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         if (sbi->s_journal) {
 865                 err = jbd2_journal_destroy(sbi->s_journal);
 866                 sbi->s_journal = NULL;
 867                 if (err < 0)
 868                         ext4_abort(sb, "Couldn't clean up the journal");
 869         }
 870
 871         del_timer(&sbi->s_err_report);
 872         ext4_release_system_zone(sb);
 873         ext4_mb_release(sb);
 874         ext4_ext_release(sb);
 875         ext4_xattr_put_super(sb);
 876
 877         if (!(sb->s_flags & MS_RDONLY)) {
 878                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 879                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 880         }
 881         if (!(sb->s_flags & MS_RDONLY))
 882                 ext4_commit_super(sb, 1);
 883
 884         if (sbi->s_proc) {
 885                 remove_proc_entry("options", sbi->s_proc);
 886                 remove_proc_entry(sb->s_id, ext4_proc_root);
 887         }
 888         kobject_del(&sbi->s_kobj);
 889
 890         for (i = 0; i < sbi->s_gdb_count; i++)
 891                 brelse(sbi->s_group_desc[i]);
 892         ext4_kvfree(sbi->s_group_desc);
 893         ext4_kvfree(sbi->s_flex_groups);
 894         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 895         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 896         percpu_counter_destroy(&sbi->s_dirs_counter);
 897         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 898         brelse(sbi->s_sbh);
 899 #ifdef CONFIG_QUOTA
 900         for (i = 0; i < MAXQUOTAS; i++)
 901                 kfree(sbi->s_qf_names[i]);
 902 #endif
 903
 904         /* Debugging code just in case the in-memory inode orphan list
 905          * isn't empty.  The on-disk one can be non-empty if we've
 906          * detected an error and taken the fs readonly, but the
 907          * in-memory list had better be clean by this point. */
 908         if (!list_empty(&sbi->s_orphan))
 909                 dump_orphan_list(sb, sbi);
 910         J_ASSERT(list_empty(&sbi->s_orphan));
 911
 912         invalidate_bdev(sb->s_bdev);
 913         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 914                 /*
 915                  * Invalidate the journal device's buffers.  We don't want them
 916                  * floating about in memory - the physical journal device may
 917                  * hotswapped, and it breaks the `ro-after' testing code.
 918                  */
 919                 sync_blockdev(sbi->journal_bdev);
 920                 invalidate_bdev(sbi->journal_bdev);
 921                 ext4_blkdev_remove(sbi);
 922         }
 923         if (sbi->s_mmp_tsk)
 924                 kthread_stop(sbi->s_mmp_tsk);
 925         sb->s_fs_info = NULL;
 926         /*
 927          * Now that we are completely done shutting down the
 928          * superblock, we need to actually destroy the kobject.
 929          */
 930         kobject_put(&sbi->s_kobj);
 931         wait_for_completion(&sbi->s_kobj_unregister);
 932         if (sbi->s_chksum_driver)
 933                 crypto_free_shash(sbi->s_chksum_driver);
 934         kfree(sbi->s_blockgroup_lock);
 935         kfree(sbi);
 936 }
 937
 938 static struct kmem_cache *ext4_inode_cachep;
 939
 940 /*
 941  * Called inside transaction, so use GFP_NOFS
 942  */
 943 static struct inode *ext4_alloc_inode(struct super_block *sb)
 944 {
 945         struct ext4_inode_info *ei;
 946
 947         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 948         if (!ei)
 949                 return NULL;
 950
 951         ei->vfs_inode.i_version = 1;
 952         ei->vfs_inode.i_data.writeback_index = 0;
 953         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 954         INIT_LIST_HEAD(&ei->i_prealloc_list);
 955         spin_lock_init(&ei->i_prealloc_lock);
 956         ei->i_reserved_data_blocks = 0;
 957         ei->i_reserved_meta_blocks = 0;
 958         ei->i_allocated_meta_blocks = 0;
 959         ei->i_da_metadata_calc_len = 0;
 960         ei->i_da_metadata_calc_last_lblock = 0;
 961         spin_lock_init(&(ei->i_block_reservation_lock));
 962 #ifdef CONFIG_QUOTA
 963         ei->i_reserved_quota = 0;
 964 #endif
 965         ei->jinode = NULL;
 966         INIT_LIST_HEAD(&ei->i_completed_io_list);
 967         spin_lock_init(&ei->i_completed_io_lock);
 968         ei->cur_aio_dio = NULL;
 969         ei->i_sync_tid = 0;
 970         ei->i_datasync_tid = 0;
 971         atomic_set(&ei->i_ioend_count, 0);
 972         atomic_set(&ei->i_aiodio_unwritten, 0);
 973
 974         return &ei->vfs_inode;
 975 }
 976
 977 static int ext4_drop_inode(struct inode *inode)
 978 {
 979         int drop = generic_drop_inode(inode);
 980
 981         trace_ext4_drop_inode(inode, drop);
 982         return drop;
 983 }
 984
 985 static void ext4_i_callback(struct rcu_head *head)
 986 {
 987         struct inode *inode = container_of(head, struct inode, i_rcu);
 988         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 989 }
 990
 991 static void ext4_destroy_inode(struct inode *inode)
 992 {
 993         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 994                 ext4_msg(inode->i_sb, KERN_ERR,
 995                          "Inode %lu (%p): orphan list check failed!",
 996                          inode->i_ino, EXT4_I(inode));
 997                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 998                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 999                                 true);
1000                 dump_stack();
1001         }
1002         call_rcu(&inode->i_rcu, ext4_i_callback);
1003 }
1004
1005 static void init_once(void *foo)
1006 {
1007         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
1008
1009         INIT_LIST_HEAD(&ei->i_orphan);
1010 #ifdef CONFIG_EXT4_FS_XATTR
1011         init_rwsem(&ei->xattr_sem);
1012 #endif
1013         init_rwsem(&ei->i_data_sem);
1014         inode_init_once(&ei->vfs_inode);
1015 }
1016
1017 static int init_inodecache(void)
1018 {
1019         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1020                                              sizeof(struct ext4_inode_info),
1021                                              0, (SLAB_RECLAIM_ACCOUNT|
1022                                                 SLAB_MEM_SPREAD),
1023                                              init_once);
1024         if (ext4_inode_cachep == NULL)
1025                 return -ENOMEM;
1026         return 0;
1027 }
1028
1029 static void destroy_inodecache(void)
1030 {
1031         kmem_cache_destroy(ext4_inode_cachep);
1032 }
1033
1034 void ext4_clear_inode(struct inode *inode)
1035 {
1036         invalidate_inode_buffers(inode);
1037         clear_inode(inode);
1038         dquot_drop(inode);
1039         ext4_discard_preallocations(inode);
1040         if (EXT4_I(inode)->jinode) {
1041                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1042                                                EXT4_I(inode)->jinode);
1043                 jbd2_free_inode(EXT4_I(inode)->jinode);
1044                 EXT4_I(inode)->jinode = NULL;
1045         }
1046 }
1047
1048 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1049                                         u64 ino, u32 generation)
1050 {
1051         struct inode *inode;
1052
1053         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1054                 return ERR_PTR(-ESTALE);
1055         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1056                 return ERR_PTR(-ESTALE);
1057
1058         /* iget isn't really right if the inode is currently unallocated!!
1059          *
1060          * ext4_read_inode will return a bad_inode if the inode had been
1061          * deleted, so we should be safe.
1062          *
1063          * Currently we don't know the generation for parent directory, so
1064          * a generation of 0 means "accept any"
1065          */
1066         inode = ext4_iget(sb, ino);
1067         if (IS_ERR(inode))
1068                 return ERR_CAST(inode);
1069         if (generation && inode->i_generation != generation) {
1070                 iput(inode);
1071                 return ERR_PTR(-ESTALE);
1072         }
1073
1074         return inode;
1075 }
1076
1077 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1078                                         int fh_len, int fh_type)
1079 {
1080         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1081                                     ext4_nfs_get_inode);
1082 }
1083
1084 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1085                                         int fh_len, int fh_type)
1086 {
1087         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1088                                     ext4_nfs_get_inode);
1089 }
1090
1091 /*
1092  * Try to release metadata pages (indirect blocks, directories) which are
1093  * mapped via the block device.  Since these pages could have journal heads
1094  * which would prevent try_to_free_buffers() from freeing them, we must use
1095  * jbd2 layer's try_to_free_buffers() function to release them.
1096  */
1097 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1098                                  gfp_t wait)
1099 {
1100         journal_t *journal = EXT4_SB(sb)->s_journal;
1101
1102         WARN_ON(PageChecked(page));
1103         if (!page_has_buffers(page))
1104                 return 0;
1105         if (journal)
1106                 return jbd2_journal_try_to_free_buffers(journal, page,
1107                                                         wait & ~__GFP_WAIT);
1108         return try_to_free_buffers(page);
1109 }
1110
1111 #ifdef CONFIG_QUOTA
1112 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1113 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1114
1115 static int ext4_write_dquot(struct dquot *dquot);
1116 static int ext4_acquire_dquot(struct dquot *dquot);
1117 static int ext4_release_dquot(struct dquot *dquot);
1118 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1119 static int ext4_write_info(struct super_block *sb, int type);
1120 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1121                          struct path *path);
1122 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1123                                  int format_id);
1124 static int ext4_quota_off(struct super_block *sb, int type);
1125 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1126 static int ext4_quota_on_mount(struct super_block *sb, int type);
1127 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1128                                size_t len, loff_t off);
1129 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1130                                 const char *data, size_t len, loff_t off);
1131 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1132                              unsigned int flags);
1133 static int ext4_enable_quotas(struct super_block *sb);
1134
1135 static const struct dquot_operations ext4_quota_operations = {
1136         .get_reserved_space = ext4_get_reserved_space,
1137         .write_dquot    = ext4_write_dquot,
1138         .acquire_dquot  = ext4_acquire_dquot,
1139         .release_dquot  = ext4_release_dquot,
1140         .mark_dirty     = ext4_mark_dquot_dirty,
1141         .write_info     = ext4_write_info,
1142         .alloc_dquot    = dquot_alloc,
1143         .destroy_dquot  = dquot_destroy,
1144 };
1145
1146 static const struct quotactl_ops ext4_qctl_operations = {
1147         .quota_on       = ext4_quota_on,
1148         .quota_off      = ext4_quota_off,
1149         .quota_sync     = dquot_quota_sync,
1150         .get_info       = dquot_get_dqinfo,
1151         .set_info       = dquot_set_dqinfo,
1152         .get_dqblk      = dquot_get_dqblk,
1153         .set_dqblk      = dquot_set_dqblk
1154 };
1155
1156 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1157         .quota_on_meta  = ext4_quota_on_sysfile,
1158         .quota_off      = ext4_quota_off_sysfile,
1159         .quota_sync     = dquot_quota_sync,
1160         .get_info       = dquot_get_dqinfo,
1161         .set_info       = dquot_set_dqinfo,
1162         .get_dqblk      = dquot_get_dqblk,
1163         .set_dqblk      = dquot_set_dqblk
1164 };
1165 #endif
1166
1167 static const struct super_operations ext4_sops = {
1168         .alloc_inode    = ext4_alloc_inode,
1169         .destroy_inode  = ext4_destroy_inode,
1170         .write_inode    = ext4_write_inode,
1171         .dirty_inode    = ext4_dirty_inode,
1172         .drop_inode     = ext4_drop_inode,
1173         .evict_inode    = ext4_evict_inode,
1174         .put_super      = ext4_put_super,
1175         .sync_fs        = ext4_sync_fs,
1176         .freeze_fs      = ext4_freeze,
1177         .unfreeze_fs    = ext4_unfreeze,
1178         .statfs         = ext4_statfs,
1179         .remount_fs     = ext4_remount,
1180         .show_options   = ext4_show_options,
1181 #ifdef CONFIG_QUOTA
1182         .quota_read     = ext4_quota_read,
1183         .quota_write    = ext4_quota_write,
1184 #endif
1185         .bdev_try_to_free_page = bdev_try_to_free_page,
1186 };
1187
1188 static const struct super_operations ext4_nojournal_sops = {
1189         .alloc_inode    = ext4_alloc_inode,
1190         .destroy_inode  = ext4_destroy_inode,
1191         .write_inode    = ext4_write_inode,
1192         .dirty_inode    = ext4_dirty_inode,
1193         .drop_inode     = ext4_drop_inode,
1194         .evict_inode    = ext4_evict_inode,
1195         .put_super      = ext4_put_super,
1196         .statfs         = ext4_statfs,
1197         .remount_fs     = ext4_remount,
1198         .show_options   = ext4_show_options,
1199 #ifdef CONFIG_QUOTA
1200         .quota_read     = ext4_quota_read,
1201         .quota_write    = ext4_quota_write,
1202 #endif
1203         .bdev_try_to_free_page = bdev_try_to_free_page,
1204 };
1205
1206 static const struct export_operations ext4_export_ops = {
1207         .fh_to_dentry = ext4_fh_to_dentry,
1208         .fh_to_parent = ext4_fh_to_parent,
1209         .get_parent = ext4_get_parent,
1210 };
1211
1212 enum {
1213         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1214         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1215         Opt_nouid32, Opt_debug, Opt_removed,
1216         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1217         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1218         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1219         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1220         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1221         Opt_data_err_abort, Opt_data_err_ignore,
1222         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1223         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1224         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1225         Opt_usrquota, Opt_grpquota, Opt_i_version,
1226         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1227         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1228         Opt_inode_readahead_blks, Opt_journal_ioprio,
1229         Opt_dioread_nolock, Opt_dioread_lock,
1230         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1231         Opt_max_dir_size_kb,
1232 };
1233
1234 static const match_table_t tokens = {
1235         {Opt_bsd_df, "bsddf"},
1236         {Opt_minix_df, "minixdf"},
1237         {Opt_grpid, "grpid"},
1238         {Opt_grpid, "bsdgroups"},
1239         {Opt_nogrpid, "nogrpid"},
1240         {Opt_nogrpid, "sysvgroups"},
1241         {Opt_resgid, "resgid=%u"},
1242         {Opt_resuid, "resuid=%u"},
1243         {Opt_sb, "sb=%u"},
1244         {Opt_err_cont, "errors=continue"},
1245         {Opt_err_panic, "errors=panic"},
1246         {Opt_err_ro, "errors=remount-ro"},
1247         {Opt_nouid32, "nouid32"},
1248         {Opt_debug, "debug"},
1249         {Opt_removed, "oldalloc"},
1250         {Opt_removed, "orlov"},
1251         {Opt_user_xattr, "user_xattr"},
1252         {Opt_nouser_xattr, "nouser_xattr"},
1253         {Opt_acl, "acl"},
1254         {Opt_noacl, "noacl"},
1255         {Opt_noload, "norecovery"},
1256         {Opt_noload, "noload"},
1257         {Opt_removed, "nobh"},
1258         {Opt_removed, "bh"},
1259         {Opt_commit, "commit=%u"},
1260         {Opt_min_batch_time, "min_batch_time=%u"},
1261         {Opt_max_batch_time, "max_batch_time=%u"},
1262         {Opt_journal_dev, "journal_dev=%u"},
1263         {Opt_journal_checksum, "journal_checksum"},
1264         {Opt_journal_async_commit, "journal_async_commit"},
1265         {Opt_abort, "abort"},
1266         {Opt_data_journal, "data=journal"},
1267         {Opt_data_ordered, "data=ordered"},
1268         {Opt_data_writeback, "data=writeback"},
1269         {Opt_data_err_abort, "data_err=abort"},
1270         {Opt_data_err_ignore, "data_err=ignore"},
1271         {Opt_offusrjquota, "usrjquota="},
1272         {Opt_usrjquota, "usrjquota=%s"},
1273         {Opt_offgrpjquota, "grpjquota="},
1274         {Opt_grpjquota, "grpjquota=%s"},
1275         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1276         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1277         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1278         {Opt_grpquota, "grpquota"},
1279         {Opt_noquota, "noquota"},
1280         {Opt_quota, "quota"},
1281         {Opt_usrquota, "usrquota"},
1282         {Opt_barrier, "barrier=%u"},
1283         {Opt_barrier, "barrier"},
1284         {Opt_nobarrier, "nobarrier"},
1285         {Opt_i_version, "i_version"},
1286         {Opt_stripe, "stripe=%u"},
1287         {Opt_delalloc, "delalloc"},
1288         {Opt_nodelalloc, "nodelalloc"},
