fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static const char *ext4_decode_error(struct super_block *sb, int errno,
  73                                      char nbuf[16]);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  96 #else
  97 #define IS_EXT2_SB(sb) (0)
  98 #endif
  99
 100
 101 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 102 static struct file_system_type ext3_fs_type = {
 103         .owner          = THIS_MODULE,
 104         .name           = "ext3",
 105         .mount          = ext4_mount,
 106         .kill_sb        = kill_block_super,
 107         .fs_flags       = FS_REQUIRES_DEV,
 108 };
 109 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 110 #else
 111 #define IS_EXT3_SB(sb) (0)
 112 #endif
 113
 114 static int ext4_verify_csum_type(struct super_block *sb,
 115                                  struct ext4_super_block *es)
 116 {
 117         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 118                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 119                 return 1;
 120
 121         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 122 }
 123
 124 static __le32 ext4_superblock_csum(struct super_block *sb,
 125                                    struct ext4_super_block *es)
 126 {
 127         struct ext4_sb_info *sbi = EXT4_SB(sb);
 128         int offset = offsetof(struct ext4_super_block, s_checksum);
 129         __u32 csum;
 130
 131         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 132
 133         return cpu_to_le32(csum);
 134 }
 135
 136 int ext4_superblock_csum_verify(struct super_block *sb,
 137                                 struct ext4_super_block *es)
 138 {
 139         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 140                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 141                 return 1;
 142
 143         return es->s_checksum == ext4_superblock_csum(sb, es);
 144 }
 145
 146 void ext4_superblock_csum_set(struct super_block *sb,
 147                               struct ext4_super_block *es)
 148 {
 149         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 150                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 151                 return;
 152
 153         es->s_checksum = ext4_superblock_csum(sb, es);
 154 }
 155
 156 void *ext4_kvmalloc(size_t size, gfp_t flags)
 157 {
 158         void *ret;
 159
 160         ret = kmalloc(size, flags);
 161         if (!ret)
 162                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 163         return ret;
 164 }
 165
 166 void *ext4_kvzalloc(size_t size, gfp_t flags)
 167 {
 168         void *ret;
 169
 170         ret = kzalloc(size, flags);
 171         if (!ret)
 172                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 173         return ret;
 174 }
 175
 176 void ext4_kvfree(void *ptr)
 177 {
 178         if (is_vmalloc_addr(ptr))
 179                 vfree(ptr);
 180         else
 181                 kfree(ptr);
 182
 183 }
 184
 185 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 186                                struct ext4_group_desc *bg)
 187 {
 188         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 189                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 190                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 191 }
 192
 193 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 194                                struct ext4_group_desc *bg)
 195 {
 196         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 197                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 198                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 199 }
 200
 201 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 202                               struct ext4_group_desc *bg)
 203 {
 204         return le32_to_cpu(bg->bg_inode_table_lo) |
 205                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 206                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 207 }
 208
 209 __u32 ext4_free_group_clusters(struct super_block *sb,
 210                                struct ext4_group_desc *bg)
 211 {
 212         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 213                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 214                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 215 }
 216
 217 __u32 ext4_free_inodes_count(struct super_block *sb,
 218                               struct ext4_group_desc *bg)
 219 {
 220         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 221                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 222                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 223 }
 224
 225 __u32 ext4_used_dirs_count(struct super_block *sb,
 226                               struct ext4_group_desc *bg)
 227 {
 228         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 229                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 230                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 231 }
 232
 233 __u32 ext4_itable_unused_count(struct super_block *sb,
 234                               struct ext4_group_desc *bg)
 235 {
 236         return le16_to_cpu(bg->bg_itable_unused_lo) |
 237                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 238                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 239 }
 240
 241 void ext4_block_bitmap_set(struct super_block *sb,
 242                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 243 {
 244         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 245         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 246                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 247 }
 248
 249 void ext4_inode_bitmap_set(struct super_block *sb,
 250                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 251 {
 252         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 253         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 254                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 255 }
 256
 257 void ext4_inode_table_set(struct super_block *sb,
 258                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 259 {
 260         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 261         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 262                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 263 }
 264
 265 void ext4_free_group_clusters_set(struct super_block *sb,
 266                                   struct ext4_group_desc *bg, __u32 count)
 267 {
 268         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 269         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 270                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 271 }
 272
 273 void ext4_free_inodes_set(struct super_block *sb,
 274                           struct ext4_group_desc *bg, __u32 count)
 275 {
 276         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 277         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 278                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 279 }
 280
 281 void ext4_used_dirs_set(struct super_block *sb,
 282                           struct ext4_group_desc *bg, __u32 count)
 283 {
 284         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 285         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 286                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 287 }
 288
 289 void ext4_itable_unused_set(struct super_block *sb,
 290                           struct ext4_group_desc *bg, __u32 count)
 291 {
 292         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 293         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 294                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 295 }
 296
 297
 298 /* Just increment the non-pointer handle value */
 299 static handle_t *ext4_get_nojournal(void)
 300 {
 301         handle_t *handle = current->journal_info;
 302         unsigned long ref_cnt = (unsigned long)handle;
 303
 304         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 305
 306         ref_cnt++;
 307         handle = (handle_t *)ref_cnt;
 308
 309         current->journal_info = handle;
 310         return handle;
 311 }
 312
 313
 314 /* Decrement the non-pointer handle value */
 315 static void ext4_put_nojournal(handle_t *handle)
 316 {
 317         unsigned long ref_cnt = (unsigned long)handle;
 318
 319         BUG_ON(ref_cnt == 0);
 320
 321         ref_cnt--;
 322         handle = (handle_t *)ref_cnt;
 323
 324         current->journal_info = handle;
 325 }
 326
 327 /*
 328  * Wrappers for jbd2_journal_start/end.
 329  *
 330  * The only special thing we need to do here is to make sure that all
 331  * journal_end calls result in the superblock being marked dirty, so
 332  * that sync() will call the filesystem's write_super callback if
 333  * appropriate.
 334  */
 335 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 336 {
 337         journal_t *journal;
 338
 339         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 340         if (sb->s_flags & MS_RDONLY)
 341                 return ERR_PTR(-EROFS);
 342
 343         WARN_ON(sb->s_writers.frozen == SB_FREEZE_COMPLETE);
 344         journal = EXT4_SB(sb)->s_journal;
 345         if (!journal)
 346                 return ext4_get_nojournal();
 347         /*
 348          * Special case here: if the journal has aborted behind our
 349          * backs (eg. EIO in the commit thread), then we still need to
 350          * take the FS itself readonly cleanly.
 351          */
 352         if (is_journal_aborted(journal)) {
 353                 ext4_abort(sb, "Detected aborted journal");
 354                 return ERR_PTR(-EROFS);
 355         }
 356         return jbd2_journal_start(journal, nblocks);
 357 }
 358
 359 /*
 360  * The only special thing we need to do here is to make sure that all
 361  * jbd2_journal_stop calls result in the superblock being marked dirty, so
 362  * that sync() will call the filesystem's write_super callback if
 363  * appropriate.
 364  */
 365 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 366 {
 367         struct super_block *sb;
 368         int err;
 369         int rc;
 370
 371         if (!ext4_handle_valid(handle)) {
 372                 ext4_put_nojournal(handle);
 373                 return 0;
 374         }
 375         sb = handle->h_transaction->t_journal->j_private;
 376         err = handle->h_err;
 377         rc = jbd2_journal_stop(handle);
 378
 379         if (!err)
 380                 err = rc;
 381         if (err)
 382                 __ext4_std_error(sb, where, line, err);
 383         return err;
 384 }
 385
 386 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 387                                const char *err_fn, struct buffer_head *bh,
 388                                handle_t *handle, int err)
 389 {
 390         char nbuf[16];
 391         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 392
 393         BUG_ON(!ext4_handle_valid(handle));
 394
 395         if (bh)
 396                 BUFFER_TRACE(bh, "abort");
 397
 398         if (!handle->h_err)
 399                 handle->h_err = err;
 400
 401         if (is_handle_aborted(handle))
 402                 return;
 403
 404         printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
 405                caller, line, errstr, err_fn);
 406
 407         jbd2_journal_abort_handle(handle);
 408 }
 409
 410 static void __save_error_info(struct super_block *sb, const char *func,
 411                             unsigned int line)
 412 {
 413         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 414
 415         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 416         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 417         es->s_last_error_time = cpu_to_le32(get_seconds());
 418         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 419         es->s_last_error_line = cpu_to_le32(line);
 420         if (!es->s_first_error_time) {
 421                 es->s_first_error_time = es->s_last_error_time;
 422                 strncpy(es->s_first_error_func, func,
 423                         sizeof(es->s_first_error_func));
 424                 es->s_first_error_line = cpu_to_le32(line);
 425                 es->s_first_error_ino = es->s_last_error_ino;
 426                 es->s_first_error_block = es->s_last_error_block;
 427         }
 428         /*
 429          * Start the daily error reporting function if it hasn't been
 430          * started already
 431          */
 432         if (!es->s_error_count)
 433                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 434         es->s_error_count = cpu_to_le32(le32_to_cpu(es->s_error_count) + 1);
 435 }
 436
 437 static void save_error_info(struct super_block *sb, const char *func,
 438                             unsigned int line)
 439 {
 440         __save_error_info(sb, func, line);
 441         ext4_commit_super(sb, 1);
 442 }
 443
 444 /*
 445  * The del_gendisk() function uninitializes the disk-specific data
 446  * structures, including the bdi structure, without telling anyone
 447  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 448  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 449  * This is a kludge to prevent these oops until we can put in a proper
 450  * hook in del_gendisk() to inform the VFS and file system layers.
 451  */
 452 static int block_device_ejected(struct super_block *sb)
 453 {
 454         struct inode *bd_inode = sb->s_bdev->bd_inode;
 455         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 456
 457         return bdi->dev == NULL;
 458 }
 459
 460 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 461 {
 462         struct super_block              *sb = journal->j_private;
 463         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 464         int                             error = is_journal_aborted(journal);
 465         struct ext4_journal_cb_entry    *jce, *tmp;
 466
 467         spin_lock(&sbi->s_md_lock);
 468         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 469                 list_del_init(&jce->jce_list);
 470                 spin_unlock(&sbi->s_md_lock);
 471                 jce->jce_func(sb, jce, error);
 472                 spin_lock(&sbi->s_md_lock);
 473         }
 474         spin_unlock(&sbi->s_md_lock);
 475 }
 476
 477 /* Deal with the reporting of failure conditions on a filesystem such as
 478  * inconsistencies detected or read IO failures.
 479  *
 480  * On ext2, we can store the error state of the filesystem in the
 481  * superblock.  That is not possible on ext4, because we may have other
 482  * write ordering constraints on the superblock which prevent us from
 483  * writing it out straight away; and given that the journal is about to
 484  * be aborted, we can't rely on the current, or future, transactions to
 485  * write out the superblock safely.
 486  *
 487  * We'll just use the jbd2_journal_abort() error code to record an error in
 488  * the journal instead.  On recovery, the journal will complain about
 489  * that error until we've noted it down and cleared it.
 490  */
 491
 492 static void ext4_handle_error(struct super_block *sb)
 493 {
 494         if (sb->s_flags & MS_RDONLY)
 495                 return;
 496
 497         if (!test_opt(sb, ERRORS_CONT)) {
 498                 journal_t *journal = EXT4_SB(sb)->s_journal;
 499
 500                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 501                 if (journal)
 502                         jbd2_journal_abort(journal, -EIO);
 503         }
 504         if (test_opt(sb, ERRORS_RO)) {
 505                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 506                 sb->s_flags |= MS_RDONLY;
 507         }
 508         if (test_opt(sb, ERRORS_PANIC))
 509                 panic("EXT4-fs (device %s): panic forced after error\n",
 510                         sb->s_id);
 511 }
 512
 513 void __ext4_error(struct super_block *sb, const char *function,
 514                   unsigned int line, const char *fmt, ...)
 515 {
 516         struct va_format vaf;
 517         va_list args;
 518
 519         va_start(args, fmt);
 520         vaf.fmt = fmt;
 521         vaf.va = &args;
 522         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 523                sb->s_id, function, line, current->comm, &vaf);
 524         va_end(args);
 525         save_error_info(sb, function, line);
 526
 527         ext4_handle_error(sb);
 528 }
 529
 530 void ext4_error_inode(struct inode *inode, const char *function,
 531                       unsigned int line, ext4_fsblk_t block,
 532                       const char *fmt, ...)
 533 {
 534         va_list args;
 535         struct va_format vaf;
 536         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 537
 538         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 539         es->s_last_error_block = cpu_to_le64(block);
 540         save_error_info(inode->i_sb, function, line);
 541         va_start(args, fmt);
 542         vaf.fmt = fmt;
 543         vaf.va = &args;
 544         if (block)
 545                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 546                        "inode #%lu: block %llu: comm %s: %pV\n",
 547                        inode->i_sb->s_id, function, line, inode->i_ino,
 548                        block, current->comm, &vaf);
 549         else
 550                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 551                        "inode #%lu: comm %s: %pV\n",
 552                        inode->i_sb->s_id, function, line, inode->i_ino,
 553                        current->comm, &vaf);
 554         va_end(args);
 555
 556         ext4_handle_error(inode->i_sb);
 557 }
 558
 559 void ext4_error_file(struct file *file, const char *function,
 560                      unsigned int line, ext4_fsblk_t block,
 561                      const char *fmt, ...)
 562 {
 563         va_list args;
 564         struct va_format vaf;
 565         struct ext4_super_block *es;
 566         struct inode *inode = file->f_dentry->d_inode;
 567         char pathname[80], *path;
 568
 569         es = EXT4_SB(inode->i_sb)->s_es;
 570         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 571         save_error_info(inode->i_sb, function, line);
 572         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 573         if (IS_ERR(path))
 574                 path = "(unknown)";
 575         va_start(args, fmt);
 576         vaf.fmt = fmt;
 577         vaf.va = &args;
 578         if (block)
 579                 printk(KERN_CRIT
 580                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 581                        "block %llu: comm %s: path %s: %pV\n",
 582                        inode->i_sb->s_id, function, line, inode->i_ino,
 583                        block, current->comm, path, &vaf);
 584         else
 585                 printk(KERN_CRIT
 586                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 587                        "comm %s: path %s: %pV\n",
 588                        inode->i_sb->s_id, function, line, inode->i_ino,
 589                        current->comm, path, &vaf);
 590         va_end(args);
 591
 592         ext4_handle_error(inode->i_sb);
 593 }
 594
 595 static const char *ext4_decode_error(struct super_block *sb, int errno,
 596                                      char nbuf[16])
 597 {
 598         char *errstr = NULL;
 599
 600         switch (errno) {
 601         case -EIO:
 602                 errstr = "IO failure";
 603                 break;
 604         case -ENOMEM:
 605                 errstr = "Out of memory";
 606                 break;
 607         case -EROFS:
 608                 if (!sb || (EXT4_SB(sb)->s_journal &&
 609                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 610                         errstr = "Journal has aborted";
 611                 else
 612                         errstr = "Readonly filesystem";
 613                 break;
 614         default:
 615                 /* If the caller passed in an extra buffer for unknown
 616                  * errors, textualise them now.  Else we just return
 617                  * NULL. */
 618                 if (nbuf) {
 619                         /* Check for truncated error codes... */
 620                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 621                                 errstr = nbuf;
 622                 }
 623                 break;
 624         }
 625
 626         return errstr;
 627 }
 628
 629 /* __ext4_std_error decodes expected errors from journaling functions
 630  * automatically and invokes the appropriate error response.  */
 631
 632 void __ext4_std_error(struct super_block *sb, const char *function,
 633                       unsigned int line, int errno)
 634 {
 635         char nbuf[16];
 636         const char *errstr;
 637
 638         /* Special case: if the error is EROFS, and we're not already
 639          * inside a transaction, then there's really no point in logging
 640          * an error. */
 641         if (errno == -EROFS && journal_current_handle() == NULL &&
 642             (sb->s_flags & MS_RDONLY))
 643                 return;
 644
 645         errstr = ext4_decode_error(sb, errno, nbuf);
 646         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 647                sb->s_id, function, line, errstr);
 648         save_error_info(sb, function, line);
 649
 650         ext4_handle_error(sb);
 651 }
 652
 653 /*
 654  * ext4_abort is a much stronger failure handler than ext4_error.  The
 655  * abort function may be used to deal with unrecoverable failures such
 656  * as journal IO errors or ENOMEM at a critical moment in log management.
 657  *
 658  * We unconditionally force the filesystem into an ABORT|READONLY state,
 659  * unless the error response on the fs has been set to panic in which
 660  * case we take the easy way out and panic immediately.
 661  */
 662
 663 void __ext4_abort(struct super_block *sb, const char *function,
 664                 unsigned int line, const char *fmt, ...)
 665 {
 666         va_list args;
 667
 668         save_error_info(sb, function, line);
 669         va_start(args, fmt);
 670         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 671                function, line);
 672         vprintk(fmt, args);
 673         printk("\n");
 674         va_end(args);
 675
 676         if ((sb->s_flags & MS_RDONLY) == 0) {
 677                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 678                 sb->s_flags |= MS_RDONLY;
 679                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 680                 if (EXT4_SB(sb)->s_journal)
 681                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 682                 save_error_info(sb, function, line);
 683         }
 684         if (test_opt(sb, ERRORS_PANIC))
 685                 panic("EXT4-fs panic from previous error\n");
 686 }
 687
 688 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 689 {
 690         struct va_format vaf;
 691         va_list args;
 692
 693         va_start(args, fmt);
 694         vaf.fmt = fmt;
 695         vaf.va = &args;
 696         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 697         va_end(args);
 698 }
 699
 700 void __ext4_warning(struct super_block *sb, const char *function,
 701                     unsigned int line, const char *fmt, ...)
 702 {
 703         struct va_format vaf;
 704         va_list args;
 705
 706         va_start(args, fmt);
 707         vaf.fmt = fmt;
 708         vaf.va = &args;
 709         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 710                sb->s_id, function, line, &vaf);
 711         va_end(args);
 712 }
 713
 714 void __ext4_grp_locked_error(const char *function, unsigned int line,
 715                              struct super_block *sb, ext4_group_t grp,
 716                              unsigned long ino, ext4_fsblk_t block,
 717                              const char *fmt, ...)
 718 __releases(bitlock)
 719 __acquires(bitlock)
 720 {
 721         struct va_format vaf;
 722         va_list args;
 723         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 724
 725         es->s_last_error_ino = cpu_to_le32(ino);
 726         es->s_last_error_block = cpu_to_le64(block);
 727         __save_error_info(sb, function, line);
 728
 729         va_start(args, fmt);
 730
 731         vaf.fmt = fmt;
 732         vaf.va = &args;
 733         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 734                sb->s_id, function, line, grp);
 735         if (ino)
 736                 printk(KERN_CONT "inode %lu: ", ino);
 737         if (block)
 738                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 739         printk(KERN_CONT "%pV\n", &vaf);
 740         va_end(args);
 741
 742         if (test_opt(sb, ERRORS_CONT)) {
 743                 ext4_commit_super(sb, 0);
 744                 return;
 745         }
 746
 747         ext4_unlock_group(sb, grp);
 748         ext4_handle_error(sb);
 749         /*
 750          * We only get here in the ERRORS_RO case; relocking the group
 751          * may be dangerous, but nothing bad will happen since the
 752          * filesystem will have already been marked read/only and the
 753          * journal has been aborted.  We return 1 as a hint to callers
 754          * who might what to use the return value from
 755          * ext4_grp_locked_error() to distinguish between the
 756          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 757          * aggressively from the ext4 function in question, with a
 758          * more appropriate error code.
