
4411
4412 /*
4413  * Force the running and committing transactions to commit,
4414  * and wait on the commit.
4415  */
4416 int ext4_force_commit(struct super_block *sb)
4417 {
4418         journal_t *journal;
4419         int ret = 0;
4420
4421         if (sb->s_flags & MS_RDONLY)
4422                 return 0;
4423
4424         journal = EXT4_SB(sb)->s_journal;
4425         if (journal) {
4426                 vfs_check_frozen(sb, SB_FREEZE_TRANS);
4427                 ret = ext4_journal_force_commit(journal);
4428         }
4429
4430         return ret;
4431 }
4432
4433 static void ext4_write_super(struct super_block *sb)
4434 {
4435         lock_super(sb);
4436         ext4_commit_super(sb, 1);
4437         unlock_super(sb);
4438 }
4439
4440 static int ext4_sync_fs(struct super_block *sb, int wait)
4441 {
4442         int ret = 0;
4443         tid_t target;
4444         struct ext4_sb_info *sbi = EXT4_SB(sb);
4445
4446         trace_ext4_sync_fs(sb, wait);
4447         flush_workqueue(sbi->dio_unwritten_wq);
4448         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4449                 if (wait)
4450                         jbd2_log_wait_commit(sbi->s_journal, target);
4451         }
4452         return ret;
4453 }
4454
4455 /*
4456  * LVM calls this function before a (read-only) snapshot is created.  This
4457  * gives us a chance to flush the journal completely and mark the fs clean.
4458  *
4459  * Note that only this function cannot bring a filesystem to be in a clean
4460  * state independently, because ext4 prevents a new handle from being started
4461  * by @sb->s_frozen, which stays in an upper layer.  It thus needs help from
4462  * the upper layer.
4463  */
4464 static int ext4_freeze(struct super_block *sb)
4465 {
4466         int error = 0;
4467         journal_t *journal;
4468
4469         if (sb->s_flags & MS_RDONLY)
4470                 return 0;
4471
4472         journal = EXT4_SB(sb)->s_journal;
4473
4474         /* Now we set up the journal barrier. */
4475         jbd2_journal_lock_updates(journal);
4476
4477         /*
4478          * Don't clear the needs_recovery flag if we failed to flush
4479          * the journal.
4480          */
4481         error = jbd2_journal_flush(journal);
4482         if (error < 0)
4483                 goto out;
4484
4485         /* Journal blocked and flushed, clear needs_recovery flag. */
4486         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4487         error = ext4_commit_super(sb, 1);
4488 out:
4489         /* we rely on s_frozen to stop further updates */
4490         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4491         return error;
4492 }
4493
4494 /*
4495  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4496  * flag here, even though the filesystem is not technically dirty yet.
4497  */
4498 static int ext4_unfreeze(struct super_block *sb)
4499 {
4500         if (sb->s_flags & MS_RDONLY)
4501                 return 0;
4502
4503         lock_super(sb);
4504         /* Reset the needs_recovery flag before the fs is unlocked. */
4505         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4506         ext4_commit_super(sb, 1);
4507         unlock_super(sb);
4508         return 0;
4509 }
4510
4511 /*
4512  * Structure to save mount options for ext4_remount's benefit
4513  */
4514 struct ext4_mount_options {
4515         unsigned long s_mount_opt;
4516         unsigned long s_mount_opt2;
4517         kuid_t s_resuid;
4518         kgid_t s_resgid;
4519         unsigned long s_commit_interval;
4520         u32 s_min_batch_time, s_max_batch_time;
4521 #ifdef CONFIG_QUOTA
4522         int s_jquota_fmt;
4523         char *s_qf_names[MAXQUOTAS];
4524 #endif
4525 };
4526
4527 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4528 {
4529         struct ext4_super_block *es;
4530         struct ext4_sb_info *sbi = EXT4_SB(sb);
4531         unsigned long old_sb_flags;
4532         struct ext4_mount_options old_opts;
4533         int enable_quota = 0;
4534         ext4_group_t g;
4535         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4536         int err = 0;
4537 #ifdef CONFIG_QUOTA
4538         int i;
4539 #endif
4540         char *orig_data = kstrdup(data, GFP_KERNEL);
4541
4542         /* Store the original options */
4543         lock_super(sb);
4544         old_sb_flags = sb->s_flags;
4545         old_opts.s_mount_opt = sbi->s_mount_opt;
4546         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4547         old_opts.s_resuid = sbi->s_resuid;
4548         old_opts.s_resgid = sbi->s_resgid;
4549         old_opts.s_commit_interval = sbi->s_commit_interval;
4550         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4551         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4552 #ifdef CONFIG_QUOTA
4553         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4554         for (i = 0; i < MAXQUOTAS; i++)
4555                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4556 #endif
4557         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4558                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4559
4560         /*
4561          * Allow the "check" option to be passed as a remount option.