1289         {Opt_mblk_io_submit, "mblk_io_submit"},
1290         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1291         {Opt_block_validity, "block_validity"},
1292         {Opt_noblock_validity, "noblock_validity"},
1293         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1294         {Opt_journal_ioprio, "journal_ioprio=%u"},
1295         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1296         {Opt_auto_da_alloc, "auto_da_alloc"},
1297         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1298         {Opt_dioread_nolock, "dioread_nolock"},
1299         {Opt_dioread_lock, "dioread_lock"},
1300         {Opt_discard, "discard"},
1301         {Opt_nodiscard, "nodiscard"},
1302         {Opt_init_itable, "init_itable=%u"},
1303         {Opt_init_itable, "init_itable"},
1304         {Opt_noinit_itable, "noinit_itable"},
1305         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1306         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1307         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1308         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1309         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1310         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1311         {Opt_err, NULL},
1312 };
1313
1314 static ext4_fsblk_t get_sb_block(void **data)
1315 {
1316         ext4_fsblk_t    sb_block;
1317         char            *options = (char *) *data;
1318
1319         if (!options || strncmp(options, "sb=", 3) != 0)
1320                 return 1;       /* Default location */
1321
1322         options += 3;
1323         /* TODO: use simple_strtoll with >32bit ext4 */
1324         sb_block = simple_strtoul(options, &options, 0);
1325         if (*options && *options != ',') {
1326                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1327                        (char *) *data);
1328                 return 1;
1329         }
1330         if (*options == ',')
1331                 options++;
1332         *data = (void *) options;
1333
1334         return sb_block;
1335 }
1336
1337 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1338 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1339         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1340
1341 #ifdef CONFIG_QUOTA
1342 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1343 {
1344         struct ext4_sb_info *sbi = EXT4_SB(sb);
1345         char *qname;
1346
1347         if (sb_any_quota_loaded(sb) &&
1348                 !sbi->s_qf_names[qtype]) {
1349                 ext4_msg(sb, KERN_ERR,
1350                         "Cannot change journaled "
1351                         "quota options when quota turned on");
1352                 return -1;
1353         }
1354         qname = match_strdup(args);
1355         if (!qname) {
1356                 ext4_msg(sb, KERN_ERR,
1357                         "Not enough memory for storing quotafile name");
1358                 return -1;
1359         }
1360         if (sbi->s_qf_names[qtype] &&
1361                 strcmp(sbi->s_qf_names[qtype], qname)) {
1362                 ext4_msg(sb, KERN_ERR,
1363                         "%s quota file already specified", QTYPE2NAME(qtype));
1364                 kfree(qname);
1365                 return -1;
1366         }
1367         sbi->s_qf_names[qtype] = qname;
1368         if (strchr(sbi->s_qf_names[qtype], '/')) {
1369                 ext4_msg(sb, KERN_ERR,
1370                         "quotafile must be on filesystem root");
1371                 kfree(sbi->s_qf_names[qtype]);
1372                 sbi->s_qf_names[qtype] = NULL;
1373                 return -1;
1374         }
1375         set_opt(sb, QUOTA);
1376         return 1;
1377 }
1378
1379 static int clear_qf_name(struct super_block *sb, int qtype)
1380 {
1381
1382         struct ext4_sb_info *sbi = EXT4_SB(sb);
1383
1384         if (sb_any_quota_loaded(sb) &&
1385                 sbi->s_qf_names[qtype]) {
1386                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1387                         " when quota turned on");
1388                 return -1;
1389         }
1390         /*
1391          * The space will be released later when all options are confirmed
1392          * to be correct
1393          */
1394         sbi->s_qf_names[qtype] = NULL;
1395         return 1;
1396 }
1397 #endif
1398
1399 #define MOPT_SET        0x0001
1400 #define MOPT_CLEAR      0x0002
1401 #define MOPT_NOSUPPORT  0x0004
1402 #define MOPT_EXPLICIT   0x0008
1403 #define MOPT_CLEAR_ERR  0x0010
1404 #define MOPT_GTE0       0x0020
1405 #ifdef CONFIG_QUOTA
1406 #define MOPT_Q          0
1407 #define MOPT_QFMT       0x0040
1408 #else
1409 #define MOPT_Q          MOPT_NOSUPPORT
1410 #define MOPT_QFMT       MOPT_NOSUPPORT
1411 #endif
1412 #define MOPT_DATAJ      0x0080
1413
1414 static const struct mount_opts {
1415         int     token;
1416         int     mount_opt;
1417         int     flags;
1418 } ext4_mount_opts[] = {
1419         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1420         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1421         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1422         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1423         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1424         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1425         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1426         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1427         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1428         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1429         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1430         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1431         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1432         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1433         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1434         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1435                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1436         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1437         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1438         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1439         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1440         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1441         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1442         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1443         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1444         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1445         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1446         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1447         {Opt_commit, 0, MOPT_GTE0},
1448         {Opt_max_batch_time, 0, MOPT_GTE0},
1449         {Opt_min_batch_time, 0, MOPT_GTE0},
1450         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1451         {Opt_init_itable, 0, MOPT_GTE0},
1452         {Opt_stripe, 0, MOPT_GTE0},
1453         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1454         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1455         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1456 #ifdef CONFIG_EXT4_FS_XATTR
1457         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1458         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1459 #else
1460         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1461         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1462 #endif
1463 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1464         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1465         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1466 #else
1467         {Opt_acl, 0, MOPT_NOSUPPORT},
1468         {Opt_noacl, 0, MOPT_NOSUPPORT},
1469 #endif
1470         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1471         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1472         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1473         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1474                                                         MOPT_SET | MOPT_Q},
1475         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1476                                                         MOPT_SET | MOPT_Q},
1477         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1478                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1479         {Opt_usrjquota, 0, MOPT_Q},
1480         {Opt_grpjquota, 0, MOPT_Q},
1481         {Opt_offusrjquota, 0, MOPT_Q},
1482         {Opt_offgrpjquota, 0, MOPT_Q},
1483         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1484         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1485         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1486         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1487         {Opt_err, 0, 0}
1488 };
1489
1490 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1491                             substring_t *args, unsigned long *journal_devnum,
1492                             unsigned int *journal_ioprio, int is_remount)
1493 {
1494         struct ext4_sb_info *sbi = EXT4_SB(sb);
1495         const struct mount_opts *m;
1496         kuid_t uid;
1497         kgid_t gid;
1498         int arg = 0;
1499
1500 #ifdef CONFIG_QUOTA
1501         if (token == Opt_usrjquota)
1502                 return set_qf_name(sb, USRQUOTA, &args[0]);
1503         else if (token == Opt_grpjquota)
1504                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1505         else if (token == Opt_offusrjquota)
1506                 return clear_qf_name(sb, USRQUOTA);
1507         else if (token == Opt_offgrpjquota)
1508                 return clear_qf_name(sb, GRPQUOTA);
1509 #endif
1510         if (args->from && match_int(args, &arg))
1511                 return -1;
1512         switch (token) {
1513         case Opt_noacl:
1514         case Opt_nouser_xattr:
1515                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1516                 break;
1517         case Opt_sb:
1518                 return 1;       /* handled by get_sb_block() */
1519         case Opt_removed:
1520                 ext4_msg(sb, KERN_WARNING,
1521                          "Ignoring removed %s option", opt);
1522                 return 1;
1523         case Opt_resuid:
1524                 uid = make_kuid(current_user_ns(), arg);
1525                 if (!uid_valid(uid)) {
1526                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1527                         return -1;
1528                 }
1529                 sbi->s_resuid = uid;
1530                 return 1;
1531         case Opt_resgid:
1532                 gid = make_kgid(current_user_ns(), arg);
1533                 if (!gid_valid(gid)) {
1534                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1535                         return -1;
1536                 }
1537                 sbi->s_resgid = gid;
1538                 return 1;
1539         case Opt_abort:
1540                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1541                 return 1;
1542         case Opt_i_version:
1543                 sb->s_flags |= MS_I_VERSION;
1544                 return 1;
1545         case Opt_journal_dev:
1546                 if (is_remount) {
1547                         ext4_msg(sb, KERN_ERR,
1548                                  "Cannot specify journal on remount");
1549                         return -1;
1550                 }
1551                 *journal_devnum = arg;
1552                 return 1;
1553         case Opt_journal_ioprio:
1554                 if (arg < 0 || arg > 7)
1555                         return -1;
1556                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1557                 return 1;
1558         }
1559
1560         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1561                 if (token != m->token)
1562                         continue;
1563                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1564                         return -1;
1565                 if (m->flags & MOPT_EXPLICIT)
1566                         set_opt2(sb, EXPLICIT_DELALLOC);
1567                 if (m->flags & MOPT_CLEAR_ERR)
1568                         clear_opt(sb, ERRORS_MASK);
1569                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1570                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1571                                  "options when quota turned on");
1572                         return -1;
1573                 }
1574
1575                 if (m->flags & MOPT_NOSUPPORT) {
1576                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1577                 } else if (token == Opt_commit) {
1578                         if (arg == 0)
1579                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1580                         sbi->s_commit_interval = HZ * arg;
1581                 } else if (token == Opt_max_batch_time) {
1582                         if (arg == 0)
1583                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1584                         sbi->s_max_batch_time = arg;
1585                 } else if (token == Opt_min_batch_time) {
1586                         sbi->s_min_batch_time = arg;
1587                 } else if (token == Opt_inode_readahead_blks) {
1588                         if (arg > (1 << 30))
1589                                 return -1;
1590                         if (arg && !is_power_of_2(arg)) {
1591                                 ext4_msg(sb, KERN_ERR,
1592                                          "EXT4-fs: inode_readahead_blks"
1593                                          " must be a power of 2");
1594                                 return -1;
1595                         }
1596                         sbi->s_inode_readahead_blks = arg;
1597                 } else if (token == Opt_init_itable) {
1598                         set_opt(sb, INIT_INODE_TABLE);
1599                         if (!args->from)
1600                                 arg = EXT4_DEF_LI_WAIT_MULT;
1601                         sbi->s_li_wait_mult = arg;
1602                 } else if (token == Opt_max_dir_size_kb) {
1603                         sbi->s_max_dir_size_kb = arg;
1604                 } else if (token == Opt_stripe) {
1605                         sbi->s_stripe = arg;
1606                 } else if (m->flags & MOPT_DATAJ) {
1607                         if (is_remount) {
1608                                 if (!sbi->s_journal)
1609                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1610                                 else if (test_opt(sb, DATA_FLAGS) !=
1611                                          m->mount_opt) {
1612                                         ext4_msg(sb, KERN_ERR,
1613                                          "Cannot change data mode on remount");
1614                                         return -1;
1615                                 }
1616                         } else {
1617                                 clear_opt(sb, DATA_FLAGS);
1618                                 sbi->s_mount_opt |= m->mount_opt;
1619                         }
1620 #ifdef CONFIG_QUOTA
1621                 } else if (m->flags & MOPT_QFMT) {
1622                         if (sb_any_quota_loaded(sb) &&
1623                             sbi->s_jquota_fmt != m->mount_opt) {
1624                                 ext4_msg(sb, KERN_ERR, "Cannot "
1625                                          "change journaled quota options "
1626                                          "when quota turned on");
1627                                 return -1;
1628                         }
1629                         sbi->s_jquota_fmt = m->mount_opt;
1630 #endif
1631                 } else {
1632                         if (!args->from)
1633                                 arg = 1;
1634                         if (m->flags & MOPT_CLEAR)
1635                                 arg = !arg;
1636                         else if (unlikely(!(m->flags & MOPT_SET))) {
1637                                 ext4_msg(sb, KERN_WARNING,
1638                                          "buggy handling of option %s", opt);
1639                                 WARN_ON(1);
1640                                 return -1;
1641                         }
1642                         if (arg != 0)
1643                                 sbi->s_mount_opt |= m->mount_opt;
1644                         else
1645                                 sbi->s_mount_opt &= ~m->mount_opt;
1646                 }
1647                 return 1;
1648         }
1649         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1650                  "or missing value", opt);
1651         return -1;
1652 }
1653
1654 static int parse_options(char *options, struct super_block *sb,
1655                          unsigned long *journal_devnum,
1656                          unsigned int *journal_ioprio,
1657                          int is_remount)
1658 {
1659 #ifdef CONFIG_QUOTA
1660         struct ext4_sb_info *sbi = EXT4_SB(sb);
1661 #endif
1662         char *p;
1663         substring_t args[MAX_OPT_ARGS];
1664         int token;
1665
1666         if (!options)
1667                 return 1;
1668
1669         while ((p = strsep(&options, ",")) != NULL) {
1670                 if (!*p)
1671                         continue;
1672                 /*
1673                  * Initialize args struct so we know whether arg was
1674                  * found; some options take optional arguments.