 759          */
 760         ext4_lock_group(sb, grp);
 761         return;
 762 }
 763
 764 void ext4_update_dynamic_rev(struct super_block *sb)
 765 {
 766         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 767
 768         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 769                 return;
 770
 771         ext4_warning(sb,
 772                      "updating to rev %d because of new feature flag, "
 773                      "running e2fsck is recommended",
 774                      EXT4_DYNAMIC_REV);
 775
 776         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 777         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 778         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 779         /* leave es->s_feature_*compat flags alone */
 780         /* es->s_uuid will be set by e2fsck if empty */
 781
 782         /*
 783          * The rest of the superblock fields should be zero, and if not it
 784          * means they are likely already in use, so leave them alone.  We
 785          * can leave it up to e2fsck to clean up any inconsistencies there.
 786          */
 787 }
 788
 789 /*
 790  * Open the external journal device
 791  */
 792 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 793 {
 794         struct block_device *bdev;
 795         char b[BDEVNAME_SIZE];
 796
 797         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 798         if (IS_ERR(bdev))
 799                 goto fail;
 800         return bdev;
 801
 802 fail:
 803         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 804                         __bdevname(dev, b), PTR_ERR(bdev));
 805         return NULL;
 806 }
 807
 808 /*
 809  * Release the journal device
 810  */
 811 static int ext4_blkdev_put(struct block_device *bdev)
 812 {
 813         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 814 }
 815
 816 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 817 {
 818         struct block_device *bdev;
 819         int ret = -ENODEV;
 820
 821         bdev = sbi->journal_bdev;
 822         if (bdev) {
 823                 ret = ext4_blkdev_put(bdev);
 824                 sbi->journal_bdev = NULL;
 825         }
 826         return ret;
 827 }
 828
 829 static inline struct inode *orphan_list_entry(struct list_head *l)
 830 {
 831         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 832 }
 833
 834 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 835 {
 836         struct list_head *l;
 837
 838         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 839                  le32_to_cpu(sbi->s_es->s_last_orphan));
 840
 841         printk(KERN_ERR "sb_info orphan list:\n");
 842         list_for_each(l, &sbi->s_orphan) {
 843                 struct inode *inode = orphan_list_entry(l);
 844                 printk(KERN_ERR "  "
 845                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 846                        inode->i_sb->s_id, inode->i_ino, inode,
 847                        inode->i_mode, inode->i_nlink,
 848                        NEXT_ORPHAN(inode));
 849         }
 850 }
 851
 852 static void ext4_put_super(struct super_block *sb)
 853 {
 854         struct ext4_sb_info *sbi = EXT4_SB(sb);
 855         struct ext4_super_block *es = sbi->s_es;
 856         int i, err;
 857
 858         ext4_unregister_li_request(sb);
 859         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 860
 861         flush_workqueue(sbi->dio_unwritten_wq);
 862         destroy_workqueue(sbi->dio_unwritten_wq);
 863
 864         lock_super(sb);
 865         if (sbi->s_journal) {
 866                 err = jbd2_journal_destroy(sbi->s_journal);
 867                 sbi->s_journal = NULL;
 868                 if (err < 0)
 869                         ext4_abort(sb, "Couldn't clean up the journal");
 870         }
 871
 872         del_timer(&sbi->s_err_report);
 873         ext4_release_system_zone(sb);
 874         ext4_mb_release(sb);
 875         ext4_ext_release(sb);
 876         ext4_xattr_put_super(sb);
 877
 878         if (!(sb->s_flags & MS_RDONLY)) {
 879                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 880                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 881         }
 882         if (!(sb->s_flags & MS_RDONLY))
 883                 ext4_commit_super(sb, 1);
 884
 885         if (sbi->s_proc) {
 886                 remove_proc_entry("options", sbi->s_proc);
 887                 remove_proc_entry(sb->s_id, ext4_proc_root);
 888         }
 889         kobject_del(&sbi->s_kobj);
 890
 891         for (i = 0; i < sbi->s_gdb_count; i++)
 892                 brelse(sbi->s_group_desc[i]);
 893         ext4_kvfree(sbi->s_group_desc);
 894         ext4_kvfree(sbi->s_flex_groups);
 895         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 896         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 897         percpu_counter_destroy(&sbi->s_dirs_counter);
 898         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 899         brelse(sbi->s_sbh);
 900 #ifdef CONFIG_QUOTA
 901         for (i = 0; i < MAXQUOTAS; i++)
 902                 kfree(sbi->s_qf_names[i]);
 903 #endif
 904
 905         /* Debugging code just in case the in-memory inode orphan list
 906          * isn't empty.  The on-disk one can be non-empty if we've
 907          * detected an error and taken the fs readonly, but the
 908          * in-memory list had better be clean by this point. */
 909         if (!list_empty(&sbi->s_orphan))
 910                 dump_orphan_list(sb, sbi);
 911         J_ASSERT(list_empty(&sbi->s_orphan));
 912
 913         invalidate_bdev(sb->s_bdev);
 914         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 915                 /*
 916                  * Invalidate the journal device's buffers.  We don't want them
 917                  * floating about in memory - the physical journal device may
 918                  * hotswapped, and it breaks the `ro-after' testing code.
 919                  */
 920                 sync_blockdev(sbi->journal_bdev);
 921                 invalidate_bdev(sbi->journal_bdev);
 922                 ext4_blkdev_remove(sbi);
 923         }
 924         if (sbi->s_mmp_tsk)
 925                 kthread_stop(sbi->s_mmp_tsk);
 926         sb->s_fs_info = NULL;
 927         /*
 928          * Now that we are completely done shutting down the
 929          * superblock, we need to actually destroy the kobject.
 930          */
 931         unlock_super(sb);
 932         kobject_put(&sbi->s_kobj);
 933         wait_for_completion(&sbi->s_kobj_unregister);
 934         if (sbi->s_chksum_driver)
 935                 crypto_free_shash(sbi->s_chksum_driver);
 936         kfree(sbi->s_blockgroup_lock);
 937         kfree(sbi);
 938 }
 939
 940 static struct kmem_cache *ext4_inode_cachep;
 941
 942 /*
 943  * Called inside transaction, so use GFP_NOFS
 944  */
 945 static struct inode *ext4_alloc_inode(struct super_block *sb)
 946 {
 947         struct ext4_inode_info *ei;
 948
 949         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 950         if (!ei)
 951                 return NULL;
 952
 953         ei->vfs_inode.i_version = 1;
 954         ei->vfs_inode.i_data.writeback_index = 0;
 955         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 956         INIT_LIST_HEAD(&ei->i_prealloc_list);
 957         spin_lock_init(&ei->i_prealloc_lock);
 958         ei->i_reserved_data_blocks = 0;
 959         ei->i_reserved_meta_blocks = 0;
 960         ei->i_allocated_meta_blocks = 0;
 961         ei->i_da_metadata_calc_len = 0;
 962         ei->i_da_metadata_calc_last_lblock = 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_on_sysfile(struct super_block *sb, int type,
1125                                  int format_id);
1126 static int ext4_quota_off(struct super_block *sb, int type);
1127 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1128 static int ext4_quota_on_mount(struct super_block *sb, int type);
1129 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1130                                size_t len, loff_t off);
1131 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1132                                 const char *data, size_t len, loff_t off);
1133 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1134                              unsigned int flags);
1135 static int ext4_enable_quotas(struct super_block *sb);
1136
1137 static const struct dquot_operations ext4_quota_operations = {
1138         .get_reserved_space = ext4_get_reserved_space,
1139         .write_dquot    = ext4_write_dquot,
1140         .acquire_dquot  = ext4_acquire_dquot,
1141         .release_dquot  = ext4_release_dquot,
1142         .mark_dirty     = ext4_mark_dquot_dirty,
1143         .write_info     = ext4_write_info,
1144         .alloc_dquot    = dquot_alloc,
1145         .destroy_dquot  = dquot_destroy,
1146 };
1147
1148 static const struct quotactl_ops ext4_qctl_operations = {
1149         .quota_on       = ext4_quota_on,
1150         .quota_off      = ext4_quota_off,
1151         .quota_sync     = dquot_quota_sync,
1152         .get_info       = dquot_get_dqinfo,
1153         .set_info       = dquot_set_dqinfo,
1154         .get_dqblk      = dquot_get_dqblk,
1155         .set_dqblk      = dquot_set_dqblk
1156 };
1157
1158 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1159         .quota_on_meta  = ext4_quota_on_sysfile,
1160         .quota_off      = ext4_quota_off_sysfile,
1161         .quota_sync     = dquot_quota_sync,
1162         .get_info       = dquot_get_dqinfo,
1163         .set_info       = dquot_set_dqinfo,
1164         .get_dqblk      = dquot_get_dqblk,
1165         .set_dqblk      = dquot_set_dqblk
1166 };
1167 #endif
1168
1169 static const struct super_operations ext4_sops = {
1170         .alloc_inode    = ext4_alloc_inode,
1171         .destroy_inode  = ext4_destroy_inode,
1172         .write_inode    = ext4_write_inode,
1173         .dirty_inode    = ext4_dirty_inode,
1174         .drop_inode     = ext4_drop_inode,
1175         .evict_inode    = ext4_evict_inode,
1176         .put_super      = ext4_put_super,
1177         .sync_fs        = ext4_sync_fs,
1178         .freeze_fs      = ext4_freeze,
1179         .unfreeze_fs    = ext4_unfreeze,
1180         .statfs         = ext4_statfs,
1181         .remount_fs     = ext4_remount,
1182         .show_options   = ext4_show_options,
1183 #ifdef CONFIG_QUOTA
1184         .quota_read     = ext4_quota_read,
1185         .quota_write    = ext4_quota_write,
1186 #endif
1187         .bdev_try_to_free_page = bdev_try_to_free_page,
1188 };
1189
1190 static const struct super_operations ext4_nojournal_sops = {
1191         .alloc_inode    = ext4_alloc_inode,
1192         .destroy_inode  = ext4_destroy_inode,
1193         .write_inode    = ext4_write_inode,
1194         .dirty_inode    = ext4_dirty_inode,
1195         .drop_inode     = ext4_drop_inode,
1196         .evict_inode    = ext4_evict_inode,
1197         .put_super      = ext4_put_super,
1198         .statfs         = ext4_statfs,
1199         .remount_fs     = ext4_remount,
1200         .show_options   = ext4_show_options,
1201 #ifdef CONFIG_QUOTA
1202         .quota_read     = ext4_quota_read,
1203         .quota_write    = ext4_quota_write,
1204 #endif
1205         .bdev_try_to_free_page = bdev_try_to_free_page,
1206 };
1207
1208 static const struct export_operations ext4_export_ops = {
1209         .fh_to_dentry = ext4_fh_to_dentry,
1210         .fh_to_parent = ext4_fh_to_parent,
1211         .get_parent = ext4_get_parent,
1212 };
1213
1214 enum {
1215         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1216         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1217         Opt_nouid32, Opt_debug, Opt_removed,
1218         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1219         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1220         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1221         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1222         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1223         Opt_data_err_abort, Opt_data_err_ignore,
1224         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1225         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1226         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1227         Opt_usrquota, Opt_grpquota, Opt_i_version,
1228         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1229         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1230         Opt_inode_readahead_blks, Opt_journal_ioprio,
1231         Opt_dioread_nolock, Opt_dioread_lock,
1232         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1233 };
1234
1235 static const match_table_t tokens = {
1236         {Opt_bsd_df, "bsddf"},
1237         {Opt_minix_df, "minixdf"},
1238         {Opt_grpid, "grpid"},
1239         {Opt_grpid, "bsdgroups"},
1240         {Opt_nogrpid, "nogrpid"},
1241         {Opt_nogrpid, "sysvgroups"},
1242         {Opt_resgid, "resgid=%u"},
1243         {Opt_resuid, "resuid=%u"},
1244         {Opt_sb, "sb=%u"},
1245         {Opt_err_cont, "errors=continue"},
1246         {Opt_err_panic, "errors=panic"},
1247         {Opt_err_ro, "errors=remount-ro"},
1248         {Opt_nouid32, "nouid32"},
1249         {Opt_debug, "debug"},
1250         {Opt_removed, "oldalloc"},
1251         {Opt_removed, "orlov"},
1252         {Opt_user_xattr, "user_xattr"},
1253         {Opt_nouser_xattr, "nouser_xattr"},
1254         {Opt_acl, "acl"},
1255         {Opt_noacl, "noacl"},
1256         {Opt_noload, "norecovery"},
1257         {Opt_noload, "noload"},
1258         {Opt_removed, "nobh"},
1259         {Opt_removed, "bh"},
1260         {Opt_commit, "commit=%u"},
1261         {Opt_min_batch_time, "min_batch_time=%u"},
1262         {Opt_max_batch_time, "max_batch_time=%u"},
1263         {Opt_journal_dev, "journal_dev=%u"},
1264         {Opt_journal_checksum, "journal_checksum"},
1265         {Opt_journal_async_commit, "journal_async_commit"},
1266         {Opt_abort, "abort"},
1267         {Opt_data_journal, "data=journal"},
1268         {Opt_data_ordered, "data=ordered"},
1269         {Opt_data_writeback, "data=writeback"},
1270         {Opt_data_err_abort, "data_err=abort"},
1271         {Opt_data_err_ignore, "data_err=ignore"},
1272         {Opt_offusrjquota, "usrjquota="},
1273         {Opt_usrjquota, "usrjquota=%s"},
1274         {Opt_offgrpjquota, "grpjquota="},
1275         {Opt_grpjquota, "grpjquota=%s"},
1276         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1277         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1278         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1279         {Opt_grpquota, "grpquota"},
1280         {Opt_noquota, "noquota"},
1281         {Opt_quota, "quota"},
1282         {Opt_usrquota, "usrquota"},
1283         {Opt_barrier, "barrier=%u"},
1284         {Opt_barrier, "barrier"},
1285         {Opt_nobarrier, "nobarrier"},
1286         {Opt_i_version, "i_version"},
1287         {Opt_stripe, "stripe=%u"},
1288         {Opt_delalloc, "delalloc"},
1289         {Opt_nodelalloc, "nodelalloc"},
1290         {Opt_mblk_io_submit, "mblk_io_submit"},
1291         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1292         {Opt_block_validity, "block_validity"},
1293         {Opt_noblock_validity, "noblock_validity"},
1294         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1295         {Opt_journal_ioprio, "journal_ioprio=%u"},
1296         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1297         {Opt_auto_da_alloc, "auto_da_alloc"},
1298         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1299         {Opt_dioread_nolock, "dioread_nolock"},
1300         {Opt_dioread_lock, "dioread_lock"},
1301         {Opt_discard, "discard"},
1302         {Opt_nodiscard, "nodiscard"},
1303         {Opt_init_itable, "init_itable=%u"},
1304         {Opt_init_itable, "init_itable"},
1305         {Opt_noinit_itable, "noinit_itable"},
1306         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1307         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1308         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1309         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1310         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1311         {Opt_err, NULL},
1312 };
1313
1314 static ext4_fsblk_t get_sb_block(void **data)
1315 {
1316         ext4_fsblk_t    sb_block;
1317         char            *options = (char *) *data;
1318
1319         if (!options || strncmp(options, "sb=", 3) != 0)
1320                 return 1;       /* Default location */
1321
1322         options += 3;
1323         /* TODO: use simple_strtoll with >32bit ext4 */
1324         sb_block = simple_strtoul(options, &options, 0);
1325         if (*options && *options != ',') {
1326                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1327                        (char *) *data);
1328                 return 1;
1329         }
1330         if (*options == ',')
1331                 options++;
1332         *data = (void *) options;
1333
1334         return sb_block;
1335 }
1336
1337 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1338 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1339         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1340
1341 #ifdef CONFIG_QUOTA
1342 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1343 {
1344         struct ext4_sb_info *sbi = EXT4_SB(sb);
1345         char *qname;
1346
1347         if (sb_any_quota_loaded(sb) &&
1348                 !sbi->s_qf_names[qtype]) {
1349                 ext4_msg(sb, KERN_ERR,
1350                         "Cannot change journaled "
1351                         "quota options when quota turned on");
1352                 return -1;
1353         }
1354         qname = match_strdup(args);
1355         if (!qname) {
1356                 ext4_msg(sb, KERN_ERR,
1357                         "Not enough memory for storing quotafile name");
1358                 return -1;
1359         }
1360         if (sbi->s_qf_names[qtype] &&
1361                 strcmp(sbi->s_qf_names[qtype], qname)) {
1362                 ext4_msg(sb, KERN_ERR,
1363                         "%s quota file already specified", QTYPE2NAME(qtype));
1364                 kfree(qname);
1365                 return -1;
1366         }
1367         sbi->s_qf_names[qtype] = qname;
1368         if (strchr(sbi->s_qf_names[qtype], '/')) {
1369                 ext4_msg(sb, KERN_ERR,
1370                         "quotafile must be on filesystem root");
1371                 kfree(sbi->s_qf_names[qtype]);
1372                 sbi->s_qf_names[qtype] = NULL;
1373                 return -1;
1374         }
1375         set_opt(sb, QUOTA);
1376         return 1;
1377 }
1378
1379 static int clear_qf_name(struct super_block *sb, int qtype)
1380 {
1381
1382         struct ext4_sb_info *sbi = EXT4_SB(sb);
1383
1384         if (sb_any_quota_loaded(sb) &&
1385                 sbi->s_qf_names[qtype]) {
1386                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1387                         " when quota turned on");
1388                 return -1;
1389         }
1390         /*
1391          * The space will be released later when all options are confirmed
1392          * to be correct
1393          */
1394         sbi->s_qf_names[qtype] = NULL;
1395         return 1;
1396 }
1397 #endif
1398
1399 #define MOPT_SET        0x0001
1400 #define MOPT_CLEAR      0x0002
1401 #define MOPT_NOSUPPORT  0x0004
1402 #define MOPT_EXPLICIT   0x0008
1403 #define MOPT_CLEAR_ERR  0x0010
1404 #define MOPT_GTE0       0x0020
1405 #ifdef CONFIG_QUOTA
1406 #define MOPT_Q          0
1407 #define MOPT_QFMT       0x0040
1408 #else
1409 #define MOPT_Q          MOPT_NOSUPPORT
1410 #define MOPT_QFMT       MOPT_NOSUPPORT
1411 #endif
1412 #define MOPT_DATAJ      0x0080
1413
1414 static const struct mount_opts {
1415         int     token;
1416         int     mount_opt;
1417         int     flags;
1418 } ext4_mount_opts[] = {
1419         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1420         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1421         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1422         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1423         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1424         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1425         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1426         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1427         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1428         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1429         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1430         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1431         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1432         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1433         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1434         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1435                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1436         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1437         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1438         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1439         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1440         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1441         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1442         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1443         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1444         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1445         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1446         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1447         {Opt_commit, 0, MOPT_GTE0},
1448         {Opt_max_batch_time, 0, MOPT_GTE0},
1449         {Opt_min_batch_time, 0, MOPT_GTE0},
1450         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1451         {Opt_init_itable, 0, MOPT_GTE0},
1452         {Opt_stripe, 0, MOPT_GTE0},
1453         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1454         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1455         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1456 #ifdef CONFIG_EXT4_FS_XATTR
1457         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1458         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1459 #else
1460         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1461         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1462 #endif
1463 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1464         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1465         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1466 #else
1467         {Opt_acl, 0, MOPT_NOSUPPORT},
1468         {Opt_noacl, 0, MOPT_NOSUPPORT},
1469 #endif
1470         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1471         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1472         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1473         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1474                                                         MOPT_SET | MOPT_Q},
1475         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1476                                                         MOPT_SET | MOPT_Q},
1477         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1478                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1479         {Opt_usrjquota, 0, MOPT_Q},
1480         {Opt_grpjquota, 0, MOPT_Q},
1481         {Opt_offusrjquota, 0, MOPT_Q},
1482         {Opt_offgrpjquota, 0, MOPT_Q},
1483         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1484         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1485         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1486         {Opt_err, 0, 0}
1487 };
1488
1489 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1490                             substring_t *args, unsigned long *journal_devnum,
1491                             unsigned int *journal_ioprio, int is_remount)
1492 {
1493         struct ext4_sb_info *sbi = EXT4_SB(sb);
1494         const struct mount_opts *m;
1495         kuid_t uid;
1496         kgid_t gid;
1497         int arg = 0;
1498
1499 #ifdef CONFIG_QUOTA
1500         if (token == Opt_usrjquota)
1501                 return set_qf_name(sb, USRQUOTA, &args[0]);
1502         else if (token == Opt_grpjquota)
1503                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1504         else if (token == Opt_offusrjquota)
1505                 return clear_qf_name(sb, USRQUOTA);
1506         else if (token == Opt_offgrpjquota)
1507                 return clear_qf_name(sb, GRPQUOTA);
1508 #endif
1509         if (args->from && match_int(args, &arg))
1510                 return -1;
1511         switch (token) {
1512         case Opt_noacl:
1513         case Opt_nouser_xattr:
1514                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1515                 break;
1516         case Opt_sb:
1517                 return 1;       /* handled by get_sb_block() */
1518         case Opt_removed:
1519                 ext4_msg(sb, KERN_WARNING,
1520                          "Ignoring removed %s option", opt);
1521                 return 1;
1522         case Opt_resuid:
1523                 uid = make_kuid(current_user_ns(), arg);
1524                 if (!uid_valid(uid)) {
1525                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1526                         return -1;
1527                 }
1528                 sbi->s_resuid = uid;
1529                 return 1;
1530         case Opt_resgid:
1531                 gid = make_kgid(current_user_ns(), arg);
1532                 if (!gid_valid(gid)) {
1533                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1534                         return -1;
1535                 }
1536                 sbi->s_resgid = gid;
