fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static const char *ext4_decode_error(struct super_block *sb, int errno,
  73                                      char nbuf[16]);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static void ext4_write_super(struct super_block *sb);
  78 static int ext4_freeze(struct super_block *sb);
  79 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  80                        const char *dev_name, void *data);
  81 static inline int ext2_feature_set_ok(struct super_block *sb);
  82 static inline int ext3_feature_set_ok(struct super_block *sb);
  83 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  84 static void ext4_destroy_lazyinit_thread(void);
  85 static void ext4_unregister_li_request(struct super_block *sb);
  86 static void ext4_clear_request_list(void);
  87
  88 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  89 static struct file_system_type ext2_fs_type = {
  90         .owner          = THIS_MODULE,
  91         .name           = "ext2",
  92         .mount          = ext4_mount,
  93         .kill_sb        = kill_block_super,
  94         .fs_flags       = FS_REQUIRES_DEV,
  95 };
  96 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  97 #else
  98 #define IS_EXT2_SB(sb) (0)
  99 #endif
 100
 101
 102 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 103 static struct file_system_type ext3_fs_type = {
 104         .owner          = THIS_MODULE,
 105         .name           = "ext3",
 106         .mount          = ext4_mount,
 107         .kill_sb        = kill_block_super,
 108         .fs_flags       = FS_REQUIRES_DEV,
 109 };
 110 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 111 #else
 112 #define IS_EXT3_SB(sb) (0)
 113 #endif
 114
 115 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 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 337 {
 338         journal_t *journal;
 339
 340         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 341         if (sb->s_flags & MS_RDONLY)
 342                 return ERR_PTR(-EROFS);
 343
 344         WARN_ON(sb->s_writers.frozen == SB_FREEZE_COMPLETE);
 345         journal = EXT4_SB(sb)->s_journal;
 346         if (!journal)
 347                 return ext4_get_nojournal();
 348         /*
 349          * Special case here: if the journal has aborted behind our
 350          * backs (eg. EIO in the commit thread), then we still need to
 351          * take the FS itself readonly cleanly.
 352          */
 353         if (is_journal_aborted(journal)) {
 354                 ext4_abort(sb, "Detected aborted journal");
 355                 return ERR_PTR(-EROFS);
 356         }
 357         return jbd2_journal_start(journal, nblocks);
 358 }
 359
 360 /*
 361  * The only special thing we need to do here is to make sure that all
 362  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 363  * that sync() will call the filesystem's write_super callback if
 364  * appropriate.
 365  */
 366 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 367 {
 368         struct super_block *sb;
 369         int err;
 370         int rc;
 371
 372         if (!ext4_handle_valid(handle)) {
 373                 ext4_put_nojournal(handle);
 374                 return 0;
 375         }
 376         sb = handle->h_transaction->t_journal->j_private;
 377         err = handle->h_err;
 378         rc = jbd2_journal_stop(handle);
 379
 380         if (!err)
 381                 err = rc;
 382         if (err)
 383                 __ext4_std_error(sb, where, line, err);
 384         return err;
 385 }
 386
 387 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 388                                const char *err_fn, struct buffer_head *bh,
 389                                handle_t *handle, int err)
 390 {
 391         char nbuf[16];
 392         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 393
 394         BUG_ON(!ext4_handle_valid(handle));
 395
 396         if (bh)
 397                 BUFFER_TRACE(bh, "abort");
 398
 399         if (!handle->h_err)
 400                 handle->h_err = err;
 401
 402         if (is_handle_aborted(handle))
 403                 return;
 404
 405         printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
 406                caller, line, errstr, err_fn);
 407
 408         jbd2_journal_abort_handle(handle);
 409 }
 410
 411 static void __save_error_info(struct super_block *sb, const char *func,
 412                             unsigned int line)
 413 {
 414         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 415
 416         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 417         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 418         es->s_last_error_time = cpu_to_le32(get_seconds());
 419         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 420         es->s_last_error_line = cpu_to_le32(line);
 421         if (!es->s_first_error_time) {
 422                 es->s_first_error_time = es->s_last_error_time;
 423                 strncpy(es->s_first_error_func, func,
 424                         sizeof(es->s_first_error_func));
 425                 es->s_first_error_line = cpu_to_le32(line);
 426                 es->s_first_error_ino = es->s_last_error_ino;
 427                 es->s_first_error_block = es->s_last_error_block;
 428         }
 429         /*
 430          * Start the daily error reporting function if it hasn't been
 431          * started already
 432          */
 433         if (!es->s_error_count)
 434                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 435         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 436 }
 437
 438 static void save_error_info(struct super_block *sb, const char *func,
 439                             unsigned int line)
 440 {
 441         __save_error_info(sb, func, line);
 442         ext4_commit_super(sb, 1);
 443 }
 444
 445 /*
 446  * The del_gendisk() function uninitializes the disk-specific data
 447  * structures, including the bdi structure, without telling anyone
 448  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 449  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 450  * This is a kludge to prevent these oops until we can put in a proper
 451  * hook in del_gendisk() to inform the VFS and file system layers.
 452  */
 453 static int block_device_ejected(struct super_block *sb)
 454 {
 455         struct inode *bd_inode = sb->s_bdev->bd_inode;
 456         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 457
 458         return bdi->dev == NULL;
 459 }
 460
 461 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 462 {
 463         struct super_block              *sb = journal->j_private;
 464         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 465         int                             error = is_journal_aborted(journal);
 466         struct ext4_journal_cb_entry    *jce, *tmp;
 467
 468         spin_lock(&sbi->s_md_lock);
 469         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 470                 list_del_init(&jce->jce_list);
 471                 spin_unlock(&sbi->s_md_lock);
 472                 jce->jce_func(sb, jce, error);
 473                 spin_lock(&sbi->s_md_lock);
 474         }
 475         spin_unlock(&sbi->s_md_lock);
 476 }
 477
 478 /* Deal with the reporting of failure conditions on a filesystem such as
 479  * inconsistencies detected or read IO failures.
 480  *
 481  * On ext2, we can store the error state of the filesystem in the
 482  * superblock.  That is not possible on ext4, because we may have other
 483  * write ordering constraints on the superblock which prevent us from
 484  * writing it out straight away; and given that the journal is about to
 485  * be aborted, we can't rely on the current, or future, transactions to
 486  * write out the superblock safely.
 487  *
 488  * We'll just use the jbd2_journal_abort() error code to record an error in
 489  * the journal instead.  On recovery, the journal will complain about
 490  * that error until we've noted it down and cleared it.
 491  */
 492
 493 static void ext4_handle_error(struct super_block *sb)
 494 {
 495         if (sb->s_flags & MS_RDONLY)
 496                 return;
 497
 498         if (!test_opt(sb, ERRORS_CONT)) {
 499                 journal_t *journal = EXT4_SB(sb)->s_journal;
 500
 501                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 502                 if (journal)
 503                         jbd2_journal_abort(journal, -EIO);
 504         }
 505         if (test_opt(sb, ERRORS_RO)) {
 506                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 507                 sb->s_flags |= MS_RDONLY;
 508         }
 509         if (test_opt(sb, ERRORS_PANIC))
 510                 panic("EXT4-fs (device %s): panic forced after error\n",
 511                         sb->s_id);
 512 }
 513
 514 void __ext4_error(struct super_block *sb, const char *function,
 515                   unsigned int line, const char *fmt, ...)
 516 {
 517         struct va_format vaf;
 518         va_list args;
 519
 520         va_start(args, fmt);
 521         vaf.fmt = fmt;
 522         vaf.va = &args;
 523         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 524                sb->s_id, function, line, current->comm, &vaf);
 525         va_end(args);
 526         save_error_info(sb, function, line);
 527
 528         ext4_handle_error(sb);
 529 }
 530
 531 void ext4_error_inode(struct inode *inode, const char *function,
 532                       unsigned int line, ext4_fsblk_t block,
 533                       const char *fmt, ...)
 534 {
 535         va_list args;
 536         struct va_format vaf;
 537         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 538
 539         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 540         es->s_last_error_block = cpu_to_le64(block);
 541         save_error_info(inode->i_sb, function, line);
 542         va_start(args, fmt);
 543         vaf.fmt = fmt;
 544         vaf.va = &args;
 545         if (block)
 546                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 547                        "inode #%lu: block %llu: comm %s: %pV\n",
 548                        inode->i_sb->s_id, function, line, inode->i_ino,
 549                        block, current->comm, &vaf);
 550         else
 551                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 552                        "inode #%lu: comm %s: %pV\n",
 553                        inode->i_sb->s_id, function, line, inode->i_ino,
 554                        current->comm, &vaf);
 555         va_end(args);
 556
 557         ext4_handle_error(inode->i_sb);
 558 }
 559
 560 void ext4_error_file(struct file *file, const char *function,
 561                      unsigned int line, ext4_fsblk_t block,
 562                      const char *fmt, ...)
 563 {
 564         va_list args;
 565         struct va_format vaf;
 566         struct ext4_super_block *es;
 567         struct inode *inode = file->f_dentry->d_inode;
 568         char pathname[80], *path;
 569
 570         es = EXT4_SB(inode->i_sb)->s_es;
 571         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 572         save_error_info(inode->i_sb, function, line);
 573         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 574         if (IS_ERR(path))
 575                 path = "(unknown)";
 576         va_start(args, fmt);
 577         vaf.fmt = fmt;
 578         vaf.va = &args;
 579         if (block)
 580                 printk(KERN_CRIT
 581                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 582                        "block %llu: comm %s: path %s: %pV\n",
 583                        inode->i_sb->s_id, function, line, inode->i_ino,
 584                        block, current->comm, path, &vaf);
 585         else
 586                 printk(KERN_CRIT
 587                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 588                        "comm %s: path %s: %pV\n",
 589                        inode->i_sb->s_id, function, line, inode->i_ino,
 590                        current->comm, path, &vaf);
 591         va_end(args);
 592
 593         ext4_handle_error(inode->i_sb);
 594 }
 595
 596 static const char *ext4_decode_error(struct super_block *sb, int errno,
 597                                      char nbuf[16])
 598 {
 599         char *errstr = NULL;
 600
 601         switch (errno) {
 602         case -EIO:
 603                 errstr = "IO failure";
 604                 break;
 605         case -ENOMEM:
 606                 errstr = "Out of memory";
 607                 break;
 608         case -EROFS:
 609                 if (!sb || (EXT4_SB(sb)->s_journal &&
 610                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 611                         errstr = "Journal has aborted";
 612                 else
 613                         errstr = "Readonly filesystem";
 614                 break;
 615         default:
 616                 /* If the caller passed in an extra buffer for unknown
 617                  * errors, textualise them now.  Else we just return
 618                  * NULL. */
 619                 if (nbuf) {
 620                         /* Check for truncated error codes... */
 621                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 622                                 errstr = nbuf;
 623                 }
 624                 break;
 625         }
 626
 627         return errstr;
 628 }
 629
 630 /* __ext4_std_error decodes expected errors from journaling functions
 631  * automatically and invokes the appropriate error response.  */
 632
 633 void __ext4_std_error(struct super_block *sb, const char *function,
 634                       unsigned int line, int errno)
 635 {
 636         char nbuf[16];
 637         const char *errstr;
 638
 639         /* Special case: if the error is EROFS, and we're not already
 640          * inside a transaction, then there's really no point in logging
 641          * an error. */
 642         if (errno == -EROFS && journal_current_handle() == NULL &&
 643             (sb->s_flags & MS_RDONLY))
 644                 return;
 645
 646         errstr = ext4_decode_error(sb, errno, nbuf);
 647         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 648                sb->s_id, function, line, errstr);
 649         save_error_info(sb, function, line);
 650
 651         ext4_handle_error(sb);
 652 }
 653
 654 /*
 655  * ext4_abort is a much stronger failure handler than ext4_error.  The
 656  * abort function may be used to deal with unrecoverable failures such
 657  * as journal IO errors or ENOMEM at a critical moment in log management.
 658  *
 659  * We unconditionally force the filesystem into an ABORT|READONLY state,
 660  * unless the error response on the fs has been set to panic in which
 661  * case we take the easy way out and panic immediately.
 662  */
 663
 664 void __ext4_abort(struct super_block *sb, const char *function,
 665                 unsigned int line, const char *fmt, ...)
 666 {
 667         va_list args;
 668
 669         save_error_info(sb, function, line);
 670         va_start(args, fmt);
 671         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 672                function, line);
 673         vprintk(fmt, args);
 674         printk("\n");
 675         va_end(args);
 676
 677         if ((sb->s_flags & MS_RDONLY) == 0) {
 678                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 679                 sb->s_flags |= MS_RDONLY;
 680                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 681                 if (EXT4_SB(sb)->s_journal)
 682                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 683                 save_error_info(sb, function, line);
 684         }
 685         if (test_opt(sb, ERRORS_PANIC))
 686                 panic("EXT4-fs panic from previous error\n");
 687 }
 688
 689 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 690 {
 691         struct va_format vaf;
 692         va_list args;
 693
 694         va_start(args, fmt);
 695         vaf.fmt = fmt;
 696         vaf.va = &args;
 697         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 698         va_end(args);
 699 }
 700
 701 void __ext4_warning(struct super_block *sb, const char *function,
 702                     unsigned int line, const char *fmt, ...)
 703 {
 704         struct va_format vaf;
 705         va_list args;
 706
 707         va_start(args, fmt);
 708         vaf.fmt = fmt;
 709         vaf.va = &args;
 710         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 711                sb->s_id, function, line, &vaf);
 712         va_end(args);
 713 }
 714
 715 void __ext4_grp_locked_error(const char *function, unsigned int line,
 716                              struct super_block *sb, ext4_group_t grp,
 717                              unsigned long ino, ext4_fsblk_t block,
 718                              const char *fmt, ...)
 719 __releases(bitlock)
 720 __acquires(bitlock)
 721 {
 722         struct va_format vaf;
 723         va_list args;
 724         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 725
 726         es->s_last_error_ino = cpu_to_le32(ino);
 727         es->s_last_error_block = cpu_to_le64(block);
 728         __save_error_info(sb, function, line);
 729
 730         va_start(args, fmt);
 731
 732         vaf.fmt = fmt;
 733         vaf.va = &args;
 734         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 735                sb->s_id, function, line, grp);
 736         if (ino)
 737                 printk(KERN_CONT "inode %lu: ", ino);
 738         if (block)
 739                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 740         printk(KERN_CONT "%pV\n", &vaf);
 741         va_end(args);
 742
 743         if (test_opt(sb, ERRORS_CONT)) {
 744                 ext4_commit_super(sb, 0);
 745                 return;
 746         }
 747
 748         ext4_unlock_group(sb, grp);
 749         ext4_handle_error(sb);
 750         /*
 751          * We only get here in the ERRORS_RO case; relocking the group
 752          * may be dangerous, but nothing bad will happen since the
 753          * filesystem will have already been marked read/only and the
 754          * journal has been aborted.  We return 1 as a hint to callers
 755          * who might what to use the return value from
 756          * ext4_grp_locked_error() to distinguish between the
 757          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 758          * aggressively from the ext4 function in question, with a
 759          * more appropriate error code.
 760          */
 761         ext4_lock_group(sb, grp);
 762         return;
 763 }
 764
 765 void ext4_update_dynamic_rev(struct super_block *sb)
 766 {
 767         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 768
 769         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 770                 return;
 771
 772         ext4_warning(sb,
 773                      "updating to rev %d because of new feature flag, "
 774                      "running e2fsck is recommended",
 775                      EXT4_DYNAMIC_REV);
 776
 777         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 778         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 779         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 780         /* leave es->s_feature_*compat flags alone */
 781         /* es->s_uuid will be set by e2fsck if empty */
 782
 783         /*
 784          * The rest of the superblock fields should be zero, and if not it
 785          * means they are likely already in use, so leave them alone.  We
 786          * can leave it up to e2fsck to clean up any inconsistencies there.
 787          */
 788 }
 789
 790 /*
 791  * Open the external journal device
 792  */
 793 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 794 {
 795         struct block_device *bdev;
 796         char b[BDEVNAME_SIZE];
 797
 798         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 799         if (IS_ERR(bdev))
 800                 goto fail;
 801         return bdev;
 802
 803 fail:
 804         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 805                         __bdevname(dev, b), PTR_ERR(bdev));
 806         return NULL;
 807 }
 808
 809 /*
 810  * Release the journal device
 811  */
 812 static int ext4_blkdev_put(struct block_device *bdev)
 813 {
 814         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 815 }
 816
 817 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 818 {
 819         struct block_device *bdev;
 820         int ret = -ENODEV;
 821
 822         bdev = sbi->journal_bdev;
 823         if (bdev) {
 824                 ret = ext4_blkdev_put(bdev);
 825                 sbi->journal_bdev = NULL;
 826         }
 827         return ret;
 828 }
 829
 830 static inline struct inode *orphan_list_entry(struct list_head *l)
 831 {
 832         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 833 }
 834
 835 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 836 {
 837         struct list_head *l;
 838
 839         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 840                  le32_to_cpu(sbi->s_es->s_last_orphan));
 841
 842         printk(KERN_ERR "sb_info orphan list:\n");
 843         list_for_each(l, &sbi->s_orphan) {
 844                 struct inode *inode = orphan_list_entry(l);
 845                 printk(KERN_ERR "  "
 846                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 847                        inode->i_sb->s_id, inode->i_ino, inode,
 848                        inode->i_mode, inode->i_nlink,
 849                        NEXT_ORPHAN(inode));
 850         }
 851 }
 852
 853 static void ext4_put_super(struct super_block *sb)
 854 {
 855         struct ext4_sb_info *sbi = EXT4_SB(sb);
 856         struct ext4_super_block *es = sbi->s_es;
 857         int i, err;
 858
 859         ext4_unregister_li_request(sb);
 860         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 861
 862         flush_workqueue(sbi->dio_unwritten_wq);
 863         destroy_workqueue(sbi->dio_unwritten_wq);
 864
 865         lock_super(sb);
 866         if (sbi->s_journal) {
 867                 err = jbd2_journal_destroy(sbi->s_journal);
 868                 sbi->s_journal = NULL;
 869                 if (err < 0)
 870                         ext4_abort(sb, "Couldn't clean up the journal");
 871         }
 872
 873         del_timer(&sbi->s_err_report);
 874         ext4_release_system_zone(sb);
 875         ext4_mb_release(sb);
 876         ext4_ext_release(sb);
 877         ext4_xattr_put_super(sb);
 878
 879         if (!(sb->s_flags & MS_RDONLY)) {
 880                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 881                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 882         }
 883         if (sb->s_dirt || !(sb->s_flags & MS_RDONLY))
 884                 ext4_commit_super(sb, 1);
 885
 886         if (sbi->s_proc) {
 887                 remove_proc_entry("options", sbi->s_proc);
 888                 remove_proc_entry(sb->s_id, ext4_proc_root);
 889         }
 890         kobject_del(&sbi->s_kobj);
 891
 892         for (i = 0; i < sbi->s_gdb_count; i++)
 893                 brelse(sbi->s_group_desc[i]);
 894         ext4_kvfree(sbi->s_group_desc);
 895         ext4_kvfree(sbi->s_flex_groups);
 896         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 897         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 898         percpu_counter_destroy(&sbi->s_dirs_counter);
 899         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 900         brelse(sbi->s_sbh);
 901 #ifdef CONFIG_QUOTA
 902         for (i = 0; i < MAXQUOTAS; i++)
 903                 kfree(sbi->s_qf_names[i]);
 904 #endif
 905
 906         /* Debugging code just in case the in-memory inode orphan list
 907          * isn't empty.  The on-disk one can be non-empty if we've
 908          * detected an error and taken the fs readonly, but the
 909          * in-memory list had better be clean by this point. */
 910         if (!list_empty(&sbi->s_orphan))
 911                 dump_orphan_list(sb, sbi);
 912         J_ASSERT(list_empty(&sbi->s_orphan));
 913
 914         invalidate_bdev(sb->s_bdev);
 915         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 916                 /*
 917                  * Invalidate the journal device's buffers.  We don't want them
 918                  * floating about in memory - the physical journal device may
 919                  * hotswapped, and it breaks the `ro-after' testing code.