1675                  */
1676                 args[0].to = args[0].from = NULL;
1677                 token = match_token(p, tokens, args);
1678                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1679                                      journal_ioprio, is_remount) < 0)
1680                         return 0;
1681         }
1682 #ifdef CONFIG_QUOTA
1683         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1684                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1685                         clear_opt(sb, USRQUOTA);
1686
1687                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1688                         clear_opt(sb, GRPQUOTA);
1689
1690                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1691                         ext4_msg(sb, KERN_ERR, "old and new quota "
1692                                         "format mixing");
1693                         return 0;
1694                 }
1695
1696                 if (!sbi->s_jquota_fmt) {
1697                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1698                                         "not specified");
1699                         return 0;
1700                 }
1701         } else {
1702                 if (sbi->s_jquota_fmt) {
1703                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1704                                         "specified with no journaling "
1705                                         "enabled");
1706                         return 0;
1707                 }
1708         }
1709 #endif
1710         return 1;
1711 }
1712
1713 static inline void ext4_show_quota_options(struct seq_file *seq,
1714                                            struct super_block *sb)
1715 {
1716 #if defined(CONFIG_QUOTA)
1717         struct ext4_sb_info *sbi = EXT4_SB(sb);
1718
1719         if (sbi->s_jquota_fmt) {
1720                 char *fmtname = "";
1721
1722                 switch (sbi->s_jquota_fmt) {
1723                 case QFMT_VFS_OLD:
1724                         fmtname = "vfsold";
1725                         break;
1726                 case QFMT_VFS_V0:
1727                         fmtname = "vfsv0";
1728                         break;
1729                 case QFMT_VFS_V1:
1730                         fmtname = "vfsv1";
1731                         break;
1732                 }
1733                 seq_printf(seq, ",jqfmt=%s", fmtname);
1734         }
1735
1736         if (sbi->s_qf_names[USRQUOTA])
1737                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1738
1739         if (sbi->s_qf_names[GRPQUOTA])
1740                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1741
1742         if (test_opt(sb, USRQUOTA))
1743                 seq_puts(seq, ",usrquota");
1744
1745         if (test_opt(sb, GRPQUOTA))
1746                 seq_puts(seq, ",grpquota");
1747 #endif
1748 }
1749
1750 static const char *token2str(int token)
1751 {
1752         static const struct match_token *t;
1753
1754         for (t = tokens; t->token != Opt_err; t++)
1755                 if (t->token == token && !strchr(t->pattern, '='))
1756                         break;
1757         return t->pattern;
1758 }
1759
1760 /*
1761  * Show an option if
1762  *  - it's set to a non-default value OR
1763  *  - if the per-sb default is different from the global default
1764  */
1765 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1766                               int nodefs)
1767 {
1768         struct ext4_sb_info *sbi = EXT4_SB(sb);
1769         struct ext4_super_block *es = sbi->s_es;
1770         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1771         const struct mount_opts *m;
1772         char sep = nodefs ? '\n' : ',';
1773
1774 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1775 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1776
1777         if (sbi->s_sb_block != 1)
1778                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1779
1780         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1781                 int want_set = m->flags & MOPT_SET;
1782                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1783                     (m->flags & MOPT_CLEAR_ERR))
1784                         continue;
1785                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1786                         continue; /* skip if same as the default */
1787                 if ((want_set &&
1788                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1789                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1790                         continue; /* select Opt_noFoo vs Opt_Foo */
1791                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1792         }
1793
1794         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1795             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1796                 SEQ_OPTS_PRINT("resuid=%u",
1797                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1798         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1799             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1800                 SEQ_OPTS_PRINT("resgid=%u",
1801                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1802         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1803         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1804                 SEQ_OPTS_PUTS("errors=remount-ro");
1805         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1806                 SEQ_OPTS_PUTS("errors=continue");
1807         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1808                 SEQ_OPTS_PUTS("errors=panic");
1809         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1810                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1811         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1812                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1813         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1814                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1815         if (sb->s_flags & MS_I_VERSION)
1816                 SEQ_OPTS_PUTS("i_version");
1817         if (nodefs || sbi->s_stripe)
1818                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1819         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1820                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1821                         SEQ_OPTS_PUTS("data=journal");
1822                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1823                         SEQ_OPTS_PUTS("data=ordered");
1824                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1825                         SEQ_OPTS_PUTS("data=writeback");
1826         }
1827         if (nodefs ||
1828             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1829                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1830                                sbi->s_inode_readahead_blks);
1831
1832         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1833                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1834                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1835         if (nodefs || sbi->s_max_dir_size_kb)
1836                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1837
1838         ext4_show_quota_options(seq, sb);
1839         return 0;
1840 }
1841
1842 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1843 {
1844         return _ext4_show_options(seq, root->d_sb, 0);
1845 }
1846
1847 static int options_seq_show(struct seq_file *seq, void *offset)
1848 {
1849         struct super_block *sb = seq->private;
1850         int rc;
1851
1852         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1853         rc = _ext4_show_options(seq, sb, 1);
1854         seq_puts(seq, "\n");
1855         return rc;
1856 }
1857
1858 static int options_open_fs(struct inode *inode, struct file *file)
1859 {
1860         return single_open(file, options_seq_show, PDE(inode)->data);
1861 }
1862
1863 static const struct file_operations ext4_seq_options_fops = {
1864         .owner = THIS_MODULE,
1865         .open = options_open_fs,
1866         .read = seq_read,
1867         .llseek = seq_lseek,
1868         .release = single_release,
1869 };
1870
1871 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1872                             int read_only)
1873 {
1874         struct ext4_sb_info *sbi = EXT4_SB(sb);
1875         int res = 0;
1876
1877         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1878                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1879                          "forcing read-only mode");
1880                 res = MS_RDONLY;
1881         }
1882         if (read_only)
1883                 goto done;
1884         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1885                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1886                          "running e2fsck is recommended");
1887         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1888                 ext4_msg(sb, KERN_WARNING,
1889                          "warning: mounting fs with errors, "
1890                          "running e2fsck is recommended");
1891         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1892                  le16_to_cpu(es->s_mnt_count) >=
1893                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1894                 ext4_msg(sb, KERN_WARNING,
1895                          "warning: maximal mount count reached, "
1896                          "running e2fsck is recommended");
1897         else if (le32_to_cpu(es->s_checkinterval) &&
1898                 (le32_to_cpu(es->s_lastcheck) +
1899                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1900                 ext4_msg(sb, KERN_WARNING,
1901                          "warning: checktime reached, "
1902                          "running e2fsck is recommended");
1903         if (!sbi->s_journal)
1904                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1905         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1906                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1907         le16_add_cpu(&es->s_mnt_count, 1);
1908         es->s_mtime = cpu_to_le32(get_seconds());
1909         ext4_update_dynamic_rev(sb);
1910         if (sbi->s_journal)
1911                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1912
1913         ext4_commit_super(sb, 1);
1914 done:
1915         if (test_opt(sb, DEBUG))
1916                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1917                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1918                         sb->s_blocksize,
1919                         sbi->s_groups_count,
1920                         EXT4_BLOCKS_PER_GROUP(sb),
1921                         EXT4_INODES_PER_GROUP(sb),
1922                         sbi->s_mount_opt, sbi->s_mount_opt2);
1923
1924         cleancache_init_fs(sb);
1925         return res;
1926 }
1927
1928 static int ext4_fill_flex_info(struct super_block *sb)
1929 {
1930         struct ext4_sb_info *sbi = EXT4_SB(sb);
1931         struct ext4_group_desc *gdp = NULL;
1932         ext4_group_t flex_group_count;
1933         ext4_group_t flex_group;
1934         unsigned int groups_per_flex = 0;
1935         size_t size;
1936         int i;
1937
1938         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1939         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1940                 sbi->s_log_groups_per_flex = 0;
1941                 return 1;
1942         }
1943         groups_per_flex = 1 << sbi->s_log_groups_per_flex;
1944
1945         /* We allocate both existing and potentially added groups */
1946         flex_group_count = ((sbi->s_groups_count + groups_per_flex - 1) +
1947                         ((le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) + 1) <<
1948                               EXT4_DESC_PER_BLOCK_BITS(sb))) / groups_per_flex;
1949         size = flex_group_count * sizeof(struct flex_groups);
1950         sbi->s_flex_groups = ext4_kvzalloc(size, GFP_KERNEL);
1951         if (sbi->s_flex_groups == NULL) {
1952                 ext4_msg(sb, KERN_ERR, "not enough memory for %u flex groups",
1953                          flex_group_count);
1954                 goto failed;
1955         }
1956
1957         for (i = 0; i < sbi->s_groups_count; i++) {
1958                 gdp = ext4_get_group_desc(sb, i, NULL);
1959
1960                 flex_group = ext4_flex_group(sbi, i);
1961                 atomic_add(ext4_free_inodes_count(sb, gdp),
1962                            &sbi->s_flex_groups[flex_group].free_inodes);
1963                 atomic_add(ext4_free_group_clusters(sb, gdp),
1964                            &sbi->s_flex_groups[flex_group].free_clusters);
1965                 atomic_add(ext4_used_dirs_count(sb, gdp),
1966                            &sbi->s_flex_groups[flex_group].used_dirs);
1967         }
1968
1969         return 1;
1970 failed:
1971         return 0;
1972 }
1973
1974 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1975                                    struct ext4_group_desc *gdp)
1976 {
1977         int offset;
1978         __u16 crc = 0;
1979         __le32 le_group = cpu_to_le32(block_group);
1980
1981         if ((sbi->s_es->s_feature_ro_compat &
1982              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1983                 /* Use new metadata_csum algorithm */
1984                 __u16 old_csum;
1985                 __u32 csum32;
1986
1987                 old_csum = gdp->bg_checksum;
1988                 gdp->bg_checksum = 0;
1989                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1990                                      sizeof(le_group));
1991                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1992                                      sbi->s_desc_size);
1993                 gdp->bg_checksum = old_csum;
1994
1995                 crc = csum32 & 0xFFFF;
1996                 goto out;
1997         }
1998
1999         /* old crc16 code */
2000         offset = offsetof(struct ext4_group_desc, bg_checksum);
2001
2002         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2003         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2004         crc = crc16(crc, (__u8 *)gdp, offset);
2005         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2006         /* for checksum of struct ext4_group_desc do the rest...*/
2007         if ((sbi->s_es->s_feature_incompat &
2008              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2009             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2010                 crc = crc16(crc, (__u8 *)gdp + offset,
2011                             le16_to_cpu(sbi->s_es->s_desc_size) -
2012                                 offset);
2013
2014 out:
2015         return cpu_to_le16(crc);
2016 }
2017
2018 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2019                                 struct ext4_group_desc *gdp)
2020 {
2021         if (ext4_has_group_desc_csum(sb) &&
2022             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2023                                                       block_group, gdp)))
2024                 return 0;
2025
2026         return 1;
2027 }
2028
2029 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2030                               struct ext4_group_desc *gdp)
2031 {
2032         if (!ext4_has_group_desc_csum(sb))
2033                 return;
2034         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2035 }
2036
2037 /* Called at mount-time, super-block is locked */
2038 static int ext4_check_descriptors(struct super_block *sb,
2039                                   ext4_group_t *first_not_zeroed)
2040 {
2041         struct ext4_sb_info *sbi = EXT4_SB(sb);
2042         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2043         ext4_fsblk_t last_block;
2044         ext4_fsblk_t block_bitmap;
2045         ext4_fsblk_t inode_bitmap;
2046         ext4_fsblk_t inode_table;
2047         int flexbg_flag = 0;
2048         ext4_group_t i, grp = sbi->s_groups_count;
2049
2050         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2051                 flexbg_flag = 1;
2052
2053         ext4_debug("Checking group descriptors");
2054
2055         for (i = 0; i < sbi->s_groups_count; i++) {
2056                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2057
2058                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2059                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2060                 else
2061                         last_block = first_block +
2062                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2063
2064                 if ((grp == sbi->s_groups_count) &&
2065                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2066                         grp = i;
2067
2068                 block_bitmap = ext4_block_bitmap(sb, gdp);
2069                 if (block_bitmap < first_block || block_bitmap > last_block) {
2070                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2071                                "Block bitmap for group %u not in group "
2072                                "(block %llu)!", i, block_bitmap);
2073                         return 0;
2074                 }
2075                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2076                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2077                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2078                                "Inode bitmap for group %u not in group "
2079                                "(block %llu)!", i, inode_bitmap);
2080                         return 0;
2081                 }
2082                 inode_table = ext4_inode_table(sb, gdp);
2083                 if (inode_table < first_block ||
2084                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2085                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2086                                "Inode table for group %u not in group "
2087                                "(block %llu)!", i, inode_table);
2088                         return 0;
2089                 }
2090                 ext4_lock_group(sb, i);
2091                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2092                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2093                                  "Checksum for group %u failed (%u!=%u)",
2094                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2095                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2096                         if (!(sb->s_flags & MS_RDONLY)) {
2097                                 ext4_unlock_group(sb, i);
2098                                 return 0;
2099                         }
2100                 }
2101                 ext4_unlock_group(sb, i);
2102                 if (!flexbg_flag)
2103                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2104         }
2105         if (NULL != first_not_zeroed)
2106                 *first_not_zeroed = grp;
2107
2108         ext4_free_blocks_count_set(sbi->s_es,
2109                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2110         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2111         return 1;
2112 }
2113
2114 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2115  * the superblock) which were deleted from all directories, but held open by
2116  * a process at the time of a crash.  We walk the list and try to delete these
2117  * inodes at recovery time (only with a read-write filesystem).
2118  *
2119  * In order to keep the orphan inode chain consistent during traversal (in
2120  * case of crash during recovery), we link each inode into the superblock
2121  * orphan list_head and handle it the same way as an inode deletion during
2122  * normal operation (which journals the operations for us).
2123  *
2124  * We only do an iget() and an iput() on each inode, which is very safe if we
2125  * accidentally point at an in-use or already deleted inode.  The worst that
2126  * can happen in this case is that we get a "bit already cleared" message from
2127  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2128  * e2fsck was run on this filesystem, and it must have already done the orphan
2129  * inode cleanup for us, so we can safely abort without any further action.