1537                 return 1;
1538         case Opt_abort:
1539                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1540                 return 1;
1541         case Opt_i_version:
1542                 sb->s_flags |= MS_I_VERSION;
1543                 return 1;
1544         case Opt_journal_dev:
1545                 if (is_remount) {
1546                         ext4_msg(sb, KERN_ERR,
1547                                  "Cannot specify journal on remount");
1548                         return -1;
1549                 }
1550                 *journal_devnum = arg;
1551                 return 1;
1552         case Opt_journal_ioprio:
1553                 if (arg < 0 || arg > 7)
1554                         return -1;
1555                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1556                 return 1;
1557         }
1558
1559         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1560                 if (token != m->token)
1561                         continue;
1562                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1563                         return -1;
1564                 if (m->flags & MOPT_EXPLICIT)
1565                         set_opt2(sb, EXPLICIT_DELALLOC);
1566                 if (m->flags & MOPT_CLEAR_ERR)
1567                         clear_opt(sb, ERRORS_MASK);
1568                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1569                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1570                                  "options when quota turned on");
1571                         return -1;
1572                 }
1573
1574                 if (m->flags & MOPT_NOSUPPORT) {
1575                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1576                 } else if (token == Opt_commit) {
1577                         if (arg == 0)
1578                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1579                         sbi->s_commit_interval = HZ * arg;
1580                 } else if (token == Opt_max_batch_time) {
1581                         if (arg == 0)
1582                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1583                         sbi->s_max_batch_time = arg;
1584                 } else if (token == Opt_min_batch_time) {
1585                         sbi->s_min_batch_time = arg;
1586                 } else if (token == Opt_inode_readahead_blks) {
1587                         if (arg > (1 << 30))
1588                                 return -1;
1589                         if (arg && !is_power_of_2(arg)) {
1590                                 ext4_msg(sb, KERN_ERR,
1591                                          "EXT4-fs: inode_readahead_blks"
1592                                          " must be a power of 2");
1593                                 return -1;
1594                         }
1595                         sbi->s_inode_readahead_blks = arg;
1596                 } else if (token == Opt_init_itable) {
1597                         set_opt(sb, INIT_INODE_TABLE);
1598                         if (!args->from)
1599                                 arg = EXT4_DEF_LI_WAIT_MULT;
1600                         sbi->s_li_wait_mult = arg;
1601                 } else if (token == Opt_stripe) {
1602                         sbi->s_stripe = arg;
1603                 } else if (m->flags & MOPT_DATAJ) {
1604                         if (is_remount) {
1605                                 if (!sbi->s_journal)
1606                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1607                                 else if (test_opt(sb, DATA_FLAGS) !=
1608                                          m->mount_opt) {
1609                                         ext4_msg(sb, KERN_ERR,
1610                                          "Cannot change data mode on remount");
1611                                         return -1;
1612                                 }
1613                         } else {
1614                                 clear_opt(sb, DATA_FLAGS);
1615                                 sbi->s_mount_opt |= m->mount_opt;
1616                         }
1617 #ifdef CONFIG_QUOTA
1618                 } else if (m->flags & MOPT_QFMT) {
1619                         if (sb_any_quota_loaded(sb) &&
1620                             sbi->s_jquota_fmt != m->mount_opt) {
1621                                 ext4_msg(sb, KERN_ERR, "Cannot "
1622                                          "change journaled quota options "
1623                                          "when quota turned on");
1624                                 return -1;
1625                         }
1626                         sbi->s_jquota_fmt = m->mount_opt;
1627 #endif
1628                 } else {
1629                         if (!args->from)
1630                                 arg = 1;
1631                         if (m->flags & MOPT_CLEAR)
1632                                 arg = !arg;
1633                         else if (unlikely(!(m->flags & MOPT_SET))) {
1634                                 ext4_msg(sb, KERN_WARNING,
1635                                          "buggy handling of option %s", opt);
1636                                 WARN_ON(1);
1637                                 return -1;
1638                         }
1639                         if (arg != 0)
1640                                 sbi->s_mount_opt |= m->mount_opt;
1641                         else
1642                                 sbi->s_mount_opt &= ~m->mount_opt;
1643                 }
1644                 return 1;
1645         }
1646         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1647                  "or missing value", opt);
1648         return -1;
1649 }
1650
1651 static int parse_options(char *options, struct super_block *sb,
1652                          unsigned long *journal_devnum,
1653                          unsigned int *journal_ioprio,
1654                          int is_remount)
1655 {
1656 #ifdef CONFIG_QUOTA
1657         struct ext4_sb_info *sbi = EXT4_SB(sb);
1658 #endif
1659         char *p;
1660         substring_t args[MAX_OPT_ARGS];
1661         int token;
1662
1663         if (!options)
1664                 return 1;
1665
1666         while ((p = strsep(&options, ",")) != NULL) {
1667                 if (!*p)
1668                         continue;
1669                 /*
1670                  * Initialize args struct so we know whether arg was
1671                  * found; some options take optional arguments.
1672                  */
1673                 args[0].to = args[0].from = 0;
1674                 token = match_token(p, tokens, args);
1675                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1676                                      journal_ioprio, is_remount) < 0)
1677                         return 0;
1678         }
1679 #ifdef CONFIG_QUOTA
1680         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1681                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1682                         clear_opt(sb, USRQUOTA);
1683
1684                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1685                         clear_opt(sb, GRPQUOTA);
1686
1687                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1688                         ext4_msg(sb, KERN_ERR, "old and new quota "
1689                                         "format mixing");
1690                         return 0;
1691                 }
1692
1693                 if (!sbi->s_jquota_fmt) {
1694                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1695                                         "not specified");
1696                         return 0;
1697                 }
1698         } else {
1699                 if (sbi->s_jquota_fmt) {
1700                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1701                                         "specified with no journaling "
1702                                         "enabled");
1703                         return 0;
1704                 }
1705         }
1706 #endif
1707         return 1;
1708 }
1709
1710 static inline void ext4_show_quota_options(struct seq_file *seq,
1711                                            struct super_block *sb)
1712 {
1713 #if defined(CONFIG_QUOTA)
1714         struct ext4_sb_info *sbi = EXT4_SB(sb);
1715
1716         if (sbi->s_jquota_fmt) {
1717                 char *fmtname = "";
1718
1719                 switch (sbi->s_jquota_fmt) {
1720                 case QFMT_VFS_OLD:
1721                         fmtname = "vfsold";
1722                         break;
1723                 case QFMT_VFS_V0:
1724                         fmtname = "vfsv0";
1725                         break;
1726                 case QFMT_VFS_V1:
1727                         fmtname = "vfsv1";
1728                         break;
1729                 }
1730                 seq_printf(seq, ",jqfmt=%s", fmtname);
1731         }
1732
1733         if (sbi->s_qf_names[USRQUOTA])
1734                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1735
1736         if (sbi->s_qf_names[GRPQUOTA])
1737                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1738
1739         if (test_opt(sb, USRQUOTA))
1740                 seq_puts(seq, ",usrquota");
1741
1742         if (test_opt(sb, GRPQUOTA))
1743                 seq_puts(seq, ",grpquota");
1744 #endif
1745 }
1746
1747 static const char *token2str(int token)
1748 {
1749         static const struct match_token *t;
1750
1751         for (t = tokens; t->token != Opt_err; t++)
1752                 if (t->token == token && !strchr(t->pattern, '='))
1753                         break;
1754         return t->pattern;
1755 }
1756
1757 /*
1758  * Show an option if
1759  *  - it's set to a non-default value OR
1760  *  - if the per-sb default is different from the global default
1761  */
1762 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1763                               int nodefs)
1764 {
1765         struct ext4_sb_info *sbi = EXT4_SB(sb);
1766         struct ext4_super_block *es = sbi->s_es;
1767         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1768         const struct mount_opts *m;
1769         char sep = nodefs ? '\n' : ',';
1770
1771 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1772 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1773
1774         if (sbi->s_sb_block != 1)
1775                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1776
1777         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1778                 int want_set = m->flags & MOPT_SET;
1779                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1780                     (m->flags & MOPT_CLEAR_ERR))
1781                         continue;
1782                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1783                         continue; /* skip if same as the default */
1784                 if ((want_set &&
1785                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1786                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1787                         continue; /* select Opt_noFoo vs Opt_Foo */
1788                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1789         }
1790
1791         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1792             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1793                 SEQ_OPTS_PRINT("resuid=%u",
1794                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1795         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1796             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1797                 SEQ_OPTS_PRINT("resgid=%u",
1798                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1799         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1800         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1801                 SEQ_OPTS_PUTS("errors=remount-ro");
1802         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1803                 SEQ_OPTS_PUTS("errors=continue");
1804         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1805                 SEQ_OPTS_PUTS("errors=panic");
1806         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1807                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1808         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1809                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1810         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1811                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1812         if (sb->s_flags & MS_I_VERSION)
1813                 SEQ_OPTS_PUTS("i_version");
1814         if (nodefs || sbi->s_stripe)
1815                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1816         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1817                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1818                         SEQ_OPTS_PUTS("data=journal");
1819                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1820                         SEQ_OPTS_PUTS("data=ordered");
1821                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1822                         SEQ_OPTS_PUTS("data=writeback");
1823         }
1824         if (nodefs ||
1825             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1826                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1827                                sbi->s_inode_readahead_blks);
1828
1829         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1830                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1831                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1832
1833         ext4_show_quota_options(seq, sb);
1834         return 0;
1835 }
1836
1837 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1838 {
1839         return _ext4_show_options(seq, root->d_sb, 0);
1840 }
1841
1842 static int options_seq_show(struct seq_file *seq, void *offset)
1843 {
1844         struct super_block *sb = seq->private;
1845         int rc;
1846
1847         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1848         rc = _ext4_show_options(seq, sb, 1);
1849         seq_puts(seq, "\n");
1850         return rc;
1851 }
1852
1853 static int options_open_fs(struct inode *inode, struct file *file)
1854 {
1855         return single_open(file, options_seq_show, PDE(inode)->data);
1856 }
1857
1858 static const struct file_operations ext4_seq_options_fops = {
1859         .owner = THIS_MODULE,
1860         .open = options_open_fs,
1861         .read = seq_read,
1862         .llseek = seq_lseek,
1863         .release = single_release,
1864 };
1865
1866 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1867                             int read_only)
1868 {
1869         struct ext4_sb_info *sbi = EXT4_SB(sb);
1870         int res = 0;
1871
1872         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1873                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1874                          "forcing read-only mode");
1875                 res = MS_RDONLY;
1876         }
1877         if (read_only)
1878                 goto done;
1879         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1880                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1881                          "running e2fsck is recommended");
1882         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1883                 ext4_msg(sb, KERN_WARNING,
1884                          "warning: mounting fs with errors, "
1885                          "running e2fsck is recommended");
1886         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1887                  le16_to_cpu(es->s_mnt_count) >=
1888                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1889                 ext4_msg(sb, KERN_WARNING,
1890                          "warning: maximal mount count reached, "
1891                          "running e2fsck is recommended");
1892         else if (le32_to_cpu(es->s_checkinterval) &&
1893                 (le32_to_cpu(es->s_lastcheck) +
1894                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1895                 ext4_msg(sb, KERN_WARNING,
1896                          "warning: checktime reached, "
1897                          "running e2fsck is recommended");
1898         if (!sbi->s_journal)
1899                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1900         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1901                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1902         le16_add_cpu(&es->s_mnt_count, 1);
1903         es->s_mtime = cpu_to_le32(get_seconds());
1904         ext4_update_dynamic_rev(sb);
1905         if (sbi->s_journal)
1906                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1907
1908         ext4_commit_super(sb, 1);
1909 done:
1910         if (test_opt(sb, DEBUG))
1911                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1912                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1913                         sb->s_blocksize,
1914                         sbi->s_groups_count,
1915                         EXT4_BLOCKS_PER_GROUP(sb),
1916                         EXT4_INODES_PER_GROUP(sb),
1917                         sbi->s_mount_opt, sbi->s_mount_opt2);
1918
1919         cleancache_init_fs(sb);
1920         return res;
1921 }
1922
1923 static int ext4_fill_flex_info(struct super_block *sb)
1924 {
1925         struct ext4_sb_info *sbi = EXT4_SB(sb);
1926         struct ext4_group_desc *gdp = NULL;
1927         ext4_group_t flex_group_count;
1928         ext4_group_t flex_group;
1929         unsigned int groups_per_flex = 0;
1930         size_t size;
1931         int i;
1932
1933         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1934         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1935                 sbi->s_log_groups_per_flex = 0;
1936                 return 1;
1937         }
1938         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1939
1940         /* We allocate both existing and potentially added groups */
1941         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1942                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1943                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1944         size = flex_group_count * sizeof(struct flex_groups);
1945         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1946         if (sbi->s_flex_groups == NULL) {
1947                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1948                          flex_group_count);
1949                 goto failed;
1950         }
1951
1952         for (i = 0; i < sbi->s_groups_count; i++) {
1953                 gdp = ext4_get_group_desc(sb, i, NULL);
1954
1955                 flex_group = ext4_flex_group(sbi, i);
1956                 atomic_add(ext4_free_inodes_count(sb, gdp),
1957                            &sbi->s_flex_groups[flex_group].free_inodes);
1958                 atomic_add(ext4_free_group_clusters(sb, gdp),
1959                            &sbi->s_flex_groups[flex_group].free_clusters);
1960                 atomic_add(ext4_used_dirs_count(sb, gdp),
1961                            &sbi->s_flex_groups[flex_group].used_dirs);
1962         }
1963
1964         return 1;
1965 failed:
1966         return 0;
1967 }
1968
1969 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1970                                    struct ext4_group_desc *gdp)
1971 {
1972         int offset;
1973         __u16 crc = 0;
1974         __le32 le_group = cpu_to_le32(block_group);
1975
1976         if ((sbi->s_es->s_feature_ro_compat &
1977              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1978                 /* Use new metadata_csum algorithm */
1979                 __u16 old_csum;
1980                 __u32 csum32;
1981
1982                 old_csum = gdp->bg_checksum;
1983                 gdp->bg_checksum = 0;
1984                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1985                                      sizeof(le_group));
1986                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1987                                      sbi->s_desc_size);
1988                 gdp->bg_checksum = old_csum;
1989
1990                 crc = csum32 & 0xFFFF;
1991                 goto out;
1992         }
1993
1994         /* old crc16 code */
1995         offset = offsetof(struct ext4_group_desc, bg_checksum);
1996
1997         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1998         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1999         crc = crc16(crc, (__u8 *)gdp, offset);
2000         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2001         /* for checksum of struct ext4_group_desc do the rest...*/
2002         if ((sbi->s_es->s_feature_incompat &
2003              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2004             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2005                 crc = crc16(crc, (__u8 *)gdp + offset,
2006                             le16_to_cpu(sbi->s_es->s_desc_size) -
2007                                 offset);
2008
2009 out:
2010         return cpu_to_le16(crc);
2011 }
2012
2013 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2014                                 struct ext4_group_desc *gdp)
2015 {
2016         if (ext4_has_group_desc_csum(sb) &&
2017             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2018                                                       block_group, gdp)))
2019                 return 0;
2020
2021         return 1;
2022 }
2023
2024 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2025                               struct ext4_group_desc *gdp)
2026 {
2027         if (!ext4_has_group_desc_csum(sb))
2028                 return;
2029         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2030 }
2031
2032 /* Called at mount-time, super-block is locked */
2033 static int ext4_check_descriptors(struct super_block *sb,
2034                                   ext4_group_t *first_not_zeroed)
2035 {
2036         struct ext4_sb_info *sbi = EXT4_SB(sb);
2037         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2038         ext4_fsblk_t last_block;
2039         ext4_fsblk_t block_bitmap;
2040         ext4_fsblk_t inode_bitmap;
2041         ext4_fsblk_t inode_table;
2042         int flexbg_flag = 0;
2043         ext4_group_t i, grp = sbi->s_groups_count;
2044
2045         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2046                 flexbg_flag = 1;
2047
2048         ext4_debug("Checking group descriptors");
2049
2050         for (i = 0; i < sbi->s_groups_count; i++) {
2051                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2052
2053                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2054                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2055                 else
2056                         last_block = first_block +
2057                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2058
2059                 if ((grp == sbi->s_groups_count) &&
2060                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2061                         grp = i;
2062
2063                 block_bitmap = ext4_block_bitmap(sb, gdp);
2064                 if (block_bitmap < first_block || block_bitmap > last_block) {
2065                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2066                                "Block bitmap for group %u not in group "
2067                                "(block %llu)!", i, block_bitmap);
2068                         return 0;
2069                 }
2070                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2071                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2072                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2073                                "Inode bitmap for group %u not in group "
2074                                "(block %llu)!", i, inode_bitmap);
2075                         return 0;
2076                 }
2077                 inode_table = ext4_inode_table(sb, gdp);
2078                 if (inode_table < first_block ||
2079                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2080                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2081                                "Inode table for group %u not in group "
2082                                "(block %llu)!", i, inode_table);
2083                         return 0;
2084                 }
2085                 ext4_lock_group(sb, i);
2086                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2087                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2088                                  "Checksum for group %u failed (%u!=%u)",
2089                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2090                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2091                         if (!(sb->s_flags & MS_RDONLY)) {
2092                                 ext4_unlock_group(sb, i);
2093                                 return 0;
2094                         }
2095                 }
2096                 ext4_unlock_group(sb, i);
2097                 if (!flexbg_flag)
2098                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2099         }
2100         if (NULL != first_not_zeroed)
2101                 *first_not_zeroed = grp;
2102
2103         ext4_free_blocks_count_set(sbi->s_es,
2104                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2105         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2106         return 1;
2107 }
2108
2109 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2110  * the superblock) which were deleted from all directories, but held open by
2111  * a process at the time of a crash.  We walk the list and try to delete these
2112  * inodes at recovery time (only with a read-write filesystem).