 920                  */
 921                 sync_blockdev(sbi->journal_bdev);
 922                 invalidate_bdev(sbi->journal_bdev);
 923                 ext4_blkdev_remove(sbi);
 924         }
 925         if (sbi->s_mmp_tsk)
 926                 kthread_stop(sbi->s_mmp_tsk);
 927         sb->s_fs_info = NULL;
 928         /*
 929          * Now that we are completely done shutting down the
 930          * superblock, we need to actually destroy the kobject.
 931          */
 932         unlock_super(sb);
 933         kobject_put(&sbi->s_kobj);
 934         wait_for_completion(&sbi->s_kobj_unregister);
 935         if (sbi->s_chksum_driver)
 936                 crypto_free_shash(sbi->s_chksum_driver);
 937         kfree(sbi->s_blockgroup_lock);
 938         kfree(sbi);
 939 }
 940
 941 static struct kmem_cache *ext4_inode_cachep;
 942
 943 /*
 944  * Called inside transaction, so use GFP_NOFS
 945  */
 946 static struct inode *ext4_alloc_inode(struct super_block *sb)
 947 {
 948         struct ext4_inode_info *ei;
 949
 950         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 951         if (!ei)
 952                 return NULL;
 953
 954         ei->vfs_inode.i_version = 1;
 955         ei->vfs_inode.i_data.writeback_index = 0;
 956         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 957         INIT_LIST_HEAD(&ei->i_prealloc_list);
 958         spin_lock_init(&ei->i_prealloc_lock);
 959         ei->i_reserved_data_blocks = 0;
 960         ei->i_reserved_meta_blocks = 0;
 961         ei->i_allocated_meta_blocks = 0;
 962         ei->i_da_metadata_calc_len = 0;
 963         spin_lock_init(&(ei->i_block_reservation_lock));
 964 #ifdef CONFIG_QUOTA
 965         ei->i_reserved_quota = 0;
 966 #endif
 967         ei->jinode = NULL;
 968         INIT_LIST_HEAD(&ei->i_completed_io_list);
 969         spin_lock_init(&ei->i_completed_io_lock);
 970         ei->cur_aio_dio = NULL;
 971         ei->i_sync_tid = 0;
 972         ei->i_datasync_tid = 0;
 973         atomic_set(&ei->i_ioend_count, 0);
 974         atomic_set(&ei->i_aiodio_unwritten, 0);
 975
 976         return &ei->vfs_inode;
 977 }
 978
 979 static int ext4_drop_inode(struct inode *inode)
 980 {
 981         int drop = generic_drop_inode(inode);
 982
 983         trace_ext4_drop_inode(inode, drop);
 984         return drop;
 985 }
 986
 987 static void ext4_i_callback(struct rcu_head *head)
 988 {
 989         struct inode *inode = container_of(head, struct inode, i_rcu);
 990         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 991 }
 992
 993 static void ext4_destroy_inode(struct inode *inode)
 994 {
 995         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 996                 ext4_msg(inode->i_sb, KERN_ERR,
 997                          "Inode %lu (%p): orphan list check failed!",
 998                          inode->i_ino, EXT4_I(inode));
 999                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
1000                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
1001                                 true);
1002                 dump_stack();
1003         }
1004         call_rcu(&inode->i_rcu, ext4_i_callback);
1005 }
1006
1007 static void init_once(void *foo)
1008 {
1009         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1010
1011         INIT_LIST_HEAD(&ei->i_orphan);
1012 #ifdef CONFIG_EXT4_FS_XATTR
1013         init_rwsem(&ei->xattr_sem);
1014 #endif
1015         init_rwsem(&ei->i_data_sem);
1016         inode_init_once(&ei->vfs_inode);
1017 }
1018
1019 static int init_inodecache(void)
1020 {
1021         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1022                                              sizeof(struct ext4_inode_info),
1023                                              0, (SLAB_RECLAIM_ACCOUNT|
1024                                                 SLAB_MEM_SPREAD),
1025                                              init_once);
1026         if (ext4_inode_cachep == NULL)
1027                 return -ENOMEM;
1028         return 0;
1029 }
1030
1031 static void destroy_inodecache(void)
1032 {
1033         kmem_cache_destroy(ext4_inode_cachep);
1034 }
1035
1036 void ext4_clear_inode(struct inode *inode)
1037 {
1038         invalidate_inode_buffers(inode);
1039         clear_inode(inode);
1040         dquot_drop(inode);
1041         ext4_discard_preallocations(inode);
1042         if (EXT4_I(inode)->jinode) {
1043                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1044                                                EXT4_I(inode)->jinode);
1045                 jbd2_free_inode(EXT4_I(inode)->jinode);
1046                 EXT4_I(inode)->jinode = NULL;
1047         }
1048 }
1049
1050 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1051                                         u64 ino, u32 generation)
1052 {
1053         struct inode *inode;
1054
1055         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1056                 return ERR_PTR(-ESTALE);
1057         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1058                 return ERR_PTR(-ESTALE);
1059
1060         /* iget isn't really right if the inode is currently unallocated!!
1061          *
1062          * ext4_read_inode will return a bad_inode if the inode had been
1063          * deleted, so we should be safe.
1064          *
1065          * Currently we don't know the generation for parent directory, so
1066          * a generation of 0 means "accept any"
1067          */
1068         inode = ext4_iget(sb, ino);
1069         if (IS_ERR(inode))
1070                 return ERR_CAST(inode);
1071         if (generation && inode->i_generation != generation) {
1072                 iput(inode);
1073                 return ERR_PTR(-ESTALE);
1074         }
1075
1076         return inode;
1077 }
1078
1079 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1080                                         int fh_len, int fh_type)
1081 {
1082         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1083                                     ext4_nfs_get_inode);
1084 }
1085
1086 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1087                                         int fh_len, int fh_type)
1088 {
1089         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1090                                     ext4_nfs_get_inode);
1091 }
1092
1093 /*
1094  * Try to release metadata pages (indirect blocks, directories) which are
1095  * mapped via the block device.  Since these pages could have journal heads
1096  * which would prevent try_to_free_buffers() from freeing them, we must use
1097  * jbd2 layer's try_to_free_buffers() function to release them.
1098  */
1099 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1100                                  gfp_t wait)
1101 {
1102         journal_t *journal = EXT4_SB(sb)->s_journal;
1103
1104         WARN_ON(PageChecked(page));
1105         if (!page_has_buffers(page))
1106                 return 0;
1107         if (journal)
1108                 return jbd2_journal_try_to_free_buffers(journal, page,
1109                                                         wait & ~__GFP_WAIT);
1110         return try_to_free_buffers(page);
1111 }
1112
1113 #ifdef CONFIG_QUOTA
1114 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1115 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1116
1117 static int ext4_write_dquot(struct dquot *dquot);
1118 static int ext4_acquire_dquot(struct dquot *dquot);
1119 static int ext4_release_dquot(struct dquot *dquot);
1120 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1121 static int ext4_write_info(struct super_block *sb, int type);
1122 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1123                          struct path *path);
1124 static int ext4_quota_off(struct super_block *sb, int type);
1125 static int ext4_quota_on_mount(struct super_block *sb, int type);
1126 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1127                                size_t len, loff_t off);
1128 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1129                                 const char *data, size_t len, loff_t off);
1130
1131 static const struct dquot_operations ext4_quota_operations = {
1132         .get_reserved_space = ext4_get_reserved_space,
1133         .write_dquot    = ext4_write_dquot,
1134         .acquire_dquot  = ext4_acquire_dquot,
1135         .release_dquot  = ext4_release_dquot,
1136         .mark_dirty     = ext4_mark_dquot_dirty,
1137         .write_info     = ext4_write_info,
1138         .alloc_dquot    = dquot_alloc,
1139         .destroy_dquot  = dquot_destroy,
1140 };
1141
1142 static const struct quotactl_ops ext4_qctl_operations = {
1143         .quota_on       = ext4_quota_on,
1144         .quota_off      = ext4_quota_off,
1145         .quota_sync     = dquot_quota_sync,
1146         .get_info       = dquot_get_dqinfo,
1147         .set_info       = dquot_set_dqinfo,
1148         .get_dqblk      = dquot_get_dqblk,
1149         .set_dqblk      = dquot_set_dqblk
1150 };
1151 #endif
1152
1153 static const struct super_operations ext4_sops = {
1154         .alloc_inode    = ext4_alloc_inode,
1155         .destroy_inode  = ext4_destroy_inode,
1156         .write_inode    = ext4_write_inode,
1157         .dirty_inode    = ext4_dirty_inode,
1158         .drop_inode     = ext4_drop_inode,
1159         .evict_inode    = ext4_evict_inode,
1160         .put_super      = ext4_put_super,
1161         .sync_fs        = ext4_sync_fs,
1162         .freeze_fs      = ext4_freeze,
1163         .unfreeze_fs    = ext4_unfreeze,
1164         .statfs         = ext4_statfs,
1165         .remount_fs     = ext4_remount,
1166         .show_options   = ext4_show_options,
1167 #ifdef CONFIG_QUOTA
1168         .quota_read     = ext4_quota_read,
1169         .quota_write    = ext4_quota_write,
1170 #endif
1171         .bdev_try_to_free_page = bdev_try_to_free_page,
1172 };
1173
1174 static const struct super_operations ext4_nojournal_sops = {
1175         .alloc_inode    = ext4_alloc_inode,
1176         .destroy_inode  = ext4_destroy_inode,
1177         .write_inode    = ext4_write_inode,
1178         .dirty_inode    = ext4_dirty_inode,
1179         .drop_inode     = ext4_drop_inode,
1180         .evict_inode    = ext4_evict_inode,
1181         .write_super    = ext4_write_super,
1182         .put_super      = ext4_put_super,
1183         .statfs         = ext4_statfs,
1184         .remount_fs     = ext4_remount,
1185         .show_options   = ext4_show_options,
1186 #ifdef CONFIG_QUOTA
1187         .quota_read     = ext4_quota_read,
1188         .quota_write    = ext4_quota_write,
1189 #endif
1190         .bdev_try_to_free_page = bdev_try_to_free_page,
1191 };
1192
1193 static const struct export_operations ext4_export_ops = {
1194         .fh_to_dentry = ext4_fh_to_dentry,
1195         .fh_to_parent = ext4_fh_to_parent,
1196         .get_parent = ext4_get_parent,
1197 };
1198
1199 enum {
1200         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1201         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1202         Opt_nouid32, Opt_debug, Opt_removed,
1203         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1204         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1205         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1206         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1207         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1208         Opt_data_err_abort, Opt_data_err_ignore,
1209         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1210         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1211         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1212         Opt_usrquota, Opt_grpquota, Opt_i_version,
1213         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1214         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1215         Opt_inode_readahead_blks, Opt_journal_ioprio,
1216         Opt_dioread_nolock, Opt_dioread_lock,
1217         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1218 };
1219
1220 static const match_table_t tokens = {
1221         {Opt_bsd_df, "bsddf"},
1222         {Opt_minix_df, "minixdf"},
1223         {Opt_grpid, "grpid"},
1224         {Opt_grpid, "bsdgroups"},
1225         {Opt_nogrpid, "nogrpid"},
1226         {Opt_nogrpid, "sysvgroups"},
1227         {Opt_resgid, "resgid=%u"},
1228         {Opt_resuid, "resuid=%u"},
1229         {Opt_sb, "sb=%u"},
1230         {Opt_err_cont, "errors=continue"},
1231         {Opt_err_panic, "errors=panic"},
1232         {Opt_err_ro, "errors=remount-ro"},
1233         {Opt_nouid32, "nouid32"},
1234         {Opt_debug, "debug"},
1235         {Opt_removed, "oldalloc"},
1236         {Opt_removed, "orlov"},
1237         {Opt_user_xattr, "user_xattr"},
1238         {Opt_nouser_xattr, "nouser_xattr"},
1239         {Opt_acl, "acl"},
1240         {Opt_noacl, "noacl"},
1241         {Opt_noload, "norecovery"},
1242         {Opt_noload, "noload"},
1243         {Opt_removed, "nobh"},
1244         {Opt_removed, "bh"},
1245         {Opt_commit, "commit=%u"},
1246         {Opt_min_batch_time, "min_batch_time=%u"},
1247         {Opt_max_batch_time, "max_batch_time=%u"},
1248         {Opt_journal_dev, "journal_dev=%u"},
1249         {Opt_journal_checksum, "journal_checksum"},
1250         {Opt_journal_async_commit, "journal_async_commit"},
1251         {Opt_abort, "abort"},
1252         {Opt_data_journal, "data=journal"},
1253         {Opt_data_ordered, "data=ordered"},
1254         {Opt_data_writeback, "data=writeback"},
1255         {Opt_data_err_abort, "data_err=abort"},
1256         {Opt_data_err_ignore, "data_err=ignore"},
1257         {Opt_offusrjquota, "usrjquota="},
1258         {Opt_usrjquota, "usrjquota=%s"},
1259         {Opt_offgrpjquota, "grpjquota="},
1260         {Opt_grpjquota, "grpjquota=%s"},
1261         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1262         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1263         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1264         {Opt_grpquota, "grpquota"},
1265         {Opt_noquota, "noquota"},
1266         {Opt_quota, "quota"},
1267         {Opt_usrquota, "usrquota"},
1268         {Opt_barrier, "barrier=%u"},
1269         {Opt_barrier, "barrier"},
1270         {Opt_nobarrier, "nobarrier"},
1271         {Opt_i_version, "i_version"},
1272         {Opt_stripe, "stripe=%u"},
1273         {Opt_delalloc, "delalloc"},
1274         {Opt_nodelalloc, "nodelalloc"},
1275         {Opt_mblk_io_submit, "mblk_io_submit"},
1276         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1277         {Opt_block_validity, "block_validity"},
1278         {Opt_noblock_validity, "noblock_validity"},
1279         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1280         {Opt_journal_ioprio, "journal_ioprio=%u"},
1281         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1282         {Opt_auto_da_alloc, "auto_da_alloc"},
1283         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1284         {Opt_dioread_nolock, "dioread_nolock"},
1285         {Opt_dioread_lock, "dioread_lock"},
1286         {Opt_discard, "discard"},
1287         {Opt_nodiscard, "nodiscard"},
1288         {Opt_init_itable, "init_itable=%u"},
1289         {Opt_init_itable, "init_itable"},
1290         {Opt_noinit_itable, "noinit_itable"},
1291         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1292         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1293         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1294         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1295         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1296         {Opt_err, NULL},
1297 };
1298
1299 static ext4_fsblk_t get_sb_block(void **data)
1300 {
1301         ext4_fsblk_t    sb_block;
1302         char            *options = (char *) *data;
1303
1304         if (!options || strncmp(options, "sb=", 3) != 0)
1305                 return 1;       /* Default location */
1306
1307         options += 3;
1308         /* TODO: use simple_strtoll with >32bit ext4 */
1309         sb_block = simple_strtoul(options, &options, 0);
1310         if (*options && *options != ',') {
1311                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1312                        (char *) *data);
1313                 return 1;
1314         }
1315         if (*options == ',')
1316                 options++;
1317         *data = (void *) options;
1318
1319         return sb_block;
1320 }
1321
1322 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1323 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1324         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1325
1326 #ifdef CONFIG_QUOTA
1327 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1328 {
1329         struct ext4_sb_info *sbi = EXT4_SB(sb);
1330         char *qname;
1331
1332         if (sb_any_quota_loaded(sb) &&
1333                 !sbi->s_qf_names[qtype]) {
1334                 ext4_msg(sb, KERN_ERR,
1335                         "Cannot change journaled "
1336                         "quota options when quota turned on");
1337                 return -1;
1338         }
1339         qname = match_strdup(args);
1340         if (!qname) {
1341                 ext4_msg(sb, KERN_ERR,
1342                         "Not enough memory for storing quotafile name");
1343                 return -1;
1344         }
1345         if (sbi->s_qf_names[qtype] &&
1346                 strcmp(sbi->s_qf_names[qtype], qname)) {
1347                 ext4_msg(sb, KERN_ERR,
1348                         "%s quota file already specified", QTYPE2NAME(qtype));
1349                 kfree(qname);
1350                 return -1;
1351         }
1352         sbi->s_qf_names[qtype] = qname;
1353         if (strchr(sbi->s_qf_names[qtype], '/')) {
1354                 ext4_msg(sb, KERN_ERR,
1355                         "quotafile must be on filesystem root");
1356                 kfree(sbi->s_qf_names[qtype]);
1357                 sbi->s_qf_names[qtype] = NULL;
1358                 return -1;
1359         }
1360         set_opt(sb, QUOTA);
1361         return 1;
1362 }
1363
1364 static int clear_qf_name(struct super_block *sb, int qtype)
1365 {
1366
1367         struct ext4_sb_info *sbi = EXT4_SB(sb);
1368
1369         if (sb_any_quota_loaded(sb) &&
1370                 sbi->s_qf_names[qtype]) {
1371                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1372                         " when quota turned on");
1373                 return -1;
1374         }
1375         /*
1376          * The space will be released later when all options are confirmed
1377          * to be correct
1378          */
1379         sbi->s_qf_names[qtype] = NULL;
1380         return 1;
1381 }
1382 #endif
1383
1384 #define MOPT_SET        0x0001
1385 #define MOPT_CLEAR      0x0002
1386 #define MOPT_NOSUPPORT  0x0004
1387 #define MOPT_EXPLICIT   0x0008
1388 #define MOPT_CLEAR_ERR  0x0010
1389 #define MOPT_GTE0       0x0020
1390 #ifdef CONFIG_QUOTA
1391 #define MOPT_Q          0
1392 #define MOPT_QFMT       0x0040
1393 #else
1394 #define MOPT_Q          MOPT_NOSUPPORT
1395 #define MOPT_QFMT       MOPT_NOSUPPORT
1396 #endif
1397 #define MOPT_DATAJ      0x0080
1398
1399 static const struct mount_opts {
1400         int     token;
1401         int     mount_opt;
1402         int     flags;
1403 } ext4_mount_opts[] = {
1404         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1405         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1406         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1407         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1408         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1409         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1410         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1411         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1412         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1413         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1414         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1415         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1416         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1417         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1418         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1419         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1420                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1421         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1422         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1423         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1424         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1425         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1426         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1427         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1428         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1429         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1430         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1431         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1432         {Opt_commit, 0, MOPT_GTE0},
1433         {Opt_max_batch_time, 0, MOPT_GTE0},
1434         {Opt_min_batch_time, 0, MOPT_GTE0},
1435         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1436         {Opt_init_itable, 0, MOPT_GTE0},
1437         {Opt_stripe, 0, MOPT_GTE0},
1438         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1439         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1440         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1441 #ifdef CONFIG_EXT4_FS_XATTR
1442         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1443         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1444 #else
1445         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1446         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1447 #endif
1448 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1449         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1450         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1451 #else
1452         {Opt_acl, 0, MOPT_NOSUPPORT},
1453         {Opt_noacl, 0, MOPT_NOSUPPORT},
1454 #endif
1455         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1456         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1457         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1458         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1459                                                         MOPT_SET | MOPT_Q},
1460         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1461                                                         MOPT_SET | MOPT_Q},
1462         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1463                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1464         {Opt_usrjquota, 0, MOPT_Q},
1465         {Opt_grpjquota, 0, MOPT_Q},
1466         {Opt_offusrjquota, 0, MOPT_Q},
1467         {Opt_offgrpjquota, 0, MOPT_Q},
1468         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1469         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1470         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1471         {Opt_err, 0, 0}
1472 };
1473
1474 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1475                             substring_t *args, unsigned long *journal_devnum,
1476                             unsigned int *journal_ioprio, int is_remount)
1477 {
1478         struct ext4_sb_info *sbi = EXT4_SB(sb);
1479         const struct mount_opts *m;
1480         kuid_t uid;
1481         kgid_t gid;
1482         int arg = 0;
1483
1484 #ifdef CONFIG_QUOTA
1485         if (token == Opt_usrjquota)
1486                 return set_qf_name(sb, USRQUOTA, &args[0]);
1487         else if (token == Opt_grpjquota)
1488                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1489         else if (token == Opt_offusrjquota)
1490                 return clear_qf_name(sb, USRQUOTA);
1491         else if (token == Opt_offgrpjquota)
1492                 return clear_qf_name(sb, GRPQUOTA);
1493 #endif
1494         if (args->from && match_int(args, &arg))
1495                 return -1;
1496         switch (token) {
1497         case Opt_noacl:
1498         case Opt_nouser_xattr:
1499                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1500                 break;
1501         case Opt_sb:
1502                 return 1;       /* handled by get_sb_block() */
1503         case Opt_removed:
1504                 ext4_msg(sb, KERN_WARNING,
1505                          "Ignoring removed %s option", opt);
1506                 return 1;
1507         case Opt_resuid:
1508                 uid = make_kuid(current_user_ns(), arg);
1509                 if (!uid_valid(uid)) {
1510                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1511                         return -1;
1512                 }
1513                 sbi->s_resuid = uid;
1514                 return 1;
1515         case Opt_resgid:
1516                 gid = make_kgid(current_user_ns(), arg);
1517                 if (!gid_valid(gid)) {
1518                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1519                         return -1;
1520                 }
1521                 sbi->s_resgid = gid;
1522                 return 1;
1523         case Opt_abort:
1524                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1525                 return 1;
1526         case Opt_i_version:
1527                 sb->s_flags |= MS_I_VERSION;
1528                 return 1;
1529         case Opt_journal_dev:
1530                 if (is_remount) {
1531                         ext4_msg(sb, KERN_ERR,
1532                                  "Cannot specify journal on remount");
1533                         return -1;
1534                 }
1535                 *journal_devnum = arg;
1536                 return 1;
1537         case Opt_journal_ioprio:
1538                 if (arg < 0 || arg > 7)
1539                         return -1;
1540                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1541                 return 1;
1542         }
1543
1544         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1545                 if (token != m->token)
1546                         continue;
1547                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1548                         return -1;
1549                 if (m->flags & MOPT_EXPLICIT)
1550                         set_opt2(sb, EXPLICIT_DELALLOC);
1551                 if (m->flags & MOPT_CLEAR_ERR)
1552                         clear_opt(sb, ERRORS_MASK);
1553                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1554                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1555                                  "options when quota turned on");
1556                         return -1;
1557                 }
1558
1559                 if (m->flags & MOPT_NOSUPPORT) {
1560                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1561                 } else if (token == Opt_commit) {
1562                         if (arg == 0)
1563                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1564                         sbi->s_commit_interval = HZ * arg;
1565                 } else if (token == Opt_max_batch_time) {
1566                         if (arg == 0)
1567                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1568                         sbi->s_max_batch_time = arg;
1569                 } else if (token == Opt_min_batch_time) {
1570                         sbi->s_min_batch_time = arg;
1571                 } else if (token == Opt_inode_readahead_blks) {
1572                         if (arg > (1 << 30))
1573                                 return -1;
1574                         if (arg && !is_power_of_2(arg)) {
1575                                 ext4_msg(sb, KERN_ERR,
1576                                          "EXT4-fs: inode_readahead_blks"
1577                                          " must be a power of 2");
1578                                 return -1;
1579                         }
1580                         sbi->s_inode_readahead_blks = arg;
1581                 } else if (token == Opt_init_itable) {
1582                         set_opt(sb, INIT_INODE_TABLE);
1583                         if (!args->from)
1584                                 arg = EXT4_DEF_LI_WAIT_MULT;
1585                         sbi->s_li_wait_mult = arg;
1586                 } else if (token == Opt_stripe) {
1587                         sbi->s_stripe = arg;
1588                 } else if (m->flags & MOPT_DATAJ) {
1589                         if (is_remount) {
1590                                 if (!sbi->s_journal)
1591                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1592                                 else if (test_opt(sb, DATA_FLAGS) !=
1593                                          m->mount_opt) {
1594                                         ext4_msg(sb, KERN_ERR,
1595                                          "Cannot change data mode on remount");
1596                                         return -1;
1597                                 }
1598                         } else {
1599                                 clear_opt(sb, DATA_FLAGS);
1600                                 sbi->s_mount_opt |= m->mount_opt;
1601                         }
1602 #ifdef CONFIG_QUOTA
1603                 } else if (m->flags & MOPT_QFMT) {
1604                         if (sb_any_quota_loaded(sb) &&
1605                             sbi->s_jquota_fmt != m->mount_opt) {
1606                                 ext4_msg(sb, KERN_ERR, "Cannot "
1607                                          "change journaled quota options "
1608                                          "when quota turned on");
1609                                 return -1;
1610                         }
1611                         sbi->s_jquota_fmt = m->mount_opt;
1612 #endif
1613                 } else {
1614                         if (!args->from)
1615                                 arg = 1;
1616                         if (m->flags & MOPT_CLEAR)
1617                                 arg = !arg;
1618                         else if (unlikely(!(m->flags & MOPT_SET))) {
1619                                 ext4_msg(sb, KERN_WARNING,
1620                                          "buggy handling of option %s", opt);
1621                                 WARN_ON(1);
1622                                 return -1;
1623                         }
1624                         if (arg != 0)
1625                                 sbi->s_mount_opt |= m->mount_opt;
1626                         else
1627                                 sbi->s_mount_opt &= ~m->mount_opt;
1628                 }
1629                 return 1;
1630         }
1631         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1632                  "or missing value", opt);
1633         return -1;
1634 }
1635
1636 static int parse_options(char *options, struct super_block *sb,
1637                          unsigned long *journal_devnum,
1638                          unsigned int *journal_ioprio,
1639                          int is_remount)
1640 {
1641 #ifdef CONFIG_QUOTA
1642         struct ext4_sb_info *sbi = EXT4_SB(sb);
1643 #endif
1644         char *p;
1645         substring_t args[MAX_OPT_ARGS];
1646         int token;
1647
1648         if (!options)
1649                 return 1;
1650
1651         while ((p = strsep(&options, ",")) != NULL) {
1652                 if (!*p)
1653                         continue;
1654                 /*
1655                  * Initialize args struct so we know whether arg was
1656                  * found; some options take optional arguments.
1657                  */
1658                 args[0].to = args[0].from = 0;
1659                 token = match_token(p, tokens, args);
1660                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1661                                      journal_ioprio, is_remount) < 0)
1662                         return 0;
1663         }
1664 #ifdef CONFIG_QUOTA
1665         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1666                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1667                         clear_opt(sb, USRQUOTA);
1668
1669                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1670                         clear_opt(sb, GRPQUOTA);
1671
1672                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1673                         ext4_msg(sb, KERN_ERR, "old and new quota "
1674                                         "format mixing");
1675                         return 0;
1676                 }
1677
1678                 if (!sbi->s_jquota_fmt) {
1679                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1680                                         "not specified");
1681                         return 0;
1682                 }
1683         } else {
1684                 if (sbi->s_jquota_fmt) {
1685                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1686                                         "specified with no journaling "
1687                                         "enabled");
1688                         return 0;
1689                 }
1690         }
1691 #endif
1692         return 1;
1693 }
1694
1695 static inline void ext4_show_quota_options(struct seq_file *seq,
1696                                            struct super_block *sb)
1697 {
1698 #if defined(CONFIG_QUOTA)
1699         struct ext4_sb_info *sbi = EXT4_SB(sb);
1700
1701         if (sbi->s_jquota_fmt) {
1702                 char *fmtname = "";
1703
1704                 switch (sbi->s_jquota_fmt) {
1705                 case QFMT_VFS_OLD:
1706                         fmtname = "vfsold";
1707                         break;
1708                 case QFMT_VFS_V0:
1709                         fmtname = "vfsv0";
1710                         break;
1711                 case QFMT_VFS_V1:
1712                         fmtname = "vfsv1";
1713                         break;
1714                 }
1715                 seq_printf(seq, ",jqfmt=%s", fmtname);
1716         }
1717
1718         if (sbi->s_qf_names[USRQUOTA])
1719                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1720
1721         if (sbi->s_qf_names[GRPQUOTA])
1722                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1723
1724         if (test_opt(sb, USRQUOTA))
1725                 seq_puts(seq, ",usrquota");
1726
1727         if (test_opt(sb, GRPQUOTA))
1728                 seq_puts(seq, ",grpquota");
1729 #endif
1730 }
1731
1732 static const char *token2str(int token)
1733 {
1734         static const struct match_token *t;
1735
1736         for (t = tokens; t->token != Opt_err; t++)
1737                 if (t->token == token && !strchr(t->pattern, '='))
1738                         break;
1739         return t->pattern;
1740 }
1741
1742 /*
1743  * Show an option if
1744  *  - it's set to a non-default value OR
1745  *  - if the per-sb default is different from the global default
1746  */
1747 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1748                               int nodefs)
1749 {
1750         struct ext4_sb_info *sbi = EXT4_SB(sb);
1751         struct ext4_super_block *es = sbi->s_es;
1752         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1753         const struct mount_opts *m;
1754         char sep = nodefs ? '\n' : ',';
1755
1756 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1757 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1758
1759         if (sbi->s_sb_block != 1)
1760                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1761
1762         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1763                 int want_set = m->flags & MOPT_SET;
1764                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1765                     (m->flags & MOPT_CLEAR_ERR))
1766                         continue;
1767                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1768                         continue; /* skip if same as the default */
1769                 if ((want_set &&
1770                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1771                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1772                         continue; /* select Opt_noFoo vs Opt_Foo */
1773                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1774         }
1775
1776         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1777             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1778                 SEQ_OPTS_PRINT("resuid=%u",
1779                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1780         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1781             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1782                 SEQ_OPTS_PRINT("resgid=%u",
1783                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1784         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1785         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1786                 SEQ_OPTS_PUTS("errors=remount-ro");
1787         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1788                 SEQ_OPTS_PUTS("errors=continue");
1789         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1790                 SEQ_OPTS_PUTS("errors=panic");
1791         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1792                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1793         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1794                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1795         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1796                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1797         if (sb->s_flags & MS_I_VERSION)
1798                 SEQ_OPTS_PUTS("i_version");
1799         if (nodefs || sbi->s_stripe)
1800                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1801         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1802                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1803                         SEQ_OPTS_PUTS("data=journal");
1804                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1805                         SEQ_OPTS_PUTS("data=ordered");
1806                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1807                         SEQ_OPTS_PUTS("data=writeback");
1808         }
1809         if (nodefs ||
1810             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1811                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1812                                sbi->s_inode_readahead_blks);
1813
1814         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1815                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1816                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1817
1818         ext4_show_quota_options(seq, sb);
1819         return 0;
1820 }
1821
1822 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1823 {
1824         return _ext4_show_options(seq, root->d_sb, 0);
1825 }
1826
1827 static int options_seq_show(struct seq_file *seq, void *offset)
1828 {
1829         struct super_block *sb = seq->private;
1830         int rc;
1831
1832         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1833         rc = _ext4_show_options(seq, sb, 1);
1834         seq_puts(seq, "\n");
1835         return rc;
1836 }
1837
1838 static int options_open_fs(struct inode *inode, struct file *file)
1839 {
1840         return single_open(file, options_seq_show, PDE(inode)->data);
1841 }
1842
1843 static const struct file_operations ext4_seq_options_fops = {
1844         .owner = THIS_MODULE,
1845         .open = options_open_fs,
1846         .read = seq_read,
1847         .llseek = seq_lseek,
1848         .release = single_release,
1849 };
1850
1851 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1852                             int read_only)
1853 {
1854         struct ext4_sb_info *sbi = EXT4_SB(sb);
1855         int res = 0;
1856
1857         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1858                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1859                          "forcing read-only mode");
1860                 res = MS_RDONLY;
1861         }
1862         if (read_only)
1863                 goto done;
1864         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1865                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1866                          "running e2fsck is recommended");
1867         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1868                 ext4_msg(sb, KERN_WARNING,
1869                          "warning: mounting fs with errors, "
1870                          "running e2fsck is recommended");
1871         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1872                  le16_to_cpu(es->s_mnt_count) >=
1873                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1874                 ext4_msg(sb, KERN_WARNING,
1875                          "warning: maximal mount count reached, "
1876                          "running e2fsck is recommended");
1877         else if (le32_to_cpu(es->s_checkinterval) &&
1878                 (le32_to_cpu(es->s_lastcheck) +
1879                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1880                 ext4_msg(sb, KERN_WARNING,
1881                          "warning: checktime reached, "
1882                          "running e2fsck is recommended");
1883         if (!sbi->s_journal)
1884                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1885         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1886                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1887         le16_add_cpu(&es->s_mnt_count, 1);
1888         es->s_mtime = cpu_to_le32(get_seconds());
1889         ext4_update_dynamic_rev(sb);
1890         if (sbi->s_journal)
1891                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1892
1893         ext4_commit_super(sb, 1);
1894 done:
1895         if (test_opt(sb, DEBUG))
1896                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1897                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1898                         sb->s_blocksize,
1899                         sbi->s_groups_count,
1900                         EXT4_BLOCKS_PER_GROUP(sb),
1901                         EXT4_INODES_PER_GROUP(sb),
1902                         sbi->s_mount_opt, sbi->s_mount_opt2);
1903
1904         cleancache_init_fs(sb);
1905         return res;
1906 }
1907
1908 static int ext4_fill_flex_info(struct super_block *sb)
1909 {
1910         struct ext4_sb_info *sbi = EXT4_SB(sb);
1911         struct ext4_group_desc *gdp = NULL;
1912         ext4_group_t flex_group_count;
1913         ext4_group_t flex_group;
1914         unsigned int groups_per_flex = 0;
1915         size_t size;
1916         int i;
1917
1918         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1919         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1920                 sbi->s_log_groups_per_flex = 0;
1921                 return 1;
1922         }
1923         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1924
1925         /* We allocate both existing and potentially added groups */
1926         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1927                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1928                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1929         size = flex_group_count * sizeof(struct flex_groups);
1930         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1931         if (sbi->s_flex_groups == NULL) {
1932                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1933                          flex_group_count);
1934                 goto failed;
1935         }
1936
1937         for (i = 0; i < sbi->s_groups_count; i++) {
1938                 gdp = ext4_get_group_desc(sb, i, NULL);
1939
1940                 flex_group = ext4_flex_group(sbi, i);
1941                 atomic_add(ext4_free_inodes_count(sb, gdp),
1942                            &sbi->s_flex_groups[flex_group].free_inodes);
1943                 atomic_add(ext4_free_group_clusters(sb, gdp),
1944                            &sbi->s_flex_groups[flex_group].free_clusters);
1945                 atomic_add(ext4_used_dirs_count(sb, gdp),
1946                            &sbi->s_flex_groups[flex_group].used_dirs);
1947         }
1948
1949         return 1;
1950 failed:
1951         return 0;
1952 }
1953
1954 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1955                                    struct ext4_group_desc *gdp)
1956 {
1957         int offset;
1958         __u16 crc = 0;
1959         __le32 le_group = cpu_to_le32(block_group);
1960
1961         if ((sbi->s_es->s_feature_ro_compat &
1962              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1963                 /* Use new metadata_csum algorithm */
1964                 __u16 old_csum;
1965                 __u32 csum32;
1966
1967                 old_csum = gdp->bg_checksum;
1968                 gdp->bg_checksum = 0;
1969                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1970                                      sizeof(le_group));
1971                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1972                                      sbi->s_desc_size);
1973                 gdp->bg_checksum = old_csum;
1974
1975                 crc = csum32 & 0xFFFF;
1976                 goto out;
1977         }
1978
1979         /* old crc16 code */
1980         offset = offsetof(struct ext4_group_desc, bg_checksum);
1981
1982         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1983         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1984         crc = crc16(crc, (__u8 *)gdp, offset);
1985         offset += sizeof(gdp->bg_checksum); /* skip checksum */
1986         /* for checksum of struct ext4_group_desc do the rest...*/
1987         if ((sbi->s_es->s_feature_incompat &
1988              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1989             offset < le16_to_cpu(sbi->s_es->s_desc_size))
1990                 crc = crc16(crc, (__u8 *)gdp + offset,
1991                             le16_to_cpu(sbi->s_es->s_desc_size) -
1992                                 offset);
1993
1994 out:
1995         return cpu_to_le16(crc);
1996 }
1997
1998 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
1999                                 struct ext4_group_desc *gdp)
2000 {
2001         if (ext4_has_group_desc_csum(sb) &&
2002             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2003                                                       block_group, gdp)))
2004                 return 0;
2005
2006         return 1;
2007 }
2008
2009 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2010                               struct ext4_group_desc *gdp)
2011 {
2012         if (!ext4_has_group_desc_csum(sb))
2013                 return;
2014         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2015 }
2016
2017 /* Called at mount-time, super-block is locked */
2018 static int ext4_check_descriptors(struct super_block *sb,
2019                                   ext4_group_t *first_not_zeroed)
2020 {
2021         struct ext4_sb_info *sbi = EXT4_SB(sb);
2022         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2023         ext4_fsblk_t last_block;
2024         ext4_fsblk_t block_bitmap;
2025         ext4_fsblk_t inode_bitmap;
2026         ext4_fsblk_t inode_table;
2027         int flexbg_flag = 0;
2028         ext4_group_t i, grp = sbi->s_groups_count;
2029
2030         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2031                 flexbg_flag = 1;
2032
2033         ext4_debug("Checking group descriptors");
2034
2035         for (i = 0; i < sbi->s_groups_count; i++) {
2036                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2037
2038                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2039                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2040                 else
2041                         last_block = first_block +
2042                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2043
2044                 if ((grp == sbi->s_groups_count) &&
2045                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2046                         grp = i;
2047
2048                 block_bitmap = ext4_block_bitmap(sb, gdp);
2049                 if (block_bitmap < first_block || block_bitmap > last_block) {
2050                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2051                                "Block bitmap for group %u not in group "
2052                                "(block %llu)!", i, block_bitmap);
2053                         return 0;
2054                 }
2055                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2056                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2057                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2058                                "Inode bitmap for group %u not in group "
2059                                "(block %llu)!", i, inode_bitmap);
2060                         return 0;
2061                 }
2062                 inode_table = ext4_inode_table(sb, gdp);
2063                 if (inode_table < first_block ||
2064                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2065                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2066                                "Inode table for group %u not in group "
2067                                "(block %llu)!", i, inode_table);
2068                         return 0;
2069                 }
2070                 ext4_lock_group(sb, i);
2071                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2072                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2073                                  "Checksum for group %u failed (%u!=%u)",
2074                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2075                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2076                         if (!(sb->s_flags & MS_RDONLY)) {
2077                                 ext4_unlock_group(sb, i);
2078                                 return 0;
2079                         }
2080                 }
2081                 ext4_unlock_group(sb, i);
2082                 if (!flexbg_flag)
2083                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2084         }
2085         if (NULL != first_not_zeroed)
2086                 *first_not_zeroed = grp;
2087
2088         ext4_free_blocks_count_set(sbi->s_es,
2089                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2090         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2091         return 1;
2092 }
2093
2094 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2095  * the superblock) which were deleted from all directories, but held open by
2096  * a process at the time of a crash.  We walk the list and try to delete these
2097  * inodes at recovery time (only with a read-write filesystem).