4562          */
4563         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4564                 err = -EINVAL;
4565                 goto restore_opts;
4566         }
4567
4568         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4569                 ext4_abort(sb, "Abort forced by user");
4570
4571         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4572                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4573
4574         es = sbi->s_es;
4575
4576         if (sbi->s_journal) {
4577                 ext4_init_journal_params(sb, sbi->s_journal);
4578                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4579         }
4580
4581         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4582                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4583                         err = -EROFS;
4584                         goto restore_opts;
4585                 }
4586
4587                 if (*flags & MS_RDONLY) {
4588                         err = dquot_suspend(sb, -1);
4589                         if (err < 0)
4590                                 goto restore_opts;
4591
4592                         /*
4593                          * First of all, the unconditional stuff we have to do
4594                          * to disable replay of the journal when we next remount
4595                          */
4596                         sb->s_flags |= MS_RDONLY;
4597
4598                         /*
4599                          * OK, test if we are remounting a valid rw partition
4600                          * readonly, and if so set the rdonly flag and then
4601                          * mark the partition as valid again.
4602                          */
4603                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4604                             (sbi->s_mount_state & EXT4_VALID_FS))
4605                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4606
4607                         if (sbi->s_journal)
4608                                 ext4_mark_recovery_complete(sb, es);
4609                 } else {
4610                         /* Make sure we can mount this feature set readwrite */
4611                         if (!ext4_feature_set_ok(sb, 0)) {
4612                                 err = -EROFS;
4613                                 goto restore_opts;
4614                         }
4615                         /*
4616                          * Make sure the group descriptor checksums
4617                          * are sane.  If they aren't, refuse to remount r/w.
4618                          */
4619                         for (g = 0; g < sbi->s_groups_count; g++) {
4620                                 struct ext4_group_desc *gdp =
4621                                         ext4_get_group_desc(sb, g, NULL);
4622
4623                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4624                                         ext4_msg(sb, KERN_ERR,
4625                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4626                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4627                                                le16_to_cpu(gdp->bg_checksum));
4628                                         err = -EINVAL;
4629                                         goto restore_opts;
4630                                 }
4631                         }
4632
4633                         /*
4634                          * If we have an unprocessed orphan list hanging
4635                          * around from a previously readonly bdev mount,
4636                          * require a full umount/remount for now.
4637                          */
4638                         if (es->s_last_orphan) {
4639                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4640                                        "remount RDWR because of unprocessed "
4641                                        "orphan inode list.  Please "
4642                                        "umount/remount instead");
4643                                 err = -EINVAL;
4644                                 goto restore_opts;
4645                         }
4646
4647                         /*
4648                          * Mounting a RDONLY partition read-write, so reread
4649                          * and store the current valid flag.  (It may have
4650                          * been changed by e2fsck since we originally mounted
4651                          * the partition.)