2130  */
2131 static void ext4_orphan_cleanup(struct super_block *sb,
2132                                 struct ext4_super_block *es)
2133 {
2134         unsigned int s_flags = sb->s_flags;
2135         int nr_orphans = 0, nr_truncates = 0;
2136 #ifdef CONFIG_QUOTA
2137         int i;
2138 #endif
2139         if (!es->s_last_orphan) {
2140                 jbd_debug(4, "no orphan inodes to clean up\n");
2141                 return;
2142         }
2143
2144         if (bdev_read_only(sb->s_bdev)) {
2145                 ext4_msg(sb, KERN_ERR, "write access "
2146                         "unavailable, skipping orphan cleanup");
2147                 return;
2148         }
2149
2150         /* Check if feature set would not allow a r/w mount */
2151         if (!ext4_feature_set_ok(sb, 0)) {
2152                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2153                          "unknown ROCOMPAT features");
2154                 return;
2155         }
2156
2157         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2158                 if (es->s_last_orphan)
2159                         jbd_debug(1, "Errors on filesystem, "
2160                                   "clearing orphan list.\n");
2161                 es->s_last_orphan = 0;
2162                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2163                 return;
2164         }
2165
2166         if (s_flags & MS_RDONLY) {
2167                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2168                 sb->s_flags &= ~MS_RDONLY;
2169         }
2170 #ifdef CONFIG_QUOTA
2171         /* Needed for iput() to work correctly and not trash data */
2172         sb->s_flags |= MS_ACTIVE;
2173         /* Turn on quotas so that they are updated correctly */
2174         for (i = 0; i < MAXQUOTAS; i++) {
2175                 if (EXT4_SB(sb)->s_qf_names[i]) {
2176                         int ret = ext4_quota_on_mount(sb, i);
2177                         if (ret < 0)
2178                                 ext4_msg(sb, KERN_ERR,
2179                                         "Cannot turn on journaled "
2180                                         "quota: error %d", ret);
2181                 }
2182         }
2183 #endif
2184
2185         while (es->s_last_orphan) {
2186                 struct inode *inode;
2187
2188                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2189                 if (IS_ERR(inode)) {
2190                         es->s_last_orphan = 0;
2191                         break;
2192                 }
2193
2194                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2195                 dquot_initialize(inode);
2196                 if (inode->i_nlink) {
2197                         ext4_msg(sb, KERN_DEBUG,
2198                                 "%s: truncating inode %lu to %lld bytes",
2199                                 __func__, inode->i_ino, inode->i_size);
2200                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2201                                   inode->i_ino, inode->i_size);
2202                         ext4_truncate(inode);
2203                         nr_truncates++;
2204                 } else {
2205                         ext4_msg(sb, KERN_DEBUG,
2206                                 "%s: deleting unreferenced inode %lu",
2207                                 __func__, inode->i_ino);
2208                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2209                                   inode->i_ino);
2210                         nr_orphans++;
2211                 }
2212                 iput(inode);  /* The delete magic happens here! */
2213         }
2214
2215 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2216
2217         if (nr_orphans)
2218                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2219                        PLURAL(nr_orphans));
2220         if (nr_truncates)
2221                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2222                        PLURAL(nr_truncates));
2223 #ifdef CONFIG_QUOTA
2224         /* Turn quotas off */
2225         for (i = 0; i < MAXQUOTAS; i++) {
2226                 if (sb_dqopt(sb)->files[i])
2227                         dquot_quota_off(sb, i);
2228         }
2229 #endif
2230         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2231 }
2232
2233 /*
2234  * Maximal extent format file size.
2235  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2236  * extent format containers, within a sector_t, and within i_blocks
2237  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2238  * so that won't be a limiting factor.
2239  *
2240  * However there is other limiting factor. We do store extents in the form
2241  * of starting block and length, hence the resulting length of the extent
2242  * covering maximum file size must fit into on-disk format containers as
2243  * well. Given that length is always by 1 unit bigger than max unit (because
2244  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2245  *
2246  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2247  */
2248 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2249 {
2250         loff_t res;
2251         loff_t upper_limit = MAX_LFS_FILESIZE;
2252
2253         /* small i_blocks in vfs inode? */
2254         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2255                 /*
2256                  * CONFIG_LBDAF is not enabled implies the inode
2257                  * i_block represent total blocks in 512 bytes
2258                  * 32 == size of vfs inode i_blocks * 8
2259                  */
2260                 upper_limit = (1LL << 32) - 1;
2261
2262                 /* total blocks in file system block size */
2263                 upper_limit >>= (blkbits - 9);
2264                 upper_limit <<= blkbits;
2265         }
2266
2267         /*
2268          * 32-bit extent-start container, ee_block. We lower the maxbytes
2269          * by one fs block, so ee_len can cover the extent of maximum file
2270          * size
2271          */
2272         res = (1LL << 32) - 1;
2273         res <<= blkbits;
2274
2275         /* Sanity check against vm- & vfs- imposed limits */
2276         if (res > upper_limit)
2277                 res = upper_limit;
2278
2279         return res;
2280 }
2281
2282 /*
2283  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2284  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2285  * We need to be 1 filesystem block less than the 2^48 sector limit.
2286  */
2287 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2288 {
2289         loff_t res = EXT4_NDIR_BLOCKS;
2290         int meta_blocks;
2291         loff_t upper_limit;
2292         /* This is calculated to be the largest file size for a dense, block
2293          * mapped file such that the file's total number of 512-byte sectors,
2294          * including data and all indirect blocks, does not exceed (2^48 - 1).
2295          *
2296          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2297          * number of 512-byte sectors of the file.
2298          */
2299
2300         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2301                 /*
2302                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2303                  * the inode i_block field represents total file blocks in
2304                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2305                  */
2306                 upper_limit = (1LL << 32) - 1;
2307
2308                 /* total blocks in file system block size */
2309                 upper_limit >>= (bits - 9);
2310
2311         } else {
2312                 /*
2313                  * We use 48 bit ext4_inode i_blocks
2314                  * With EXT4_HUGE_FILE_FL set the i_blocks
2315                  * represent total number of blocks in
2316                  * file system block size
2317                  */
2318                 upper_limit = (1LL << 48) - 1;
2319
2320         }
2321
2322         /* indirect blocks */
2323         meta_blocks = 1;
2324         /* double indirect blocks */
2325         meta_blocks += 1 + (1LL << (bits-2));
2326         /* tripple indirect blocks */
2327         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2328
2329         upper_limit -= meta_blocks;
2330         upper_limit <<= bits;
2331
2332         res += 1LL << (bits-2);
2333         res += 1LL << (2*(bits-2));
2334         res += 1LL << (3*(bits-2));
2335         res <<= bits;
2336         if (res > upper_limit)
2337                 res = upper_limit;
2338
2339         if (res > MAX_LFS_FILESIZE)
2340                 res = MAX_LFS_FILESIZE;
2341
2342         return res;
2343 }
2344
2345 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2346                                    ext4_fsblk_t logical_sb_block, int nr)
2347 {
2348         struct ext4_sb_info *sbi = EXT4_SB(sb);
2349         ext4_group_t bg, first_meta_bg;
2350         int has_super = 0;
2351
2352         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2353
2354         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2355             nr < first_meta_bg)
2356                 return logical_sb_block + nr + 1;
2357         bg = sbi->s_desc_per_block * nr;
2358         if (ext4_bg_has_super(sb, bg))
2359                 has_super = 1;
2360
2361         return (has_super + ext4_group_first_block_no(sb, bg));
2362 }
2363
2364 /**
2365  * ext4_get_stripe_size: Get the stripe size.
2366  * @sbi: In memory super block info
2367  *
2368  * If we have specified it via mount option, then
2369  * use the mount option value. If the value specified at mount time is
2370  * greater than the blocks per group use the super block value.
2371  * If the super block value is greater than blocks per group return 0.
2372  * Allocator needs it be less than blocks per group.
2373  *
2374  */
2375 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2376 {
2377         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2378         unsigned long stripe_width =
2379                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2380         int ret;
2381
2382         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2383                 ret = sbi->s_stripe;
2384         else if (stripe_width <= sbi->s_blocks_per_group)
2385                 ret = stripe_width;
2386         else if (stride <= sbi->s_blocks_per_group)
2387                 ret = stride;
2388         else
2389                 ret = 0;
2390
2391         /*
2392          * If the stripe width is 1, this makes no sense and
2393          * we set it to 0 to turn off stripe handling code.
2394          */
2395         if (ret <= 1)
2396                 ret = 0;
2397
2398         return ret;
2399 }
2400
2401 /* sysfs supprt */
2402
2403 struct ext4_attr {
2404         struct attribute attr;
2405         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2406         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2407                          const char *, size_t);
2408         int offset;
2409 };
2410
2411 static int parse_strtoul(const char *buf,
2412                 unsigned long max, unsigned long *value)
2413 {
2414         char *endp;
2415
2416         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2417         endp = skip_spaces(endp);
2418         if (*endp || *value > max)
2419                 return -EINVAL;
2420
2421         return 0;
2422 }
2423
2424 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2425                                               struct ext4_sb_info *sbi,
2426                                               char *buf)
2427 {
2428         return snprintf(buf, PAGE_SIZE, "%llu\n",
2429                 (s64) EXT4_C2B(sbi,
2430                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2431 }
2432
2433 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2434                                          struct ext4_sb_info *sbi, char *buf)
2435 {
2436         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2437
2438         if (!sb->s_bdev->bd_part)
2439                 return snprintf(buf, PAGE_SIZE, "0\n");
2440         return snprintf(buf, PAGE_SIZE, "%lu\n",
2441                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2442                          sbi->s_sectors_written_start) >> 1);
2443 }
2444
2445 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2446                                           struct ext4_sb_info *sbi, char *buf)
2447 {
2448         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2449
2450         if (!sb->s_bdev->bd_part)
2451                 return snprintf(buf, PAGE_SIZE, "0\n");
2452         return snprintf(buf, PAGE_SIZE, "%llu\n",
2453                         (unsigned long long)(sbi->s_kbytes_written +
2454                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2455                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2456 }
2457
2458 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2459                                           struct ext4_sb_info *sbi,
2460                                           const char *buf, size_t count)
2461 {
2462         unsigned long t;
2463
2464         if (parse_strtoul(buf, 0x40000000, &t))
2465                 return -EINVAL;
2466
2467         if (t && !is_power_of_2(t))
2468                 return -EINVAL;
2469
2470         sbi->s_inode_readahead_blks = t;
2471         return count;
2472 }
2473
2474 static ssize_t sbi_ui_show(struct ext4_attr *a,
2475                            struct ext4_sb_info *sbi, char *buf)
2476 {
2477         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2478
2479         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2480 }
2481
2482 static ssize_t sbi_ui_store(struct ext4_attr *a,
2483                             struct ext4_sb_info *sbi,
2484                             const char *buf, size_t count)
2485 {
2486         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2487         unsigned long t;
2488
2489         if (parse_strtoul(buf, 0xffffffff, &t))
2490                 return -EINVAL;
2491         *ui = t;
2492         return count;
2493 }
2494
2495 static ssize_t trigger_test_error(struct ext4_attr *a,
2496                                   struct ext4_sb_info *sbi,
2497                                   const char *buf, size_t count)
2498 {
2499         int len = count;
2500
2501         if (!capable(CAP_SYS_ADMIN))
2502                 return -EPERM;
2503
2504         if (len && buf[len-1] == '\n')
2505                 len--;
2506
2507         if (len)
2508                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2509         return count;
2510 }
2511
2512 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2513 static struct ext4_attr ext4_attr_##_name = {                   \
2514         .attr = {.name = __stringify(_name), .mode = _mode },   \
2515         .show   = _show,                                        \
2516         .store  = _store,                                       \
2517         .offset = offsetof(struct ext4_sb_info, _elname),       \
2518 }
2519 #define EXT4_ATTR(name, mode, show, store) \
2520 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2521
2522 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2523 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2524 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2525 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2526         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2527 #define ATTR_LIST(name) &ext4_attr_##name.attr
2528
2529 EXT4_RO_ATTR(delayed_allocation_blocks);
2530 EXT4_RO_ATTR(session_write_kbytes);
2531 EXT4_RO_ATTR(lifetime_write_kbytes);
2532 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2533                  inode_readahead_blks_store, s_inode_readahead_blks);
2534 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2535 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2536 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2537 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2538 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2539 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2540 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2541 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2542 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2543 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2544
2545 static struct attribute *ext4_attrs[] = {
2546         ATTR_LIST(delayed_allocation_blocks),
2547         ATTR_LIST(session_write_kbytes),
2548         ATTR_LIST(lifetime_write_kbytes),
2549         ATTR_LIST(inode_readahead_blks),
2550         ATTR_LIST(inode_goal),
2551         ATTR_LIST(mb_stats),
2552         ATTR_LIST(mb_max_to_scan),
2553         ATTR_LIST(mb_min_to_scan),
2554         ATTR_LIST(mb_order2_req),
2555         ATTR_LIST(mb_stream_req),
2556         ATTR_LIST(mb_group_prealloc),
2557         ATTR_LIST(max_writeback_mb_bump),
2558         ATTR_LIST(extent_max_zeroout_kb),
2559         ATTR_LIST(trigger_fs_error),
2560         NULL,
2561 };
2562
2563 /* Features this copy of ext4 supports */
2564 EXT4_INFO_ATTR(lazy_itable_init);
2565 EXT4_INFO_ATTR(batched_discard);
2566
2567 static struct attribute *ext4_feat_attrs[] = {
2568         ATTR_LIST(lazy_itable_init),
2569         ATTR_LIST(batched_discard),
2570         NULL,
2571 };
2572
2573 static ssize_t ext4_attr_show(struct kobject *kobj,
2574                               struct attribute *attr, char *buf)
2575 {
2576         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2577                                                 s_kobj);
2578         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2579
2580         return a->show ? a->show(a, sbi, buf) : 0;
2581 }
2582
2583 static ssize_t ext4_attr_store(struct kobject *kobj,
2584                                struct attribute *attr,
2585                                const char *buf, size_t len)
2586 {
2587         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2588                                                 s_kobj);
2589         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2590
2591         return a->store ? a->store(a, sbi, buf, len) : 0;
2592 }
2593
2594 static void ext4_sb_release(struct kobject *kobj)
2595 {
2596         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2597                                                 s_kobj);
2598         complete(&sbi->s_kobj_unregister);
2599 }
2600
2601 static const struct sysfs_ops ext4_attr_ops = {
2602         .show   = ext4_attr_show,
2603         .store  = ext4_attr_store,
2604 };
2605
2606 static struct kobj_type ext4_ktype = {
2607         .default_attrs  = ext4_attrs,
2608         .sysfs_ops      = &ext4_attr_ops,
2609         .release        = ext4_sb_release,
2610 };
2611
2612 static void ext4_feat_release(struct kobject *kobj)
2613 {
2614         complete(&ext4_feat->f_kobj_unregister);
2615 }
2616
2617 static struct kobj_type ext4_feat_ktype = {
2618         .default_attrs  = ext4_feat_attrs,
2619         .sysfs_ops      = &ext4_attr_ops,
2620         .release        = ext4_feat_release,
2621 };
2622
2623 /*
2624  * Check whether this filesystem can be mounted based on
2625  * the features present and the RDONLY/RDWR mount requested.