2098  *
2099  * In order to keep the orphan inode chain consistent during traversal (in
2100  * case of crash during recovery), we link each inode into the superblock
2101  * orphan list_head and handle it the same way as an inode deletion during
2102  * normal operation (which journals the operations for us).
2103  *
2104  * We only do an iget() and an iput() on each inode, which is very safe if we
2105  * accidentally point at an in-use or already deleted inode.  The worst that
2106  * can happen in this case is that we get a "bit already cleared" message from
2107  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2108  * e2fsck was run on this filesystem, and it must have already done the orphan
2109  * inode cleanup for us, so we can safely abort without any further action.
2110  */
2111 static void ext4_orphan_cleanup(struct super_block *sb,
2112                                 struct ext4_super_block *es)
2113 {
2114         unsigned int s_flags = sb->s_flags;
2115         int nr_orphans = 0, nr_truncates = 0;
2116 #ifdef CONFIG_QUOTA
2117         int i;
2118 #endif
2119         if (!es->s_last_orphan) {
2120                 jbd_debug(4, "no orphan inodes to clean up\n");
2121                 return;
2122         }
2123
2124         if (bdev_read_only(sb->s_bdev)) {
2125                 ext4_msg(sb, KERN_ERR, "write access "
2126                         "unavailable, skipping orphan cleanup");
2127                 return;
2128         }
2129
2130         /* Check if feature set would not allow a r/w mount */
2131         if (!ext4_feature_set_ok(sb, 0)) {
2132                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2133                          "unknown ROCOMPAT features");
2134                 return;
2135         }
2136
2137         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2138                 if (es->s_last_orphan)
2139                         jbd_debug(1, "Errors on filesystem, "
2140                                   "clearing orphan list.\n");
2141                 es->s_last_orphan = 0;
2142                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2143                 return;
2144         }
2145
2146         if (s_flags & MS_RDONLY) {
2147                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2148                 sb->s_flags &= ~MS_RDONLY;
2149         }
2150 #ifdef CONFIG_QUOTA
2151         /* Needed for iput() to work correctly and not trash data */
2152         sb->s_flags |= MS_ACTIVE;
2153         /* Turn on quotas so that they are updated correctly */
2154         for (i = 0; i < MAXQUOTAS; i++) {
2155                 if (EXT4_SB(sb)->s_qf_names[i]) {
2156                         int ret = ext4_quota_on_mount(sb, i);
2157                         if (ret < 0)
2158                                 ext4_msg(sb, KERN_ERR,
2159                                         "Cannot turn on journaled "
2160                                         "quota: error %d", ret);
2161                 }
2162         }
2163 #endif
2164
2165         while (es->s_last_orphan) {
2166                 struct inode *inode;
2167
2168                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2169                 if (IS_ERR(inode)) {
2170                         es->s_last_orphan = 0;
2171                         break;
2172                 }
2173
2174                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2175                 dquot_initialize(inode);
2176                 if (inode->i_nlink) {
2177                         ext4_msg(sb, KERN_DEBUG,
2178                                 "%s: truncating inode %lu to %lld bytes",
2179                                 __func__, inode->i_ino, inode->i_size);
2180                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2181                                   inode->i_ino, inode->i_size);
2182                         ext4_truncate(inode);
2183                         nr_truncates++;
2184                 } else {
2185                         ext4_msg(sb, KERN_DEBUG,
2186                                 "%s: deleting unreferenced inode %lu",
2187                                 __func__, inode->i_ino);
2188                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2189                                   inode->i_ino);
2190                         nr_orphans++;
2191                 }
2192                 iput(inode);  /* The delete magic happens here! */
2193         }
2194
2195 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2196
2197         if (nr_orphans)
2198                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2199                        PLURAL(nr_orphans));
2200         if (nr_truncates)
2201                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2202                        PLURAL(nr_truncates));
2203 #ifdef CONFIG_QUOTA
2204         /* Turn quotas off */
2205         for (i = 0; i < MAXQUOTAS; i++) {
2206                 if (sb_dqopt(sb)->files[i])
2207                         dquot_quota_off(sb, i);
2208         }
2209 #endif
2210         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2211 }
2212
2213 /*
2214  * Maximal extent format file size.
2215  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2216  * extent format containers, within a sector_t, and within i_blocks
2217  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2218  * so that won't be a limiting factor.
2219  *
2220  * However there is other limiting factor. We do store extents in the form
2221  * of starting block and length, hence the resulting length of the extent
2222  * covering maximum file size must fit into on-disk format containers as
2223  * well. Given that length is always by 1 unit bigger than max unit (because
2224  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2225  *
2226  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2227  */
2228 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2229 {
2230         loff_t res;
2231         loff_t upper_limit = MAX_LFS_FILESIZE;
2232
2233         /* small i_blocks in vfs inode? */
2234         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2235                 /*
2236                  * CONFIG_LBDAF is not enabled implies the inode
2237                  * i_block represent total blocks in 512 bytes
2238                  * 32 == size of vfs inode i_blocks * 8
2239                  */
2240                 upper_limit = (1LL << 32) - 1;
2241
2242                 /* total blocks in file system block size */
2243                 upper_limit >>= (blkbits - 9);
2244                 upper_limit <<= blkbits;
2245         }
2246
2247         /*
2248          * 32-bit extent-start container, ee_block. We lower the maxbytes
2249          * by one fs block, so ee_len can cover the extent of maximum file
2250          * size
2251          */
2252         res = (1LL << 32) - 1;
2253         res <<= blkbits;
2254
2255         /* Sanity check against vm- & vfs- imposed limits */
2256         if (res > upper_limit)
2257                 res = upper_limit;
2258
2259         return res;
2260 }
2261
2262 /*
2263  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2264  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2265  * We need to be 1 filesystem block less than the 2^48 sector limit.
2266  */
2267 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2268 {
2269         loff_t res = EXT4_NDIR_BLOCKS;
2270         int meta_blocks;
2271         loff_t upper_limit;
2272         /* This is calculated to be the largest file size for a dense, block
2273          * mapped file such that the file's total number of 512-byte sectors,
2274          * including data and all indirect blocks, does not exceed (2^48 - 1).
2275          *
2276          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2277          * number of 512-byte sectors of the file.
2278          */
2279
2280         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2281                 /*
2282                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2283                  * the inode i_block field represents total file blocks in
2284                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2285                  */
2286                 upper_limit = (1LL << 32) - 1;
2287
2288                 /* total blocks in file system block size */
2289                 upper_limit >>= (bits - 9);
2290
2291         } else {
2292                 /*
2293                  * We use 48 bit ext4_inode i_blocks
2294                  * With EXT4_HUGE_FILE_FL set the i_blocks
2295                  * represent total number of blocks in
2296                  * file system block size
2297                  */
2298                 upper_limit = (1LL << 48) - 1;
2299
2300         }
2301
2302         /* indirect blocks */
2303         meta_blocks = 1;
2304         /* double indirect blocks */
2305         meta_blocks += 1 + (1LL << (bits-2));
2306         /* tripple indirect blocks */
2307         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2308
2309         upper_limit -= meta_blocks;
2310         upper_limit <<= bits;
2311
2312         res += 1LL << (bits-2);
2313         res += 1LL << (2*(bits-2));
2314         res += 1LL << (3*(bits-2));
2315         res <<= bits;
2316         if (res > upper_limit)
2317                 res = upper_limit;
2318
2319         if (res > MAX_LFS_FILESIZE)
2320                 res = MAX_LFS_FILESIZE;
2321
2322         return res;
2323 }
2324
2325 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2326                                    ext4_fsblk_t logical_sb_block, int nr)
2327 {
2328         struct ext4_sb_info *sbi = EXT4_SB(sb);
2329         ext4_group_t bg, first_meta_bg;
2330         int has_super = 0;
2331
2332         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2333
2334         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2335             nr < first_meta_bg)
2336                 return logical_sb_block + nr + 1;
2337         bg = sbi->s_desc_per_block * nr;
2338         if (ext4_bg_has_super(sb, bg))
2339                 has_super = 1;
2340
2341         return (has_super + ext4_group_first_block_no(sb, bg));
2342 }
2343
2344 /**
2345  * ext4_get_stripe_size: Get the stripe size.
2346  * @sbi: In memory super block info
2347  *
2348  * If we have specified it via mount option, then
2349  * use the mount option value. If the value specified at mount time is
2350  * greater than the blocks per group use the super block value.
2351  * If the super block value is greater than blocks per group return 0.
2352  * Allocator needs it be less than blocks per group.
2353  *
2354  */
2355 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2356 {
2357         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2358         unsigned long stripe_width =
2359                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2360         int ret;
2361
2362         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2363                 ret = sbi->s_stripe;
2364         else if (stripe_width <= sbi->s_blocks_per_group)
2365                 ret = stripe_width;
2366         else if (stride <= sbi->s_blocks_per_group)
2367                 ret = stride;
2368         else
2369                 ret = 0;
2370
2371         /*
2372          * If the stripe width is 1, this makes no sense and
2373          * we set it to 0 to turn off stripe handling code.
2374          */
2375         if (ret <= 1)
2376                 ret = 0;
2377
2378         return ret;
2379 }
2380
2381 /* sysfs supprt */
2382
2383 struct ext4_attr {
2384         struct attribute attr;
2385         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2386         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2387                          const char *, size_t);
2388         int offset;
2389 };
2390
2391 static int parse_strtoul(const char *buf,
2392                 unsigned long max, unsigned long *value)
2393 {
2394         char *endp;
2395
2396         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2397         endp = skip_spaces(endp);
2398         if (*endp || *value > max)
2399                 return -EINVAL;
2400
2401         return 0;
2402 }
2403
2404 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2405                                               struct ext4_sb_info *sbi,
2406                                               char *buf)
2407 {
2408         return snprintf(buf, PAGE_SIZE, "%llu\n",
2409                 (s64) EXT4_C2B(sbi,
2410                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2411 }
2412
2413 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2414                                          struct ext4_sb_info *sbi, char *buf)
2415 {
2416         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2417
2418         if (!sb->s_bdev->bd_part)
2419                 return snprintf(buf, PAGE_SIZE, "0\n");
2420         return snprintf(buf, PAGE_SIZE, "%lu\n",
2421                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2422                          sbi->s_sectors_written_start) >> 1);
2423 }
2424
2425 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2426                                           struct ext4_sb_info *sbi, char *buf)
2427 {
2428         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2429
2430         if (!sb->s_bdev->bd_part)
2431                 return snprintf(buf, PAGE_SIZE, "0\n");
2432         return snprintf(buf, PAGE_SIZE, "%llu\n",
2433                         (unsigned long long)(sbi->s_kbytes_written +
2434                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2435                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2436 }
2437
2438 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2439                                           struct ext4_sb_info *sbi,
2440                                           const char *buf, size_t count)
2441 {
2442         unsigned long t;
2443
2444         if (parse_strtoul(buf, 0x40000000, &t))
2445                 return -EINVAL;
2446
2447         if (t && !is_power_of_2(t))
2448                 return -EINVAL;
2449
2450         sbi->s_inode_readahead_blks = t;
2451         return count;
2452 }
2453
2454 static ssize_t sbi_ui_show(struct ext4_attr *a,
2455                            struct ext4_sb_info *sbi, char *buf)
2456 {
2457         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2458
2459         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2460 }
2461
2462 static ssize_t sbi_ui_store(struct ext4_attr *a,
2463                             struct ext4_sb_info *sbi,
2464                             const char *buf, size_t count)
2465 {
2466         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2467         unsigned long t;
2468
2469         if (parse_strtoul(buf, 0xffffffff, &t))
2470                 return -EINVAL;
2471         *ui = t;
2472         return count;
2473 }
2474
2475 static ssize_t trigger_test_error(struct ext4_attr *a,
2476                                   struct ext4_sb_info *sbi,
2477                                   const char *buf, size_t count)
2478 {
2479         int len = count;
2480
2481         if (!capable(CAP_SYS_ADMIN))
2482                 return -EPERM;
2483
2484         if (len && buf[len-1] == '\n')
2485                 len--;
2486
2487         if (len)
2488                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2489         return count;
2490 }
2491
2492 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2493 static struct ext4_attr ext4_attr_##_name = {                   \
2494         .attr = {.name = __stringify(_name), .mode = _mode },   \
2495         .show   = _show,                                        \
2496         .store  = _store,                                       \
2497         .offset = offsetof(struct ext4_sb_info, _elname),       \
2498 }
2499 #define EXT4_ATTR(name, mode, show, store) \
2500 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2501
2502 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2503 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2504 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2505 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2506         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2507 #define ATTR_LIST(name) &ext4_attr_##name.attr
2508
2509 EXT4_RO_ATTR(delayed_allocation_blocks);
2510 EXT4_RO_ATTR(session_write_kbytes);
2511 EXT4_RO_ATTR(lifetime_write_kbytes);
2512 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2513                  inode_readahead_blks_store, s_inode_readahead_blks);
2514 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2515 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2516 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2517 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2518 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2519 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2520 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2521 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2522 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2523
2524 static struct attribute *ext4_attrs[] = {
2525         ATTR_LIST(delayed_allocation_blocks),
2526         ATTR_LIST(session_write_kbytes),
2527         ATTR_LIST(lifetime_write_kbytes),
2528         ATTR_LIST(inode_readahead_blks),
2529         ATTR_LIST(inode_goal),
2530         ATTR_LIST(mb_stats),
2531         ATTR_LIST(mb_max_to_scan),
2532         ATTR_LIST(mb_min_to_scan),
2533         ATTR_LIST(mb_order2_req),
2534         ATTR_LIST(mb_stream_req),
2535         ATTR_LIST(mb_group_prealloc),
2536         ATTR_LIST(max_writeback_mb_bump),
2537         ATTR_LIST(trigger_fs_error),
2538         NULL,
2539 };
2540
2541 /* Features this copy of ext4 supports */
2542 EXT4_INFO_ATTR(lazy_itable_init);
2543 EXT4_INFO_ATTR(batched_discard);
2544
2545 static struct attribute *ext4_feat_attrs[] = {
2546         ATTR_LIST(lazy_itable_init),
2547         ATTR_LIST(batched_discard),
2548         NULL,
2549 };
2550
2551 static ssize_t ext4_attr_show(struct kobject *kobj,
2552                               struct attribute *attr, char *buf)
2553 {
2554         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2555                                                 s_kobj);
2556         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2557
2558         return a->show ? a->show(a, sbi, buf) : 0;
2559 }
2560
2561 static ssize_t ext4_attr_store(struct kobject *kobj,
2562                                struct attribute *attr,
2563                                const char *buf, size_t len)
2564 {
2565         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2566                                                 s_kobj);
2567         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2568
2569         return a->store ? a->store(a, sbi, buf, len) : 0;
2570 }
2571
2572 static void ext4_sb_release(struct kobject *kobj)
2573 {
2574         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2575                                                 s_kobj);
2576         complete(&sbi->s_kobj_unregister);
2577 }
2578
2579 static const struct sysfs_ops ext4_attr_ops = {
2580         .show   = ext4_attr_show,
2581         .store  = ext4_attr_store,
2582 };
2583
2584 static struct kobj_type ext4_ktype = {
2585         .default_attrs  = ext4_attrs,
2586         .sysfs_ops      = &ext4_attr_ops,
2587         .release        = ext4_sb_release,
2588 };
2589
2590 static void ext4_feat_release(struct kobject *kobj)
2591 {
2592         complete(&ext4_feat->f_kobj_unregister);
2593 }
2594
2595 static struct kobj_type ext4_feat_ktype = {
2596         .default_attrs  = ext4_feat_attrs,
2597         .sysfs_ops      = &ext4_attr_ops,
2598         .release        = ext4_feat_release,
2599 };
2600
2601 /*
2602  * Check whether this filesystem can be mounted based on
2603  * the features present and the RDONLY/RDWR mount requested.
2604  * Returns 1 if this filesystem can be mounted as requested,
2605  * 0 if it cannot be.