4652                          */
4653                         if (sbi->s_journal)
4654                                 ext4_clear_journal_err(sb, es);
4655                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4656                         if (!ext4_setup_super(sb, es, 0))
4657                                 sb->s_flags &= ~MS_RDONLY;
4658                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4659                                                      EXT4_FEATURE_INCOMPAT_MMP))
4660                                 if (ext4_multi_mount_protect(sb,
4661                                                 le64_to_cpu(es->s_mmp_block))) {
4662                                         err = -EROFS;
4663                                         goto restore_opts;
4664                                 }
4665                         enable_quota = 1;
4666                 }
4667         }
4668
4669         /*
4670          * Reinitialize lazy itable initialization thread based on
4671          * current settings
4672          */
4673         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4674                 ext4_unregister_li_request(sb);
4675         else {
4676                 ext4_group_t first_not_zeroed;
4677                 first_not_zeroed = ext4_has_uninit_itable(sb);
4678                 ext4_register_li_request(sb, first_not_zeroed);
4679         }
4680
4681         ext4_setup_system_zone(sb);
4682         if (sbi->s_journal == NULL)
4683                 ext4_commit_super(sb, 1);
4684
4685 #ifdef CONFIG_QUOTA
4686         /* Release old quota file names */
4687         for (i = 0; i < MAXQUOTAS; i++)
4688                 if (old_opts.s_qf_names[i] &&
4689                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4690                         kfree(old_opts.s_qf_names[i]);
4691 #endif
4692         unlock_super(sb);
4693         if (enable_quota)
4694                 dquot_resume(sb, -1);
4695
4696         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4697         kfree(orig_data);
4698         return 0;
4699
4700 restore_opts:
4701         sb->s_flags = old_sb_flags;
4702         sbi->s_mount_opt = old_opts.s_mount_opt;
4703         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4704         sbi->s_resuid = old_opts.s_resuid;
4705         sbi->s_resgid = old_opts.s_resgid;
4706         sbi->s_commit_interval = old_opts.s_commit_interval;
4707         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4708         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4709 #ifdef CONFIG_QUOTA
4710         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4711         for (i = 0; i < MAXQUOTAS; i++) {
4712                 if (sbi->s_qf_names[i] &&
4713                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4714                         kfree(sbi->s_qf_names[i]);
4715                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4716         }
4717 #endif
4718         unlock_super(sb);
4719         kfree(orig_data);
4720         return err;
4721 }
4722
4723 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4724 {
4725         struct super_block *sb = dentry->d_sb;
4726         struct ext4_sb_info *sbi = EXT4_SB(sb);
4727         struct ext4_super_block *es = sbi->s_es;
4728         ext4_fsblk_t overhead = 0;
4729         u64 fsid;
4730         s64 bfree;
4731
4732         if (!test_opt(sb, MINIX_DF))
4733                 overhead = sbi->s_overhead;
4734
4735         buf->f_type = EXT4_SUPER_MAGIC;
4736         buf->f_bsize = sb->s_blocksize;
4737         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, sbi->s_overhead);
4738         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4739                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4740         /* prevent underflow in case that few free space is available */
4741         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4742         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4743         if (buf->f_bfree < ext4_r_blocks_count(es))
4744                 buf->f_bavail = 0;
4745         buf->f_files = le32_to_cpu(es->s_inodes_count);
4746         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4747         buf->f_namelen = EXT4_NAME_LEN;
4748         fsid = le64_to_cpup((void *)es->s_uuid) ^
4749                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4750         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4751         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4752
4753         return 0;
4754 }
4755
4756 /* Helper function for writing quotas on sync - we need to start transaction
4757  * before quota file is locked for write. Otherwise the are possible deadlocks:
4758  * Process 1                         Process 2
4759  * ext4_create()                     quota_sync()
4760  *   jbd2_journal_start()                  write_dquot()
4761  *   dquot_initialize()                         down(dqio_mutex)
4762  *     down(dqio_mutex)                    jbd2_journal_start()
4763  *
4764  */
4765
4766 #ifdef CONFIG_QUOTA
4767
4768 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4769 {
4770         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4771 }
4772
4773 static int ext4_write_dquot(struct dquot *dquot)
4774 {
4775         int ret, err;
4776         handle_t *handle;
4777         struct inode *inode;
4778
4779         inode = dquot_to_inode(dquot);
4780         handle = ext4_journal_start(inode,
4781                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4782         if (IS_ERR(handle))
4783                 return PTR_ERR(handle);
4784         ret = dquot_commit(dquot);
4785         err = ext4_journal_stop(handle);
4786         if (!ret)
4787                 ret = err;
4788         return ret;
4789 }
4790
4791 static int ext4_acquire_dquot(struct dquot *dquot)
4792 {
4793         int ret, err;
4794         handle_t *handle;
4795