2113  *
2114  * In order to keep the orphan inode chain consistent during traversal (in
2115  * case of crash during recovery), we link each inode into the superblock
2116  * orphan list_head and handle it the same way as an inode deletion during
2117  * normal operation (which journals the operations for us).
2118  *
2119  * We only do an iget() and an iput() on each inode, which is very safe if we
2120  * accidentally point at an in-use or already deleted inode.  The worst that
2121  * can happen in this case is that we get a "bit already cleared" message from
2122  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2123  * e2fsck was run on this filesystem, and it must have already done the orphan
2124  * inode cleanup for us, so we can safely abort without any further action.
2125  */
2126 static void ext4_orphan_cleanup(struct super_block *sb,
2127                                 struct ext4_super_block *es)
2128 {
2129         unsigned int s_flags = sb->s_flags;
2130         int nr_orphans = 0, nr_truncates = 0;
2131 #ifdef CONFIG_QUOTA
2132         int i;
2133 #endif
2134         if (!es->s_last_orphan) {
2135                 jbd_debug(4, "no orphan inodes to clean up\n");
2136                 return;
2137         }
2138
2139         if (bdev_read_only(sb->s_bdev)) {
2140                 ext4_msg(sb, KERN_ERR, "write access "
2141                         "unavailable, skipping orphan cleanup");
2142                 return;
2143         }
2144
2145         /* Check if feature set would not allow a r/w mount */
2146         if (!ext4_feature_set_ok(sb, 0)) {
2147                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2148                          "unknown ROCOMPAT features");
2149                 return;
2150         }
2151
2152         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2153                 if (es->s_last_orphan)
2154                         jbd_debug(1, "Errors on filesystem, "
2155                                   "clearing orphan list.\n");
2156                 es->s_last_orphan = 0;
2157                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2158                 return;
2159         }
2160
2161         if (s_flags & MS_RDONLY) {
2162                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2163                 sb->s_flags &= ~MS_RDONLY;
2164         }
2165 #ifdef CONFIG_QUOTA
2166         /* Needed for iput() to work correctly and not trash data */
2167         sb->s_flags |= MS_ACTIVE;
2168         /* Turn on quotas so that they are updated correctly */
2169         for (i = 0; i < MAXQUOTAS; i++) {
2170                 if (EXT4_SB(sb)->s_qf_names[i]) {
2171                         int ret = ext4_quota_on_mount(sb, i);
2172                         if (ret < 0)
2173                                 ext4_msg(sb, KERN_ERR,
2174                                         "Cannot turn on journaled "
2175                                         "quota: error %d", ret);
2176                 }
2177         }
2178 #endif
2179
2180         while (es->s_last_orphan) {
2181                 struct inode *inode;
2182
2183                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2184                 if (IS_ERR(inode)) {
2185                         es->s_last_orphan = 0;
2186                         break;
2187                 }
2188
2189                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2190                 dquot_initialize(inode);
2191                 if (inode->i_nlink) {
2192                         ext4_msg(sb, KERN_DEBUG,
2193                                 "%s: truncating inode %lu to %lld bytes",
2194                                 __func__, inode->i_ino, inode->i_size);
2195                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2196                                   inode->i_ino, inode->i_size);
2197                         ext4_truncate(inode);
2198                         nr_truncates++;
2199                 } else {
2200                         ext4_msg(sb, KERN_DEBUG,
2201                                 "%s: deleting unreferenced inode %lu",
2202                                 __func__, inode->i_ino);
2203                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2204                                   inode->i_ino);
2205                         nr_orphans++;
2206                 }
2207                 iput(inode);  /* The delete magic happens here! */
2208         }
2209
2210 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2211
2212         if (nr_orphans)
2213                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2214                        PLURAL(nr_orphans));
2215         if (nr_truncates)
2216                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2217                        PLURAL(nr_truncates));
2218 #ifdef CONFIG_QUOTA
2219         /* Turn quotas off */
2220         for (i = 0; i < MAXQUOTAS; i++) {
2221                 if (sb_dqopt(sb)->files[i])
2222                         dquot_quota_off(sb, i);
2223         }
2224 #endif
2225         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2226 }
2227
2228 /*
2229  * Maximal extent format file size.
2230  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2231  * extent format containers, within a sector_t, and within i_blocks
2232  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2233  * so that won't be a limiting factor.
2234  *
2235  * However there is other limiting factor. We do store extents in the form
2236  * of starting block and length, hence the resulting length of the extent
2237  * covering maximum file size must fit into on-disk format containers as
2238  * well. Given that length is always by 1 unit bigger than max unit (because
2239  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2240  *
2241  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2242  */
2243 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2244 {
2245         loff_t res;
2246         loff_t upper_limit = MAX_LFS_FILESIZE;
2247
2248         /* small i_blocks in vfs inode? */
2249         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2250                 /*
2251                  * CONFIG_LBDAF is not enabled implies the inode
2252                  * i_block represent total blocks in 512 bytes
2253                  * 32 == size of vfs inode i_blocks * 8
2254                  */
2255                 upper_limit = (1LL << 32) - 1;
2256
2257                 /* total blocks in file system block size */
2258                 upper_limit >>= (blkbits - 9);
2259                 upper_limit <<= blkbits;
2260         }
2261
2262         /*
2263          * 32-bit extent-start container, ee_block. We lower the maxbytes
2264          * by one fs block, so ee_len can cover the extent of maximum file
2265          * size
2266          */
2267         res = (1LL << 32) - 1;
2268         res <<= blkbits;
2269
2270         /* Sanity check against vm- & vfs- imposed limits */
2271         if (res > upper_limit)
2272                 res = upper_limit;
2273
2274         return res;
2275 }
2276
2277 /*
2278  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2279  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2280  * We need to be 1 filesystem block less than the 2^48 sector limit.
2281  */
2282 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2283 {
2284         loff_t res = EXT4_NDIR_BLOCKS;
2285         int meta_blocks;
2286         loff_t upper_limit;
2287         /* This is calculated to be the largest file size for a dense, block
2288          * mapped file such that the file's total number of 512-byte sectors,
2289          * including data and all indirect blocks, does not exceed (2^48 - 1).
2290          *
2291          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2292          * number of 512-byte sectors of the file.
2293          */
2294
2295         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2296                 /*
2297                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2298                  * the inode i_block field represents total file blocks in
2299                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2300                  */
2301                 upper_limit = (1LL << 32) - 1;
2302
2303                 /* total blocks in file system block size */
2304                 upper_limit >>= (bits - 9);
2305
2306         } else {
2307                 /*
2308                  * We use 48 bit ext4_inode i_blocks
2309                  * With EXT4_HUGE_FILE_FL set the i_blocks
2310                  * represent total number of blocks in
2311                  * file system block size
2312                  */
2313                 upper_limit = (1LL << 48) - 1;
2314
2315         }
2316
2317         /* indirect blocks */
2318         meta_blocks = 1;
2319         /* double indirect blocks */
2320         meta_blocks += 1 + (1LL << (bits-2));
2321         /* tripple indirect blocks */
2322         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2323
2324         upper_limit -= meta_blocks;
2325         upper_limit <<= bits;
2326
2327         res += 1LL << (bits-2);
2328         res += 1LL << (2*(bits-2));
2329         res += 1LL << (3*(bits-2));
2330         res <<= bits;
2331         if (res > upper_limit)
2332                 res = upper_limit;
2333
2334         if (res > MAX_LFS_FILESIZE)
2335                 res = MAX_LFS_FILESIZE;
2336
2337         return res;
2338 }
2339
2340 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2341                                    ext4_fsblk_t logical_sb_block, int nr)
2342 {
2343         struct ext4_sb_info *sbi = EXT4_SB(sb);
2344         ext4_group_t bg, first_meta_bg;
2345         int has_super = 0;
2346
2347         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2348
2349         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2350             nr < first_meta_bg)
2351                 return logical_sb_block + nr + 1;
2352         bg = sbi->s_desc_per_block * nr;
2353         if (ext4_bg_has_super(sb, bg))
2354                 has_super = 1;
2355
2356         return (has_super + ext4_group_first_block_no(sb, bg));
2357 }
2358
2359 /**
2360  * ext4_get_stripe_size: Get the stripe size.
2361  * @sbi: In memory super block info
2362  *
2363  * If we have specified it via mount option, then
2364  * use the mount option value. If the value specified at mount time is
2365  * greater than the blocks per group use the super block value.
2366  * If the super block value is greater than blocks per group return 0.
2367  * Allocator needs it be less than blocks per group.
2368  *
2369  */
2370 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2371 {
2372         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2373         unsigned long stripe_width =
2374                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2375         int ret;
2376
2377         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2378                 ret = sbi->s_stripe;
2379         else if (stripe_width <= sbi->s_blocks_per_group)
2380                 ret = stripe_width;
2381         else if (stride <= sbi->s_blocks_per_group)
2382                 ret = stride;
2383         else
2384                 ret = 0;
2385
2386         /*
2387          * If the stripe width is 1, this makes no sense and
2388          * we set it to 0 to turn off stripe handling code.
2389          */
2390         if (ret <= 1)
2391                 ret = 0;
2392
2393         return ret;
2394 }
2395
2396 /* sysfs supprt */
2397
2398 struct ext4_attr {
2399         struct attribute attr;
2400         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2401         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2402                          const char *, size_t);
2403         int offset;
2404 };
2405
2406 static int parse_strtoul(const char *buf,
2407                 unsigned long max, unsigned long *value)
2408 {
2409         char *endp;
2410
2411         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2412         endp = skip_spaces(endp);
2413         if (*endp || *value > max)
2414                 return -EINVAL;
2415
2416         return 0;
2417 }
2418
2419 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2420                                               struct ext4_sb_info *sbi,
2421                                               char *buf)
2422 {
2423         return snprintf(buf, PAGE_SIZE, "%llu\n",
2424                 (s64) EXT4_C2B(sbi,
2425                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2426 }
2427
2428 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2429                                          struct ext4_sb_info *sbi, char *buf)
2430 {
2431         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2432
2433         if (!sb->s_bdev->bd_part)
2434                 return snprintf(buf, PAGE_SIZE, "0\n");
2435         return snprintf(buf, PAGE_SIZE, "%lu\n",
2436                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2437                          sbi->s_sectors_written_start) >> 1);
2438 }
2439
2440 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2441                                           struct ext4_sb_info *sbi, char *buf)
2442 {
2443         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2444
2445         if (!sb->s_bdev->bd_part)
2446                 return snprintf(buf, PAGE_SIZE, "0\n");
2447         return snprintf(buf, PAGE_SIZE, "%llu\n",
2448                         (unsigned long long)(sbi->s_kbytes_written +
2449                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2450                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2451 }
2452
2453 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2454                                           struct ext4_sb_info *sbi,
2455                                           const char *buf, size_t count)
2456 {
2457         unsigned long t;
2458
2459         if (parse_strtoul(buf, 0x40000000, &t))
2460                 return -EINVAL;
2461
2462         if (t && !is_power_of_2(t))
2463                 return -EINVAL;
2464
2465         sbi->s_inode_readahead_blks = t;
2466         return count;
2467 }
2468
2469 static ssize_t sbi_ui_show(struct ext4_attr *a,
2470                            struct ext4_sb_info *sbi, char *buf)
2471 {
2472         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2473
2474         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2475 }
2476
2477 static ssize_t sbi_ui_store(struct ext4_attr *a,
2478                             struct ext4_sb_info *sbi,
2479                             const char *buf, size_t count)
2480 {
2481         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2482         unsigned long t;
2483
2484         if (parse_strtoul(buf, 0xffffffff, &t))
2485                 return -EINVAL;
2486         *ui = t;
2487         return count;
2488 }
2489
2490 static ssize_t trigger_test_error(struct ext4_attr *a,
2491                                   struct ext4_sb_info *sbi,
2492                                   const char *buf, size_t count)
2493 {
2494         int len = count;
2495
2496         if (!capable(CAP_SYS_ADMIN))
2497                 return -EPERM;
2498
2499         if (len && buf[len-1] == '\n')
2500                 len--;
2501
2502         if (len)
2503                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2504         return count;
2505 }
2506
2507 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2508 static struct ext4_attr ext4_attr_##_name = {                   \
2509         .attr = {.name = __stringify(_name), .mode = _mode },   \
2510         .show   = _show,                                        \
2511         .store  = _store,                                       \
2512         .offset = offsetof(struct ext4_sb_info, _elname),       \
2513 }
2514 #define EXT4_ATTR(name, mode, show, store) \
2515 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2516
2517 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2518 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2519 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2520 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2521         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2522 #define ATTR_LIST(name) &ext4_attr_##name.attr
2523
2524 EXT4_RO_ATTR(delayed_allocation_blocks);
2525 EXT4_RO_ATTR(session_write_kbytes);
2526 EXT4_RO_ATTR(lifetime_write_kbytes);
2527 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2528                  inode_readahead_blks_store, s_inode_readahead_blks);
2529 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2530 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2531 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2532 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2533 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2534 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2535 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2536 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2537 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2538
2539 static struct attribute *ext4_attrs[] = {
2540         ATTR_LIST(delayed_allocation_blocks),
2541         ATTR_LIST(session_write_kbytes),
2542         ATTR_LIST(lifetime_write_kbytes),
2543         ATTR_LIST(inode_readahead_blks),
2544         ATTR_LIST(inode_goal),
2545         ATTR_LIST(mb_stats),
2546         ATTR_LIST(mb_max_to_scan),
2547         ATTR_LIST(mb_min_to_scan),
2548         ATTR_LIST(mb_order2_req),
2549         ATTR_LIST(mb_stream_req),
2550         ATTR_LIST(mb_group_prealloc),
2551         ATTR_LIST(max_writeback_mb_bump),
2552         ATTR_LIST(trigger_fs_error),
2553         NULL,
2554 };
2555
2556 /* Features this copy of ext4 supports */
2557 EXT4_INFO_ATTR(lazy_itable_init);
2558 EXT4_INFO_ATTR(batched_discard);
2559
2560 static struct attribute *ext4_feat_attrs[] = {
2561         ATTR_LIST(lazy_itable_init),
2562         ATTR_LIST(batched_discard),
2563         NULL,
2564 };
2565
2566 static ssize_t ext4_attr_show(struct kobject *kobj,
2567                               struct attribute *attr, char *buf)
2568 {
2569         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2570                                                 s_kobj);
2571         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2572
2573         return a->show ? a->show(a, sbi, buf) : 0;
2574 }
2575
2576 static ssize_t ext4_attr_store(struct kobject *kobj,
2577                                struct attribute *attr,
2578                                const char *buf, size_t len)
2579 {
2580         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2581                                                 s_kobj);
2582         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2583
2584         return a->store ? a->store(a, sbi, buf, len) : 0;
2585 }
2586
2587 static void ext4_sb_release(struct kobject *kobj)
2588 {
2589         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2590                                                 s_kobj);
2591         complete(&sbi->s_kobj_unregister);
2592 }
2593
2594 static const struct sysfs_ops ext4_attr_ops = {
2595         .show   = ext4_attr_show,
2596         .store  = ext4_attr_store,
2597 };
2598
2599 static struct kobj_type ext4_ktype = {
2600         .default_attrs  = ext4_attrs,
2601         .sysfs_ops      = &ext4_attr_ops,
2602         .release        = ext4_sb_release,
2603 };
2604
2605 static void ext4_feat_release(struct kobject *kobj)
2606 {
2607         complete(&ext4_feat->f_kobj_unregister);
2608 }
2609
2610 static struct kobj_type ext4_feat_ktype = {
2611         .default_attrs  = ext4_feat_attrs,
2612         .sysfs_ops      = &ext4_attr_ops,
2613         .release        = ext4_feat_release,
2614 };
2615
2616 /*
2617  * Check whether this filesystem can be mounted based on
2618  * the features present and the RDONLY/RDWR mount requested.
2619  * Returns 1 if this filesystem can be mounted as requested,
2620  * 0 if it cannot be.
2621  */
2622 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2623 {
2624         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2625                 ext4_msg(sb, KERN_ERR,
2626                         "Couldn't mount because of "
2627                         "unsupported optional features (%x)",
2628                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2629                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2630                 return 0;
2631         }
2632
2633         if (readonly)
2634                 return 1;
2635
2636         /* Check that feature set is OK for a read-write mount */
2637         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2638                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2639                          "unsupported optional features (%x)",
2640                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2641                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2642                 return 0;
2643         }
2644         /*
2645          * Large file size enabled file system can only be mounted
2646          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2647          */
2648         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2649                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2650                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2651                                  "cannot be mounted RDWR without "
2652                                  "CONFIG_LBDAF");
2653                         return 0;
2654                 }
2655         }
2656         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2657             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2658                 ext4_msg(sb, KERN_ERR,
2659                          "Can't support bigalloc feature without "
2660                          "extents feature\n");
2661                 return 0;
2662         }
2663
2664 #ifndef CONFIG_QUOTA
2665         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2666             !readonly) {
2667                 ext4_msg(sb, KERN_ERR,
2668                          "Filesystem with quota feature cannot be mounted RDWR "
2669                          "without CONFIG_QUOTA");
2670                 return 0;
2671         }
2672 #endif  /* CONFIG_QUOTA */
2673         return 1;
2674 }
2675
2676 /*
2677  * This function is called once a day if we have errors logged
2678  * on the file system
2679  */
2680 static void print_daily_error_info(unsigned long arg)
2681 {
2682         struct super_block *sb = (struct super_block *) arg;
2683         struct ext4_sb_info *sbi;
2684         struct ext4_super_block *es;
2685
2686         sbi = EXT4_SB(sb);
2687         es = sbi->s_es;
2688
2689         if (es->s_error_count)
2690                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2691                          le32_to_cpu(es->s_error_count));
2692         if (es->s_first_error_time) {
2693                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2694                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2695                        (int) sizeof(es->s_first_error_func),
2696                        es->s_first_error_func,
2697                        le32_to_cpu(es->s_first_error_line));
2698                 if (es->s_first_error_ino)
2699                         printk(": inode %u",
2700                                le32_to_cpu(es->s_first_error_ino));
2701                 if (es->s_first_error_block)
2702                         printk(": block %llu", (unsigned long long)
2703                                le64_to_cpu(es->s_first_error_block));
2704                 printk("\n");
2705         }
2706         if (es->s_last_error_time) {
2707                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2708                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2709                        (int) sizeof(es->s_last_error_func),
2710                        es->s_last_error_func,
2711                        le32_to_cpu(es->s_last_error_line));
2712                 if (es->s_last_error_ino)
2713                         printk(": inode %u",
2714                                le32_to_cpu(es->s_last_error_ino));
2715                 if (es->s_last_error_block)
2716                         printk(": block %llu", (unsigned long long)
2717                                le64_to_cpu(es->s_last_error_block));
2718                 printk("\n");
2719         }
2720         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2721 }
2722
2723 /* Find next suitable group and run ext4_init_inode_table */
2724 static int ext4_run_li_request(struct ext4_li_request *elr)
2725 {
2726         struct ext4_group_desc *gdp = NULL;
2727         ext4_group_t group, ngroups;
2728         struct super_block *sb;
2729         unsigned long timeout = 0;
2730         int ret = 0;
2731
2732         sb = elr->lr_super;
2733         ngroups = EXT4_SB(sb)->s_groups_count;
2734
2735         sb_start_write(sb);
2736         for (group = elr->lr_next_group; group < ngroups; group++) {
2737                 gdp = ext4_get_group_desc(sb, group, NULL);
2738                 if (!gdp) {
2739                         ret = 1;
2740                         break;
2741                 }
2742
2743                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2744                         break;
2745         }
2746
2747         if (group == ngroups)
2748                 ret = 1;
2749
2750         if (!ret) {
2751                 timeout = jiffies;
2752                 ret = ext4_init_inode_table(sb, group,
2753                                             elr->lr_timeout ? 0 : 1);
2754                 if (elr->lr_timeout == 0) {
2755                         timeout = (jiffies - timeout) *
2756                                   elr->lr_sbi->s_li_wait_mult;
2757                         elr->lr_timeout = timeout;
2758                 }
2759                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2760                 elr->lr_next_group = group + 1;
2761         }
2762         sb_end_write(sb);
2763
2764         return ret;
2765 }
2766
2767 /*
2768  * Remove lr_request from the list_request and free the
2769  * request structure. Should be called with li_list_mtx held
2770  */
2771 static void ext4_remove_li_request(struct ext4_li_request *elr)
2772 {
2773         struct ext4_sb_info *sbi;
2774
2775         if (!elr)
2776                 return;
2777
2778         sbi = elr->lr_sbi;
2779
2780         list_del(&elr->lr_request);
2781         sbi->s_li_request = NULL;
2782         kfree(elr);
2783 }
2784
2785 static void ext4_unregister_li_request(struct super_block *sb)
2786 {
2787         mutex_lock(&ext4_li_mtx);
2788         if (!ext4_li_info) {
2789                 mutex_unlock(&ext4_li_mtx);
2790                 return;
2791         }
2792
2793         mutex_lock(&ext4_li_info->li_list_mtx);
2794         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2795         mutex_unlock(&ext4_li_info->li_list_mtx);
2796         mutex_unlock(&ext4_li_mtx);
2797 }
2798
2799 static struct task_struct *ext4_lazyinit_task;
2800
2801 /*
2802  * This is the function where ext4lazyinit thread lives. It walks
2803  * through the request list searching for next scheduled filesystem.