2626  * Returns 1 if this filesystem can be mounted as requested,
2627  * 0 if it cannot be.
2628  */
2629 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2630 {
2631         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2632                 ext4_msg(sb, KERN_ERR,
2633                         "Couldn't mount because of "
2634                         "unsupported optional features (%x)",
2635                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2636                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2637                 return 0;
2638         }
2639
2640         if (readonly)
2641                 return 1;
2642
2643         /* Check that feature set is OK for a read-write mount */
2644         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2645                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2646                          "unsupported optional features (%x)",
2647                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2648                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2649                 return 0;
2650         }
2651         /*
2652          * Large file size enabled file system can only be mounted
2653          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2654          */
2655         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2656                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2657                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2658                                  "cannot be mounted RDWR without "
2659                                  "CONFIG_LBDAF");
2660                         return 0;
2661                 }
2662         }
2663         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2664             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2665                 ext4_msg(sb, KERN_ERR,
2666                          "Can't support bigalloc feature without "
2667                          "extents feature\n");
2668                 return 0;
2669         }
2670
2671 #ifndef CONFIG_QUOTA
2672         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2673             !readonly) {
2674                 ext4_msg(sb, KERN_ERR,
2675                          "Filesystem with quota feature cannot be mounted RDWR "
2676                          "without CONFIG_QUOTA");
2677                 return 0;
2678         }
2679 #endif  /* CONFIG_QUOTA */
2680         return 1;
2681 }
2682
2683 /*
2684  * This function is called once a day if we have errors logged
2685  * on the file system
2686  */
2687 static void print_daily_error_info(unsigned long arg)
2688 {
2689         struct super_block *sb = (struct super_block *) arg;
2690         struct ext4_sb_info *sbi;
2691         struct ext4_super_block *es;
2692
2693         sbi = EXT4_SB(sb);
2694         es = sbi->s_es;
2695
2696         if (es->s_error_count)
2697                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2698                          le32_to_cpu(es->s_error_count));
2699         if (es->s_first_error_time) {
2700                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2701                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2702                        (int) sizeof(es->s_first_error_func),
2703                        es->s_first_error_func,
2704                        le32_to_cpu(es->s_first_error_line));
2705                 if (es->s_first_error_ino)
2706                         printk(": inode %u",
2707                                le32_to_cpu(es->s_first_error_ino));
2708                 if (es->s_first_error_block)
2709                         printk(": block %llu", (unsigned long long)
2710                                le64_to_cpu(es->s_first_error_block));
2711                 printk("\n");
2712         }
2713         if (es->s_last_error_time) {
2714                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2715                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2716                        (int) sizeof(es->s_last_error_func),
2717                        es->s_last_error_func,
2718                        le32_to_cpu(es->s_last_error_line));
2719                 if (es->s_last_error_ino)
2720                         printk(": inode %u",
2721                                le32_to_cpu(es->s_last_error_ino));
2722                 if (es->s_last_error_block)
2723                         printk(": block %llu", (unsigned long long)
2724                                le64_to_cpu(es->s_last_error_block));
2725                 printk("\n");
2726         }
2727         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2728 }
2729
2730 /* Find next suitable group and run ext4_init_inode_table */
2731 static int ext4_run_li_request(struct ext4_li_request *elr)
2732 {
2733         struct ext4_group_desc *gdp = NULL;
2734         ext4_group_t group, ngroups;
2735         struct super_block *sb;
2736         unsigned long timeout = 0;
2737         int ret = 0;
2738
2739         sb = elr->lr_super;
2740         ngroups = EXT4_SB(sb)->s_groups_count;
2741
2742         sb_start_write(sb);
2743         for (group = elr->lr_next_group; group < ngroups; group++) {
2744                 gdp = ext4_get_group_desc(sb, group, NULL);
2745                 if (!gdp) {
2746                         ret = 1;
2747                         break;
2748                 }
2749
2750                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2751                         break;
2752         }
2753
2754         if (group == ngroups)
2755                 ret = 1;
2756
2757         if (!ret) {
2758                 timeout = jiffies;
2759                 ret = ext4_init_inode_table(sb, group,
2760                                             elr->lr_timeout ? 0 : 1);
2761                 if (elr->lr_timeout == 0) {
2762                         timeout = (jiffies - timeout) *
2763                                   elr->lr_sbi->s_li_wait_mult;
2764                         elr->lr_timeout = timeout;
2765                 }
2766                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2767                 elr->lr_next_group = group + 1;
2768         }
2769         sb_end_write(sb);
2770
2771         return ret;
2772 }
2773
2774 /*
2775  * Remove lr_request from the list_request and free the
2776  * request structure. Should be called with li_list_mtx held
2777  */
2778 static void ext4_remove_li_request(struct ext4_li_request *elr)
2779 {
2780         struct ext4_sb_info *sbi;
2781
2782         if (!elr)
2783                 return;
2784
2785         sbi = elr->lr_sbi;
2786
2787         list_del(&elr->lr_request);
2788         sbi->s_li_request = NULL;
2789         kfree(elr);
2790 }
2791
2792 static void ext4_unregister_li_request(struct super_block *sb)
2793 {
2794         mutex_lock(&ext4_li_mtx);
2795         if (!ext4_li_info) {
2796                 mutex_unlock(&ext4_li_mtx);
2797                 return;
2798         }
2799
2800         mutex_lock(&ext4_li_info->li_list_mtx);
2801         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2802         mutex_unlock(&ext4_li_info->li_list_mtx);
2803         mutex_unlock(&ext4_li_mtx);
2804 }
2805
2806 static struct task_struct *ext4_lazyinit_task;
2807
2808 /*
2809  * This is the function where ext4lazyinit thread lives. It walks
2810  * through the request list searching for next scheduled filesystem.
2811  * When such a fs is found, run the lazy initialization request
2812  * (ext4_rn_li_request) and keep track of the time spend in this
2813  * function. Based on that time we compute next schedule time of
2814  * the request. When walking through the list is complete, compute
2815  * next waking time and put itself into sleep.
2816  */
2817 static int ext4_lazyinit_thread(void *arg)
2818 {
2819         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2820         struct list_head *pos, *n;
2821         struct ext4_li_request *elr;
2822         unsigned long next_wakeup, cur;
2823
2824         BUG_ON(NULL == eli);
2825
2826 cont_thread:
2827         while (true) {
2828                 next_wakeup = MAX_JIFFY_OFFSET;
2829
2830                 mutex_lock(&eli->li_list_mtx);
2831                 if (list_empty(&eli->li_request_list)) {
2832                         mutex_unlock(&eli->li_list_mtx);
2833                         goto exit_thread;
2834                 }
2835
2836                 list_for_each_safe(pos, n, &eli->li_request_list) {
2837                         elr = list_entry(pos, struct ext4_li_request,
2838                                          lr_request);
2839
2840                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2841                                 if (ext4_run_li_request(elr) != 0) {
2842                                         /* error, remove the lazy_init job */
2843                                         ext4_remove_li_request(elr);
2844                                         continue;
2845                                 }
2846                         }
2847
2848                         if (time_before(elr->lr_next_sched, next_wakeup))
2849                                 next_wakeup = elr->lr_next_sched;
2850                 }
2851                 mutex_unlock(&eli->li_list_mtx);
2852
2853                 try_to_freeze();
2854
2855                 cur = jiffies;
2856                 if ((time_after_eq(cur, next_wakeup)) ||
2857                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2858                         cond_resched();
2859                         continue;
2860                 }
2861
2862                 schedule_timeout_interruptible(next_wakeup - cur);
2863
2864                 if (kthread_should_stop()) {
2865                         ext4_clear_request_list();
2866                         goto exit_thread;
2867                 }
2868         }
2869
2870 exit_thread:
2871         /*
2872          * It looks like the request list is empty, but we need
2873          * to check it under the li_list_mtx lock, to prevent any
2874          * additions into it, and of course we should lock ext4_li_mtx
2875          * to atomically free the list and ext4_li_info, because at
2876          * this point another ext4 filesystem could be registering
2877          * new one.
2878          */
2879         mutex_lock(&ext4_li_mtx);
2880         mutex_lock(&eli->li_list_mtx);
2881         if (!list_empty(&eli->li_request_list)) {
2882                 mutex_unlock(&eli->li_list_mtx);
2883                 mutex_unlock(&ext4_li_mtx);
2884                 goto cont_thread;
2885         }
2886         mutex_unlock(&eli->li_list_mtx);
2887         kfree(ext4_li_info);
2888         ext4_li_info = NULL;
2889         mutex_unlock(&ext4_li_mtx);
2890
2891         return 0;
2892 }
2893
2894 static void ext4_clear_request_list(void)
2895 {
2896         struct list_head *pos, *n;
2897         struct ext4_li_request *elr;
2898
2899         mutex_lock(&ext4_li_info->li_list_mtx);
2900         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2901                 elr = list_entry(pos, struct ext4_li_request,
2902                                  lr_request);
2903                 ext4_remove_li_request(elr);
2904         }
2905         mutex_unlock(&ext4_li_info->li_list_mtx);
2906 }
2907
2908 static int ext4_run_lazyinit_thread(void)
2909 {
2910         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2911                                          ext4_li_info, "ext4lazyinit");
2912         if (IS_ERR(ext4_lazyinit_task)) {
2913                 int err = PTR_ERR(ext4_lazyinit_task);
2914                 ext4_clear_request_list();
2915                 kfree(ext4_li_info);
2916                 ext4_li_info = NULL;
2917                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2918                                  "initialization thread\n",
2919                                  err);
2920                 return err;
2921         }
2922         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2923         return 0;
2924 }
2925
2926 /*
2927  * Check whether it make sense to run itable init. thread or not.
2928  * If there is at least one uninitialized inode table, return
2929  * corresponding group number, else the loop goes through all
2930  * groups and return total number of groups.
2931  */
2932 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2933 {
2934         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2935         struct ext4_group_desc *gdp = NULL;
2936
2937         for (group = 0; group < ngroups; group++) {
2938                 gdp = ext4_get_group_desc(sb, group, NULL);
2939                 if (!gdp)
2940                         continue;
2941
2942                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2943                         break;
2944         }
2945
2946         return group;
2947 }
2948
2949 static int ext4_li_info_new(void)
2950 {
2951         struct ext4_lazy_init *eli = NULL;
2952
2953         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2954         if (!eli)
2955                 return -ENOMEM;
2956
2957         INIT_LIST_HEAD(&eli->li_request_list);
2958         mutex_init(&eli->li_list_mtx);
2959
2960         eli->li_state |= EXT4_LAZYINIT_QUIT;
2961
2962         ext4_li_info = eli;
2963
2964         return 0;
2965 }
2966
2967 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2968                                             ext4_group_t start)
2969 {
2970         struct ext4_sb_info *sbi = EXT4_SB(sb);
2971         struct ext4_li_request *elr;
2972         unsigned long rnd;
2973
2974         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2975         if (!elr)
2976                 return NULL;
2977
2978         elr->lr_super = sb;
2979         elr->lr_sbi = sbi;
2980         elr->lr_next_group = start;
2981
2982         /*
2983          * Randomize first schedule time of the request to
2984          * spread the inode table initialization requests
2985          * better.
2986          */
2987         get_random_bytes(&rnd, sizeof(rnd));
2988         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2989                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2990
2991         return elr;
2992 }
2993
2994 static int ext4_register_li_request(struct super_block *sb,
2995                                     ext4_group_t first_not_zeroed)
2996 {
2997         struct ext4_sb_info *sbi = EXT4_SB(sb);
2998         struct ext4_li_request *elr;
2999         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3000         int ret = 0;
3001
3002         if (sbi->s_li_request != NULL) {
3003                 /*
3004                  * Reset timeout so it can be computed again, because
3005                  * s_li_wait_mult might have changed.
3006                  */
3007                 sbi->s_li_request->lr_timeout = 0;
3008                 return 0;
3009         }
3010
3011         if (first_not_zeroed == ngroups ||
3012             (sb->s_flags & MS_RDONLY) ||
3013             !test_opt(sb, INIT_INODE_TABLE))
3014                 return 0;
3015
3016         elr = ext4_li_request_new(sb, first_not_zeroed);
3017         if (!elr)
3018                 return -ENOMEM;
3019
3020         mutex_lock(&ext4_li_mtx);
3021
3022         if (NULL == ext4_li_info) {
3023                 ret = ext4_li_info_new();
3024                 if (ret)
3025                         goto out;
3026         }
3027
3028         mutex_lock(&ext4_li_info->li_list_mtx);
3029         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3030         mutex_unlock(&ext4_li_info->li_list_mtx);
3031
3032         sbi->s_li_request = elr;
3033         /*
3034          * set elr to NULL here since it has been inserted to
3035          * the request_list and the removal and free of it is
3036          * handled by ext4_clear_request_list from now on.
3037          */
3038         elr = NULL;
3039
3040         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3041                 ret = ext4_run_lazyinit_thread();
3042                 if (ret)
3043                         goto out;
3044         }
3045 out:
3046         mutex_unlock(&ext4_li_mtx);
3047         if (ret)
3048                 kfree(elr);
3049         return ret;
3050 }
3051
3052 /*
3053  * We do not need to lock anything since this is called on
3054  * module unload.
3055  */
3056 static void ext4_destroy_lazyinit_thread(void)
3057 {
3058         /*
3059          * If thread exited earlier
3060          * there's nothing to be done.