2606  */
2607 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2608 {
2609         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2610                 ext4_msg(sb, KERN_ERR,
2611                         "Couldn't mount because of "
2612                         "unsupported optional features (%x)",
2613                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2614                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2615                 return 0;
2616         }
2617
2618         if (readonly)
2619                 return 1;
2620
2621         /* Check that feature set is OK for a read-write mount */
2622         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2623                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2624                          "unsupported optional features (%x)",
2625                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2626                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2627                 return 0;
2628         }
2629         /*
2630          * Large file size enabled file system can only be mounted
2631          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2632          */
2633         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2634                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2635                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2636                                  "cannot be mounted RDWR without "
2637                                  "CONFIG_LBDAF");
2638                         return 0;
2639                 }
2640         }
2641         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2642             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2643                 ext4_msg(sb, KERN_ERR,
2644                          "Can't support bigalloc feature without "
2645                          "extents feature\n");
2646                 return 0;
2647         }
2648         return 1;
2649 }
2650
2651 /*
2652  * This function is called once a day if we have errors logged
2653  * on the file system
2654  */
2655 static void print_daily_error_info(unsigned long arg)
2656 {
2657         struct super_block *sb = (struct super_block *) arg;
2658         struct ext4_sb_info *sbi;
2659         struct ext4_super_block *es;
2660
2661         sbi = EXT4_SB(sb);
2662         es = sbi->s_es;
2663
2664         if (es->s_error_count)
2665                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2666                          le32_to_cpu(es->s_error_count));
2667         if (es->s_first_error_time) {
2668                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2669                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2670                        (int) sizeof(es->s_first_error_func),
2671                        es->s_first_error_func,
2672                        le32_to_cpu(es->s_first_error_line));
2673                 if (es->s_first_error_ino)
2674                         printk(": inode %u",
2675                                le32_to_cpu(es->s_first_error_ino));
2676                 if (es->s_first_error_block)
2677                         printk(": block %llu", (unsigned long long)
2678                                le64_to_cpu(es->s_first_error_block));
2679                 printk("\n");
2680         }
2681         if (es->s_last_error_time) {
2682                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2683                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2684                        (int) sizeof(es->s_last_error_func),
2685                        es->s_last_error_func,
2686                        le32_to_cpu(es->s_last_error_line));
2687                 if (es->s_last_error_ino)
2688                         printk(": inode %u",
2689                                le32_to_cpu(es->s_last_error_ino));
2690                 if (es->s_last_error_block)
2691                         printk(": block %llu", (unsigned long long)
2692                                le64_to_cpu(es->s_last_error_block));
2693                 printk("\n");
2694         }
2695         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2696 }
2697
2698 /* Find next suitable group and run ext4_init_inode_table */
2699 static int ext4_run_li_request(struct ext4_li_request *elr)
2700 {
2701         struct ext4_group_desc *gdp = NULL;
2702         ext4_group_t group, ngroups;
2703         struct super_block *sb;
2704         unsigned long timeout = 0;
2705         int ret = 0;
2706
2707         sb = elr->lr_super;
2708         ngroups = EXT4_SB(sb)->s_groups_count;
2709
2710         sb_start_write(sb);
2711         for (group = elr->lr_next_group; group < ngroups; group++) {
2712                 gdp = ext4_get_group_desc(sb, group, NULL);
2713                 if (!gdp) {
2714                         ret = 1;
2715                         break;
2716                 }
2717
2718                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2719                         break;
2720         }
2721
2722         if (group == ngroups)
2723                 ret = 1;
2724
2725         if (!ret) {
2726                 timeout = jiffies;
2727                 ret = ext4_init_inode_table(sb, group,
2728                                             elr->lr_timeout ? 0 : 1);
2729                 if (elr->lr_timeout == 0) {
2730                         timeout = (jiffies - timeout) *
2731                                   elr->lr_sbi->s_li_wait_mult;
2732                         elr->lr_timeout = timeout;
2733                 }
2734                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2735                 elr->lr_next_group = group + 1;
2736         }
2737         sb_end_write(sb);
2738
2739         return ret;
2740 }
2741
2742 /*
2743  * Remove lr_request from the list_request and free the
2744  * request structure. Should be called with li_list_mtx held
2745  */
2746 static void ext4_remove_li_request(struct ext4_li_request *elr)
2747 {
2748         struct ext4_sb_info *sbi;
2749
2750         if (!elr)
2751                 return;
2752
2753         sbi = elr->lr_sbi;
2754
2755         list_del(&elr->lr_request);
2756         sbi->s_li_request = NULL;
2757         kfree(elr);
2758 }
2759
2760 static void ext4_unregister_li_request(struct super_block *sb)
2761 {
2762         mutex_lock(&ext4_li_mtx);
2763         if (!ext4_li_info) {
2764                 mutex_unlock(&ext4_li_mtx);
2765                 return;
2766         }
2767
2768         mutex_lock(&ext4_li_info->li_list_mtx);
2769         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2770         mutex_unlock(&ext4_li_info->li_list_mtx);
2771         mutex_unlock(&ext4_li_mtx);
2772 }
2773
2774 static struct task_struct *ext4_lazyinit_task;
2775
2776 /*
2777  * This is the function where ext4lazyinit thread lives. It walks
2778  * through the request list searching for next scheduled filesystem.
2779  * When such a fs is found, run the lazy initialization request
2780  * (ext4_rn_li_request) and keep track of the time spend in this
2781  * function. Based on that time we compute next schedule time of
2782  * the request. When walking through the list is complete, compute
2783  * next waking time and put itself into sleep.
2784  */
2785 static int ext4_lazyinit_thread(void *arg)
2786 {
2787         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2788         struct list_head *pos, *n;
2789         struct ext4_li_request *elr;
2790         unsigned long next_wakeup, cur;
2791
2792         BUG_ON(NULL == eli);
2793
2794 cont_thread:
2795         while (true) {
2796                 next_wakeup = MAX_JIFFY_OFFSET;
2797
2798                 mutex_lock(&eli->li_list_mtx);
2799                 if (list_empty(&eli->li_request_list)) {
2800                         mutex_unlock(&eli->li_list_mtx);
2801                         goto exit_thread;
2802                 }
2803
2804                 list_for_each_safe(pos, n, &eli->li_request_list) {
2805                         elr = list_entry(pos, struct ext4_li_request,
2806                                          lr_request);
2807
2808                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2809                                 if (ext4_run_li_request(elr) != 0) {
2810                                         /* error, remove the lazy_init job */
2811                                         ext4_remove_li_request(elr);
2812                                         continue;
2813                                 }
2814                         }
2815
2816                         if (time_before(elr->lr_next_sched, next_wakeup))
2817                                 next_wakeup = elr->lr_next_sched;
2818                 }
2819                 mutex_unlock(&eli->li_list_mtx);
2820
2821                 try_to_freeze();
2822
2823                 cur = jiffies;
2824                 if ((time_after_eq(cur, next_wakeup)) ||
2825                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2826                         cond_resched();
2827                         continue;
2828                 }
2829
2830                 schedule_timeout_interruptible(next_wakeup - cur);
2831
2832                 if (kthread_should_stop()) {
2833                         ext4_clear_request_list();
2834                         goto exit_thread;
2835                 }
2836         }
2837
2838 exit_thread:
2839         /*
2840          * It looks like the request list is empty, but we need
2841          * to check it under the li_list_mtx lock, to prevent any
2842          * additions into it, and of course we should lock ext4_li_mtx
2843          * to atomically free the list and ext4_li_info, because at
2844          * this point another ext4 filesystem could be registering
2845          * new one.
2846          */
2847         mutex_lock(&ext4_li_mtx);
2848         mutex_lock(&eli->li_list_mtx);
2849         if (!list_empty(&eli->li_request_list)) {
2850                 mutex_unlock(&eli->li_list_mtx);
2851                 mutex_unlock(&ext4_li_mtx);
2852                 goto cont_thread;
2853         }
2854         mutex_unlock(&eli->li_list_mtx);
2855         kfree(ext4_li_info);
2856         ext4_li_info = NULL;
2857         mutex_unlock(&ext4_li_mtx);
2858
2859         return 0;
2860 }
2861
2862 static void ext4_clear_request_list(void)
2863 {
2864         struct list_head *pos, *n;
2865         struct ext4_li_request *elr;
2866
2867         mutex_lock(&ext4_li_info->li_list_mtx);
2868         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2869                 elr = list_entry(pos, struct ext4_li_request,
2870                                  lr_request);
2871                 ext4_remove_li_request(elr);
2872         }
2873         mutex_unlock(&ext4_li_info->li_list_mtx);
2874 }
2875
2876 static int ext4_run_lazyinit_thread(void)
2877 {
2878         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2879                                          ext4_li_info, "ext4lazyinit");
2880         if (IS_ERR(ext4_lazyinit_task)) {
2881                 int err = PTR_ERR(ext4_lazyinit_task);
2882                 ext4_clear_request_list();
2883                 kfree(ext4_li_info);
2884                 ext4_li_info = NULL;
2885                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2886                                  "initialization thread\n",
2887                                  err);
2888                 return err;
2889         }
2890         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2891         return 0;
2892 }
2893
2894 /*
2895  * Check whether it make sense to run itable init. thread or not.
2896  * If there is at least one uninitialized inode table, return
2897  * corresponding group number, else the loop goes through all
2898  * groups and return total number of groups.
2899  */
2900 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2901 {
2902         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2903         struct ext4_group_desc *gdp = NULL;
2904
2905         for (group = 0; group < ngroups; group++) {
2906                 gdp = ext4_get_group_desc(sb, group, NULL);
2907                 if (!gdp)
2908                         continue;
2909
2910                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2911                         break;
2912         }
2913
2914         return group;
2915 }
2916
2917 static int ext4_li_info_new(void)
2918 {
2919         struct ext4_lazy_init *eli = NULL;
2920
2921         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2922         if (!eli)
2923                 return -ENOMEM;
2924
2925         INIT_LIST_HEAD(&eli->li_request_list);
2926         mutex_init(&eli->li_list_mtx);
2927
2928         eli->li_state |= EXT4_LAZYINIT_QUIT;
2929
2930         ext4_li_info = eli;
2931
2932         return 0;
2933 }
2934
2935 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2936                                             ext4_group_t start)
2937 {
2938         struct ext4_sb_info *sbi = EXT4_SB(sb);
2939         struct ext4_li_request *elr;
2940         unsigned long rnd;
2941
2942         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2943         if (!elr)
2944                 return NULL;
2945
2946         elr->lr_super = sb;
2947         elr->lr_sbi = sbi;
2948         elr->lr_next_group = start;
2949
2950         /*
2951          * Randomize first schedule time of the request to
2952          * spread the inode table initialization requests
2953          * better.
2954          */
2955         get_random_bytes(&rnd, sizeof(rnd));
2956         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2957                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2958
2959         return elr;
2960 }
2961
2962 static int ext4_register_li_request(struct super_block *sb,
2963                                     ext4_group_t first_not_zeroed)
2964 {
2965         struct ext4_sb_info *sbi = EXT4_SB(sb);
2966         struct ext4_li_request *elr;
2967         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2968         int ret = 0;
2969
2970         if (sbi->s_li_request != NULL) {
2971                 /*
2972                  * Reset timeout so it can be computed again, because
2973                  * s_li_wait_mult might have changed.
2974                  */
2975                 sbi->s_li_request->lr_timeout = 0;
2976                 return 0;
2977         }
2978
2979         if (first_not_zeroed == ngroups ||
2980             (sb->s_flags & MS_RDONLY) ||
2981             !test_opt(sb, INIT_INODE_TABLE))
2982                 return 0;
2983
2984         elr = ext4_li_request_new(sb, first_not_zeroed);
2985         if (!elr)
2986                 return -ENOMEM;
2987
2988         mutex_lock(&ext4_li_mtx);
2989
2990         if (NULL == ext4_li_info) {
2991                 ret = ext4_li_info_new();
2992                 if (ret)
2993                         goto out;
2994         }
2995
2996         mutex_lock(&ext4_li_info->li_list_mtx);
2997         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2998         mutex_unlock(&ext4_li_info->li_list_mtx);
2999
3000         sbi->s_li_request = elr;
3001         /*
3002          * set elr to NULL here since it has been inserted to
3003          * the request_list and the removal and free of it is
3004          * handled by ext4_clear_request_list from now on.
3005          */
3006         elr = NULL;
3007
3008         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3009                 ret = ext4_run_lazyinit_thread();
3010                 if (ret)
3011                         goto out;
3012         }
3013 out:
3014         mutex_unlock(&ext4_li_mtx);
3015         if (ret)
3016                 kfree(elr);
3017         return ret;
3018 }
3019
3020 /*
3021  * We do not need to lock anything since this is called on
3022  * module unload.
3023  */
3024 static void ext4_destroy_lazyinit_thread(void)
3025 {
3026         /*
3027          * If thread exited earlier
3028          * there's nothing to be done.
3029          */
3030         if (!ext4_li_info || !ext4_lazyinit_task)
3031                 return;
3032
3033         kthread_stop(ext4_lazyinit_task);
3034 }
3035
3036 static int set_journal_csum_feature_set(struct super_block *sb)
3037 {
3038         int ret = 1;
3039         int compat, incompat;
3040         struct ext4_sb_info *sbi = EXT4_SB(sb);
3041
3042         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3043                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3044                 /* journal checksum v2 */
3045                 compat = 0;
3046                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3047         } else {
3048                 /* journal checksum v1 */
3049                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3050                 incompat = 0;
3051         }
3052
3053         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3054                 ret = jbd2_journal_set_features(sbi->s_journal,
3055                                 compat, 0,
3056                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3057                                 incompat);
3058         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3059                 ret = jbd2_journal_set_features(sbi->s_journal,
3060                                 compat, 0,
3061                                 incompat);
3062                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3063                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3064         } else {
3065                 jbd2_journal_clear_features(sbi->s_journal,
3066                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3067                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3068                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3069         }
3070
3071         return ret;
3072 }
3073
3074 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3075 {
3076         char *orig_data = kstrdup(data, GFP_KERNEL);
3077         struct buffer_head *bh;
3078         struct ext4_super_block *es = NULL;
3079         struct ext4_sb_info *sbi;
3080         ext4_fsblk_t block;
3081         ext4_fsblk_t sb_block = get_sb_block(&data);
3082         ext4_fsblk_t logical_sb_block;
3083         unsigned long offset = 0;
3084         unsigned long journal_devnum = 0;
3085         unsigned long def_mount_opts;
3086         struct inode *root;
3087         char *cp;
3088         const char *descr;
3089         int ret = -ENOMEM;
3090         int blocksize, clustersize;
3091         unsigned int db_count;
3092         unsigned int i;
3093         int needs_recovery, has_huge_files, has_bigalloc;
3094         __u64 blocks_count;
3095         int err;
3096         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3097         ext4_group_t first_not_zeroed;
3098
3099         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3100         if (!sbi)
3101                 goto out_free_orig;
3102
3103         sbi->s_blockgroup_lock =
3104                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3105         if (!sbi->s_blockgroup_lock) {
3106                 kfree(sbi);
3107                 goto out_free_orig;
3108         }
3109         sb->s_fs_info = sbi;
3110         sbi->s_sb = sb;
3111         sbi->s_mount_opt = 0;
3112         sbi->s_resuid = make_kuid(&init_user_ns, EXT4_DEF_RESUID);
3113         sbi->s_resgid = make_kgid(&init_user_ns, EXT4_DEF_RESGID);
3114         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3115         sbi->s_sb_block = sb_block;
3116         if (sb->s_bdev->bd_part)
3117                 sbi->s_sectors_written_start =
3118                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3119
3120         /* Cleanup superblock name */
3121         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3122                 *cp = '!';
3123
3124         ret = -EINVAL;
3125         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3126         if (!blocksize) {
3127                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3128                 goto out_fail;
3129         }
3130
3131         /*
3132          * The ext4 superblock will not be buffer aligned for other than 1kB
3133          * block sizes.  We need to calculate the offset from buffer start.
3134          */
3135         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3136                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3137                 offset = do_div(logical_sb_block, blocksize);
3138         } else {
3139                 logical_sb_block = sb_block;
3140         }
3141
3142         if (!(bh = sb_bread(sb, logical_sb_block))) {
3143                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3144                 goto out_fail;
3145         }
3146         /*
3147          * Note: s_es must be initialized as soon as possible because
3148          *       some ext4 macro-instructions depend on its value
3149          */
3150         es = (struct ext4_super_block *) (bh->b_data + offset);
3151         sbi->s_es = es;
3152         sb->s_magic = le16_to_cpu(es->s_magic);
3153         if (sb->s_magic != EXT4_SUPER_MAGIC)
3154                 goto cantfind_ext4;
3155         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3156
3157         /* Warn if metadata_csum and gdt_csum are both set. */
3158         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3159                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3160             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3161                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3162                              "redundant flags; please run fsck.");
3163
3164         /* Check for a known checksum algorithm */
3165         if (!ext4_verify_csum_type(sb, es)) {
3166                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3167                          "unknown checksum algorithm.");
3168                 silent = 1;
3169                 goto cantfind_ext4;
3170         }
3171
3172         /* Load the checksum driver */
3173         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3174                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3175                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3176                 if (IS_ERR(sbi->s_chksum_driver)) {
3177                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3178                         ret = PTR_ERR(sbi->s_chksum_driver);
3179                         sbi->s_chksum_driver = NULL;
3180                         goto failed_mount;
3181                 }
3182         }
3183
3184         /* Check superblock checksum */
3185         if (!ext4_superblock_csum_verify(sb, es)) {
3186                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3187                          "invalid superblock checksum.  Run e2fsck?");
3188                 silent = 1;
3189                 goto cantfind_ext4;
3190         }
3191
3192         /* Precompute checksum seed for all metadata */
3193         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3194                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3195                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3196                                                sizeof(es->s_uuid));
3197
3198         /* Set defaults before we parse the mount options */
3199         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3200         set_opt(sb, INIT_INODE_TABLE);
3201         if (def_mount_opts & EXT4_DEFM_DEBUG)
3202                 set_opt(sb, DEBUG);
3203         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3204                 set_opt(sb, GRPID);
3205         if (def_mount_opts & EXT4_DEFM_UID16)
3206                 set_opt(sb, NO_UID32);
3207         /* xattr user namespace & acls are now defaulted on */
3208 #ifdef CONFIG_EXT4_FS_XATTR
3209         set_opt(sb, XATTR_USER);
3210 #endif
3211 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3212         set_opt(sb, POSIX_ACL);
3213 #endif
3214         set_opt(sb, MBLK_IO_SUBMIT);
3215         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3216                 set_opt(sb, JOURNAL_DATA);
3217         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3218                 set_opt(sb, ORDERED_DATA);
3219         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3220                 set_opt(sb, WRITEBACK_DATA);
3221
3222         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3223                 set_opt(sb, ERRORS_PANIC);
3224         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3225                 set_opt(sb, ERRORS_CONT);
3226         else
3227                 set_opt(sb, ERRORS_RO);
3228         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3229                 set_opt(sb, BLOCK_VALIDITY);
3230         if (def_mount_opts & EXT4_DEFM_DISCARD)
3231                 set_opt(sb, DISCARD);
3232
3233         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3234         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3235         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3236         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3237         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3238
3239         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3240                 set_opt(sb, BARRIER);
3241
3242         /*
3243          * enable delayed allocation by default
3244          * Use -o nodelalloc to turn it off
3245          */
3246         if (!IS_EXT3_SB(sb) &&
3247             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3248                 set_opt(sb, DELALLOC);
3249
3250         /*
3251          * set default s_li_wait_mult for lazyinit, for the case there is
3252          * no mount option specified.