2804  * When such a fs is found, run the lazy initialization request
2805  * (ext4_rn_li_request) and keep track of the time spend in this
2806  * function. Based on that time we compute next schedule time of
2807  * the request. When walking through the list is complete, compute
2808  * next waking time and put itself into sleep.
2809  */
2810 static int ext4_lazyinit_thread(void *arg)
2811 {
2812         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2813         struct list_head *pos, *n;
2814         struct ext4_li_request *elr;
2815         unsigned long next_wakeup, cur;
2816
2817         BUG_ON(NULL == eli);
2818
2819 cont_thread:
2820         while (true) {
2821                 next_wakeup = MAX_JIFFY_OFFSET;
2822
2823                 mutex_lock(&eli->li_list_mtx);
2824                 if (list_empty(&eli->li_request_list)) {
2825                         mutex_unlock(&eli->li_list_mtx);
2826                         goto exit_thread;
2827                 }
2828
2829                 list_for_each_safe(pos, n, &eli->li_request_list) {
2830                         elr = list_entry(pos, struct ext4_li_request,
2831                                          lr_request);
2832
2833                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2834                                 if (ext4_run_li_request(elr) != 0) {
2835                                         /* error, remove the lazy_init job */
2836                                         ext4_remove_li_request(elr);
2837                                         continue;
2838                                 }
2839                         }
2840
2841                         if (time_before(elr->lr_next_sched, next_wakeup))
2842                                 next_wakeup = elr->lr_next_sched;
2843                 }
2844                 mutex_unlock(&eli->li_list_mtx);
2845
2846                 try_to_freeze();
2847
2848                 cur = jiffies;
2849                 if ((time_after_eq(cur, next_wakeup)) ||
2850                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2851                         cond_resched();
2852                         continue;
2853                 }
2854
2855                 schedule_timeout_interruptible(next_wakeup - cur);
2856
2857                 if (kthread_should_stop()) {
2858                         ext4_clear_request_list();
2859                         goto exit_thread;
2860                 }
2861         }
2862
2863 exit_thread:
2864         /*
2865          * It looks like the request list is empty, but we need
2866          * to check it under the li_list_mtx lock, to prevent any
2867          * additions into it, and of course we should lock ext4_li_mtx
2868          * to atomically free the list and ext4_li_info, because at
2869          * this point another ext4 filesystem could be registering
2870          * new one.
2871          */
2872         mutex_lock(&ext4_li_mtx);
2873         mutex_lock(&eli->li_list_mtx);
2874         if (!list_empty(&eli->li_request_list)) {
2875                 mutex_unlock(&eli->li_list_mtx);
2876                 mutex_unlock(&ext4_li_mtx);
2877                 goto cont_thread;
2878         }
2879         mutex_unlock(&eli->li_list_mtx);
2880         kfree(ext4_li_info);
2881         ext4_li_info = NULL;
2882         mutex_unlock(&ext4_li_mtx);
2883
2884         return 0;
2885 }
2886
2887 static void ext4_clear_request_list(void)
2888 {
2889         struct list_head *pos, *n;
2890         struct ext4_li_request *elr;
2891
2892         mutex_lock(&ext4_li_info->li_list_mtx);
2893         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2894                 elr = list_entry(pos, struct ext4_li_request,
2895                                  lr_request);
2896                 ext4_remove_li_request(elr);
2897         }
2898         mutex_unlock(&ext4_li_info->li_list_mtx);
2899 }
2900
2901 static int ext4_run_lazyinit_thread(void)
2902 {
2903         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2904                                          ext4_li_info, "ext4lazyinit");
2905         if (IS_ERR(ext4_lazyinit_task)) {
2906                 int err = PTR_ERR(ext4_lazyinit_task);
2907                 ext4_clear_request_list();
2908                 kfree(ext4_li_info);
2909                 ext4_li_info = NULL;
2910                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2911                                  "initialization thread\n",
2912                                  err);
2913                 return err;
2914         }
2915         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2916         return 0;
2917 }
2918
2919 /*
2920  * Check whether it make sense to run itable init. thread or not.
2921  * If there is at least one uninitialized inode table, return
2922  * corresponding group number, else the loop goes through all
2923  * groups and return total number of groups.
2924  */
2925 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2926 {
2927         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2928         struct ext4_group_desc *gdp = NULL;
2929
2930         for (group = 0; group < ngroups; group++) {
2931                 gdp = ext4_get_group_desc(sb, group, NULL);
2932                 if (!gdp)
2933                         continue;
2934
2935                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2936                         break;
2937         }
2938
2939         return group;
2940 }
2941
2942 static int ext4_li_info_new(void)
2943 {
2944         struct ext4_lazy_init *eli = NULL;
2945
2946         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2947         if (!eli)
2948                 return -ENOMEM;
2949
2950         INIT_LIST_HEAD(&eli->li_request_list);
2951         mutex_init(&eli->li_list_mtx);
2952
2953         eli->li_state |= EXT4_LAZYINIT_QUIT;
2954
2955         ext4_li_info = eli;
2956
2957         return 0;
2958 }
2959
2960 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2961                                             ext4_group_t start)
2962 {
2963         struct ext4_sb_info *sbi = EXT4_SB(sb);
2964         struct ext4_li_request *elr;
2965         unsigned long rnd;
2966
2967         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2968         if (!elr)
2969                 return NULL;
2970
2971         elr->lr_super = sb;
2972         elr->lr_sbi = sbi;
2973         elr->lr_next_group = start;
2974
2975         /*
2976          * Randomize first schedule time of the request to
2977          * spread the inode table initialization requests
2978          * better.
2979          */
2980         get_random_bytes(&rnd, sizeof(rnd));
2981         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2982                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2983
2984         return elr;
2985 }
2986
2987 static int ext4_register_li_request(struct super_block *sb,
2988                                     ext4_group_t first_not_zeroed)
2989 {
2990         struct ext4_sb_info *sbi = EXT4_SB(sb);
2991         struct ext4_li_request *elr;
2992         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2993         int ret = 0;
2994
2995         if (sbi->s_li_request != NULL) {
2996                 /*
2997                  * Reset timeout so it can be computed again, because
2998                  * s_li_wait_mult might have changed.
2999                  */
3000                 sbi->s_li_request->lr_timeout = 0;
3001                 return 0;
3002         }
3003
3004         if (first_not_zeroed == ngroups ||
3005             (sb->s_flags & MS_RDONLY) ||
3006             !test_opt(sb, INIT_INODE_TABLE))
3007                 return 0;
3008
3009         elr = ext4_li_request_new(sb, first_not_zeroed);
3010         if (!elr)
3011                 return -ENOMEM;
3012
3013         mutex_lock(&ext4_li_mtx);
3014
3015         if (NULL == ext4_li_info) {
3016                 ret = ext4_li_info_new();
3017                 if (ret)
3018                         goto out;
3019         }
3020
3021         mutex_lock(&ext4_li_info->li_list_mtx);
3022         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3023         mutex_unlock(&ext4_li_info->li_list_mtx);
3024
3025         sbi->s_li_request = elr;
3026         /*
3027          * set elr to NULL here since it has been inserted to
3028          * the request_list and the removal and free of it is
3029          * handled by ext4_clear_request_list from now on.
3030          */
3031         elr = NULL;
3032
3033         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3034                 ret = ext4_run_lazyinit_thread();
3035                 if (ret)
3036                         goto out;
3037         }
3038 out:
3039         mutex_unlock(&ext4_li_mtx);
3040         if (ret)
3041                 kfree(elr);
3042         return ret;
3043 }
3044
3045 /*
3046  * We do not need to lock anything since this is called on
3047  * module unload.
3048  */
3049 static void ext4_destroy_lazyinit_thread(void)
3050 {
3051         /*
3052          * If thread exited earlier
3053          * there's nothing to be done.
3054          */
3055         if (!ext4_li_info || !ext4_lazyinit_task)
3056                 return;
3057
3058         kthread_stop(ext4_lazyinit_task);
3059 }
3060
3061 static int set_journal_csum_feature_set(struct super_block *sb)
3062 {
3063         int ret = 1;
3064         int compat, incompat;
3065         struct ext4_sb_info *sbi = EXT4_SB(sb);
3066
3067         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3068                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3069                 /* journal checksum v2 */
3070                 compat = 0;
3071                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3072         } else {
3073                 /* journal checksum v1 */
3074                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3075                 incompat = 0;
3076         }
3077
3078         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3079                 ret = jbd2_journal_set_features(sbi->s_journal,
3080                                 compat, 0,
3081                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3082                                 incompat);
3083         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3084                 ret = jbd2_journal_set_features(sbi->s_journal,
3085                                 compat, 0,
3086                                 incompat);
3087                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3088                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3089         } else {
3090                 jbd2_journal_clear_features(sbi->s_journal,
3091                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3092                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3093                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3094         }
3095
3096         return ret;
3097 }
3098
3099 /*
3100  * Note: calculating the overhead so we can be compatible with
3101  * historical BSD practice is quite difficult in the face of
3102  * clusters/bigalloc.  This is because multiple metadata blocks from
3103  * different block group can end up in the same allocation cluster.
3104  * Calculating the exact overhead in the face of clustered allocation
3105  * requires either O(all block bitmaps) in memory or O(number of block
3106  * groups**2) in time.  We will still calculate the superblock for
3107  * older file systems --- and if we come across with a bigalloc file
3108  * system with zero in s_overhead_clusters the estimate will be close to
3109  * correct especially for very large cluster sizes --- but for newer
3110  * file systems, it's better to calculate this figure once at mkfs
3111  * time, and store it in the superblock.  If the superblock value is
3112  * present (even for non-bigalloc file systems), we will use it.
3113  */
3114 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3115                           char *buf)
3116 {
3117         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3118         struct ext4_group_desc  *gdp;
3119         ext4_fsblk_t            first_block, last_block, b;
3120         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3121         int                     s, j, count = 0;
3122
3123         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3124                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3125         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3126         for (i = 0; i < ngroups; i++) {
3127                 gdp = ext4_get_group_desc(sb, i, NULL);
3128                 b = ext4_block_bitmap(sb, gdp);
3129                 if (b >= first_block && b <= last_block) {
3130                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3131                         count++;
3132                 }
3133                 b = ext4_inode_bitmap(sb, gdp);
3134                 if (b >= first_block && b <= last_block) {
3135                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3136                         count++;
3137                 }
3138                 b = ext4_inode_table(sb, gdp);
3139                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3140                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3141                                 int c = EXT4_B2C(sbi, b - first_block);
3142                                 ext4_set_bit(c, buf);
3143                                 count++;
3144                         }
3145                 if (i != grp)
3146                         continue;
3147                 s = 0;
3148                 if (ext4_bg_has_super(sb, grp)) {
3149                         ext4_set_bit(s++, buf);
3150                         count++;
3151                 }
3152                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3153                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3154                         count++;
3155                 }
3156         }
3157         if (!count)
3158                 return 0;
3159         return EXT4_CLUSTERS_PER_GROUP(sb) -
3160                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3161 }
3162
3163 /*
3164  * Compute the overhead and stash it in sbi->s_overhead
3165  */
3166 int ext4_calculate_overhead(struct super_block *sb)
3167 {
3168         struct ext4_sb_info *sbi = EXT4_SB(sb);
3169         struct ext4_super_block *es = sbi->s_es;
3170         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3171         ext4_fsblk_t overhead = 0;
3172         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3173
3174         memset(buf, 0, PAGE_SIZE);
3175         if (!buf)
3176                 return -ENOMEM;
3177
3178         /*
3179          * Compute the overhead (FS structures).  This is constant
3180          * for a given filesystem unless the number of block groups
3181          * changes so we cache the previous value until it does.
3182          */
3183
3184         /*
3185          * All of the blocks before first_data_block are overhead
3186          */
3187         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3188
3189         /*
3190          * Add the overhead found in each block group
3191          */
3192         for (i = 0; i < ngroups; i++) {
3193                 int blks;
3194
3195                 blks = count_overhead(sb, i, buf);
3196                 overhead += blks;
3197                 if (blks)
3198                         memset(buf, 0, PAGE_SIZE);
3199                 cond_resched();
3200         }
3201         sbi->s_overhead = overhead;
3202         smp_wmb();
3203         free_page((unsigned long) buf);
3204         return 0;
3205 }
3206
3207 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3208 {
3209         char *orig_data = kstrdup(data, GFP_KERNEL);
3210         struct buffer_head *bh;
3211         struct ext4_super_block *es = NULL;
3212         struct ext4_sb_info *sbi;
3213         ext4_fsblk_t block;
3214         ext4_fsblk_t sb_block = get_sb_block(&data);
3215         ext4_fsblk_t logical_sb_block;
3216         unsigned long offset = 0;
3217         unsigned long journal_devnum = 0;
3218         unsigned long def_mount_opts;
3219         struct inode *root;
3220         char *cp;
3221         const char *descr;
3222         int ret = -ENOMEM;
3223         int blocksize, clustersize;
3224         unsigned int db_count;
3225         unsigned int i;
3226         int needs_recovery, has_huge_files, has_bigalloc;
3227         __u64 blocks_count;
3228         int err;
3229         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3230         ext4_group_t first_not_zeroed;
3231
3232         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3233         if (!sbi)
3234                 goto out_free_orig;
3235
3236         sbi->s_blockgroup_lock =
3237                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3238         if (!sbi->s_blockgroup_lock) {
3239                 kfree(sbi);
3240                 goto out_free_orig;
3241         }
3242         sb->s_fs_info = sbi;
3243         sbi->s_sb = sb;
3244         sbi->s_mount_opt = 0;
3245         sbi->s_resuid = make_kuid(&init_user_ns, EXT4_DEF_RESUID);
3246         sbi->s_resgid = make_kgid(&init_user_ns, EXT4_DEF_RESGID);
3247         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3248         sbi->s_sb_block = sb_block;
3249         if (sb->s_bdev->bd_part)
3250                 sbi->s_sectors_written_start =
3251                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3252
3253         /* Cleanup superblock name */
3254         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3255                 *cp = '!';
3256
3257         ret = -EINVAL;
3258         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3259         if (!blocksize) {
3260                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3261                 goto out_fail;
3262         }
3263
3264         /*
3265          * The ext4 superblock will not be buffer aligned for other than 1kB
3266          * block sizes.  We need to calculate the offset from buffer start.
3267          */
3268         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3269                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3270                 offset = do_div(logical_sb_block, blocksize);
3271         } else {
3272                 logical_sb_block = sb_block;
3273         }
3274
3275         if (!(bh = sb_bread(sb, logical_sb_block))) {
3276                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3277                 goto out_fail;
3278         }
3279         /*
3280          * Note: s_es must be initialized as soon as possible because
3281          *       some ext4 macro-instructions depend on its value
3282          */
3283         es = (struct ext4_super_block *) (bh->b_data + offset);
3284         sbi->s_es = es;
3285         sb->s_magic = le16_to_cpu(es->s_magic);
3286         if (sb->s_magic != EXT4_SUPER_MAGIC)
3287                 goto cantfind_ext4;
3288         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3289
3290         /* Warn if metadata_csum and gdt_csum are both set. */
3291         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3292                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3293             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3294                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3295                              "redundant flags; please run fsck.");
3296
3297         /* Check for a known checksum algorithm */
3298         if (!ext4_verify_csum_type(sb, es)) {
3299                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3300                          "unknown checksum algorithm.");
3301                 silent = 1;
3302                 goto cantfind_ext4;
3303         }
3304
3305         /* Load the checksum driver */
3306         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3307                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3308                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3309                 if (IS_ERR(sbi->s_chksum_driver)) {
3310                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3311                         ret = PTR_ERR(sbi->s_chksum_driver);
3312                         sbi->s_chksum_driver = NULL;
3313                         goto failed_mount;
3314                 }
3315         }
3316
3317         /* Check superblock checksum */
3318         if (!ext4_superblock_csum_verify(sb, es)) {
3319                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3320                          "invalid superblock checksum.  Run e2fsck?");
3321                 silent = 1;
3322                 goto cantfind_ext4;
3323         }
3324
3325         /* Precompute checksum seed for all metadata */
3326         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3327                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3328                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3329                                                sizeof(es->s_uuid));
3330
3331         /* Set defaults before we parse the mount options */
3332         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3333         set_opt(sb, INIT_INODE_TABLE);
3334         if (def_mount_opts & EXT4_DEFM_DEBUG)
3335                 set_opt(sb, DEBUG);
3336         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3337                 set_opt(sb, GRPID);
3338         if (def_mount_opts & EXT4_DEFM_UID16)
3339                 set_opt(sb, NO_UID32);
3340         /* xattr user namespace & acls are now defaulted on */
3341 #ifdef CONFIG_EXT4_FS_XATTR
3342         set_opt(sb, XATTR_USER);
3343 #endif
3344 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3345         set_opt(sb, POSIX_ACL);
3346 #endif
3347         set_opt(sb, MBLK_IO_SUBMIT);
3348         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3349                 set_opt(sb, JOURNAL_DATA);
3350         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3351                 set_opt(sb, ORDERED_DATA);
3352         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3353                 set_opt(sb, WRITEBACK_DATA);
3354
3355         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3356                 set_opt(sb, ERRORS_PANIC);
3357         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3358                 set_opt(sb, ERRORS_CONT);
3359         else
3360                 set_opt(sb, ERRORS_RO);
3361         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3362                 set_opt(sb, BLOCK_VALIDITY);
3363         if (def_mount_opts & EXT4_DEFM_DISCARD)
3364                 set_opt(sb, DISCARD);
3365
3366         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3367         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3368         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3369         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3370         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3371
3372         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3373                 set_opt(sb, BARRIER);
3374
3375         /*
3376          * enable delayed allocation by default
3377          * Use -o nodelalloc to turn it off
3378          */
3379         if (!IS_EXT3_SB(sb) &&
3380             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3381                 set_opt(sb, DELALLOC);
3382
3383         /*
3384          * set default s_li_wait_mult for lazyinit, for the case there is
3385          * no mount option specified.