4796         handle = ext4_journal_start(dquot_to_inode(dquot),
4797                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4798         if (IS_ERR(handle))
4799                 return PTR_ERR(handle);
4800         ret = dquot_acquire(dquot);
4801         err = ext4_journal_stop(handle);
4802         if (!ret)
4803                 ret = err;
4804         return ret;
4805 }
4806
4807 static int ext4_release_dquot(struct dquot *dquot)
4808 {
4809         int ret, err;
4810         handle_t *handle;
4811
4812         handle = ext4_journal_start(dquot_to_inode(dquot),
4813                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4814         if (IS_ERR(handle)) {
4815                 /* Release dquot anyway to avoid endless cycle in dqput() */
4816                 dquot_release(dquot);
4817                 return PTR_ERR(handle);
4818         }
4819         ret = dquot_release(dquot);
4820         err = ext4_journal_stop(handle);
4821         if (!ret)
4822                 ret = err;
4823         return ret;
4824 }
4825
4826 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4827 {
4828         /* Are we journaling quotas? */
4829         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4830             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4831                 dquot_mark_dquot_dirty(dquot);
4832                 return ext4_write_dquot(dquot);
4833         } else {
4834                 return dquot_mark_dquot_dirty(dquot);
4835         }
4836 }
4837
4838 static int ext4_write_info(struct super_block *sb, int type)
4839 {
4840         int ret, err;
4841         handle_t *handle;
4842
4843         /* Data block + inode block */
4844         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4845         if (IS_ERR(handle))
4846                 return PTR_ERR(handle);
4847         ret = dquot_commit_info(sb, type);
4848         err = ext4_journal_stop(handle);
4849         if (!ret)
4850                 ret = err;
4851         return ret;
4852 }
4853
4854 /*
4855  * Turn on quotas during mount time - we need to find
4856  * the quota file and such...
4857  */
4858 static int ext4_quota_on_mount(struct super_block *sb, int type)
4859 {
4860         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4861                                         EXT4_SB(sb)->s_jquota_fmt, type);
4862 }
4863
4864 /*
4865  * Standard function to be called on quota_on
4866  */
4867 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4868                          struct path *path)
4869 {
4870         int err;
4871
4872         if (!test_opt(sb, QUOTA))
4873                 return -EINVAL;
4874
4875         /* Quotafile not on the same filesystem? */
4876         if (path->dentry->d_sb != sb)
4877                 return -EXDEV;
4878         /* Journaling quota? */
4879         if (EXT4_SB(sb)->s_qf_names[type]) {
4880                 /* Quotafile not in fs root? */
4881                 if (path->dentry->d_parent != sb->s_root)
4882                         ext4_msg(sb, KERN_WARNING,
4883                                 "Quota file not on filesystem root. "
4884                                 "Journaled quota will not work");
4885         }
4886
4887         /*
4888          * When we journal data on quota file, we have to flush journal to see
4889          * all updates to the file when we bypass pagecache...
4890          */
4891         if (EXT4_SB(sb)->s_journal &&
4892             ext4_should_journal_data(path->dentry->d_inode)) {
4893                 /*
4894                  * We don't need to lock updates but journal_flush() could
4895                  * otherwise be livelocked...
4896                  */
4897                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4898                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4899                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4900                 if (err)
4901                         return err;
4902         }
4903
4904         return dquot_quota_on(sb, type, format_id, path);
4905 }
4906
4907 static int ext4_quota_off(struct super_block *sb, int type)
4908 {
4909         struct inode *inode = sb_dqopt(sb)->files[type];
4910         handle_t *handle;
4911
4912         /* Force all delayed allocation blocks to be allocated.
4913          * Caller already holds s_umount sem */
4914         if (test_opt(sb, DELALLOC))
4915                 sync_filesystem(sb);
4916
4917         if (!inode)
4918                 goto out;
4919
4920         /* Update modification times of quota files when userspace can
4921          * start looking at them */
4922         handle = ext4_journal_start(inode, 1);
4923         if (IS_ERR(handle))
4924                 goto out;
4925         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4926         ext4_mark_inode_dirty(handle, inode);
4927         ext4_journal_stop(handle);
4928
4929 out:
4930         return dquot_quota_off(sb, type);
4931 }
4932
4933 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4934  * acquiring the locks... As quota files are never truncated and quota code
4935  * itself serializes the operations (and no one else should touch the files)
4936  * we don't have to be afraid of races */
4937 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4938                                size_t len, loff_t off)
4939 {
4940         struct inode *inode = sb_dqopt(sb)->files[type];
4941         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4942         int err = 0;
4943         int offset = off & (sb->s_blocksize - 1);
4944         int tocopy;
4945         size_t toread;
4946         struct buffer_head *bh;
4947         loff_t i_size = i_size_read(inode);
4948
4949         if (off > i_size)
4950                 return 0;
4951         if (off+len > i_size)
4952                 len = i_size-off;
4953         toread = len;
4954         while (toread > 0) {
4955                 tocopy = sb->s_blocksize - offset < toread ?