3253          */
3254         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3255
3256         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3257                            &journal_devnum, &journal_ioprio, 0)) {
3258                 ext4_msg(sb, KERN_WARNING,
3259                          "failed to parse options in superblock: %s",
3260                          sbi->s_es->s_mount_opts);
3261         }
3262         sbi->s_def_mount_opt = sbi->s_mount_opt;
3263         if (!parse_options((char *) data, sb, &journal_devnum,
3264                            &journal_ioprio, 0))
3265                 goto failed_mount;
3266
3267         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3268                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3269                             "with data=journal disables delayed "
3270                             "allocation and O_DIRECT support!\n");
3271                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3272                         ext4_msg(sb, KERN_ERR, "can't mount with "
3273                                  "both data=journal and delalloc");
3274                         goto failed_mount;
3275                 }
3276                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3277                         ext4_msg(sb, KERN_ERR, "can't mount with "
3278                                  "both data=journal and delalloc");
3279                         goto failed_mount;
3280                 }
3281                 if (test_opt(sb, DELALLOC))
3282                         clear_opt(sb, DELALLOC);
3283         }
3284
3285         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3286         if (test_opt(sb, DIOREAD_NOLOCK)) {
3287                 if (blocksize < PAGE_SIZE) {
3288                         ext4_msg(sb, KERN_ERR, "can't mount with "
3289                                  "dioread_nolock if block size != PAGE_SIZE");
3290                         goto failed_mount;
3291                 }
3292         }
3293
3294         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3295                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3296
3297         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3298             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3299              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3300              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3301                 ext4_msg(sb, KERN_WARNING,
3302                        "feature flags set on rev 0 fs, "
3303                        "running e2fsck is recommended");
3304
3305         if (IS_EXT2_SB(sb)) {
3306                 if (ext2_feature_set_ok(sb))
3307                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3308                                  "using the ext4 subsystem");
3309                 else {
3310                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3311                                  "to feature incompatibilities");
3312                         goto failed_mount;
3313                 }
3314         }
3315
3316         if (IS_EXT3_SB(sb)) {
3317                 if (ext3_feature_set_ok(sb))
3318                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3319                                  "using the ext4 subsystem");
3320                 else {
3321                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3322                                  "to feature incompatibilities");
3323                         goto failed_mount;
3324                 }
3325         }
3326
3327         /*
3328          * Check feature flags regardless of the revision level, since we
3329          * previously didn't change the revision level when setting the flags,
3330          * so there is a chance incompat flags are set on a rev 0 filesystem.
3331          */
3332         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3333                 goto failed_mount;
3334
3335         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3336             blocksize > EXT4_MAX_BLOCK_SIZE) {
3337                 ext4_msg(sb, KERN_ERR,
3338                        "Unsupported filesystem blocksize %d", blocksize);
3339                 goto failed_mount;
3340         }
3341
3342         if (sb->s_blocksize != blocksize) {
3343                 /* Validate the filesystem blocksize */
3344                 if (!sb_set_blocksize(sb, blocksize)) {
3345                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3346                                         blocksize);
3347                         goto failed_mount;
3348                 }
3349
3350                 brelse(bh);
3351                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3352                 offset = do_div(logical_sb_block, blocksize);
3353                 bh = sb_bread(sb, logical_sb_block);
3354                 if (!bh) {
3355                         ext4_msg(sb, KERN_ERR,
3356                                "Can't read superblock on 2nd try");
3357                         goto failed_mount;
3358                 }
3359                 es = (struct ext4_super_block *)(bh->b_data + offset);
3360                 sbi->s_es = es;
3361                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3362                         ext4_msg(sb, KERN_ERR,
3363                                "Magic mismatch, very weird!");
3364                         goto failed_mount;
3365                 }
3366         }
3367
3368         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3369                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3370         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3371                                                       has_huge_files);
3372         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3373
3374         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3375                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3376                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3377         } else {
3378                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3379                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3380                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3381                     (!is_power_of_2(sbi->s_inode_size)) ||
3382                     (sbi->s_inode_size > blocksize)) {
3383                         ext4_msg(sb, KERN_ERR,
3384                                "unsupported inode size: %d",
3385                                sbi->s_inode_size);
3386                         goto failed_mount;
3387                 }
3388                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3389                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3390         }
3391
3392         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3393         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3394                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3395                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3396                     !is_power_of_2(sbi->s_desc_size)) {
3397                         ext4_msg(sb, KERN_ERR,
3398                                "unsupported descriptor size %lu",
3399                                sbi->s_desc_size);
3400                         goto failed_mount;
3401                 }
3402         } else
3403                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3404
3405         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3406         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3407         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3408                 goto cantfind_ext4;
3409
3410         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3411         if (sbi->s_inodes_per_block == 0)
3412                 goto cantfind_ext4;
3413         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3414                                         sbi->s_inodes_per_block;
3415         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3416         sbi->s_sbh = bh;
3417         sbi->s_mount_state = le16_to_cpu(es->s_state);
3418         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3419         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3420
3421         for (i = 0; i < 4; i++)
3422                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3423         sbi->s_def_hash_version = es->s_def_hash_version;
3424         i = le32_to_cpu(es->s_flags);
3425         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3426                 sbi->s_hash_unsigned = 3;
3427         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3428 #ifdef __CHAR_UNSIGNED__
3429                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3430                 sbi->s_hash_unsigned = 3;
3431 #else
3432                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3433 #endif
3434         }
3435
3436         /* Handle clustersize */
3437         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3438         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3439                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3440         if (has_bigalloc) {
3441                 if (clustersize < blocksize) {
3442                         ext4_msg(sb, KERN_ERR,
3443                                  "cluster size (%d) smaller than "
3444                                  "block size (%d)", clustersize, blocksize);
3445                         goto failed_mount;
3446                 }
3447                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3448                         le32_to_cpu(es->s_log_block_size);
3449                 sbi->s_clusters_per_group =
3450                         le32_to_cpu(es->s_clusters_per_group);
3451                 if (sbi->s_clusters_per_group > blocksize * 8) {
3452                         ext4_msg(sb, KERN_ERR,
3453                                  "#clusters per group too big: %lu",
3454                                  sbi->s_clusters_per_group);
3455                         goto failed_mount;
3456                 }
3457                 if (sbi->s_blocks_per_group !=
3458                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3459                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3460                                  "clusters per group (%lu) inconsistent",
3461                                  sbi->s_blocks_per_group,
3462                                  sbi->s_clusters_per_group);
3463                         goto failed_mount;
3464                 }
3465         } else {
3466                 if (clustersize != blocksize) {
3467                         ext4_warning(sb, "fragment/cluster size (%d) != "
3468                                      "block size (%d)", clustersize,
3469                                      blocksize);
3470                         clustersize = blocksize;
3471                 }
3472                 if (sbi->s_blocks_per_group > blocksize * 8) {
3473                         ext4_msg(sb, KERN_ERR,
3474                                  "#blocks per group too big: %lu",
3475                                  sbi->s_blocks_per_group);
3476                         goto failed_mount;
3477                 }
3478                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3479                 sbi->s_cluster_bits = 0;
3480         }
3481         sbi->s_cluster_ratio = clustersize / blocksize;
3482
3483         if (sbi->s_inodes_per_group > blocksize * 8) {
3484                 ext4_msg(sb, KERN_ERR,
3485                        "#inodes per group too big: %lu",
3486                        sbi->s_inodes_per_group);
3487                 goto failed_mount;
3488         }
3489
3490         /*
3491          * Test whether we have more sectors than will fit in sector_t,
3492          * and whether the max offset is addressable by the page cache.
3493          */
3494         err = generic_check_addressable(sb->s_blocksize_bits,
3495                                         ext4_blocks_count(es));
3496         if (err) {
3497                 ext4_msg(sb, KERN_ERR, "filesystem"
3498                          " too large to mount safely on this system");
3499                 if (sizeof(sector_t) < 8)
3500                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3501                 ret = err;
3502                 goto failed_mount;
3503         }
3504
3505         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3506                 goto cantfind_ext4;
3507
3508         /* check blocks count against device size */
3509         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3510         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3511                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3512                        "exceeds size of device (%llu blocks)",
3513                        ext4_blocks_count(es), blocks_count);
3514                 goto failed_mount;
3515         }
3516
3517         /*
3518          * It makes no sense for the first data block to be beyond the end
3519          * of the filesystem.
3520          */
3521         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3522                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3523                          "block %u is beyond end of filesystem (%llu)",
3524                          le32_to_cpu(es->s_first_data_block),
3525                          ext4_blocks_count(es));
3526                 goto failed_mount;
3527         }
3528         blocks_count = (ext4_blocks_count(es) -
3529                         le32_to_cpu(es->s_first_data_block) +
3530                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3531         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3532         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3533                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3534                        "(block count %llu, first data block %u, "
3535                        "blocks per group %lu)", sbi->s_groups_count,
3536                        ext4_blocks_count(es),
3537                        le32_to_cpu(es->s_first_data_block),
3538                        EXT4_BLOCKS_PER_GROUP(sb));
3539                 goto failed_mount;
3540         }
3541         sbi->s_groups_count = blocks_count;
3542         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3543                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3544         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3545                    EXT4_DESC_PER_BLOCK(sb);
3546         sbi->s_group_desc = ext4_kvmalloc(db_count *
3547                                           sizeof(struct buffer_head *),
3548                                           GFP_KERNEL);
3549         if (sbi->s_group_desc == NULL) {
3550                 ext4_msg(sb, KERN_ERR, "not enough memory");
3551                 ret = -ENOMEM;
3552                 goto failed_mount;
3553         }
3554
3555         if (ext4_proc_root)
3556                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3557
3558         if (sbi->s_proc)
3559                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3560                                  &ext4_seq_options_fops, sb);
3561
3562         bgl_lock_init(sbi->s_blockgroup_lock);
3563
3564         for (i = 0; i < db_count; i++) {
3565                 block = descriptor_loc(sb, logical_sb_block, i);
3566                 sbi->s_group_desc[i] = sb_bread(sb, block);
3567                 if (!sbi->s_group_desc[i]) {
3568                         ext4_msg(sb, KERN_ERR,
3569                                "can't read group descriptor %d", i);
3570                         db_count = i;
3571                         goto failed_mount2;
3572                 }
3573         }
3574         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3575                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3576                 goto failed_mount2;
3577         }
3578         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3579                 if (!ext4_fill_flex_info(sb)) {
3580                         ext4_msg(sb, KERN_ERR,
3581                                "unable to initialize "
3582                                "flex_bg meta info!");
3583                         goto failed_mount2;
3584                 }
3585
3586         sbi->s_gdb_count = db_count;
3587         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3588         spin_lock_init(&sbi->s_next_gen_lock);
3589
3590         init_timer(&sbi->s_err_report);
3591         sbi->s_err_report.function = print_daily_error_info;
3592         sbi->s_err_report.data = (unsigned long) sb;
3593
3594         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3595                         ext4_count_free_clusters(sb));
3596         if (!err) {
3597                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3598                                 ext4_count_free_inodes(sb));
3599         }
3600         if (!err) {
3601                 err = percpu_counter_init(&sbi->s_dirs_counter,
3602                                 ext4_count_dirs(sb));
3603         }
3604         if (!err) {
3605                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3606         }
3607         if (err) {
3608                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3609                 ret = err;
3610                 goto failed_mount3;
3611         }
3612
3613         sbi->s_stripe = ext4_get_stripe_size(sbi);
3614         sbi->s_max_writeback_mb_bump = 128;
3615
3616         /*
3617          * set up enough so that it can read an inode
3618          */
3619         if (!test_opt(sb, NOLOAD) &&
3620             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3621                 sb->s_op = &ext4_sops;
3622         else
3623                 sb->s_op = &ext4_nojournal_sops;
3624         sb->s_export_op = &ext4_export_ops;
3625         sb->s_xattr = ext4_xattr_handlers;
3626 #ifdef CONFIG_QUOTA
3627         sb->s_qcop = &ext4_qctl_operations;
3628         sb->dq_op = &ext4_quota_operations;
3629 #endif
3630         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3631
3632         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3633         mutex_init(&sbi->s_orphan_lock);
3634         sbi->s_resize_flags = 0;
3635
3636         sb->s_root = NULL;
3637
3638         needs_recovery = (es->s_last_orphan != 0 ||
3639                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3640                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3641
3642         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3643             !(sb->s_flags & MS_RDONLY))
3644                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3645                         goto failed_mount3;
3646
3647         /*
3648          * The first inode we look at is the journal inode.  Don't try
3649          * root first: it may be modified in the journal!
3650          */
3651         if (!test_opt(sb, NOLOAD) &&
3652             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3653                 if (ext4_load_journal(sb, es, journal_devnum))
3654                         goto failed_mount3;
3655         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3656               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3657                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3658                        "suppressed and not mounted read-only");
3659                 goto failed_mount_wq;
3660         } else {
3661                 clear_opt(sb, DATA_FLAGS);
3662                 sbi->s_journal = NULL;
3663                 needs_recovery = 0;
3664                 goto no_journal;
3665         }
3666
3667         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3668             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3669                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3670                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3671                 goto failed_mount_wq;
3672         }
3673
3674         if (!set_journal_csum_feature_set(sb)) {
3675                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3676                          "feature set");
3677                 goto failed_mount_wq;
3678         }
3679
3680         /* We have now updated the journal if required, so we can
3681          * validate the data journaling mode. */
3682         switch (test_opt(sb, DATA_FLAGS)) {
3683         case 0:
3684                 /* No mode set, assume a default based on the journal
3685                  * capabilities: ORDERED_DATA if the journal can
3686                  * cope, else JOURNAL_DATA
3687                  */
3688                 if (jbd2_journal_check_available_features
3689                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3690                         set_opt(sb, ORDERED_DATA);
3691                 else
3692                         set_opt(sb, JOURNAL_DATA);
3693                 break;
3694
3695         case EXT4_MOUNT_ORDERED_DATA:
3696         case EXT4_MOUNT_WRITEBACK_DATA:
3697                 if (!jbd2_journal_check_available_features
3698                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3699                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3700                                "requested data journaling mode");
3701                         goto failed_mount_wq;
3702                 }
3703         default:
3704                 break;
3705         }
3706         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3707
3708         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3709
3710         /*
3711          * The journal may have updated the bg summary counts, so we
3712          * need to update the global counters.
3713          */
3714         percpu_counter_set(&sbi->s_freeclusters_counter,
3715                            ext4_count_free_clusters(sb));
3716         percpu_counter_set(&sbi->s_freeinodes_counter,
3717                            ext4_count_free_inodes(sb));
3718         percpu_counter_set(&sbi->s_dirs_counter,
3719                            ext4_count_dirs(sb));
3720         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3721
3722 no_journal:
3723         /*
3724          * The maximum number of concurrent works can be high and
3725          * concurrency isn't really necessary.  Limit it to 1.
3726          */
3727         EXT4_SB(sb)->dio_unwritten_wq =
3728                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3729         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3730                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3731                 goto failed_mount_wq;
3732         }
3733
3734         /*
3735          * The jbd2_journal_load will have done any necessary log recovery,
3736          * so we can safely mount the rest of the filesystem now.
3737          */
3738
3739         root = ext4_iget(sb, EXT4_ROOT_INO);
3740         if (IS_ERR(root)) {
3741                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3742                 ret = PTR_ERR(root);
3743                 root = NULL;
3744                 goto failed_mount4;
3745         }
3746         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3747                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3748                 iput(root);
3749                 goto failed_mount4;
3750         }
3751         sb->s_root = d_make_root(root);
3752         if (!sb->s_root) {
3753                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3754                 ret = -ENOMEM;
3755                 goto failed_mount4;
3756         }
3757
3758         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3759                 sb->s_flags |= MS_RDONLY;
3760
3761         /* determine the minimum size of new large inodes, if present */
3762         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3763                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3764                                                      EXT4_GOOD_OLD_INODE_SIZE;
3765                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3766                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3767                         if (sbi->s_want_extra_isize <
3768                             le16_to_cpu(es->s_want_extra_isize))
3769                                 sbi->s_want_extra_isize =
3770                                         le16_to_cpu(es->s_want_extra_isize);
3771                         if (sbi->s_want_extra_isize <
3772                             le16_to_cpu(es->s_min_extra_isize))
3773                                 sbi->s_want_extra_isize =
3774                                         le16_to_cpu(es->s_min_extra_isize);
3775                 }
3776         }
3777         /* Check if enough inode space is available */
3778         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3779                                                         sbi->s_inode_size) {
3780                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3781                                                        EXT4_GOOD_OLD_INODE_SIZE;
3782                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3783                          "available");
3784         }
3785
3786         err = ext4_setup_system_zone(sb);
3787         if (err) {
3788                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3789                          "zone (%d)", err);
3790                 goto failed_mount4a;
3791         }
3792
3793         ext4_ext_init(sb);
3794         err = ext4_mb_init(sb);
3795         if (err) {
3796                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3797                          err);
3798                 goto failed_mount5;
3799         }
3800
3801         err = ext4_register_li_request(sb, first_not_zeroed);
3802         if (err)
3803                 goto failed_mount6;
3804
3805         sbi->s_kobj.kset = ext4_kset;
3806         init_completion(&sbi->s_kobj_unregister);
3807         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3808                                    "%s", sb->s_id);
3809         if (err)
3810                 goto failed_mount7;
3811
3812         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3813         ext4_orphan_cleanup(sb, es);
3814         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3815         if (needs_recovery) {
3816                 ext4_msg(sb, KERN_INFO, "recovery complete");
3817                 ext4_mark_recovery_complete(sb, es);
3818         }
3819         if (EXT4_SB(sb)->s_journal) {
3820                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3821                         descr = " journalled data mode";
3822                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3823                         descr = " ordered data mode";
3824                 else
3825                         descr = " writeback data mode";
3826         } else
3827                 descr = "out journal";
3828
3829         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3830                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3831                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3832
3833         if (es->s_error_count)
3834                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3835
3836         kfree(orig_data);
3837         return 0;
3838
3839 cantfind_ext4:
3840         if (!silent)
3841                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3842         goto failed_mount;
3843
3844 failed_mount7:
3845         ext4_unregister_li_request(sb);
3846 failed_mount6:
3847         ext4_mb_release(sb);
3848 failed_mount5:
3849         ext4_ext_release(sb);
3850         ext4_release_system_zone(sb);
3851 failed_mount4a:
3852         dput(sb->s_root);
3853         sb->s_root = NULL;
3854 failed_mount4:
3855         ext4_msg(sb, KERN_ERR, "mount failed");
3856         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
3857 failed_mount_wq:
3858         if (sbi->s_journal) {
3859                 jbd2_journal_destroy(sbi->s_journal);
3860                 sbi->s_journal = NULL;
3861         }
3862 failed_mount3:
3863         del_timer(&sbi->s_err_report);
3864         if (sbi->s_flex_groups)
3865                 ext4_kvfree(sbi->s_flex_groups);
3866         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3867         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3868         percpu_counter_destroy(&sbi->s_dirs_counter);
3869         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3870         if (sbi->s_mmp_tsk)
3871                 kthread_stop(sbi->s_mmp_tsk);
3872 failed_mount2:
3873         for (i = 0; i < db_count; i++)
3874                 brelse(sbi->s_group_desc[i]);
3875         ext4_kvfree(sbi->s_group_desc);
3876 failed_mount:
3877         if (sbi->s_chksum_driver)
3878                 crypto_free_shash(sbi->s_chksum_driver);
3879         if (sbi->s_proc) {
3880                 remove_proc_entry("options", sbi->s_proc);
3881                 remove_proc_entry(sb->s_id, ext4_proc_root);
3882         }
3883 #ifdef CONFIG_QUOTA
3884         for (i = 0; i < MAXQUOTAS; i++)
3885                 kfree(sbi->s_qf_names[i]);
3886 #endif
3887         ext4_blkdev_remove(sbi);
3888         brelse(bh);
3889 out_fail:
3890         sb->s_fs_info = NULL;
3891         kfree(sbi->s_blockgroup_lock);
3892         kfree(sbi);
3893 out_free_orig:
3894         kfree(orig_data);
3895         return ret;
3896 }
3897
3898 /*
3899  * Setup any per-fs journal parameters now.  We'll do this both on
3900  * initial mount, once the journal has been initialised but before we've
3901  * done any recovery; and again on any subsequent remount.