3061          */
3062         if (!ext4_li_info || !ext4_lazyinit_task)
3063                 return;
3064
3065         kthread_stop(ext4_lazyinit_task);
3066 }
3067
3068 static int set_journal_csum_feature_set(struct super_block *sb)
3069 {
3070         int ret = 1;
3071         int compat, incompat;
3072         struct ext4_sb_info *sbi = EXT4_SB(sb);
3073
3074         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3075                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3076                 /* journal checksum v2 */
3077                 compat = 0;
3078                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3079         } else {
3080                 /* journal checksum v1 */
3081                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3082                 incompat = 0;
3083         }
3084
3085         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3086                 ret = jbd2_journal_set_features(sbi->s_journal,
3087                                 compat, 0,
3088                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3089                                 incompat);
3090         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3091                 ret = jbd2_journal_set_features(sbi->s_journal,
3092                                 compat, 0,
3093                                 incompat);
3094                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3095                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3096         } else {
3097                 jbd2_journal_clear_features(sbi->s_journal,
3098                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3099                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3100                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3101         }
3102
3103         return ret;
3104 }
3105
3106 /*
3107  * Note: calculating the overhead so we can be compatible with
3108  * historical BSD practice is quite difficult in the face of
3109  * clusters/bigalloc.  This is because multiple metadata blocks from
3110  * different block group can end up in the same allocation cluster.
3111  * Calculating the exact overhead in the face of clustered allocation
3112  * requires either O(all block bitmaps) in memory or O(number of block
3113  * groups**2) in time.  We will still calculate the superblock for
3114  * older file systems --- and if we come across with a bigalloc file
3115  * system with zero in s_overhead_clusters the estimate will be close to
3116  * correct especially for very large cluster sizes --- but for newer
3117  * file systems, it's better to calculate this figure once at mkfs
3118  * time, and store it in the superblock.  If the superblock value is
3119  * present (even for non-bigalloc file systems), we will use it.
3120  */
3121 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3122                           char *buf)
3123 {
3124         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3125         struct ext4_group_desc  *gdp;
3126         ext4_fsblk_t            first_block, last_block, b;
3127         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3128         int                     s, j, count = 0;
3129
3130         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3131                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3132                         sbi->s_itb_per_group + 2);
3133
3134         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3135                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3136         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3137         for (i = 0; i < ngroups; i++) {
3138                 gdp = ext4_get_group_desc(sb, i, NULL);
3139                 b = ext4_block_bitmap(sb, gdp);
3140                 if (b >= first_block && b <= last_block) {
3141                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3142                         count++;
3143                 }
3144                 b = ext4_inode_bitmap(sb, gdp);
3145                 if (b >= first_block && b <= last_block) {
3146                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3147                         count++;
3148                 }
3149                 b = ext4_inode_table(sb, gdp);
3150                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3151                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3152                                 int c = EXT4_B2C(sbi, b - first_block);
3153                                 ext4_set_bit(c, buf);
3154                                 count++;
3155                         }
3156                 if (i != grp)
3157                         continue;
3158                 s = 0;
3159                 if (ext4_bg_has_super(sb, grp)) {
3160                         ext4_set_bit(s++, buf);
3161                         count++;
3162                 }
3163                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3164                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3165                         count++;
3166                 }
3167         }
3168         if (!count)
3169                 return 0;
3170         return EXT4_CLUSTERS_PER_GROUP(sb) -
3171                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3172 }
3173
3174 /*
3175  * Compute the overhead and stash it in sbi->s_overhead
3176  */
3177 int ext4_calculate_overhead(struct super_block *sb)
3178 {
3179         struct ext4_sb_info *sbi = EXT4_SB(sb);
3180         struct ext4_super_block *es = sbi->s_es;
3181         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3182         ext4_fsblk_t overhead = 0;
3183         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3184
3185         memset(buf, 0, PAGE_SIZE);
3186         if (!buf)
3187                 return -ENOMEM;
3188
3189         /*
3190          * Compute the overhead (FS structures).  This is constant
3191          * for a given filesystem unless the number of block groups
3192          * changes so we cache the previous value until it does.
3193          */
3194
3195         /*
3196          * All of the blocks before first_data_block are overhead
3197          */
3198         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3199
3200         /*
3201          * Add the overhead found in each block group
3202          */
3203         for (i = 0; i < ngroups; i++) {
3204                 int blks;
3205
3206                 blks = count_overhead(sb, i, buf);
3207                 overhead += blks;
3208                 if (blks)
3209                         memset(buf, 0, PAGE_SIZE);
3210                 cond_resched();
3211         }
3212         sbi->s_overhead = overhead;
3213         smp_wmb();
3214         free_page((unsigned long) buf);
3215         return 0;
3216 }
3217
3218 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3219 {
3220         char *orig_data = kstrdup(data, GFP_KERNEL);
3221         struct buffer_head *bh;
3222         struct ext4_super_block *es = NULL;
3223         struct ext4_sb_info *sbi;
3224         ext4_fsblk_t block;
3225         ext4_fsblk_t sb_block = get_sb_block(&data);
3226         ext4_fsblk_t logical_sb_block;
3227         unsigned long offset = 0;
3228         unsigned long journal_devnum = 0;
3229         unsigned long def_mount_opts;
3230         struct inode *root;
3231         char *cp;
3232         const char *descr;
3233         int ret = -ENOMEM;
3234         int blocksize, clustersize;
3235         unsigned int db_count;
3236         unsigned int i;
3237         int needs_recovery, has_huge_files, has_bigalloc;
3238         __u64 blocks_count;
3239         int err;
3240         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3241         ext4_group_t first_not_zeroed;
3242
3243         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3244         if (!sbi)
3245                 goto out_free_orig;
3246
3247         sbi->s_blockgroup_lock =
3248                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3249         if (!sbi->s_blockgroup_lock) {
3250                 kfree(sbi);
3251                 goto out_free_orig;
3252         }
3253         sb->s_fs_info = sbi;
3254         sbi->s_sb = sb;
3255         sbi->s_mount_opt = 0;
3256         sbi->s_resuid = make_kuid(&init_user_ns, EXT4_DEF_RESUID);
3257         sbi->s_resgid = make_kgid(&init_user_ns, EXT4_DEF_RESGID);
3258         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3259         sbi->s_sb_block = sb_block;
3260         if (sb->s_bdev->bd_part)
3261                 sbi->s_sectors_written_start =
3262                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3263
3264         /* Cleanup superblock name */
3265         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3266                 *cp = '!';
3267
3268         ret = -EINVAL;
3269         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3270         if (!blocksize) {
3271                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3272                 goto out_fail;
3273         }
3274
3275         /*
3276          * The ext4 superblock will not be buffer aligned for other than 1kB
3277          * block sizes.  We need to calculate the offset from buffer start.
3278          */
3279         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3280                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3281                 offset = do_div(logical_sb_block, blocksize);
3282         } else {
3283                 logical_sb_block = sb_block;
3284         }
3285
3286         if (!(bh = sb_bread(sb, logical_sb_block))) {
3287                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3288                 goto out_fail;
3289         }
3290         /*
3291          * Note: s_es must be initialized as soon as possible because
3292          *       some ext4 macro-instructions depend on its value
3293          */
3294         es = (struct ext4_super_block *) (bh->b_data + offset);
3295         sbi->s_es = es;
3296         sb->s_magic = le16_to_cpu(es->s_magic);
3297         if (sb->s_magic != EXT4_SUPER_MAGIC)
3298                 goto cantfind_ext4;
3299         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3300
3301         /* Warn if metadata_csum and gdt_csum are both set. */
3302         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3303                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3304             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3305                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3306                              "redundant flags; please run fsck.");
3307
3308         /* Check for a known checksum algorithm */
3309         if (!ext4_verify_csum_type(sb, es)) {
3310                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3311                          "unknown checksum algorithm.");
3312                 silent = 1;
3313                 goto cantfind_ext4;
3314         }
3315
3316         /* Load the checksum driver */
3317         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3318                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3319                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3320                 if (IS_ERR(sbi->s_chksum_driver)) {
3321                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3322                         ret = PTR_ERR(sbi->s_chksum_driver);
3323                         sbi->s_chksum_driver = NULL;
3324                         goto failed_mount;
3325                 }
3326         }
3327
3328         /* Check superblock checksum */
3329         if (!ext4_superblock_csum_verify(sb, es)) {
3330                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3331                          "invalid superblock checksum.  Run e2fsck?");
3332                 silent = 1;
3333                 goto cantfind_ext4;
3334         }
3335
3336         /* Precompute checksum seed for all metadata */
3337         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3338                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3339                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3340                                                sizeof(es->s_uuid));
3341
3342         /* Set defaults before we parse the mount options */
3343         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3344         set_opt(sb, INIT_INODE_TABLE);
3345         if (def_mount_opts & EXT4_DEFM_DEBUG)
3346                 set_opt(sb, DEBUG);
3347         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3348                 set_opt(sb, GRPID);
3349         if (def_mount_opts & EXT4_DEFM_UID16)
3350                 set_opt(sb, NO_UID32);
3351         /* xattr user namespace & acls are now defaulted on */
3352 #ifdef CONFIG_EXT4_FS_XATTR
3353         set_opt(sb, XATTR_USER);
3354 #endif
3355 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3356         set_opt(sb, POSIX_ACL);
3357 #endif
3358         set_opt(sb, MBLK_IO_SUBMIT);
3359         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3360                 set_opt(sb, JOURNAL_DATA);
3361         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3362                 set_opt(sb, ORDERED_DATA);
3363         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3364                 set_opt(sb, WRITEBACK_DATA);
3365
3366         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3367                 set_opt(sb, ERRORS_PANIC);
3368         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3369                 set_opt(sb, ERRORS_CONT);
3370         else
3371                 set_opt(sb, ERRORS_RO);
3372         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3373                 set_opt(sb, BLOCK_VALIDITY);
3374         if (def_mount_opts & EXT4_DEFM_DISCARD)
3375                 set_opt(sb, DISCARD);
3376
3377         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3378         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3379         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3380         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3381         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3382
3383         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3384                 set_opt(sb, BARRIER);
3385
3386         /*
3387          * enable delayed allocation by default
3388          * Use -o nodelalloc to turn it off
3389          */
3390         if (!IS_EXT3_SB(sb) &&
3391             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3392                 set_opt(sb, DELALLOC);
3393
3394         /*
3395          * set default s_li_wait_mult for lazyinit, for the case there is
3396          * no mount option specified.
3397          */
3398         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3399
3400         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3401                            &journal_devnum, &journal_ioprio, 0)) {
3402                 ext4_msg(sb, KERN_WARNING,
3403                          "failed to parse options in superblock: %s",
3404                          sbi->s_es->s_mount_opts);
3405         }
3406         sbi->s_def_mount_opt = sbi->s_mount_opt;
3407         if (!parse_options((char *) data, sb, &journal_devnum,
3408                            &journal_ioprio, 0))
3409                 goto failed_mount;
3410
3411         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3412                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3413                             "with data=journal disables delayed "
3414                             "allocation and O_DIRECT support!\n");
3415                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3416                         ext4_msg(sb, KERN_ERR, "can't mount with "
3417                                  "both data=journal and delalloc");
3418                         goto failed_mount;
3419                 }
3420                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3421                         ext4_msg(sb, KERN_ERR, "can't mount with "
3422                                  "both data=journal and delalloc");
3423                         goto failed_mount;
3424                 }
3425                 if (test_opt(sb, DELALLOC))
3426                         clear_opt(sb, DELALLOC);
3427         }
3428
3429         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3430         if (test_opt(sb, DIOREAD_NOLOCK)) {
3431                 if (blocksize < PAGE_SIZE) {
3432                         ext4_msg(sb, KERN_ERR, "can't mount with "
3433                                  "dioread_nolock if block size != PAGE_SIZE");
3434                         goto failed_mount;
3435                 }
3436         }
3437
3438         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3439                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3440
3441         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3442             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3443              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3444              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3445                 ext4_msg(sb, KERN_WARNING,
3446                        "feature flags set on rev 0 fs, "
3447                        "running e2fsck is recommended");
3448
3449         if (IS_EXT2_SB(sb)) {
3450                 if (ext2_feature_set_ok(sb))
3451                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3452                                  "using the ext4 subsystem");
3453                 else {
3454                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3455                                  "to feature incompatibilities");
3456                         goto failed_mount;
3457                 }
3458         }
3459
3460         if (IS_EXT3_SB(sb)) {
3461                 if (ext3_feature_set_ok(sb))
3462                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3463                                  "using the ext4 subsystem");
3464                 else {
3465                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3466                                  "to feature incompatibilities");
3467                         goto failed_mount;
3468                 }
3469         }
3470
3471         /*
3472          * Check feature flags regardless of the revision level, since we
3473          * previously didn't change the revision level when setting the flags,
3474          * so there is a chance incompat flags are set on a rev 0 filesystem.