3386          */
3387         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3388
3389         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3390                            &journal_devnum, &journal_ioprio, 0)) {
3391                 ext4_msg(sb, KERN_WARNING,
3392                          "failed to parse options in superblock: %s",
3393                          sbi->s_es->s_mount_opts);
3394         }
3395         sbi->s_def_mount_opt = sbi->s_mount_opt;
3396         if (!parse_options((char *) data, sb, &journal_devnum,
3397                            &journal_ioprio, 0))
3398                 goto failed_mount;
3399
3400         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3401                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3402                             "with data=journal disables delayed "
3403                             "allocation and O_DIRECT support!\n");
3404                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3405                         ext4_msg(sb, KERN_ERR, "can't mount with "
3406                                  "both data=journal and delalloc");
3407                         goto failed_mount;
3408                 }
3409                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3410                         ext4_msg(sb, KERN_ERR, "can't mount with "
3411                                  "both data=journal and delalloc");
3412                         goto failed_mount;
3413                 }
3414                 if (test_opt(sb, DELALLOC))
3415                         clear_opt(sb, DELALLOC);
3416         }
3417
3418         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3419         if (test_opt(sb, DIOREAD_NOLOCK)) {
3420                 if (blocksize < PAGE_SIZE) {
3421                         ext4_msg(sb, KERN_ERR, "can't mount with "
3422                                  "dioread_nolock if block size != PAGE_SIZE");
3423                         goto failed_mount;
3424                 }
3425         }
3426
3427         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3428                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3429
3430         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3431             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3432              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3433              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3434                 ext4_msg(sb, KERN_WARNING,
3435                        "feature flags set on rev 0 fs, "
3436                        "running e2fsck is recommended");
3437
3438         if (IS_EXT2_SB(sb)) {
3439                 if (ext2_feature_set_ok(sb))
3440                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3441                                  "using the ext4 subsystem");
3442                 else {
3443                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3444                                  "to feature incompatibilities");
3445                         goto failed_mount;
3446                 }
3447         }
3448
3449         if (IS_EXT3_SB(sb)) {
3450                 if (ext3_feature_set_ok(sb))
3451                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3452                                  "using the ext4 subsystem");
3453                 else {
3454                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3455                                  "to feature incompatibilities");
3456                         goto failed_mount;
3457                 }
3458         }
3459
3460         /*
3461          * Check feature flags regardless of the revision level, since we
3462          * previously didn't change the revision level when setting the flags,
3463          * so there is a chance incompat flags are set on a rev 0 filesystem.
3464          */
3465         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3466                 goto failed_mount;
3467
3468         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3469             blocksize > EXT4_MAX_BLOCK_SIZE) {
3470                 ext4_msg(sb, KERN_ERR,
3471                        "Unsupported filesystem blocksize %d", blocksize);
3472                 goto failed_mount;
3473         }
3474
3475         if (sb->s_blocksize != blocksize) {
3476                 /* Validate the filesystem blocksize */
3477                 if (!sb_set_blocksize(sb, blocksize)) {
3478                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3479                                         blocksize);
3480                         goto failed_mount;
3481                 }
3482
3483                 brelse(bh);
3484                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3485                 offset = do_div(logical_sb_block, blocksize);
3486                 bh = sb_bread(sb, logical_sb_block);
3487                 if (!bh) {
3488                         ext4_msg(sb, KERN_ERR,
3489                                "Can't read superblock on 2nd try");
3490                         goto failed_mount;
3491                 }
3492                 es = (struct ext4_super_block *)(bh->b_data + offset);
3493                 sbi->s_es = es;
3494                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3495                         ext4_msg(sb, KERN_ERR,
3496                                "Magic mismatch, very weird!");
3497                         goto failed_mount;
3498                 }
3499         }
3500
3501         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3502                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3503         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3504                                                       has_huge_files);
3505         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3506
3507         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3508                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3509                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3510         } else {
3511                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3512                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3513                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3514                     (!is_power_of_2(sbi->s_inode_size)) ||
3515                     (sbi->s_inode_size > blocksize)) {
3516                         ext4_msg(sb, KERN_ERR,
3517                                "unsupported inode size: %d",
3518                                sbi->s_inode_size);
3519                         goto failed_mount;
3520                 }
3521                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3522                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3523         }
3524
3525         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3526         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3527                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3528                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3529                     !is_power_of_2(sbi->s_desc_size)) {
3530                         ext4_msg(sb, KERN_ERR,
3531                                "unsupported descriptor size %lu",
3532                                sbi->s_desc_size);
3533                         goto failed_mount;
3534                 }
3535         } else
3536                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3537
3538         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3539         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3540         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3541                 goto cantfind_ext4;
3542
3543         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3544         if (sbi->s_inodes_per_block == 0)
3545                 goto cantfind_ext4;
3546         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3547                                         sbi->s_inodes_per_block;
3548         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3549         sbi->s_sbh = bh;
3550         sbi->s_mount_state = le16_to_cpu(es->s_state);
3551         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3552         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3553
3554         for (i = 0; i < 4; i++)
3555                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3556         sbi->s_def_hash_version = es->s_def_hash_version;
3557         i = le32_to_cpu(es->s_flags);
3558         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3559                 sbi->s_hash_unsigned = 3;
3560         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3561 #ifdef __CHAR_UNSIGNED__
3562                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3563                 sbi->s_hash_unsigned = 3;
3564 #else
3565                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3566 #endif
3567         }
3568
3569         /* Handle clustersize */
3570         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3571         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3572                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3573         if (has_bigalloc) {
3574                 if (clustersize < blocksize) {
3575                         ext4_msg(sb, KERN_ERR,
3576                                  "cluster size (%d) smaller than "
3577                                  "block size (%d)", clustersize, blocksize);
3578                         goto failed_mount;
3579                 }
3580                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3581                         le32_to_cpu(es->s_log_block_size);
3582                 sbi->s_clusters_per_group =
3583                         le32_to_cpu(es->s_clusters_per_group);
3584                 if (sbi->s_clusters_per_group > blocksize * 8) {
3585                         ext4_msg(sb, KERN_ERR,
3586                                  "#clusters per group too big: %lu",
3587                                  sbi->s_clusters_per_group);
3588                         goto failed_mount;
3589                 }
3590                 if (sbi->s_blocks_per_group !=
3591                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3592                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3593                                  "clusters per group (%lu) inconsistent",
3594                                  sbi->s_blocks_per_group,
3595                                  sbi->s_clusters_per_group);
3596                         goto failed_mount;
3597                 }
3598         } else {
3599                 if (clustersize != blocksize) {
3600                         ext4_warning(sb, "fragment/cluster size (%d) != "
3601                                      "block size (%d)", clustersize,
3602                                      blocksize);
3603                         clustersize = blocksize;
3604                 }
3605                 if (sbi->s_blocks_per_group > blocksize * 8) {
3606                         ext4_msg(sb, KERN_ERR,
3607                                  "#blocks per group too big: %lu",
3608                                  sbi->s_blocks_per_group);
3609                         goto failed_mount;
3610                 }
3611                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3612                 sbi->s_cluster_bits = 0;
3613         }
3614         sbi->s_cluster_ratio = clustersize / blocksize;
3615
3616         if (sbi->s_inodes_per_group > blocksize * 8) {
3617                 ext4_msg(sb, KERN_ERR,
3618                        "#inodes per group too big: %lu",
3619                        sbi->s_inodes_per_group);
3620                 goto failed_mount;
3621         }
3622
3623         /*
3624          * Test whether we have more sectors than will fit in sector_t,
3625          * and whether the max offset is addressable by the page cache.
3626          */
3627         err = generic_check_addressable(sb->s_blocksize_bits,
3628                                         ext4_blocks_count(es));
3629         if (err) {
3630                 ext4_msg(sb, KERN_ERR, "filesystem"
3631                          " too large to mount safely on this system");
3632                 if (sizeof(sector_t) < 8)
3633                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3634                 ret = err;
3635                 goto failed_mount;
3636         }
3637
3638         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3639                 goto cantfind_ext4;
3640
3641         /* check blocks count against device size */
3642         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3643         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3644                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3645                        "exceeds size of device (%llu blocks)",
3646                        ext4_blocks_count(es), blocks_count);
3647                 goto failed_mount;
3648         }
3649
3650         /*
3651          * It makes no sense for the first data block to be beyond the end
3652          * of the filesystem.
3653          */
3654         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3655                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3656                          "block %u is beyond end of filesystem (%llu)",
3657                          le32_to_cpu(es->s_first_data_block),
3658                          ext4_blocks_count(es));
3659                 goto failed_mount;
3660         }
3661         blocks_count = (ext4_blocks_count(es) -
3662                         le32_to_cpu(es->s_first_data_block) +
3663                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3664         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3665         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3666                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3667                        "(block count %llu, first data block %u, "
3668                        "blocks per group %lu)", sbi->s_groups_count,
3669                        ext4_blocks_count(es),
3670                        le32_to_cpu(es->s_first_data_block),
3671                        EXT4_BLOCKS_PER_GROUP(sb));
3672                 goto failed_mount;
3673         }
3674         sbi->s_groups_count = blocks_count;
3675         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3676                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3677         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3678                    EXT4_DESC_PER_BLOCK(sb);
3679         sbi->s_group_desc = ext4_kvmalloc(db_count *
3680                                           sizeof(struct buffer_head *),
3681                                           GFP_KERNEL);
3682         if (sbi->s_group_desc == NULL) {
3683                 ext4_msg(sb, KERN_ERR, "not enough memory");
3684                 ret = -ENOMEM;
3685                 goto failed_mount;
3686         }
3687
3688         if (ext4_proc_root)
3689                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3690
3691         if (sbi->s_proc)
3692                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3693                                  &ext4_seq_options_fops, sb);
3694
3695         bgl_lock_init(sbi->s_blockgroup_lock);
3696
3697         for (i = 0; i < db_count; i++) {
3698                 block = descriptor_loc(sb, logical_sb_block, i);
3699                 sbi->s_group_desc[i] = sb_bread(sb, block);
3700                 if (!sbi->s_group_desc[i]) {
3701                         ext4_msg(sb, KERN_ERR,
3702                                "can't read group descriptor %d", i);
3703                         db_count = i;
3704                         goto failed_mount2;
3705                 }
3706         }
3707         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3708                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3709                 goto failed_mount2;
3710         }
3711         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3712                 if (!ext4_fill_flex_info(sb)) {
3713                         ext4_msg(sb, KERN_ERR,
3714                                "unable to initialize "
3715                                "flex_bg meta info!");
3716                         goto failed_mount2;
3717                 }
3718
3719         sbi->s_gdb_count = db_count;
3720         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3721         spin_lock_init(&sbi->s_next_gen_lock);
3722
3723         init_timer(&sbi->s_err_report);
3724         sbi->s_err_report.function = print_daily_error_info;
3725         sbi->s_err_report.data = (unsigned long) sb;
3726
3727         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3728                         ext4_count_free_clusters(sb));
3729         if (!err) {
3730                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3731                                 ext4_count_free_inodes(sb));
3732         }
3733         if (!err) {
3734                 err = percpu_counter_init(&sbi->s_dirs_counter,
3735                                 ext4_count_dirs(sb));
3736         }
3737         if (!err) {
3738                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3739         }
3740         if (err) {
3741                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3742                 ret = err;
3743                 goto failed_mount3;
3744         }
3745
3746         sbi->s_stripe = ext4_get_stripe_size(sbi);
3747         sbi->s_max_writeback_mb_bump = 128;
3748
3749         /*
3750          * set up enough so that it can read an inode
3751          */
3752         if (!test_opt(sb, NOLOAD) &&
3753             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3754                 sb->s_op = &ext4_sops;
3755         else
3756                 sb->s_op = &ext4_nojournal_sops;
3757         sb->s_export_op = &ext4_export_ops;
3758         sb->s_xattr = ext4_xattr_handlers;
3759 #ifdef CONFIG_QUOTA
3760         sb->s_qcop = &ext4_qctl_operations;
3761         sb->dq_op = &ext4_quota_operations;
3762
3763         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
3764                 /* Use qctl operations for hidden quota files. */
3765                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3766         }
3767 #endif
3768         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3769
3770         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3771         mutex_init(&sbi->s_orphan_lock);
3772         sbi->s_resize_flags = 0;
3773
3774         sb->s_root = NULL;
3775
3776         needs_recovery = (es->s_last_orphan != 0 ||
3777                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3778                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3779
3780         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3781             !(sb->s_flags & MS_RDONLY))
3782                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3783                         goto failed_mount3;
3784
3785         /*
3786          * The first inode we look at is the journal inode.  Don't try
3787          * root first: it may be modified in the journal!
3788          */
3789         if (!test_opt(sb, NOLOAD) &&
3790             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3791                 if (ext4_load_journal(sb, es, journal_devnum))
3792                         goto failed_mount3;
3793         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3794               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3795                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3796                        "suppressed and not mounted read-only");
3797                 goto failed_mount_wq;
3798         } else {
3799                 clear_opt(sb, DATA_FLAGS);
3800                 sbi->s_journal = NULL;
3801                 needs_recovery = 0;
3802                 goto no_journal;
3803         }
3804
3805         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3806             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3807                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3808                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3809                 goto failed_mount_wq;
3810         }
3811
3812         if (!set_journal_csum_feature_set(sb)) {
3813                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3814                          "feature set");
3815                 goto failed_mount_wq;
3816         }
3817
3818         /* We have now updated the journal if required, so we can
3819          * validate the data journaling mode. */
3820         switch (test_opt(sb, DATA_FLAGS)) {
3821         case 0:
3822                 /* No mode set, assume a default based on the journal
3823                  * capabilities: ORDERED_DATA if the journal can
3824                  * cope, else JOURNAL_DATA
3825                  */
3826                 if (jbd2_journal_check_available_features
3827                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3828                         set_opt(sb, ORDERED_DATA);
3829                 else
3830                         set_opt(sb, JOURNAL_DATA);
3831                 break;
3832
3833         case EXT4_MOUNT_ORDERED_DATA:
3834         case EXT4_MOUNT_WRITEBACK_DATA:
3835                 if (!jbd2_journal_check_available_features
3836                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3837                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3838                                "requested data journaling mode");
3839                         goto failed_mount_wq;
3840                 }
3841         default:
3842                 break;
3843         }
3844         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3845
3846         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3847
3848         /*
3849          * The journal may have updated the bg summary counts, so we
3850          * need to update the global counters.
3851          */
3852         percpu_counter_set(&sbi->s_freeclusters_counter,
3853                            ext4_count_free_clusters(sb));
3854         percpu_counter_set(&sbi->s_freeinodes_counter,
3855                            ext4_count_free_inodes(sb));
3856         percpu_counter_set(&sbi->s_dirs_counter,
3857                            ext4_count_dirs(sb));
3858         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3859
3860 no_journal:
3861         /*
3862          * Get the # of file system overhead blocks from the
3863          * superblock if present.
3864          */
3865         if (es->s_overhead_clusters)
3866                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3867         else {
3868                 ret = ext4_calculate_overhead(sb);
3869                 if (ret)
3870                         goto failed_mount_wq;
3871         }
3872
3873         /*
3874          * The maximum number of concurrent works can be high and
3875          * concurrency isn't really necessary.  Limit it to 1.
3876          */
3877         EXT4_SB(sb)->dio_unwritten_wq =
3878                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3879         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3880                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3881                 goto failed_mount_wq;
3882         }
3883
3884         /*
3885          * The jbd2_journal_load will have done any necessary log recovery,
3886          * so we can safely mount the rest of the filesystem now.
3887          */
3888
3889         root = ext4_iget(sb, EXT4_ROOT_INO);
3890         if (IS_ERR(root)) {
3891                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3892                 ret = PTR_ERR(root);
3893                 root = NULL;
3894                 goto failed_mount4;
3895         }
3896         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3897                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3898                 iput(root);
3899                 goto failed_mount4;
3900         }
3901         sb->s_root = d_make_root(root);
3902         if (!sb->s_root) {
3903                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3904                 ret = -ENOMEM;
3905                 goto failed_mount4;
3906         }
3907
3908         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3909                 sb->s_flags |= MS_RDONLY;
3910
3911         /* determine the minimum size of new large inodes, if present */
3912         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3913                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3914                                                      EXT4_GOOD_OLD_INODE_SIZE;
3915                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3916                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3917                         if (sbi->s_want_extra_isize <
3918                             le16_to_cpu(es->s_want_extra_isize))
3919                                 sbi->s_want_extra_isize =
3920                                         le16_to_cpu(es->s_want_extra_isize);
3921                         if (sbi->s_want_extra_isize <
3922                             le16_to_cpu(es->s_min_extra_isize))
3923                                 sbi->s_want_extra_isize =
3924                                         le16_to_cpu(es->s_min_extra_isize);
3925                 }
3926         }
3927         /* Check if enough inode space is available */
3928         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3929                                                         sbi->s_inode_size) {
3930                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3931                                                        EXT4_GOOD_OLD_INODE_SIZE;
3932                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3933                          "available");
3934         }
3935
3936         err = ext4_setup_system_zone(sb);
3937         if (err) {
3938                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3939                          "zone (%d)", err);
3940                 goto failed_mount4a;
3941         }
3942
3943         ext4_ext_init(sb);
3944         err = ext4_mb_init(sb);
3945         if (err) {
3946                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3947                          err);
3948                 goto failed_mount5;
3949         }
3950
3951         err = ext4_register_li_request(sb, first_not_zeroed);
3952         if (err)
3953                 goto failed_mount6;
3954
3955         sbi->s_kobj.kset = ext4_kset;
3956         init_completion(&sbi->s_kobj_unregister);
3957         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3958                                    "%s", sb->s_id);
3959         if (err)
3960                 goto failed_mount7;
3961
3962         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3963         ext4_orphan_cleanup(sb, es);
3964         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3965         if (needs_recovery) {
3966                 ext4_msg(sb, KERN_INFO, "recovery complete");
3967                 ext4_mark_recovery_complete(sb, es);
3968         }
3969         if (EXT4_SB(sb)->s_journal) {
3970                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3971                         descr = " journalled data mode";
3972                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3973                         descr = " ordered data mode";
3974                 else
3975                         descr = " writeback data mode";
3976         } else
3977                 descr = "out journal";
3978
3979 #ifdef CONFIG_QUOTA
3980         /* Enable quota usage during mount. */
3981         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
3982             !(sb->s_flags & MS_RDONLY)) {
3983                 ret = ext4_enable_quotas(sb);
3984                 if (ret)
3985                         goto failed_mount7;
3986         }
3987 #endif  /* CONFIG_QUOTA */
3988
3989         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3990                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3991                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3992
3993         if (es->s_error_count)
3994                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3995
3996         kfree(orig_data);
3997         return 0;
3998
3999 cantfind_ext4:
4000         if (!silent)
4001                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4002         goto failed_mount;
4003
4004 failed_mount7:
4005         ext4_unregister_li_request(sb);
4006 failed_mount6:
4007         ext4_mb_release(sb);
4008 failed_mount5:
4009         ext4_ext_release(sb);
4010         ext4_release_system_zone(sb);
4011 failed_mount4a:
4012         dput(sb->s_root);
4013         sb->s_root = NULL;
4014 failed_mount4:
4015         ext4_msg(sb, KERN_ERR, "mount failed");
4016         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4017 failed_mount_wq:
4018         if (sbi->s_journal) {
4019                 jbd2_journal_destroy(sbi->s_journal);
4020                 sbi->s_journal = NULL;
4021         }
4022 failed_mount3:
4023         del_timer(&sbi->s_err_report);
4024         if (sbi->s_flex_groups)
4025                 ext4_kvfree(sbi->s_flex_groups);
4026         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4027         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4028         percpu_counter_destroy(&sbi->s_dirs_counter);
4029         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4030         if (sbi->s_mmp_tsk)
4031                 kthread_stop(sbi->s_mmp_tsk);
4032 failed_mount2:
4033         for (i = 0; i < db_count; i++)
4034                 brelse(sbi->s_group_desc[i]);
4035         ext4_kvfree(sbi->s_group_desc);
4036 failed_mount:
4037         if (sbi->s_chksum_driver)
4038                 crypto_free_shash(sbi->s_chksum_driver);
4039         if (sbi->s_proc) {
4040                 remove_proc_entry("options", sbi->s_proc);
4041                 remove_proc_entry(sb->s_id, ext4_proc_root);
4042         }
4043 #ifdef CONFIG_QUOTA
4044         for (i = 0; i < MAXQUOTAS; i++)
4045                 kfree(sbi->s_qf_names[i]);
4046 #endif
4047         ext4_blkdev_remove(sbi);
4048         brelse(bh);
4049 out_fail:
4050         sb->s_fs_info = NULL;
4051         kfree(sbi->s_blockgroup_lock);
4052         kfree(sbi);
4053 out_free_orig:
4054         kfree(orig_data);
4055         return ret;
4056 }
4057
4058 /*
4059  * Setup any per-fs journal parameters now.  We'll do this both on
4060  * initial mount, once the journal has been initialised but before we've
4061  * done any recovery; and again on any subsequent remount.