3902  */
3903 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
3904 {
3905         struct ext4_sb_info *sbi = EXT4_SB(sb);
3906
3907         journal->j_commit_interval = sbi->s_commit_interval;
3908         journal->j_min_batch_time = sbi->s_min_batch_time;
3909         journal->j_max_batch_time = sbi->s_max_batch_time;
3910
3911         write_lock(&journal->j_state_lock);
3912         if (test_opt(sb, BARRIER))
3913                 journal->j_flags |= JBD2_BARRIER;
3914         else
3915                 journal->j_flags &= ~JBD2_BARRIER;
3916         if (test_opt(sb, DATA_ERR_ABORT))
3917                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
3918         else
3919                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
3920         write_unlock(&journal->j_state_lock);
3921 }
3922
3923 static journal_t *ext4_get_journal(struct super_block *sb,
3924                                    unsigned int journal_inum)
3925 {
3926         struct inode *journal_inode;
3927         journal_t *journal;
3928
3929         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3930
3931         /* First, test for the existence of a valid inode on disk.  Bad
3932          * things happen if we iget() an unused inode, as the subsequent
3933          * iput() will try to delete it. */
3934
3935         journal_inode = ext4_iget(sb, journal_inum);
3936         if (IS_ERR(journal_inode)) {
3937                 ext4_msg(sb, KERN_ERR, "no journal found");
3938                 return NULL;
3939         }
3940         if (!journal_inode->i_nlink) {
3941                 make_bad_inode(journal_inode);
3942                 iput(journal_inode);
3943                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
3944                 return NULL;
3945         }
3946
3947         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
3948                   journal_inode, journal_inode->i_size);
3949         if (!S_ISREG(journal_inode->i_mode)) {
3950                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
3951                 iput(journal_inode);
3952                 return NULL;
3953         }
3954
3955         journal = jbd2_journal_init_inode(journal_inode);
3956         if (!journal) {
3957                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
3958                 iput(journal_inode);
3959                 return NULL;
3960         }
3961         journal->j_private = sb;
3962         ext4_init_journal_params(sb, journal);
3963         return journal;
3964 }
3965
3966 static journal_t *ext4_get_dev_journal(struct super_block *sb,
3967                                        dev_t j_dev)
3968 {
3969         struct buffer_head *bh;
3970         journal_t *journal;
3971         ext4_fsblk_t start;
3972         ext4_fsblk_t len;
3973         int hblock, blocksize;
3974         ext4_fsblk_t sb_block;
3975         unsigned long offset;
3976         struct ext4_super_block *es;
3977         struct block_device *bdev;
3978
3979         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
3980
3981         bdev = ext4_blkdev_get(j_dev, sb);
3982         if (bdev == NULL)
3983                 return NULL;
3984
3985         blocksize = sb->s_blocksize;
3986         hblock = bdev_logical_block_size(bdev);
3987         if (blocksize < hblock) {
3988                 ext4_msg(sb, KERN_ERR,
3989                         "blocksize too small for journal device");
3990                 goto out_bdev;
3991         }
3992
3993         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
3994         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
3995         set_blocksize(bdev, blocksize);
3996         if (!(bh = __bread(bdev, sb_block, blocksize))) {
3997                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
3998                        "external journal");
3999                 goto out_bdev;
4000         }
4001
4002         es = (struct ext4_super_block *) (bh->b_data + offset);
4003         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4004             !(le32_to_cpu(es->s_feature_incompat) &
4005               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4006                 ext4_msg(sb, KERN_ERR, "external journal has "
4007                                         "bad superblock");
4008                 brelse(bh);
4009                 goto out_bdev;
4010         }
4011
4012         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4013                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4014                 brelse(bh);
4015                 goto out_bdev;
4016         }
4017
4018         len = ext4_blocks_count(es);
4019         start = sb_block + 1;
4020         brelse(bh);     /* we're done with the superblock */
4021
4022         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4023                                         start, len, blocksize);
4024         if (!journal) {
4025                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4026                 goto out_bdev;
4027         }
4028         journal->j_private = sb;
4029         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4030         wait_on_buffer(journal->j_sb_buffer);
4031         if (!buffer_uptodate(journal->j_sb_buffer)) {
4032                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4033                 goto out_journal;
4034         }
4035         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4036                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4037                                         "user (unsupported) - %d",
4038                         be32_to_cpu(journal->j_superblock->s_nr_users));
4039                 goto out_journal;
4040         }
4041         EXT4_SB(sb)->journal_bdev = bdev;
4042         ext4_init_journal_params(sb, journal);
4043         return journal;
4044
4045 out_journal:
4046         jbd2_journal_destroy(journal);
4047 out_bdev:
4048         ext4_blkdev_put(bdev);
4049         return NULL;
4050 }
4051
4052 static int ext4_load_journal(struct super_block *sb,
4053                              struct ext4_super_block *es,
4054                              unsigned long journal_devnum)
4055 {
4056         journal_t *journal;
4057         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4058         dev_t journal_dev;
4059         int err = 0;
4060         int really_read_only;
4061
4062         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4063
4064         if (journal_devnum &&
4065             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4066                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4067                         "numbers have changed");
4068                 journal_dev = new_decode_dev(journal_devnum);
4069         } else
4070                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4071
4072         really_read_only = bdev_read_only(sb->s_bdev);
4073
4074         /*
4075          * Are we loading a blank journal or performing recovery after a
4076          * crash?  For recovery, we need to check in advance whether we
4077          * can get read-write access to the device.
4078          */
4079         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4080                 if (sb->s_flags & MS_RDONLY) {
4081                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4082                                         "required on readonly filesystem");
4083                         if (really_read_only) {
4084                                 ext4_msg(sb, KERN_ERR, "write access "
4085                                         "unavailable, cannot proceed");
4086                                 return -EROFS;
4087                         }
4088                         ext4_msg(sb, KERN_INFO, "write access will "
4089                                "be enabled during recovery");
4090                 }
4091         }
4092
4093         if (journal_inum && journal_dev) {
4094                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4095                        "and inode journals!");
4096                 return -EINVAL;
4097         }
4098
4099         if (journal_inum) {
4100                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4101                         return -EINVAL;
4102         } else {
4103                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4104                         return -EINVAL;
4105         }
4106
4107         if (!(journal->j_flags & JBD2_BARRIER))
4108                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4109
4110         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4111                 err = jbd2_journal_wipe(journal, !really_read_only);
4112         if (!err) {
4113                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4114                 if (save)
4115                         memcpy(save, ((char *) es) +
4116                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4117                 err = jbd2_journal_load(journal);
4118                 if (save)
4119                         memcpy(((char *) es) + EXT4_S_ERR_START,
4120                                save, EXT4_S_ERR_LEN);
4121                 kfree(save);
4122         }
4123
4124         if (err) {
4125                 ext4_msg(sb, KERN_ERR, "error loading journal");
4126                 jbd2_journal_destroy(journal);
4127                 return err;
4128         }
4129
4130         EXT4_SB(sb)->s_journal = journal;
4131         ext4_clear_journal_err(sb, es);
4132
4133         if (!really_read_only && journal_devnum &&
4134             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4135                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4136
4137                 /* Make sure we flush the recovery flag to disk. */
4138                 ext4_commit_super(sb, 1);
4139         }
4140
4141         return 0;
4142 }
4143
4144 static int ext4_commit_super(struct super_block *sb, int sync)
4145 {
4146         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4147         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4148         int error = 0;
4149
4150         if (!sbh || block_device_ejected(sb))
4151                 return error;
4152         if (buffer_write_io_error(sbh)) {
4153                 /*
4154                  * Oh, dear.  A previous attempt to write the
4155                  * superblock failed.  This could happen because the
4156                  * USB device was yanked out.  Or it could happen to
4157                  * be a transient write error and maybe the block will
4158                  * be remapped.  Nothing we can do but to retry the
4159                  * write and hope for the best.
4160                  */
4161                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4162                        "superblock detected");
4163                 clear_buffer_write_io_error(sbh);
4164                 set_buffer_uptodate(sbh);
4165         }
4166         /*
4167          * If the file system is mounted read-only, don't update the
4168          * superblock write time.  This avoids updating the superblock
4169          * write time when we are mounting the root file system
4170          * read/only but we need to replay the journal; at that point,
4171          * for people who are east of GMT and who make their clock
4172          * tick in localtime for Windows bug-for-bug compatibility,
4173          * the clock is set in the future, and this will cause e2fsck
4174          * to complain and force a full file system check.
4175          */
4176         if (!(sb->s_flags & MS_RDONLY))
4177                 es->s_wtime = cpu_to_le32(get_seconds());
4178         if (sb->s_bdev->bd_part)
4179                 es->s_kbytes_written =
4180                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4181                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4182                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4183         else
4184                 es->s_kbytes_written =
4185                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4186         ext4_free_blocks_count_set(es,
4187                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4188                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4189         es->s_free_inodes_count =
4190                 cpu_to_le32(percpu_counter_sum_positive(
4191                                 &EXT4_SB(sb)->s_freeinodes_counter));
4192         sb->s_dirt = 0;
4193         BUFFER_TRACE(sbh, "marking dirty");
4194         ext4_superblock_csum_set(sb, es);
4195         mark_buffer_dirty(sbh);
4196         if (sync) {
4197                 error = sync_dirty_buffer(sbh);
4198                 if (error)
4199                         return error;
4200
4201                 error = buffer_write_io_error(sbh);
4202                 if (error) {
4203                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4204                                "superblock");
4205                         clear_buffer_write_io_error(sbh);
4206                         set_buffer_uptodate(sbh);
4207                 }
4208         }
4209         return error;
4210 }
4211
4212 /*
4213  * Have we just finished recovery?  If so, and if we are mounting (or
4214  * remounting) the filesystem readonly, then we will end up with a
4215  * consistent fs on disk.  Record that fact.
4216  */
4217 static void ext4_mark_recovery_complete(struct super_block *sb,
4218                                         struct ext4_super_block *es)
4219 {
4220         journal_t *journal = EXT4_SB(sb)->s_journal;
4221
4222         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4223                 BUG_ON(journal != NULL);
4224                 return;
4225         }
4226         jbd2_journal_lock_updates(journal);
4227         if (jbd2_journal_flush(journal) < 0)
4228                 goto out;
4229
4230         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4231             sb->s_flags & MS_RDONLY) {
4232                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4233                 ext4_commit_super(sb, 1);
4234         }
4235
4236 out:
4237         jbd2_journal_unlock_updates(journal);
4238 }
4239
4240 /*
4241  * If we are mounting (or read-write remounting) a filesystem whose journal
4242  * has recorded an error from a previous lifetime, move that error to the
4243  * main filesystem now.
4244  */
4245 static void ext4_clear_journal_err(struct super_block *sb,
4246                                    struct ext4_super_block *es)
4247 {
4248         journal_t *journal;
4249         int j_errno;
4250         const char *errstr;
4251
4252         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4253
4254         journal = EXT4_SB(sb)->s_journal;
4255
4256         /*
4257          * Now check for any error status which may have been recorded in the
4258          * journal by a prior ext4_error() or ext4_abort()
4259          */
4260
4261         j_errno = jbd2_journal_errno(journal);
4262         if (j_errno) {
4263                 char nbuf[16];
4264
4265                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4266                 ext4_warning(sb, "Filesystem error recorded "
4267                              "from previous mount: %s", errstr);
4268                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4269
4270                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4271                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4272                 ext4_commit_super(sb, 1);
4273
4274                 jbd2_journal_clear_err(journal);
4275         }
4276 }
4277
4278 /*
4279  * Force the running and committing transactions to commit,
4280  * and wait on the commit.
4281  */
4282 int ext4_force_commit(struct super_block *sb)
4283 {
4284         journal_t *journal;
4285         int ret = 0;
4286
4287         if (sb->s_flags & MS_RDONLY)
4288                 return 0;
4289
4290         journal = EXT4_SB(sb)->s_journal;
4291         if (journal)
4292                 ret = ext4_journal_force_commit(journal);
4293
4294         return ret;
4295 }
4296
4297 static void ext4_write_super(struct super_block *sb)
4298 {
4299         lock_super(sb);
4300         ext4_commit_super(sb, 1);
4301         unlock_super(sb);
4302 }
4303
4304 static int ext4_sync_fs(struct super_block *sb, int wait)
4305 {
4306         int ret = 0;
4307         tid_t target;
4308         struct ext4_sb_info *sbi = EXT4_SB(sb);
4309
4310         trace_ext4_sync_fs(sb, wait);
4311         flush_workqueue(sbi->dio_unwritten_wq);
4312         /*
4313          * Writeback quota in non-journalled quota case - journalled quota has
4314          * no dirty dquots
4315          */
4316         dquot_writeback_dquots(sb, -1);
4317         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4318                 if (wait)
4319                         jbd2_log_wait_commit(sbi->s_journal, target);
4320         }
4321         return ret;
4322 }
4323
4324 /*
4325  * LVM calls this function before a (read-only) snapshot is created.  This
4326  * gives us a chance to flush the journal completely and mark the fs clean.
4327  *
4328  * Note that only this function cannot bring a filesystem to be in a clean
4329  * state independently. It relies on upper layer to stop all data & metadata
4330  * modifications.
4331  */
4332 static int ext4_freeze(struct super_block *sb)
4333 {
4334         int error = 0;
4335         journal_t *journal;
4336
4337         if (sb->s_flags & MS_RDONLY)
4338                 return 0;
4339
4340         journal = EXT4_SB(sb)->s_journal;
4341
4342         /* Now we set up the journal barrier. */
4343         jbd2_journal_lock_updates(journal);
4344
4345         /*
4346          * Don't clear the needs_recovery flag if we failed to flush
4347          * the journal.
4348          */
4349         error = jbd2_journal_flush(journal);
4350         if (error < 0)
4351                 goto out;
4352
4353         /* Journal blocked and flushed, clear needs_recovery flag. */
4354         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4355         error = ext4_commit_super(sb, 1);
4356 out:
4357         /* we rely on upper layer to stop further updates */
4358         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4359         return error;
4360 }
4361
4362 /*
4363  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4364  * flag here, even though the filesystem is not technically dirty yet.
4365  */
4366 static int ext4_unfreeze(struct super_block *sb)
4367 {
4368         if (sb->s_flags & MS_RDONLY)
4369                 return 0;
4370
4371         lock_super(sb);
4372         /* Reset the needs_recovery flag before the fs is unlocked. */
4373         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4374         ext4_commit_super(sb, 1);
4375         unlock_super(sb);
4376         return 0;
4377 }
4378
4379 /*
4380  * Structure to save mount options for ext4_remount's benefit
4381  */
4382 struct ext4_mount_options {
4383         unsigned long s_mount_opt;
4384         unsigned long s_mount_opt2;
4385         kuid_t s_resuid;
4386         kgid_t s_resgid;
4387         unsigned long s_commit_interval;
4388         u32 s_min_batch_time, s_max_batch_time;
4389 #ifdef CONFIG_QUOTA
4390         int s_jquota_fmt;
4391         char *s_qf_names[MAXQUOTAS];
4392 #endif
4393 };
4394
4395 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4396 {
4397         struct ext4_super_block *es;
4398         struct ext4_sb_info *sbi = EXT4_SB(sb);
4399         unsigned long old_sb_flags;
4400         struct ext4_mount_options old_opts;
4401         int enable_quota = 0;
4402         ext4_group_t g;
4403         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4404         int err = 0;
4405 #ifdef CONFIG_QUOTA
4406         int i;
4407 #endif
4408         char *orig_data = kstrdup(data, GFP_KERNEL);
4409
4410         /* Store the original options */
4411         lock_super(sb);
4412         old_sb_flags = sb->s_flags;
4413         old_opts.s_mount_opt = sbi->s_mount_opt;
4414         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4415         old_opts.s_resuid = sbi->s_resuid;
4416         old_opts.s_resgid = sbi->s_resgid;
4417         old_opts.s_commit_interval = sbi->s_commit_interval;
4418         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4419         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4420 #ifdef CONFIG_QUOTA
4421         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4422         for (i = 0; i < MAXQUOTAS; i++)
4423                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4424 #endif
4425         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4426                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4427
4428         /*
4429          * Allow the "check" option to be passed as a remount option.
4430          */
4431         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4432                 err = -EINVAL;
4433                 goto restore_opts;
4434         }
4435
4436         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4437                 ext4_abort(sb, "Abort forced by user");
4438
4439         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4440                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4441
4442         es = sbi->s_es;
4443
4444         if (sbi->s_journal) {
4445                 ext4_init_journal_params(sb, sbi->s_journal);
4446                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4447         }
4448
4449         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4450                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4451                         err = -EROFS;
4452                         goto restore_opts;
4453                 }
4454
4455                 if (*flags & MS_RDONLY) {
4456                         err = dquot_suspend(sb, -1);
4457                         if (err < 0)
4458                                 goto restore_opts;
4459
4460                         /*
4461                          * First of all, the unconditional stuff we have to do
4462                          * to disable replay of the journal when we next remount
4463                          */
4464                         sb->s_flags |= MS_RDONLY;
4465
4466                         /*
4467                          * OK, test if we are remounting a valid rw partition
4468                          * readonly, and if so set the rdonly flag and then
4469                          * mark the partition as valid again.
4470                          */
4471                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4472                             (sbi->s_mount_state & EXT4_VALID_FS))
4473                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4474
4475                         if (sbi->s_journal)
4476                                 ext4_mark_recovery_complete(sb, es);
4477                 } else {
4478                         /* Make sure we can mount this feature set readwrite */
4479                         if (!ext4_feature_set_ok(sb, 0)) {
4480                                 err = -EROFS;
4481                                 goto restore_opts;
4482                         }
4483                         /*
4484                          * Make sure the group descriptor checksums
4485                          * are sane.  If they aren't, refuse to remount r/w.
4486                          */
4487                         for (g = 0; g < sbi->s_groups_count; g++) {
4488                                 struct ext4_group_desc *gdp =
4489                                         ext4_get_group_desc(sb, g, NULL);
4490
4491                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4492                                         ext4_msg(sb, KERN_ERR,
4493                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4494                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4495                                                le16_to_cpu(gdp->bg_checksum));
4496                                         err = -EINVAL;
4497                                         goto restore_opts;
4498                                 }
4499                         }
4500
4501                         /*
4502                          * If we have an unprocessed orphan list hanging
4503                          * around from a previously readonly bdev mount,
4504                          * require a full umount/remount for now.
4505                          */
4506                         if (es->s_last_orphan) {
4507                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4508                                        "remount RDWR because of unprocessed "
4509                                        "orphan inode list.  Please "
4510                                        "umount/remount instead");
4511                                 err = -EINVAL;
4512                                 goto restore_opts;
4513                         }
4514
4515                         /*
4516                          * Mounting a RDONLY partition read-write, so reread
4517                          * and store the current valid flag.  (It may have
4518                          * been changed by e2fsck since we originally mounted
4519                          * the partition.)