4956                                 sb->s_blocksize - offset : toread;
4957                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4958                 if (err)
4959                         return err;
4960                 if (!bh)        /* A hole? */
4961                         memset(data, 0, tocopy);
4962                 else
4963                         memcpy(data, bh->b_data+offset, tocopy);
4964                 brelse(bh);
4965                 offset = 0;
4966                 toread -= tocopy;
4967                 data += tocopy;
4968                 blk++;
4969         }
4970         return len;
4971 }
4972
4973 /* Write to quotafile (we know the transaction is already started and has
4974  * enough credits) */
4975 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4976                                 const char *data, size_t len, loff_t off)
4977 {
4978         struct inode *inode = sb_dqopt(sb)->files[type];
4979         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4980         int err = 0;
4981         int offset = off & (sb->s_blocksize - 1);
4982         struct buffer_head *bh;
4983         handle_t *handle = journal_current_handle();
4984
4985         if (EXT4_SB(sb)->s_journal && !handle) {
4986                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4987                         " cancelled because transaction is not started",
4988                         (unsigned long long)off, (unsigned long long)len);
4989                 return -EIO;
4990         }
4991         /*
4992          * Since we account only one data block in transaction credits,
4993          * then it is impossible to cross a block boundary.
4994          */
4995         if (sb->s_blocksize - offset < len) {
4996                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4997                         " cancelled because not block aligned",
4998                         (unsigned long long)off, (unsigned long long)len);
4999                 return -EIO;
5000         }
5001
5002         bh = ext4_bread(handle, inode, blk, 1, &err);
5003         if (!bh)
5004                 goto out;
5005         err = ext4_journal_get_write_access(handle, bh);
5006         if (err) {
5007                 brelse(bh);
5008                 goto out;
5009         }
5010         lock_buffer(bh);
5011         memcpy(bh->b_data+offset, data, len);
5012         flush_dcache_page(bh->b_page);
5013         unlock_buffer(bh);
5014         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5015         brelse(bh);
5016 out:
5017         if (err)
5018                 return err;
5019         if (inode->i_size < off + len) {
5020                 i_size_write(inode, off + len);
5021                 EXT4_I(inode)->i_disksize = inode->i_size;
5022                 ext4_mark_inode_dirty(handle, inode);
5023         }
5024         return len;
5025 }
5026
5027 #endif
5028
5029 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5030                        const char *dev_name, void *data)
5031 {
5032         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5033 }
5034
5035 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5036 static inline void register_as_ext2(void)
5037 {
5038         int err = register_filesystem(&ext2_fs_type);
5039         if (err)
5040                 printk(KERN_WARNING
5041                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5042 }
5043
5044 static inline void unregister_as_ext2(void)
5045 {
5046         unregister_filesystem(&ext2_fs_type);
5047 }
5048
5049 static inline int ext2_feature_set_ok(struct super_block *sb)
5050 {
5051         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5052                 return 0;
5053         if (sb->s_flags & MS_RDONLY)
5054                 return 1;
5055         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5056                 return 0;
5057         return 1;
5058 }
5059 MODULE_ALIAS("ext2");
5060 #else
5061 static inline void register_as_ext2(void) { }
5062 static inline void unregister_as_ext2(void) { }
5063 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5064 #endif
5065
5066 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5067 static inline void register_as_ext3(void)
5068 {
5069         int err = register_filesystem(&ext3_fs_type);
5070         if (err)
5071                 printk(KERN_WARNING
5072                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5073 }
5074
5075 static inline void unregister_as_ext3(void)
5076 {
5077         unregister_filesystem(&ext3_fs_type);
5078 }
5079
5080 static inline int ext3_feature_set_ok(struct super_block *sb)