4062  */
4063 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4064 {
4065         struct ext4_sb_info *sbi = EXT4_SB(sb);
4066
4067         journal->j_commit_interval = sbi->s_commit_interval;
4068         journal->j_min_batch_time = sbi->s_min_batch_time;
4069         journal->j_max_batch_time = sbi->s_max_batch_time;
4070
4071         write_lock(&journal->j_state_lock);
4072         if (test_opt(sb, BARRIER))
4073                 journal->j_flags |= JBD2_BARRIER;
4074         else
4075                 journal->j_flags &= ~JBD2_BARRIER;
4076         if (test_opt(sb, DATA_ERR_ABORT))
4077                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4078         else
4079                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4080         write_unlock(&journal->j_state_lock);
4081 }
4082
4083 static journal_t *ext4_get_journal(struct super_block *sb,
4084                                    unsigned int journal_inum)
4085 {
4086         struct inode *journal_inode;
4087         journal_t *journal;
4088
4089         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4090
4091         /* First, test for the existence of a valid inode on disk.  Bad
4092          * things happen if we iget() an unused inode, as the subsequent
4093          * iput() will try to delete it. */
4094
4095         journal_inode = ext4_iget(sb, journal_inum);
4096         if (IS_ERR(journal_inode)) {
4097                 ext4_msg(sb, KERN_ERR, "no journal found");
4098                 return NULL;
4099         }
4100         if (!journal_inode->i_nlink) {
4101                 make_bad_inode(journal_inode);
4102                 iput(journal_inode);
4103                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4104                 return NULL;
4105         }
4106
4107         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4108                   journal_inode, journal_inode->i_size);
4109         if (!S_ISREG(journal_inode->i_mode)) {
4110                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4111                 iput(journal_inode);
4112                 return NULL;
4113         }
4114
4115         journal = jbd2_journal_init_inode(journal_inode);
4116         if (!journal) {
4117                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4118                 iput(journal_inode);
4119                 return NULL;
4120         }
4121         journal->j_private = sb;
4122         ext4_init_journal_params(sb, journal);
4123         return journal;
4124 }
4125
4126 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4127                                        dev_t j_dev)
4128 {
4129         struct buffer_head *bh;
4130         journal_t *journal;
4131         ext4_fsblk_t start;
4132         ext4_fsblk_t len;
4133         int hblock, blocksize;
4134         ext4_fsblk_t sb_block;
4135         unsigned long offset;
4136         struct ext4_super_block *es;
4137         struct block_device *bdev;
4138
4139         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4140
4141         bdev = ext4_blkdev_get(j_dev, sb);
4142         if (bdev == NULL)
4143                 return NULL;
4144
4145         blocksize = sb->s_blocksize;
4146         hblock = bdev_logical_block_size(bdev);
4147         if (blocksize < hblock) {
4148                 ext4_msg(sb, KERN_ERR,
4149                         "blocksize too small for journal device");
4150                 goto out_bdev;
4151         }
4152
4153         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4154         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4155         set_blocksize(bdev, blocksize);
4156         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4157                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4158                        "external journal");
4159                 goto out_bdev;
4160         }
4161
4162         es = (struct ext4_super_block *) (bh->b_data + offset);
4163         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4164             !(le32_to_cpu(es->s_feature_incompat) &
4165               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4166                 ext4_msg(sb, KERN_ERR, "external journal has "
4167                                         "bad superblock");
4168                 brelse(bh);
4169                 goto out_bdev;
4170         }
4171
4172         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4173                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4174                 brelse(bh);
4175                 goto out_bdev;
4176         }
4177
4178         len = ext4_blocks_count(es);
4179         start = sb_block + 1;
4180         brelse(bh);     /* we're done with the superblock */
4181
4182         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4183                                         start, len, blocksize);
4184         if (!journal) {
4185                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4186                 goto out_bdev;
4187         }
4188         journal->j_private = sb;
4189         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4190         wait_on_buffer(journal->j_sb_buffer);
4191         if (!buffer_uptodate(journal->j_sb_buffer)) {
4192                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4193                 goto out_journal;
4194         }
4195         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4196                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4197                                         "user (unsupported) - %d",
4198                         be32_to_cpu(journal->j_superblock->s_nr_users));
4199                 goto out_journal;
4200         }
4201         EXT4_SB(sb)->journal_bdev = bdev;
4202         ext4_init_journal_params(sb, journal);
4203         return journal;
4204
4205 out_journal:
4206         jbd2_journal_destroy(journal);
4207 out_bdev:
4208         ext4_blkdev_put(bdev);
4209         return NULL;
4210 }
4211
4212 static int ext4_load_journal(struct super_block *sb,
4213                              struct ext4_super_block *es,
4214                              unsigned long journal_devnum)
4215 {
4216         journal_t *journal;
4217         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4218         dev_t journal_dev;
4219         int err = 0;
4220         int really_read_only;
4221
4222         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4223
4224         if (journal_devnum &&
4225             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4226                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4227                         "numbers have changed");
4228                 journal_dev = new_decode_dev(journal_devnum);
4229         } else
4230                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4231
4232         really_read_only = bdev_read_only(sb->s_bdev);
4233
4234         /*
4235          * Are we loading a blank journal or performing recovery after a
4236          * crash?  For recovery, we need to check in advance whether we
4237          * can get read-write access to the device.
4238          */
4239         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4240                 if (sb->s_flags & MS_RDONLY) {
4241                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4242                                         "required on readonly filesystem");
4243                         if (really_read_only) {
4244                                 ext4_msg(sb, KERN_ERR, "write access "
4245                                         "unavailable, cannot proceed");
4246                                 return -EROFS;
4247                         }
4248                         ext4_msg(sb, KERN_INFO, "write access will "
4249                                "be enabled during recovery");
4250                 }
4251         }
4252
4253         if (journal_inum && journal_dev) {
4254                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4255                        "and inode journals!");
4256                 return -EINVAL;
4257         }
4258
4259         if (journal_inum) {
4260                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4261                         return -EINVAL;
4262         } else {
4263                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4264                         return -EINVAL;
4265         }
4266
4267         if (!(journal->j_flags & JBD2_BARRIER))
4268                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4269
4270         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4271                 err = jbd2_journal_wipe(journal, !really_read_only);
4272         if (!err) {
4273                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4274                 if (save)
4275                         memcpy(save, ((char *) es) +
4276                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4277                 err = jbd2_journal_load(journal);
4278                 if (save)
4279                         memcpy(((char *) es) + EXT4_S_ERR_START,
4280                                save, EXT4_S_ERR_LEN);
4281                 kfree(save);
4282         }
4283
4284         if (err) {
4285                 ext4_msg(sb, KERN_ERR, "error loading journal");
4286                 jbd2_journal_destroy(journal);
4287                 return err;
4288         }
4289
4290         EXT4_SB(sb)->s_journal = journal;
4291         ext4_clear_journal_err(sb, es);
4292
4293         if (!really_read_only && journal_devnum &&
4294             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4295                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4296
4297                 /* Make sure we flush the recovery flag to disk. */
4298                 ext4_commit_super(sb, 1);
4299         }
4300
4301         return 0;
4302 }
4303
4304 static int ext4_commit_super(struct super_block *sb, int sync)
4305 {
4306         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4307         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4308         int error = 0;
4309
4310         if (!sbh || block_device_ejected(sb))
4311                 return error;
4312         if (buffer_write_io_error(sbh)) {
4313                 /*
4314                  * Oh, dear.  A previous attempt to write the
4315                  * superblock failed.  This could happen because the
4316                  * USB device was yanked out.  Or it could happen to
4317                  * be a transient write error and maybe the block will
4318                  * be remapped.  Nothing we can do but to retry the
4319                  * write and hope for the best.
4320                  */
4321                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4322                        "superblock detected");
4323                 clear_buffer_write_io_error(sbh);
4324                 set_buffer_uptodate(sbh);
4325         }
4326         /*
4327          * If the file system is mounted read-only, don't update the
4328          * superblock write time.  This avoids updating the superblock
4329          * write time when we are mounting the root file system
4330          * read/only but we need to replay the journal; at that point,
4331          * for people who are east of GMT and who make their clock
4332          * tick in localtime for Windows bug-for-bug compatibility,
4333          * the clock is set in the future, and this will cause e2fsck
4334          * to complain and force a full file system check.
4335          */
4336         if (!(sb->s_flags & MS_RDONLY))
4337                 es->s_wtime = cpu_to_le32(get_seconds());
4338         if (sb->s_bdev->bd_part)
4339                 es->s_kbytes_written =
4340                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4341                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4342                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4343         else
4344                 es->s_kbytes_written =
4345                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4346         ext4_free_blocks_count_set(es,
4347                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4348                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4349         es->s_free_inodes_count =
4350                 cpu_to_le32(percpu_counter_sum_positive(
4351                                 &EXT4_SB(sb)->s_freeinodes_counter));
4352         BUFFER_TRACE(sbh, "marking dirty");
4353         ext4_superblock_csum_set(sb, es);
4354         mark_buffer_dirty(sbh);
4355         if (sync) {
4356                 error = sync_dirty_buffer(sbh);
4357                 if (error)
4358                         return error;
4359
4360                 error = buffer_write_io_error(sbh);
4361                 if (error) {
4362                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4363                                "superblock");
4364                         clear_buffer_write_io_error(sbh);
4365                         set_buffer_uptodate(sbh);
4366                 }
4367         }
4368         return error;
4369 }
4370
4371 /*
4372  * Have we just finished recovery?  If so, and if we are mounting (or
4373  * remounting) the filesystem readonly, then we will end up with a
4374  * consistent fs on disk.  Record that fact.
4375  */
4376 static void ext4_mark_recovery_complete(struct super_block *sb,
4377                                         struct ext4_super_block *es)
4378 {
4379         journal_t *journal = EXT4_SB(sb)->s_journal;
4380
4381         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4382                 BUG_ON(journal != NULL);
4383                 return;
4384         }
4385         jbd2_journal_lock_updates(journal);
4386         if (jbd2_journal_flush(journal) < 0)
4387                 goto out;
4388
4389         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4390             sb->s_flags & MS_RDONLY) {
4391                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4392                 ext4_commit_super(sb, 1);
4393         }
4394
4395 out:
4396         jbd2_journal_unlock_updates(journal);
4397 }
4398
4399 /*
4400  * If we are mounting (or read-write remounting) a filesystem whose journal
4401  * has recorded an error from a previous lifetime, move that error to the
4402  * main filesystem now.
4403  */
4404 static void ext4_clear_journal_err(struct super_block *sb,
4405                                    struct ext4_super_block *es)
4406 {
4407         journal_t *journal;
4408         int j_errno;
4409         const char *errstr;
4410
4411         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4412
4413         journal = EXT4_SB(sb)->s_journal;
4414
4415         /*
4416          * Now check for any error status which may have been recorded in the
4417          * journal by a prior ext4_error() or ext4_abort()
4418          */
4419
4420         j_errno = jbd2_journal_errno(journal);
4421         if (j_errno) {
4422                 char nbuf[16];
4423
4424                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4425                 ext4_warning(sb, "Filesystem error recorded "
4426                              "from previous mount: %s", errstr);
4427                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4428
4429                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4430                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4431                 ext4_commit_super(sb, 1);
4432
4433                 jbd2_journal_clear_err(journal);
4434                 jbd2_journal_update_sb_errno(journal);
4435         }
4436 }
4437
4438 /*
4439  * Force the running and committing transactions to commit,
4440  * and wait on the commit.
4441  */
4442 int ext4_force_commit(struct super_block *sb)
4443 {
4444         journal_t *journal;
4445         int ret = 0;
4446
4447         if (sb->s_flags & MS_RDONLY)
4448                 return 0;
4449
4450         journal = EXT4_SB(sb)->s_journal;
4451         if (journal)
4452                 ret = ext4_journal_force_commit(journal);
4453
4454         return ret;
4455 }
4456
4457 static int ext4_sync_fs(struct super_block *sb, int wait)
4458 {
4459         int ret = 0;
4460         tid_t target;
4461         struct ext4_sb_info *sbi = EXT4_SB(sb);
4462
4463         trace_ext4_sync_fs(sb, wait);
4464         flush_workqueue(sbi->dio_unwritten_wq);
4465         /*
4466          * Writeback quota in non-journalled quota case - journalled quota has
4467          * no dirty dquots
4468          */
4469         dquot_writeback_dquots(sb, -1);
4470         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4471                 if (wait)
4472                         jbd2_log_wait_commit(sbi->s_journal, target);
4473         }
4474         return ret;
4475 }
4476
4477 /*
4478  * LVM calls this function before a (read-only) snapshot is created.  This
4479  * gives us a chance to flush the journal completely and mark the fs clean.
4480  *
4481  * Note that only this function cannot bring a filesystem to be in a clean
4482  * state independently. It relies on upper layer to stop all data & metadata
4483  * modifications.
4484  */
4485 static int ext4_freeze(struct super_block *sb)
4486 {
4487         int error = 0;
4488         journal_t *journal;
4489
4490         if (sb->s_flags & MS_RDONLY)
4491                 return 0;
4492
4493         journal = EXT4_SB(sb)->s_journal;
4494
4495         /* Now we set up the journal barrier. */
4496         jbd2_journal_lock_updates(journal);
4497
4498         /*
4499          * Don't clear the needs_recovery flag if we failed to flush
4500          * the journal.
4501          */
4502         error = jbd2_journal_flush(journal);
4503         if (error < 0)
4504                 goto out;
4505
4506         /* Journal blocked and flushed, clear needs_recovery flag. */
4507         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4508         error = ext4_commit_super(sb, 1);
4509 out:
4510         /* we rely on upper layer to stop further updates */
4511         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4512         return error;
4513 }
4514
4515 /*
4516  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4517  * flag here, even though the filesystem is not technically dirty yet.
4518  */
4519 static int ext4_unfreeze(struct super_block *sb)
4520 {
4521         if (sb->s_flags & MS_RDONLY)
4522                 return 0;
4523
4524         lock_super(sb);
4525         /* Reset the needs_recovery flag before the fs is unlocked. */
4526         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4527         ext4_commit_super(sb, 1);
4528         unlock_super(sb);
4529         return 0;
4530 }
4531
4532 /*
4533  * Structure to save mount options for ext4_remount's benefit
4534  */
4535 struct ext4_mount_options {
4536         unsigned long s_mount_opt;
4537         unsigned long s_mount_opt2;
4538         kuid_t s_resuid;
4539         kgid_t s_resgid;
4540         unsigned long s_commit_interval;
4541         u32 s_min_batch_time, s_max_batch_time;
4542 #ifdef CONFIG_QUOTA
4543         int s_jquota_fmt;
4544         char *s_qf_names[MAXQUOTAS];
4545 #endif
4546 };
4547
4548 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4549 {
4550         struct ext4_super_block *es;
4551         struct ext4_sb_info *sbi = EXT4_SB(sb);
4552         unsigned long old_sb_flags;
4553         struct ext4_mount_options old_opts;
4554         int enable_quota = 0;
4555         ext4_group_t g;
4556         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4557         int err = 0;
4558 #ifdef CONFIG_QUOTA
4559         int i;
4560 #endif
4561         char *orig_data = kstrdup(data, GFP_KERNEL);
4562
4563         /* Store the original options */
4564         lock_super(sb);
4565         old_sb_flags = sb->s_flags;
4566         old_opts.s_mount_opt = sbi->s_mount_opt;
4567         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4568         old_opts.s_resuid = sbi->s_resuid;
4569         old_opts.s_resgid = sbi->s_resgid;
4570         old_opts.s_commit_interval = sbi->s_commit_interval;
4571         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4572         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4573 #ifdef CONFIG_QUOTA
4574         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4575         for (i = 0; i < MAXQUOTAS; i++)
4576                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4577 #endif
4578         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4579                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4580
4581         /*
4582          * Allow the "check" option to be passed as a remount option.
4583          */
4584         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4585                 err = -EINVAL;
4586                 goto restore_opts;
4587         }
4588
4589         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4590                 ext4_abort(sb, "Abort forced by user");
4591
4592         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4593                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4594
4595         es = sbi->s_es;
4596
4597         if (sbi->s_journal) {
4598                 ext4_init_journal_params(sb, sbi->s_journal);
4599                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4600         }
4601
4602         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4603                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4604                         err = -EROFS;
4605                         goto restore_opts;
4606                 }
4607
4608                 if (*flags & MS_RDONLY) {
4609                         err = dquot_suspend(sb, -1);
4610                         if (err < 0)
4611                                 goto restore_opts;
4612
4613                         /*
4614                          * First of all, the unconditional stuff we have to do
4615                          * to disable replay of the journal when we next remount
4616                          */
4617                         sb->s_flags |= MS_RDONLY;
4618
4619                         /*
4620                          * OK, test if we are remounting a valid rw partition
4621                          * readonly, and if so set the rdonly flag and then
4622                          * mark the partition as valid again.
4623                          */
4624                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4625                             (sbi->s_mount_state & EXT4_VALID_FS))
4626                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4627
4628                         if (sbi->s_journal)
4629                                 ext4_mark_recovery_complete(sb, es);
4630                 } else {
4631                         /* Make sure we can mount this feature set readwrite */
4632                         if (!ext4_feature_set_ok(sb, 0)) {
4633                                 err = -EROFS;
4634                                 goto restore_opts;
4635                         }
4636                         /*
4637                          * Make sure the group descriptor checksums
4638                          * are sane.  If they aren't, refuse to remount r/w.
4639                          */
4640                         for (g = 0; g < sbi->s_groups_count; g++) {
4641                                 struct ext4_group_desc *gdp =
4642                                         ext4_get_group_desc(sb, g, NULL);
4643
4644                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4645                                         ext4_msg(sb, KERN_ERR,
4646                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4647                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4648                                                le16_to_cpu(gdp->bg_checksum));
4649                                         err = -EINVAL;
4650                                         goto restore_opts;
4651                                 }
4652                         }
4653
4654                         /*
4655                          * If we have an unprocessed orphan list hanging
4656                          * around from a previously readonly bdev mount,
4657                          * require a full umount/remount for now.