3475          */
3476         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3477                 goto failed_mount;
3478
3479         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3480             blocksize > EXT4_MAX_BLOCK_SIZE) {
3481                 ext4_msg(sb, KERN_ERR,
3482                        "Unsupported filesystem blocksize %d", blocksize);
3483                 goto failed_mount;
3484         }
3485
3486         if (sb->s_blocksize != blocksize) {
3487                 /* Validate the filesystem blocksize */
3488                 if (!sb_set_blocksize(sb, blocksize)) {
3489                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3490                                         blocksize);
3491                         goto failed_mount;
3492                 }
3493
3494                 brelse(bh);
3495                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3496                 offset = do_div(logical_sb_block, blocksize);
3497                 bh = sb_bread(sb, logical_sb_block);
3498                 if (!bh) {
3499                         ext4_msg(sb, KERN_ERR,
3500                                "Can't read superblock on 2nd try");
3501                         goto failed_mount;
3502                 }
3503                 es = (struct ext4_super_block *)(bh->b_data + offset);
3504                 sbi->s_es = es;
3505                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3506                         ext4_msg(sb, KERN_ERR,
3507                                "Magic mismatch, very weird!");
3508                         goto failed_mount;
3509                 }
3510         }
3511
3512         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3513                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3514         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3515                                                       has_huge_files);
3516         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3517
3518         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3519                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3520                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3521         } else {
3522                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3523                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3524                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3525                     (!is_power_of_2(sbi->s_inode_size)) ||
3526                     (sbi->s_inode_size > blocksize)) {
3527                         ext4_msg(sb, KERN_ERR,
3528                                "unsupported inode size: %d",
3529                                sbi->s_inode_size);
3530                         goto failed_mount;
3531                 }
3532                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3533                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3534         }
3535
3536         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3537         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3538                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3539                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3540                     !is_power_of_2(sbi->s_desc_size)) {
3541                         ext4_msg(sb, KERN_ERR,
3542                                "unsupported descriptor size %lu",
3543                                sbi->s_desc_size);
3544                         goto failed_mount;
3545                 }
3546         } else
3547                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3548
3549         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3550         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3551         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3552                 goto cantfind_ext4;
3553
3554         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3555         if (sbi->s_inodes_per_block == 0)
3556                 goto cantfind_ext4;
3557         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3558                                         sbi->s_inodes_per_block;
3559         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3560         sbi->s_sbh = bh;
3561         sbi->s_mount_state = le16_to_cpu(es->s_state);
3562         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3563         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3564
3565         for (i = 0; i < 4; i++)
3566                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3567         sbi->s_def_hash_version = es->s_def_hash_version;
3568         i = le32_to_cpu(es->s_flags);
3569         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3570                 sbi->s_hash_unsigned = 3;
3571         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3572 #ifdef __CHAR_UNSIGNED__
3573                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3574                 sbi->s_hash_unsigned = 3;
3575 #else
3576                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3577 #endif
3578         }
3579
3580         /* Handle clustersize */
3581         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3582         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3583                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3584         if (has_bigalloc) {
3585                 if (clustersize < blocksize) {
3586                         ext4_msg(sb, KERN_ERR,
3587                                  "cluster size (%d) smaller than "
3588                                  "block size (%d)", clustersize, blocksize);
3589                         goto failed_mount;
3590                 }
3591                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3592                         le32_to_cpu(es->s_log_block_size);
3593                 sbi->s_clusters_per_group =
3594                         le32_to_cpu(es->s_clusters_per_group);
3595                 if (sbi->s_clusters_per_group > blocksize * 8) {
3596                         ext4_msg(sb, KERN_ERR,
3597                                  "#clusters per group too big: %lu",
3598                                  sbi->s_clusters_per_group);
3599                         goto failed_mount;
3600                 }
3601                 if (sbi->s_blocks_per_group !=
3602                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3603                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3604                                  "clusters per group (%lu) inconsistent",
3605                                  sbi->s_blocks_per_group,
3606                                  sbi->s_clusters_per_group);
3607                         goto failed_mount;
3608                 }
3609         } else {
3610                 if (clustersize != blocksize) {
3611                         ext4_warning(sb, "fragment/cluster size (%d) != "
3612                                      "block size (%d)", clustersize,
3613                                      blocksize);
3614                         clustersize = blocksize;
3615                 }
3616                 if (sbi->s_blocks_per_group > blocksize * 8) {
3617                         ext4_msg(sb, KERN_ERR,
3618                                  "#blocks per group too big: %lu",
3619                                  sbi->s_blocks_per_group);
3620                         goto failed_mount;
3621                 }
3622                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3623                 sbi->s_cluster_bits = 0;
3624         }
3625         sbi->s_cluster_ratio = clustersize / blocksize;
3626
3627         if (sbi->s_inodes_per_group > blocksize * 8) {
3628                 ext4_msg(sb, KERN_ERR,
3629                        "#inodes per group too big: %lu",
3630                        sbi->s_inodes_per_group);
3631                 goto failed_mount;
3632         }
3633
3634         /*
3635          * Test whether we have more sectors than will fit in sector_t,
3636          * and whether the max offset is addressable by the page cache.
3637          */
3638         err = generic_check_addressable(sb->s_blocksize_bits,
3639                                         ext4_blocks_count(es));
3640         if (err) {
3641                 ext4_msg(sb, KERN_ERR, "filesystem"
3642                          " too large to mount safely on this system");
3643                 if (sizeof(sector_t) < 8)
3644                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3645                 ret = err;
3646                 goto failed_mount;
3647         }
3648
3649         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3650                 goto cantfind_ext4;
3651
3652         /* check blocks count against device size */
3653         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3654         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3655                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3656                        "exceeds size of device (%llu blocks)",
3657                        ext4_blocks_count(es), blocks_count);
3658                 goto failed_mount;
3659         }
3660
3661         /*
3662          * It makes no sense for the first data block to be beyond the end
3663          * of the filesystem.
3664          */
3665         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3666                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3667                          "block %u is beyond end of filesystem (%llu)",
3668                          le32_to_cpu(es->s_first_data_block),
3669                          ext4_blocks_count(es));
3670                 goto failed_mount;
3671         }
3672         blocks_count = (ext4_blocks_count(es) -
3673                         le32_to_cpu(es->s_first_data_block) +
3674                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3675         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3676         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3677                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3678                        "(block count %llu, first data block %u, "
3679                        "blocks per group %lu)", sbi->s_groups_count,
3680                        ext4_blocks_count(es),
3681                        le32_to_cpu(es->s_first_data_block),
3682                        EXT4_BLOCKS_PER_GROUP(sb));
3683                 goto failed_mount;
3684         }
3685         sbi->s_groups_count = blocks_count;
3686         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3687                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3688         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3689                    EXT4_DESC_PER_BLOCK(sb);
3690         sbi->s_group_desc = ext4_kvmalloc(db_count *
3691                                           sizeof(struct buffer_head *),
3692                                           GFP_KERNEL);
3693         if (sbi->s_group_desc == NULL) {
3694                 ext4_msg(sb, KERN_ERR, "not enough memory");
3695                 ret = -ENOMEM;
3696                 goto failed_mount;
3697         }
3698
3699         if (ext4_proc_root)
3700                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3701
3702         if (sbi->s_proc)
3703                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3704                                  &ext4_seq_options_fops, sb);
3705
3706         bgl_lock_init(sbi->s_blockgroup_lock);
3707
3708         for (i = 0; i < db_count; i++) {
3709                 block = descriptor_loc(sb, logical_sb_block, i);
3710                 sbi->s_group_desc[i] = sb_bread(sb, block);
3711                 if (!sbi->s_group_desc[i]) {
3712                         ext4_msg(sb, KERN_ERR,
3713                                "can't read group descriptor %d", i);
3714                         db_count = i;
3715                         goto failed_mount2;
3716                 }
3717         }
3718         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3719                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3720                 goto failed_mount2;
3721         }
3722         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3723                 if (!ext4_fill_flex_info(sb)) {
3724                         ext4_msg(sb, KERN_ERR,
3725                                "unable to initialize "
3726                                "flex_bg meta info!");
3727                         goto failed_mount2;
3728                 }
3729
3730         sbi->s_gdb_count = db_count;
3731         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3732         spin_lock_init(&sbi->s_next_gen_lock);
3733
3734         init_timer(&sbi->s_err_report);
3735         sbi->s_err_report.function = print_daily_error_info;
3736         sbi->s_err_report.data = (unsigned long) sb;
3737
3738         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3739                         ext4_count_free_clusters(sb));
3740         if (!err) {
3741                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3742                                 ext4_count_free_inodes(sb));
3743         }
3744         if (!err) {
3745                 err = percpu_counter_init(&sbi->s_dirs_counter,
3746                                 ext4_count_dirs(sb));
3747         }
3748         if (!err) {
3749                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3750         }
3751         if (err) {
3752                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3753                 ret = err;
3754                 goto failed_mount3;
3755         }
3756
3757         sbi->s_stripe = ext4_get_stripe_size(sbi);
3758         sbi->s_max_writeback_mb_bump = 128;
3759         sbi->s_extent_max_zeroout_kb = 32;
3760
3761         /*
3762          * set up enough so that it can read an inode
3763          */
3764         if (!test_opt(sb, NOLOAD) &&
3765             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3766                 sb->s_op = &ext4_sops;
3767         else
3768                 sb->s_op = &ext4_nojournal_sops;
3769         sb->s_export_op = &ext4_export_ops;
3770         sb->s_xattr = ext4_xattr_handlers;
3771 #ifdef CONFIG_QUOTA
3772         sb->s_qcop = &ext4_qctl_operations;
3773         sb->dq_op = &ext4_quota_operations;
3774
3775         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
3776                 /* Use qctl operations for hidden quota files. */
3777                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3778         }
3779 #endif
3780         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3781
3782         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3783         mutex_init(&sbi->s_orphan_lock);
3784         sbi->s_resize_flags = 0;
3785
3786         sb->s_root = NULL;
3787
3788         needs_recovery = (es->s_last_orphan != 0 ||
3789                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3790                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3791
3792         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3793             !(sb->s_flags & MS_RDONLY))
3794                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3795                         goto failed_mount3;
3796
3797         /*
3798          * The first inode we look at is the journal inode.  Don't try
3799          * root first: it may be modified in the journal!
3800          */
3801         if (!test_opt(sb, NOLOAD) &&
3802             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3803                 if (ext4_load_journal(sb, es, journal_devnum))
3804                         goto failed_mount3;
3805         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3806               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3807                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3808                        "suppressed and not mounted read-only");
3809                 goto failed_mount_wq;
3810         } else {
3811                 clear_opt(sb, DATA_FLAGS);
3812                 sbi->s_journal = NULL;
3813                 needs_recovery = 0;
3814                 goto no_journal;
3815         }
3816
3817         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3818             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3819                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3820                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3821                 goto failed_mount_wq;
3822         }
3823
3824         if (!set_journal_csum_feature_set(sb)) {
3825                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3826                          "feature set");
3827                 goto failed_mount_wq;
3828         }
3829
3830         /* We have now updated the journal if required, so we can
3831          * validate the data journaling mode. */
3832         switch (test_opt(sb, DATA_FLAGS)) {
3833         case 0:
3834                 /* No mode set, assume a default based on the journal
3835                  * capabilities: ORDERED_DATA if the journal can
3836                  * cope, else JOURNAL_DATA
3837                  */
3838                 if (jbd2_journal_check_available_features
3839                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3840                         set_opt(sb, ORDERED_DATA);
3841                 else
3842                         set_opt(sb, JOURNAL_DATA);
3843                 break;
3844
3845         case EXT4_MOUNT_ORDERED_DATA:
3846         case EXT4_MOUNT_WRITEBACK_DATA:
3847                 if (!jbd2_journal_check_available_features
3848                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3849                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3850                                "requested data journaling mode");
3851                         goto failed_mount_wq;
3852                 }
3853         default:
3854                 break;
3855         }
3856         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3857
3858         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3859
3860         /*
3861          * The journal may have updated the bg summary counts, so we
3862          * need to update the global counters.
3863          */
3864         percpu_counter_set(&sbi->s_freeclusters_counter,
3865                            ext4_count_free_clusters(sb));
3866         percpu_counter_set(&sbi->s_freeinodes_counter,
3867                            ext4_count_free_inodes(sb));
3868         percpu_counter_set(&sbi->s_dirs_counter,
3869                            ext4_count_dirs(sb));
3870         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3871
3872 no_journal:
3873         /*
3874          * Get the # of file system overhead blocks from the
3875          * superblock if present.
3876          */
3877         if (es->s_overhead_clusters)
3878                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3879         else {
3880                 ret = ext4_calculate_overhead(sb);
3881                 if (ret)
3882                         goto failed_mount_wq;
3883         }
3884
3885         /*
3886          * The maximum number of concurrent works can be high and
3887          * concurrency isn't really necessary.  Limit it to 1.
3888          */
3889         EXT4_SB(sb)->dio_unwritten_wq =
3890                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3891         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3892                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3893                 goto failed_mount_wq;
3894         }
3895
3896         /*
3897          * The jbd2_journal_load will have done any necessary log recovery,
3898          * so we can safely mount the rest of the filesystem now.
3899          */
3900
3901         root = ext4_iget(sb, EXT4_ROOT_INO);
3902         if (IS_ERR(root)) {
3903                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3904                 ret = PTR_ERR(root);
3905                 root = NULL;
3906                 goto failed_mount4;
3907         }
3908         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3909                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3910                 iput(root);
3911                 goto failed_mount4;
3912         }
3913         sb->s_root = d_make_root(root);
3914         if (!sb->s_root) {
3915                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3916                 ret = -ENOMEM;
3917                 goto failed_mount4;
3918         }
3919
3920         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3921                 sb->s_flags |= MS_RDONLY;
3922
3923         /* determine the minimum size of new large inodes, if present */
3924         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3925                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3926                                                      EXT4_GOOD_OLD_INODE_SIZE;
3927                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3928                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3929                         if (sbi->s_want_extra_isize <
3930                             le16_to_cpu(es->s_want_extra_isize))
3931                                 sbi->s_want_extra_isize =
3932                                         le16_to_cpu(es->s_want_extra_isize);
3933                         if (sbi->s_want_extra_isize <
3934                             le16_to_cpu(es->s_min_extra_isize))
3935                                 sbi->s_want_extra_isize =
3936                                         le16_to_cpu(es->s_min_extra_isize);
3937                 }
3938         }
3939         /* Check if enough inode space is available */
3940         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3941                                                         sbi->s_inode_size) {
3942                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3943                                                        EXT4_GOOD_OLD_INODE_SIZE;
3944                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3945                          "available");
3946         }
3947
3948         err = ext4_setup_system_zone(sb);
3949         if (err) {
3950                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3951                          "zone (%d)", err);
3952                 goto failed_mount4a;
3953         }
3954
3955         ext4_ext_init(sb);
3956         err = ext4_mb_init(sb);
3957         if (err) {
3958                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3959                          err);
3960                 goto failed_mount5;
3961         }
3962
3963         err = ext4_register_li_request(sb, first_not_zeroed);
3964         if (err)
3965                 goto failed_mount6;
3966
3967         sbi->s_kobj.kset = ext4_kset;
3968         init_completion(&sbi->s_kobj_unregister);
3969         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3970                                    "%s", sb->s_id);
3971         if (err)
3972                 goto failed_mount7;
3973
3974         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3975         ext4_orphan_cleanup(sb, es);
3976         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3977         if (needs_recovery) {
3978                 ext4_msg(sb, KERN_INFO, "recovery complete");
3979                 ext4_mark_recovery_complete(sb, es);
3980         }
3981         if (EXT4_SB(sb)->s_journal) {
3982                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3983                         descr = " journalled data mode";
3984                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3985                         descr = " ordered data mode";
3986                 else
3987                         descr = " writeback data mode";
3988         } else
3989                 descr = "out journal";
3990
3991 #ifdef CONFIG_QUOTA
3992         /* Enable quota usage during mount. */
3993         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
3994             !(sb->s_flags & MS_RDONLY)) {
3995                 ret = ext4_enable_quotas(sb);
3996                 if (ret)
3997                         goto failed_mount7;
3998         }
3999 #endif  /* CONFIG_QUOTA */
4000
4001         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4002                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4003                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4004
4005         if (es->s_error_count)
4006                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4007
4008         kfree(orig_data);
4009         return 0;
4010
4011 cantfind_ext4:
4012         if (!silent)
4013                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4014         goto failed_mount;
4015
4016 failed_mount7:
4017         ext4_unregister_li_request(sb);
4018 failed_mount6:
4019         ext4_mb_release(sb);
4020 failed_mount5:
4021         ext4_ext_release(sb);
4022         ext4_release_system_zone(sb);
4023 failed_mount4a:
4024         dput(sb->s_root);
4025         sb->s_root = NULL;
4026 failed_mount4:
4027         ext4_msg(sb, KERN_ERR, "mount failed");
4028         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4029 failed_mount_wq:
4030         if (sbi->s_journal) {
4031                 jbd2_journal_destroy(sbi->s_journal);
4032                 sbi->s_journal = NULL;
4033         }
4034 failed_mount3:
4035         del_timer(&sbi->s_err_report);
4036         if (sbi->s_flex_groups)
4037                 ext4_kvfree(sbi->s_flex_groups);
4038         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4039         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4040         percpu_counter_destroy(&sbi->s_dirs_counter);
4041         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4042         if (sbi->s_mmp_tsk)
4043                 kthread_stop(sbi->s_mmp_tsk);
4044 failed_mount2:
4045         for (i = 0; i < db_count; i++)
4046                 brelse(sbi->s_group_desc[i]);
4047         ext4_kvfree(sbi->s_group_desc);
4048 failed_mount:
4049         if (sbi->s_chksum_driver)
4050                 crypto_free_shash(sbi->s_chksum_driver);
4051         if (sbi->s_proc) {
4052                 remove_proc_entry("options", sbi->s_proc);
4053                 remove_proc_entry(sb->s_id, ext4_proc_root);
4054         }
4055 #ifdef CONFIG_QUOTA
4056         for (i = 0; i < MAXQUOTAS; i++)
4057                 kfree(sbi->s_qf_names[i]);
4058 #endif
4059         ext4_blkdev_remove(sbi);
4060         brelse(bh);
4061 out_fail:
4062         sb->s_fs_info = NULL;
4063         kfree(sbi->s_blockgroup_lock);
4064         kfree(sbi);
4065 out_free_orig:
4066         kfree(orig_data);
4067         return ret;
4068 }
4069
4070 /*
4071  * Setup any per-fs journal parameters now.  We'll do this both on
4072  * initial mount, once the journal has been initialised but before we've
4073  * done any recovery; and again on any subsequent remount.