5081 {
5082         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5083                 return 0;
5084         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5085                 return 0;
5086         if (sb->s_flags & MS_RDONLY)
5087                 return 1;
5088         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5089                 return 0;
5090         return 1;
5091 }
5092 MODULE_ALIAS("ext3");
5093 #else
5094 static inline void register_as_ext3(void) { }
5095 static inline void unregister_as_ext3(void) { }
5096 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5097 #endif
5098
5099 static struct file_system_type ext4_fs_type = {
5100         .owner          = THIS_MODULE,
5101         .name           = "ext4",
5102         .mount          = ext4_mount,
5103         .kill_sb        = kill_block_super,
5104         .fs_flags       = FS_REQUIRES_DEV,
5105 };
5106
5107 static int __init ext4_init_feat_adverts(void)
5108 {
5109         struct ext4_features *ef;
5110         int ret = -ENOMEM;
5111
5112         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5113         if (!ef)
5114                 goto out;
5115
5116         ef->f_kobj.kset = ext4_kset;
5117         init_completion(&ef->f_kobj_unregister);
5118         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5119                                    "features");
5120         if (ret) {
5121                 kfree(ef);
5122                 goto out;
5123         }
5124
5125         ext4_feat = ef;
5126         ret = 0;
5127 out:
5128         return ret;
5129 }
5130
5131 static void ext4_exit_feat_adverts(void)
5132 {
5133         kobject_put(&ext4_feat->f_kobj);
5134         wait_for_completion(&ext4_feat->f_kobj_unregister);
5135         kfree(ext4_feat);
5136 }
5137
5138 /* Shared across all ext4 file systems */
5139 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5140 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5141
5142 static int __init ext4_init_fs(void)
5143 {
5144         int i, err;
5145
5146         ext4_li_info = NULL;
5147         mutex_init(&ext4_li_mtx);
5148
5149         ext4_check_flag_values();
5150
5151         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5152                 mutex_init(&ext4__aio_mutex[i]);
5153                 init_waitqueue_head(&ext4__ioend_wq[i]);
5154         }
5155
5156         err = ext4_init_pageio();
5157         if (err)
5158                 return err;
5159         err = ext4_init_system_zone();
5160         if (err)
5161                 goto out6;
5162         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5163         if (!ext4_kset)
5164                 goto out5;
5165         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5166
5167         err = ext4_init_feat_adverts();
5168         if (err)
5169                 goto out4;
5170
5171         err = ext4_init_mballoc();
5172         if (err)
5173                 goto out3;
5174
5175         err = ext4_init_xattr();
5176         if (err)
5177                 goto out2;
5178         err = init_inodecache();
5179         if (err)
5180                 goto out1;
5181         register_as_ext3();
5182         register_as_ext2();
5183         err = register_filesystem(&ext4_fs_type);
5184         if (err)
5185                 goto out;
5186
5187         return 0;
5188 out:
5189         unregister_as_ext2();
5190         unregister_as_ext3();
5191         destroy_inodecache();
5192 out1:
5193         ext4_exit_xattr();
5194 out2:
5195         ext4_exit_mballoc();
5196 out3:
5197         ext4_exit_feat_adverts();
5198 out4:
5199         if (ext4_proc_root)
5200                 remove_proc_entry("fs/ext4", NULL);
5201         kset_unregister(ext4_kset);
5202 out5:
5203         ext4_exit_system_zone();
5204 out6:
5205         ext4_exit_pageio();
5206         return err;
5207 }
5208
5209 static void __exit ext4_exit_fs(void)
5210 {
5211         ext4_destroy_lazyinit_thread();
5212         unregister_as_ext2();
5213         unregister_as_ext3();
5214         unregister_filesystem(&ext4_fs_type);
5215         destroy_inodecache();
5216         ext4_exit_xattr();
5217         ext4_exit_mballoc();
5218         ext4_exit_feat_adverts();
5219         remove_proc_entry("fs/ext4", NULL);
5220         kset_unregister(ext4_kset);
5221         ext4_exit_system_zone();
5222         ext4_exit_pageio();
5223 }
5224
5225 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5226 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5227 MODULE_LICENSE("GPL");
5228 module_init(ext4_init_fs)
5229 module_exit(ext4_exit_fs)