4520                          */
4521                         if (sbi->s_journal)
4522                                 ext4_clear_journal_err(sb, es);
4523                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4524                         if (!ext4_setup_super(sb, es, 0))
4525                                 sb->s_flags &= ~MS_RDONLY;
4526                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4527                                                      EXT4_FEATURE_INCOMPAT_MMP))
4528                                 if (ext4_multi_mount_protect(sb,
4529                                                 le64_to_cpu(es->s_mmp_block))) {
4530                                         err = -EROFS;
4531                                         goto restore_opts;
4532                                 }
4533                         enable_quota = 1;
4534                 }
4535         }
4536
4537         /*
4538          * Reinitialize lazy itable initialization thread based on
4539          * current settings
4540          */
4541         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4542                 ext4_unregister_li_request(sb);
4543         else {
4544                 ext4_group_t first_not_zeroed;
4545                 first_not_zeroed = ext4_has_uninit_itable(sb);
4546                 ext4_register_li_request(sb, first_not_zeroed);
4547         }
4548
4549         ext4_setup_system_zone(sb);
4550         if (sbi->s_journal == NULL)
4551                 ext4_commit_super(sb, 1);
4552
4553 #ifdef CONFIG_QUOTA
4554         /* Release old quota file names */
4555         for (i = 0; i < MAXQUOTAS; i++)
4556                 if (old_opts.s_qf_names[i] &&
4557                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4558                         kfree(old_opts.s_qf_names[i]);
4559 #endif
4560         unlock_super(sb);
4561         if (enable_quota)
4562                 dquot_resume(sb, -1);
4563
4564         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4565         kfree(orig_data);
4566         return 0;
4567
4568 restore_opts:
4569         sb->s_flags = old_sb_flags;
4570         sbi->s_mount_opt = old_opts.s_mount_opt;
4571         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4572         sbi->s_resuid = old_opts.s_resuid;
4573         sbi->s_resgid = old_opts.s_resgid;
4574         sbi->s_commit_interval = old_opts.s_commit_interval;
4575         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4576         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4577 #ifdef CONFIG_QUOTA
4578         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4579         for (i = 0; i < MAXQUOTAS; i++) {
4580                 if (sbi->s_qf_names[i] &&
4581                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4582                         kfree(sbi->s_qf_names[i]);
4583                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4584         }
4585 #endif
4586         unlock_super(sb);
4587         kfree(orig_data);
4588         return err;
4589 }
4590
4591 /*
4592  * Note: calculating the overhead so we can be compatible with
4593  * historical BSD practice is quite difficult in the face of
4594  * clusters/bigalloc.  This is because multiple metadata blocks from
4595  * different block group can end up in the same allocation cluster.
4596  * Calculating the exact overhead in the face of clustered allocation
4597  * requires either O(all block bitmaps) in memory or O(number of block
4598  * groups**2) in time.  We will still calculate the superblock for
4599  * older file systems --- and if we come across with a bigalloc file
4600  * system with zero in s_overhead_clusters the estimate will be close to
4601  * correct especially for very large cluster sizes --- but for newer
4602  * file systems, it's better to calculate this figure once at mkfs
4603  * time, and store it in the superblock.  If the superblock value is
4604  * present (even for non-bigalloc file systems), we will use it.
4605  */
4606 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4607 {
4608         struct super_block *sb = dentry->d_sb;
4609         struct ext4_sb_info *sbi = EXT4_SB(sb);
4610         struct ext4_super_block *es = sbi->s_es;
4611         struct ext4_group_desc *gdp;
4612         u64 fsid;
4613         s64 bfree;
4614
4615         if (test_opt(sb, MINIX_DF)) {
4616                 sbi->s_overhead_last = 0;
4617         } else if (es->s_overhead_clusters) {
4618                 sbi->s_overhead_last = le32_to_cpu(es->s_overhead_clusters);
4619         } else if (sbi->s_blocks_last != ext4_blocks_count(es)) {
4620                 ext4_group_t i, ngroups = ext4_get_groups_count(sb);
4621                 ext4_fsblk_t overhead = 0;
4622
4623                 /*
4624                  * Compute the overhead (FS structures).  This is constant
4625                  * for a given filesystem unless the number of block groups
4626                  * changes so we cache the previous value until it does.
4627                  */
4628
4629                 /*
4630                  * All of the blocks before first_data_block are
4631                  * overhead
4632                  */
4633                 overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
4634
4635                 /*
4636                  * Add the overhead found in each block group
4637                  */
4638                 for (i = 0; i < ngroups; i++) {
4639                         gdp = ext4_get_group_desc(sb, i, NULL);
4640                         overhead += ext4_num_overhead_clusters(sb, i, gdp);
4641                         cond_resched();
4642                 }
4643                 sbi->s_overhead_last = overhead;
4644                 smp_wmb();
4645                 sbi->s_blocks_last = ext4_blocks_count(es);
4646         }
4647
4648         buf->f_type = EXT4_SUPER_MAGIC;
4649         buf->f_bsize = sb->s_blocksize;
4650         buf->f_blocks = (ext4_blocks_count(es) -
4651                          EXT4_C2B(sbi, sbi->s_overhead_last));
4652         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4653                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4654         /* prevent underflow in case that few free space is available */
4655         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4656         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4657         if (buf->f_bfree < ext4_r_blocks_count(es))
4658                 buf->f_bavail = 0;
4659         buf->f_files = le32_to_cpu(es->s_inodes_count);
4660         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4661         buf->f_namelen = EXT4_NAME_LEN;
4662         fsid = le64_to_cpup((void *)es->s_uuid) ^
4663                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4664         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4665         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4666
4667         return 0;
4668 }
4669
4670 /* Helper function for writing quotas on sync - we need to start transaction
4671  * before quota file is locked for write. Otherwise the are possible deadlocks:
4672  * Process 1                         Process 2
4673  * ext4_create()                     quota_sync()
4674  *   jbd2_journal_start()                  write_dquot()
4675  *   dquot_initialize()                         down(dqio_mutex)
4676  *     down(dqio_mutex)                    jbd2_journal_start()
4677  *
4678  */
4679
4680 #ifdef CONFIG_QUOTA
4681
4682 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4683 {
4684         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4685 }
4686
4687 static int ext4_write_dquot(struct dquot *dquot)
4688 {
4689         int ret, err;
4690         handle_t *handle;
4691         struct inode *inode;
4692
4693         inode = dquot_to_inode(dquot);
4694         handle = ext4_journal_start(inode,
4695                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4696         if (IS_ERR(handle))
4697                 return PTR_ERR(handle);
4698         ret = dquot_commit(dquot);
4699         err = ext4_journal_stop(handle);
4700         if (!ret)
4701                 ret = err;
4702         return ret;
4703 }
4704
4705 static int ext4_acquire_dquot(struct dquot *dquot)
4706 {
4707         int ret, err;
4708         handle_t *handle;
4709
4710         handle = ext4_journal_start(dquot_to_inode(dquot),
4711                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4712         if (IS_ERR(handle))
4713                 return PTR_ERR(handle);
4714         ret = dquot_acquire(dquot);
4715         err = ext4_journal_stop(handle);
4716         if (!ret)
4717                 ret = err;
4718         return ret;
4719 }
4720
4721 static int ext4_release_dquot(struct dquot *dquot)
4722 {
4723         int ret, err;
4724         handle_t *handle;
4725
4726         handle = ext4_journal_start(dquot_to_inode(dquot),
4727                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4728         if (IS_ERR(handle)) {
4729                 /* Release dquot anyway to avoid endless cycle in dqput() */
4730                 dquot_release(dquot);
4731                 return PTR_ERR(handle);
4732         }
4733         ret = dquot_release(dquot);
4734         err = ext4_journal_stop(handle);
4735         if (!ret)
4736                 ret = err;
4737         return ret;
4738 }
4739
4740 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4741 {
4742         /* Are we journaling quotas? */
4743         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4744             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4745                 dquot_mark_dquot_dirty(dquot);
4746                 return ext4_write_dquot(dquot);
4747         } else {
4748                 return dquot_mark_dquot_dirty(dquot);
4749         }
4750 }
4751
4752 static int ext4_write_info(struct super_block *sb, int type)
4753 {
4754         int ret, err;
4755         handle_t *handle;
4756
4757         /* Data block + inode block */
4758         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4759         if (IS_ERR(handle))
4760                 return PTR_ERR(handle);
4761         ret = dquot_commit_info(sb, type);
4762         err = ext4_journal_stop(handle);
4763         if (!ret)
4764                 ret = err;
4765         return ret;
4766 }
4767
4768 /*
4769  * Turn on quotas during mount time - we need to find
4770  * the quota file and such...
4771  */
4772 static int ext4_quota_on_mount(struct super_block *sb, int type)
4773 {
4774         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4775                                         EXT4_SB(sb)->s_jquota_fmt, type);
4776 }
4777
4778 /*
4779  * Standard function to be called on quota_on
4780  */
4781 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4782                          struct path *path)
4783 {
4784         int err;
4785
4786         if (!test_opt(sb, QUOTA))
4787                 return -EINVAL;
4788
4789         /* Quotafile not on the same filesystem? */
4790         if (path->dentry->d_sb != sb)
4791                 return -EXDEV;
4792         /* Journaling quota? */
4793         if (EXT4_SB(sb)->s_qf_names[type]) {
4794                 /* Quotafile not in fs root? */
4795                 if (path->dentry->d_parent != sb->s_root)
4796                         ext4_msg(sb, KERN_WARNING,
4797                                 "Quota file not on filesystem root. "
4798                                 "Journaled quota will not work");
4799         }
4800
4801         /*
4802          * When we journal data on quota file, we have to flush journal to see
4803          * all updates to the file when we bypass pagecache...
4804          */
4805         if (EXT4_SB(sb)->s_journal &&
4806             ext4_should_journal_data(path->dentry->d_inode)) {
4807                 /*
4808                  * We don't need to lock updates but journal_flush() could
4809                  * otherwise be livelocked...
4810                  */
4811                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4812                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4813                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4814                 if (err)
4815                         return err;
4816         }
4817
4818         return dquot_quota_on(sb, type, format_id, path);
4819 }
4820
4821 static int ext4_quota_off(struct super_block *sb, int type)
4822 {
4823         struct inode *inode = sb_dqopt(sb)->files[type];
4824         handle_t *handle;
4825
4826         /* Force all delayed allocation blocks to be allocated.
4827          * Caller already holds s_umount sem */
4828         if (test_opt(sb, DELALLOC))
4829                 sync_filesystem(sb);
4830
4831         if (!inode)
4832                 goto out;
4833
4834         /* Update modification times of quota files when userspace can
4835          * start looking at them */
4836         handle = ext4_journal_start(inode, 1);
4837         if (IS_ERR(handle))
4838                 goto out;
4839         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
4840         ext4_mark_inode_dirty(handle, inode);
4841         ext4_journal_stop(handle);
4842
4843 out:
4844         return dquot_quota_off(sb, type);
4845 }
4846
4847 /* Read data from quotafile - avoid pagecache and such because we cannot afford
4848  * acquiring the locks... As quota files are never truncated and quota code
4849  * itself serializes the operations (and no one else should touch the files)
4850  * we don't have to be afraid of races */
4851 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
4852                                size_t len, loff_t off)
4853 {
4854         struct inode *inode = sb_dqopt(sb)->files[type];
4855         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4856         int err = 0;
4857         int offset = off & (sb->s_blocksize - 1);
4858         int tocopy;
4859         size_t toread;
4860         struct buffer_head *bh;
4861         loff_t i_size = i_size_read(inode);
4862
4863         if (off > i_size)
4864                 return 0;
4865         if (off+len > i_size)
4866                 len = i_size-off;
4867         toread = len;
4868         while (toread > 0) {
4869                 tocopy = sb->s_blocksize - offset < toread ?
4870                                 sb->s_blocksize - offset : toread;
4871                 bh = ext4_bread(NULL, inode, blk, 0, &err);
4872                 if (err)
4873                         return err;
4874                 if (!bh)        /* A hole? */
4875                         memset(data, 0, tocopy);
4876                 else
4877                         memcpy(data, bh->b_data+offset, tocopy);
4878                 brelse(bh);
4879                 offset = 0;
4880                 toread -= tocopy;
4881                 data += tocopy;
4882                 blk++;
4883         }
4884         return len;
4885 }
4886
4887 /* Write to quotafile (we know the transaction is already started and has
4888  * enough credits) */
4889 static ssize_t ext4_quota_write(struct super_block *sb, int type,
4890                                 const char *data, size_t len, loff_t off)
4891 {
4892         struct inode *inode = sb_dqopt(sb)->files[type];
4893         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
4894         int err = 0;
4895         int offset = off & (sb->s_blocksize - 1);
4896         struct buffer_head *bh;
4897         handle_t *handle = journal_current_handle();
4898
4899         if (EXT4_SB(sb)->s_journal && !handle) {
4900                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4901                         " cancelled because transaction is not started",
4902                         (unsigned long long)off, (unsigned long long)len);
4903                 return -EIO;
4904         }
4905         /*
4906          * Since we account only one data block in transaction credits,
4907          * then it is impossible to cross a block boundary.
4908          */
4909         if (sb->s_blocksize - offset < len) {
4910                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
4911                         " cancelled because not block aligned",
4912                         (unsigned long long)off, (unsigned long long)len);
4913                 return -EIO;
4914         }
4915
4916         bh = ext4_bread(handle, inode, blk, 1, &err);
4917         if (!bh)
4918                 goto out;
4919         err = ext4_journal_get_write_access(handle, bh);
4920         if (err) {
4921                 brelse(bh);
4922                 goto out;
4923         }
4924         lock_buffer(bh);
4925         memcpy(bh->b_data+offset, data, len);
4926         flush_dcache_page(bh->b_page);
4927         unlock_buffer(bh);
4928         err = ext4_handle_dirty_metadata(handle, NULL, bh);
4929         brelse(bh);
4930 out:
4931         if (err)
4932                 return err;
4933         if (inode->i_size < off + len) {
4934                 i_size_write(inode, off + len);
4935                 EXT4_I(inode)->i_disksize = inode->i_size;
4936                 ext4_mark_inode_dirty(handle, inode);
4937         }
4938         return len;
4939 }
4940
4941 #endif
4942
4943 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
4944                        const char *dev_name, void *data)
4945 {
4946         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
4947 }
4948
4949 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4950 static inline void register_as_ext2(void)
4951 {
4952         int err = register_filesystem(&ext2_fs_type);
4953         if (err)
4954                 printk(KERN_WARNING
4955                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
4956 }
4957
4958 static inline void unregister_as_ext2(void)
4959 {
4960         unregister_filesystem(&ext2_fs_type);
4961 }
4962
4963 static inline int ext2_feature_set_ok(struct super_block *sb)
4964 {
4965         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
4966                 return 0;
4967         if (sb->s_flags & MS_RDONLY)
4968                 return 1;
4969         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
4970                 return 0;
4971         return 1;
4972 }
4973 MODULE_ALIAS("ext2");
4974 #else
4975 static inline void register_as_ext2(void) { }
4976 static inline void unregister_as_ext2(void) { }
4977 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
4978 #endif
4979
4980 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
4981 static inline void register_as_ext3(void)
4982 {
4983         int err = register_filesystem(&ext3_fs_type);
4984         if (err)
4985                 printk(KERN_WARNING
4986                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
4987 }
4988
4989 static inline void unregister_as_ext3(void)
4990 {
4991         unregister_filesystem(&ext3_fs_type);
4992 }
4993
4994 static inline int ext3_feature_set_ok(struct super_block *sb)
4995 {
4996         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
4997                 return 0;
4998         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
4999                 return 0;
5000         if (sb->s_flags & MS_RDONLY)
5001                 return 1;
5002         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5003                 return 0;
5004         return 1;
5005 }
5006 MODULE_ALIAS("ext3");
5007 #else
5008 static inline void register_as_ext3(void) { }
5009 static inline void unregister_as_ext3(void) { }
5010 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5011 #endif
5012
5013 static struct file_system_type ext4_fs_type = {
5014         .owner          = THIS_MODULE,
5015         .name           = "ext4",
5016         .mount          = ext4_mount,
5017         .kill_sb        = kill_block_super,
5018         .fs_flags       = FS_REQUIRES_DEV,
5019 };
5020
5021 static int __init ext4_init_feat_adverts(void)
5022 {
5023         struct ext4_features *ef;
5024         int ret = -ENOMEM;
5025
5026         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5027         if (!ef)
5028                 goto out;
5029
5030         ef->f_kobj.kset = ext4_kset;
5031         init_completion(&ef->f_kobj_unregister);
5032         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5033                                    "features");
5034         if (ret) {
5035                 kfree(ef);
5036                 goto out;
5037         }
5038
5039         ext4_feat = ef;
5040         ret = 0;
5041 out:
5042         return ret;
5043 }
5044
5045 static void ext4_exit_feat_adverts(void)
5046 {
5047         kobject_put(&ext4_feat->f_kobj);
5048         wait_for_completion(&ext4_feat->f_kobj_unregister);
5049         kfree(ext4_feat);
5050 }
5051
5052 /* Shared across all ext4 file systems */
5053 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5054 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5055
5056 static int __init ext4_init_fs(void)
5057 {
5058         int i, err;
5059
5060         ext4_li_info = NULL;
5061         mutex_init(&ext4_li_mtx);
5062
5063         ext4_check_flag_values();
5064
5065         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5066                 mutex_init(&ext4__aio_mutex[i]);
5067                 init_waitqueue_head(&ext4__ioend_wq[i]);
5068         }
5069
5070         err = ext4_init_pageio();
5071         if (err)
5072                 return err;
5073         err = ext4_init_system_zone();
5074         if (err)
5075                 goto out6;
5076         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5077         if (!ext4_kset)
5078                 goto out5;
5079         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5080
5081         err = ext4_init_feat_adverts();
5082         if (err)
5083                 goto out4;
5084
5085         err = ext4_init_mballoc();
5086         if (err)
5087                 goto out3;
5088
5089         err = ext4_init_xattr();
5090         if (err)
5091                 goto out2;
5092         err = init_inodecache();
5093         if (err)
5094                 goto out1;
5095         register_as_ext3();
5096         register_as_ext2();
5097         err = register_filesystem(&ext4_fs_type);
5098         if (err)
5099                 goto out;
5100
5101         return 0;
5102 out:
5103         unregister_as_ext2();
5104         unregister_as_ext3();
5105         destroy_inodecache();
5106 out1:
5107         ext4_exit_xattr();
5108 out2:
5109         ext4_exit_mballoc();
5110 out3:
5111         ext4_exit_feat_adverts();
5112 out4:
5113         if (ext4_proc_root)
5114                 remove_proc_entry("fs/ext4", NULL);
5115         kset_unregister(ext4_kset);
5116 out5:
5117         ext4_exit_system_zone();
5118 out6:
5119         ext4_exit_pageio();
5120         return err;
5121 }
5122
5123 static void __exit ext4_exit_fs(void)
5124 {
5125         ext4_destroy_lazyinit_thread();
5126         unregister_as_ext2();
5127         unregister_as_ext3();
5128         unregister_filesystem(&ext4_fs_type);
5129         destroy_inodecache();
5130         ext4_exit_xattr();
5131         ext4_exit_mballoc();
5132         ext4_exit_feat_adverts();
5133         remove_proc_entry("fs/ext4", NULL);
5134         kset_unregister(ext4_kset);
5135         ext4_exit_system_zone();
5136         ext4_exit_pageio();
5137 }
5138
5139 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5140 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5141 MODULE_LICENSE("GPL");
5142 module_init(ext4_init_fs)
5143 module_exit(ext4_exit_fs)