4658                          */
4659                         if (es->s_last_orphan) {
4660                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4661                                        "remount RDWR because of unprocessed "
4662                                        "orphan inode list.  Please "
4663                                        "umount/remount instead");
4664                                 err = -EINVAL;
4665                                 goto restore_opts;
4666                         }
4667
4668                         /*
4669                          * Mounting a RDONLY partition read-write, so reread
4670                          * and store the current valid flag.  (It may have
4671                          * been changed by e2fsck since we originally mounted
4672                          * the partition.)
4673                          */
4674                         if (sbi->s_journal)
4675                                 ext4_clear_journal_err(sb, es);
4676                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4677                         if (!ext4_setup_super(sb, es, 0))
4678                                 sb->s_flags &= ~MS_RDONLY;
4679                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4680                                                      EXT4_FEATURE_INCOMPAT_MMP))
4681                                 if (ext4_multi_mount_protect(sb,
4682                                                 le64_to_cpu(es->s_mmp_block))) {
4683                                         err = -EROFS;
4684                                         goto restore_opts;
4685                                 }
4686                         enable_quota = 1;
4687                 }
4688         }
4689
4690         /*
4691          * Reinitialize lazy itable initialization thread based on
4692          * current settings
4693          */
4694         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4695                 ext4_unregister_li_request(sb);
4696         else {
4697                 ext4_group_t first_not_zeroed;
4698                 first_not_zeroed = ext4_has_uninit_itable(sb);
4699                 ext4_register_li_request(sb, first_not_zeroed);
4700         }
4701
4702         ext4_setup_system_zone(sb);
4703         if (sbi->s_journal == NULL)
4704                 ext4_commit_super(sb, 1);
4705
4706         unlock_super(sb);
4707 #ifdef CONFIG_QUOTA
4708         /* Release old quota file names */
4709         for (i = 0; i < MAXQUOTAS; i++)
4710                 if (old_opts.s_qf_names[i] &&
4711                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4712                         kfree(old_opts.s_qf_names[i]);
4713         if (enable_quota) {
4714                 if (sb_any_quota_suspended(sb))
4715                         dquot_resume(sb, -1);
4716                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4717                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4718                         err = ext4_enable_quotas(sb);
4719                         if (err) {
4720                                 lock_super(sb);
4721                                 goto restore_opts;
4722                         }
4723                 }
4724         }
4725 #endif
4726
4727         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4728         kfree(orig_data);
4729         return 0;
4730
4731 restore_opts:
4732         sb->s_flags = old_sb_flags;
4733         sbi->s_mount_opt = old_opts.s_mount_opt;
4734         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4735         sbi->s_resuid = old_opts.s_resuid;
4736         sbi->s_resgid = old_opts.s_resgid;
4737         sbi->s_commit_interval = old_opts.s_commit_interval;
4738         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4739         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4740 #ifdef CONFIG_QUOTA
4741         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4742         for (i = 0; i < MAXQUOTAS; i++) {
4743                 if (sbi->s_qf_names[i] &&
4744                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4745                         kfree(sbi->s_qf_names[i]);
4746                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4747         }
4748 #endif
4749         unlock_super(sb);
4750         kfree(orig_data);
4751         return err;
4752 }
4753
4754 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4755 {
4756         struct super_block *sb = dentry->d_sb;
4757         struct ext4_sb_info *sbi = EXT4_SB(sb);
4758         struct ext4_super_block *es = sbi->s_es;
4759         ext4_fsblk_t overhead = 0;
4760         u64 fsid;
4761         s64 bfree;
4762
4763         if (!test_opt(sb, MINIX_DF))
4764                 overhead = sbi->s_overhead;
4765
4766         buf->f_type = EXT4_SUPER_MAGIC;
4767         buf->f_bsize = sb->s_blocksize;
4768         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, sbi->s_overhead);
4769         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4770                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4771         /* prevent underflow in case that few free space is available */
4772         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4773         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4774         if (buf->f_bfree < ext4_r_blocks_count(es))
4775                 buf->f_bavail = 0;
4776         buf->f_files = le32_to_cpu(es->s_inodes_count);
4777         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4778         buf->f_namelen = EXT4_NAME_LEN;
4779         fsid = le64_to_cpup((void *)es->s_uuid) ^
4780                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4781         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4782         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4783
4784         return 0;
4785 }
4786
4787 /* Helper function for writing quotas on sync - we need to start transaction
4788  * before quota file is locked for write. Otherwise the are possible deadlocks:
4789  * Process 1                         Process 2
4790  * ext4_create()                     quota_sync()
4791  *   jbd2_journal_start()                  write_dquot()
4792  *   dquot_initialize()                         down(dqio_mutex)
4793  *     down(dqio_mutex)                    jbd2_journal_start()
4794  *
4795  */
4796
4797 #ifdef CONFIG_QUOTA
4798
4799 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4800 {
4801         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4802 }
4803
4804 static int ext4_write_dquot(struct dquot *dquot)
4805 {
4806         int ret, err;
4807         handle_t *handle;
4808         struct inode *inode;
4809
4810         inode = dquot_to_inode(dquot);
4811         handle = ext4_journal_start(inode,
4812                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4813         if (IS_ERR(handle))
4814                 return PTR_ERR(handle);
4815         ret = dquot_commit(dquot);
4816         err = ext4_journal_stop(handle);
4817         if (!ret)
4818                 ret = err;
4819         return ret;
4820 }
4821
4822 static int ext4_acquire_dquot(struct dquot *dquot)
4823 {
4824         int ret, err;
4825         handle_t *handle;
4826
4827         handle = ext4_journal_start(dquot_to_inode(dquot),
4828                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4829         if (IS_ERR(handle))
4830                 return PTR_ERR(handle);
4831         ret = dquot_acquire(dquot);
4832         err = ext4_journal_stop(handle);
4833         if (!ret)
4834                 ret = err;
4835         return ret;
4836 }
4837
4838 static int ext4_release_dquot(struct dquot *dquot)
4839 {
4840         int ret, err;
4841         handle_t *handle;
4842
4843         handle = ext4_journal_start(dquot_to_inode(dquot),
4844                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4845         if (IS_ERR(handle)) {
4846                 /* Release dquot anyway to avoid endless cycle in dqput() */
4847                 dquot_release(dquot);
4848                 return PTR_ERR(handle);
4849         }
4850         ret = dquot_release(dquot);
4851         err = ext4_journal_stop(handle);
4852         if (!ret)
4853                 ret = err;
4854         return ret;
4855 }
4856
4857 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4858 {
4859         /* Are we journaling quotas? */
4860         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4861             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4862                 dquot_mark_dquot_dirty(dquot);
4863                 return ext4_write_dquot(dquot);
4864         } else {
4865                 return dquot_mark_dquot_dirty(dquot);
4866         }
4867 }
4868
4869 static int ext4_write_info(struct super_block *sb, int type)
4870 {
4871         int ret, err;
4872         handle_t *handle;
4873
4874         /* Data block + inode block */
4875         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4876         if (IS_ERR(handle))
4877                 return PTR_ERR(handle);
4878         ret = dquot_commit_info(sb, type);
4879         err = ext4_journal_stop(handle);
4880         if (!ret)
4881                 ret = err;
4882         return ret;
4883 }
4884
4885 /*
4886  * Turn on quotas during mount time - we need to find
4887  * the quota file and such...
4888  */
4889 static int ext4_quota_on_mount(struct super_block *sb, int type)
4890 {
4891         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4892                                         EXT4_SB(sb)->s_jquota_fmt, type);
4893 }
4894
4895 /*
4896  * Standard function to be called on quota_on
4897  */
4898 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4899                          struct path *path)
4900 {
4901         int err;
4902
4903         if (!test_opt(sb, QUOTA))
4904                 return -EINVAL;
4905
4906         /* Quotafile not on the same filesystem? */
4907         if (path->dentry->d_sb != sb)
4908                 return -EXDEV;
4909         /* Journaling quota? */
4910         if (EXT4_SB(sb)->s_qf_names[type]) {
4911                 /* Quotafile not in fs root? */
4912                 if (path->dentry->d_parent != sb->s_root)
4913                         ext4_msg(sb, KERN_WARNING,
4914                                 "Quota file not on filesystem root. "
4915                                 "Journaled quota will not work");
4916         }
4917
4918         /*
4919          * When we journal data on quota file, we have to flush journal to see
4920          * all updates to the file when we bypass pagecache...
4921          */
4922         if (EXT4_SB(sb)->s_journal &&
4923             ext4_should_journal_data(path->dentry->d_inode)) {
4924                 /*
4925                  * We don't need to lock updates but journal_flush() could
4926                  * otherwise be livelocked...
4927                  */
4928                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4929                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4930                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4931                 if (err)
4932                         return err;
4933         }
4934
4935         return dquot_quota_on(sb, type, format_id, path);
4936 }
4937
4938 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
4939                              unsigned int flags)
4940 {
4941         int err;
4942         struct inode *qf_inode;
4943         unsigned long qf_inums[MAXQUOTAS] = {
4944                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4945                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4946         };
4947
4948         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
4949
4950         if (!qf_inums[type])
4951                 return -EPERM;
4952
4953         qf_inode = ext4_iget(sb, qf_inums[type]);
4954         if (IS_ERR(qf_inode)) {
4955                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
4956                 return PTR_ERR(qf_inode);
4957         }
4958
4959         err = dquot_enable(qf_inode, type, format_id, flags);
4960         iput(qf_inode);
4961
4962         return err;
4963 }
4964
4965 /* Enable usage tracking for all quota types. */
4966 static int ext4_enable_quotas(struct super_block *sb)
4967 {
4968         int type, err = 0;
4969         unsigned long qf_inums[MAXQUOTAS] = {
4970                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4971                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4972         };
4973
4974         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
4975         for (type = 0; type < MAXQUOTAS; type++) {
4976                 if (qf_inums[type]) {
4977                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
4978                                                 DQUOT_USAGE_ENABLED);
4979                         if (err) {
4980                                 ext4_warning(sb,
4981                                         "Failed to enable quota (type=%d) "
4982                                         "tracking. Please run e2fsck to fix.",
4983                                         type);
4984                                 return err;
4985                         }
4986                 }
4987         }
4988         return 0;
4989 }
4990
4991 /*
4992  * quota_on function that is used when QUOTA feature is set.
4993  */
4994 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
4995                                  int format_id)
4996 {
4997         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
4998                 return -EINVAL;
4999
5000         /*
5001          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5002          */
5003         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5004 }
5005
5006 static int ext4_quota_off(struct super_block *sb, int type)
5007 {
5008         struct inode *inode = sb_dqopt(sb)->files[type];
5009         handle_t *handle;
5010
5011         /* Force all delayed allocation blocks to be allocated.
5012          * Caller already holds s_umount sem */
5013         if (test_opt(sb, DELALLOC))
5014                 sync_filesystem(sb);
5015
5016         if (!inode)
5017                 goto out;
5018
5019         /* Update modification times of quota files when userspace can
5020          * start looking at them */
5021         handle = ext4_journal_start(inode, 1);
5022         if (IS_ERR(handle))
5023                 goto out;
5024         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5025         ext4_mark_inode_dirty(handle, inode);
5026         ext4_journal_stop(handle);
5027
5028 out:
5029         return dquot_quota_off(sb, type);
5030 }
5031
5032 /*
5033  * quota_off function that is used when QUOTA feature is set.
5034  */
5035 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5036 {
5037         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5038                 return -EINVAL;
5039
5040         /* Disable only the limits. */
5041         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5042 }
5043
5044 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5045  * acquiring the locks... As quota files are never truncated and quota code
5046  * itself serializes the operations (and no one else should touch the files)
5047  * we don't have to be afraid of races */
5048 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5049                                size_t len, loff_t off)
5050 {
5051         struct inode *inode = sb_dqopt(sb)->files[type];
5052         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5053         int err = 0;
5054         int offset = off & (sb->s_blocksize - 1);
5055         int tocopy;
5056         size_t toread;
5057         struct buffer_head *bh;
5058         loff_t i_size = i_size_read(inode);
5059
5060         if (off > i_size)
5061                 return 0;
5062         if (off+len > i_size)
5063                 len = i_size-off;
5064         toread = len;
5065         while (toread > 0) {
5066                 tocopy = sb->s_blocksize - offset < toread ?
5067                                 sb->s_blocksize - offset : toread;
5068                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5069                 if (err)
5070                         return err;
5071                 if (!bh)        /* A hole? */
5072                         memset(data, 0, tocopy);
5073                 else
5074                         memcpy(data, bh->b_data+offset, tocopy);
5075                 brelse(bh);
5076                 offset = 0;
5077                 toread -= tocopy;
5078                 data += tocopy;
5079                 blk++;
5080         }
5081         return len;
5082 }
5083
5084 /* Write to quotafile (we know the transaction is already started and has
5085  * enough credits) */
5086 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5087                                 const char *data, size_t len, loff_t off)
5088 {
5089         struct inode *inode = sb_dqopt(sb)->files[type];
5090         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5091         int err = 0;
5092         int offset = off & (sb->s_blocksize - 1);
5093         struct buffer_head *bh;
5094         handle_t *handle = journal_current_handle();
5095
5096         if (EXT4_SB(sb)->s_journal && !handle) {
5097                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5098                         " cancelled because transaction is not started",
5099                         (unsigned long long)off, (unsigned long long)len);
5100                 return -EIO;
5101         }
5102         /*
5103          * Since we account only one data block in transaction credits,
5104          * then it is impossible to cross a block boundary.
5105          */
5106         if (sb->s_blocksize - offset < len) {
5107                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5108                         " cancelled because not block aligned",
5109                         (unsigned long long)off, (unsigned long long)len);
5110                 return -EIO;
5111         }
5112
5113         bh = ext4_bread(handle, inode, blk, 1, &err);
5114         if (!bh)
5115                 goto out;
5116         err = ext4_journal_get_write_access(handle, bh);
5117         if (err) {
5118                 brelse(bh);
5119                 goto out;
5120         }
5121         lock_buffer(bh);
5122         memcpy(bh->b_data+offset, data, len);
5123         flush_dcache_page(bh->b_page);
5124         unlock_buffer(bh);
5125         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5126         brelse(bh);
5127 out:
5128         if (err)
5129                 return err;
5130         if (inode->i_size < off + len) {
5131                 i_size_write(inode, off + len);
5132                 EXT4_I(inode)->i_disksize = inode->i_size;
5133                 ext4_mark_inode_dirty(handle, inode);
5134         }
5135         return len;
5136 }
5137
5138 #endif
5139
5140 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5141                        const char *dev_name, void *data)
5142 {
5143         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5144 }
5145
5146 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5147 static inline void register_as_ext2(void)
5148 {
5149         int err = register_filesystem(&ext2_fs_type);
5150         if (err)
5151                 printk(KERN_WARNING
5152                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5153 }
5154
5155 static inline void unregister_as_ext2(void)
5156 {
5157         unregister_filesystem(&ext2_fs_type);
5158 }
5159
5160 static inline int ext2_feature_set_ok(struct super_block *sb)
5161 {
5162         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5163                 return 0;
5164         if (sb->s_flags & MS_RDONLY)
5165                 return 1;
5166         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5167                 return 0;
5168         return 1;
5169 }
5170 MODULE_ALIAS("ext2");
5171 #else
5172 static inline void register_as_ext2(void) { }
5173 static inline void unregister_as_ext2(void) { }
5174 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5175 #endif
5176
5177 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5178 static inline void register_as_ext3(void)
5179 {
5180         int err = register_filesystem(&ext3_fs_type);
5181         if (err)
5182                 printk(KERN_WARNING
5183                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5184 }
5185
5186 static inline void unregister_as_ext3(void)
5187 {
5188         unregister_filesystem(&ext3_fs_type);
5189 }
5190
5191 static inline int ext3_feature_set_ok(struct super_block *sb)
5192 {
5193         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5194                 return 0;
5195         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5196                 return 0;
5197         if (sb->s_flags & MS_RDONLY)
5198                 return 1;
5199         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5200                 return 0;
5201         return 1;
5202 }
5203 MODULE_ALIAS("ext3");
5204 #else
5205 static inline void register_as_ext3(void) { }
5206 static inline void unregister_as_ext3(void) { }
5207 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5208 #endif
5209
5210 static struct file_system_type ext4_fs_type = {
5211         .owner          = THIS_MODULE,
5212         .name           = "ext4",
5213         .mount          = ext4_mount,
5214         .kill_sb        = kill_block_super,
5215         .fs_flags       = FS_REQUIRES_DEV,
5216 };
5217
5218 static int __init ext4_init_feat_adverts(void)
5219 {
5220         struct ext4_features *ef;
5221         int ret = -ENOMEM;
5222
5223         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5224         if (!ef)
5225                 goto out;
5226
5227         ef->f_kobj.kset = ext4_kset;
5228         init_completion(&ef->f_kobj_unregister);
5229         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5230                                    "features");
5231         if (ret) {
5232                 kfree(ef);
5233                 goto out;
5234         }
5235
5236         ext4_feat = ef;
5237         ret = 0;
5238 out:
5239         return ret;
5240 }
5241
5242 static void ext4_exit_feat_adverts(void)
5243 {
5244         kobject_put(&ext4_feat->f_kobj);
5245         wait_for_completion(&ext4_feat->f_kobj_unregister);
5246         kfree(ext4_feat);
5247 }
5248
5249 /* Shared across all ext4 file systems */
5250 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5251 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5252
5253 static int __init ext4_init_fs(void)
5254 {
5255         int i, err;
5256
5257         ext4_li_info = NULL;
5258         mutex_init(&ext4_li_mtx);
5259
5260         ext4_check_flag_values();
5261
5262         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5263                 mutex_init(&ext4__aio_mutex[i]);
5264                 init_waitqueue_head(&ext4__ioend_wq[i]);
5265         }
5266
5267         err = ext4_init_pageio();
5268         if (err)
5269                 return err;
5270         err = ext4_init_system_zone();
5271         if (err)
5272                 goto out6;
5273         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5274         if (!ext4_kset)
5275                 goto out5;
5276         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5277
5278         err = ext4_init_feat_adverts();
5279         if (err)
5280                 goto out4;
5281
5282         err = ext4_init_mballoc();
5283         if (err)
5284                 goto out3;
5285
5286         err = ext4_init_xattr();
5287         if (err)
5288                 goto out2;
5289         err = init_inodecache();
5290         if (err)
5291                 goto out1;
5292         register_as_ext3();
5293         register_as_ext2();
5294         err = register_filesystem(&ext4_fs_type);
5295         if (err)
5296                 goto out;
5297
5298         return 0;
5299 out:
5300         unregister_as_ext2();
5301         unregister_as_ext3();
5302         destroy_inodecache();
5303 out1:
5304         ext4_exit_xattr();
5305 out2:
5306         ext4_exit_mballoc();
5307 out3:
5308         ext4_exit_feat_adverts();
5309 out4:
5310         if (ext4_proc_root)
5311                 remove_proc_entry("fs/ext4", NULL);
5312         kset_unregister(ext4_kset);
5313 out5:
5314         ext4_exit_system_zone();
5315 out6:
5316         ext4_exit_pageio();
5317         return err;
5318 }
5319
5320 static void __exit ext4_exit_fs(void)
5321 {
5322         ext4_destroy_lazyinit_thread();
5323         unregister_as_ext2();
5324         unregister_as_ext3();
5325         unregister_filesystem(&ext4_fs_type);
5326         destroy_inodecache();
5327         ext4_exit_xattr();
5328         ext4_exit_mballoc();
5329         ext4_exit_feat_adverts();
5330         remove_proc_entry("fs/ext4", NULL);
5331         kset_unregister(ext4_kset);
5332         ext4_exit_system_zone();
5333         ext4_exit_pageio();
5334 }
5335
5336 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5337 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5338 MODULE_LICENSE("GPL");
5339 module_init(ext4_init_fs)
5340 module_exit(ext4_exit_fs)