4074  */
4075 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4076 {
4077         struct ext4_sb_info *sbi = EXT4_SB(sb);
4078
4079         journal->j_commit_interval = sbi->s_commit_interval;
4080         journal->j_min_batch_time = sbi->s_min_batch_time;
4081         journal->j_max_batch_time = sbi->s_max_batch_time;
4082
4083         write_lock(&journal->j_state_lock);
4084         if (test_opt(sb, BARRIER))
4085                 journal->j_flags |= JBD2_BARRIER;
4086         else
4087                 journal->j_flags &= ~JBD2_BARRIER;
4088         if (test_opt(sb, DATA_ERR_ABORT))
4089                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4090         else
4091                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4092         write_unlock(&journal->j_state_lock);
4093 }
4094
4095 static journal_t *ext4_get_journal(struct super_block *sb,
4096                                    unsigned int journal_inum)
4097 {
4098         struct inode *journal_inode;
4099         journal_t *journal;
4100
4101         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4102
4103         /* First, test for the existence of a valid inode on disk.  Bad
4104          * things happen if we iget() an unused inode, as the subsequent
4105          * iput() will try to delete it. */
4106
4107         journal_inode = ext4_iget(sb, journal_inum);
4108         if (IS_ERR(journal_inode)) {
4109                 ext4_msg(sb, KERN_ERR, "no journal found");
4110                 return NULL;
4111         }
4112         if (!journal_inode->i_nlink) {
4113                 make_bad_inode(journal_inode);
4114                 iput(journal_inode);
4115                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4116                 return NULL;
4117         }
4118
4119         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4120                   journal_inode, journal_inode->i_size);
4121         if (!S_ISREG(journal_inode->i_mode)) {
4122                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4123                 iput(journal_inode);
4124                 return NULL;
4125         }
4126
4127         journal = jbd2_journal_init_inode(journal_inode);
4128         if (!journal) {
4129                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4130                 iput(journal_inode);
4131                 return NULL;
4132         }
4133         journal->j_private = sb;
4134         ext4_init_journal_params(sb, journal);
4135         return journal;
4136 }
4137
4138 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4139                                        dev_t j_dev)
4140 {
4141         struct buffer_head *bh;
4142         journal_t *journal;
4143         ext4_fsblk_t start;
4144         ext4_fsblk_t len;
4145         int hblock, blocksize;
4146         ext4_fsblk_t sb_block;
4147         unsigned long offset;
4148         struct ext4_super_block *es;
4149         struct block_device *bdev;
4150
4151         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4152
4153         bdev = ext4_blkdev_get(j_dev, sb);
4154         if (bdev == NULL)
4155                 return NULL;
4156
4157         blocksize = sb->s_blocksize;
4158         hblock = bdev_logical_block_size(bdev);
4159         if (blocksize < hblock) {
4160                 ext4_msg(sb, KERN_ERR,
4161                         "blocksize too small for journal device");
4162                 goto out_bdev;
4163         }
4164
4165         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4166         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4167         set_blocksize(bdev, blocksize);
4168         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4169                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4170                        "external journal");
4171                 goto out_bdev;
4172         }
4173
4174         es = (struct ext4_super_block *) (bh->b_data + offset);
4175         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4176             !(le32_to_cpu(es->s_feature_incompat) &
4177               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4178                 ext4_msg(sb, KERN_ERR, "external journal has "
4179                                         "bad superblock");
4180                 brelse(bh);
4181                 goto out_bdev;
4182         }
4183
4184         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4185                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4186                 brelse(bh);
4187                 goto out_bdev;
4188         }
4189
4190         len = ext4_blocks_count(es);
4191         start = sb_block + 1;
4192         brelse(bh);     /* we're done with the superblock */
4193
4194         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4195                                         start, len, blocksize);
4196         if (!journal) {
4197                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4198                 goto out_bdev;
4199         }
4200         journal->j_private = sb;
4201         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4202         wait_on_buffer(journal->j_sb_buffer);
4203         if (!buffer_uptodate(journal->j_sb_buffer)) {
4204                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4205                 goto out_journal;
4206         }
4207         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4208                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4209                                         "user (unsupported) - %d",
4210                         be32_to_cpu(journal->j_superblock->s_nr_users));
4211                 goto out_journal;
4212         }
4213         EXT4_SB(sb)->journal_bdev = bdev;
4214         ext4_init_journal_params(sb, journal);
4215         return journal;
4216
4217 out_journal:
4218         jbd2_journal_destroy(journal);
4219 out_bdev:
4220         ext4_blkdev_put(bdev);
4221         return NULL;
4222 }
4223
4224 static int ext4_load_journal(struct super_block *sb,
4225                              struct ext4_super_block *es,
4226                              unsigned long journal_devnum)
4227 {
4228         journal_t *journal;
4229         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4230         dev_t journal_dev;
4231         int err = 0;
4232         int really_read_only;
4233
4234         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4235
4236         if (journal_devnum &&
4237             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4238                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4239                         "numbers have changed");
4240                 journal_dev = new_decode_dev(journal_devnum);
4241         } else
4242                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4243
4244         really_read_only = bdev_read_only(sb->s_bdev);
4245
4246         /*
4247          * Are we loading a blank journal or performing recovery after a
4248          * crash?  For recovery, we need to check in advance whether we
4249          * can get read-write access to the device.
4250          */
4251         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4252                 if (sb->s_flags & MS_RDONLY) {
4253                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4254                                         "required on readonly filesystem");
4255                         if (really_read_only) {
4256                                 ext4_msg(sb, KERN_ERR, "write access "
4257                                         "unavailable, cannot proceed");
4258                                 return -EROFS;
4259                         }
4260                         ext4_msg(sb, KERN_INFO, "write access will "
4261                                "be enabled during recovery");
4262                 }
4263         }
4264
4265         if (journal_inum && journal_dev) {
4266                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4267                        "and inode journals!");
4268                 return -EINVAL;
4269         }
4270
4271         if (journal_inum) {
4272                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4273                         return -EINVAL;
4274         } else {
4275                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4276                         return -EINVAL;
4277         }
4278
4279         if (!(journal->j_flags & JBD2_BARRIER))
4280                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4281
4282         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4283                 err = jbd2_journal_wipe(journal, !really_read_only);
4284         if (!err) {
4285                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4286                 if (save)
4287                         memcpy(save, ((char *) es) +
4288                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4289                 err = jbd2_journal_load(journal);
4290                 if (save)
4291                         memcpy(((char *) es) + EXT4_S_ERR_START,
4292                                save, EXT4_S_ERR_LEN);
4293                 kfree(save);
4294         }
4295
4296         if (err) {
4297                 ext4_msg(sb, KERN_ERR, "error loading journal");
4298                 jbd2_journal_destroy(journal);
4299                 return err;
4300         }
4301
4302         EXT4_SB(sb)->s_journal = journal;
4303         ext4_clear_journal_err(sb, es);
4304
4305         if (!really_read_only && journal_devnum &&
4306             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4307                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4308
4309                 /* Make sure we flush the recovery flag to disk. */
4310                 ext4_commit_super(sb, 1);
4311         }
4312
4313         return 0;
4314 }
4315
4316 static int ext4_commit_super(struct super_block *sb, int sync)
4317 {
4318         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4319         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4320         int error = 0;
4321
4322         if (!sbh || block_device_ejected(sb))
4323                 return error;
4324         if (buffer_write_io_error(sbh)) {
4325                 /*
4326                  * Oh, dear.  A previous attempt to write the
4327                  * superblock failed.  This could happen because the
4328                  * USB device was yanked out.  Or it could happen to
4329                  * be a transient write error and maybe the block will
4330                  * be remapped.  Nothing we can do but to retry the
4331                  * write and hope for the best.
4332                  */
4333                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4334                        "superblock detected");
4335                 clear_buffer_write_io_error(sbh);
4336                 set_buffer_uptodate(sbh);
4337         }
4338         /*
4339          * If the file system is mounted read-only, don't update the
4340          * superblock write time.  This avoids updating the superblock
4341          * write time when we are mounting the root file system
4342          * read/only but we need to replay the journal; at that point,
4343          * for people who are east of GMT and who make their clock
4344          * tick in localtime for Windows bug-for-bug compatibility,
4345          * the clock is set in the future, and this will cause e2fsck
4346          * to complain and force a full file system check.
4347          */
4348         if (!(sb->s_flags & MS_RDONLY))
4349                 es->s_wtime = cpu_to_le32(get_seconds());
4350         if (sb->s_bdev->bd_part)
4351                 es->s_kbytes_written =
4352                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4353                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4354                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4355         else
4356                 es->s_kbytes_written =
4357                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4358         ext4_free_blocks_count_set(es,
4359                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4360                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4361         es->s_free_inodes_count =
4362                 cpu_to_le32(percpu_counter_sum_positive(
4363                                 &EXT4_SB(sb)->s_freeinodes_counter));
4364         BUFFER_TRACE(sbh, "marking dirty");
4365         ext4_superblock_csum_set(sb, es);
4366         mark_buffer_dirty(sbh);
4367         if (sync) {
4368                 error = sync_dirty_buffer(sbh);
4369                 if (error)
4370                         return error;
4371
4372                 error = buffer_write_io_error(sbh);
4373                 if (error) {
4374                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4375                                "superblock");
4376                         clear_buffer_write_io_error(sbh);
4377                         set_buffer_uptodate(sbh);
4378                 }
4379         }
4380         return error;
4381 }
4382
4383 /*
4384  * Have we just finished recovery?  If so, and if we are mounting (or
4385  * remounting) the filesystem readonly, then we will end up with a
4386  * consistent fs on disk.  Record that fact.
4387  */
4388 static void ext4_mark_recovery_complete(struct super_block *sb,
4389                                         struct ext4_super_block *es)
4390 {
4391         journal_t *journal = EXT4_SB(sb)->s_journal;
4392
4393         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4394                 BUG_ON(journal != NULL);
4395                 return;
4396         }
4397         jbd2_journal_lock_updates(journal);
4398         if (jbd2_journal_flush(journal) < 0)
4399                 goto out;
4400
4401         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4402             sb->s_flags & MS_RDONLY) {
4403                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4404                 ext4_commit_super(sb, 1);
4405         }
4406
4407 out:
4408         jbd2_journal_unlock_updates(journal);
4409 }
4410
4411 /*
4412  * If we are mounting (or read-write remounting) a filesystem whose journal
4413  * has recorded an error from a previous lifetime, move that error to the
4414  * main filesystem now.
4415  */
4416 static void ext4_clear_journal_err(struct super_block *sb,
4417                                    struct ext4_super_block *es)
4418 {
4419         journal_t *journal;
4420         int j_errno;
4421         const char *errstr;
4422
4423         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4424
4425         journal = EXT4_SB(sb)->s_journal;
4426
4427         /*
4428          * Now check for any error status which may have been recorded in the
4429          * journal by a prior ext4_error() or ext4_abort()
4430          */
4431
4432         j_errno = jbd2_journal_errno(journal);
4433         if (j_errno) {
4434                 char nbuf[16];
4435
4436                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4437                 ext4_warning(sb, "Filesystem error recorded "
4438                              "from previous mount: %s", errstr);
4439                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4440
4441                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4442                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4443                 ext4_commit_super(sb, 1);
4444
4445                 jbd2_journal_clear_err(journal);
4446                 jbd2_journal_update_sb_errno(journal);
4447         }
4448 }
4449
4450 /*
4451  * Force the running and committing transactions to commit,
4452  * and wait on the commit.
4453  */
4454 int ext4_force_commit(struct super_block *sb)
4455 {
4456         journal_t *journal;
4457         int ret = 0;
4458
4459         if (sb->s_flags & MS_RDONLY)
4460                 return 0;
4461
4462         journal = EXT4_SB(sb)->s_journal;
4463         if (journal)
4464                 ret = ext4_journal_force_commit(journal);
4465
4466         return ret;
4467 }
4468
4469 static int ext4_sync_fs(struct super_block *sb, int wait)
4470 {
4471         int ret = 0;
4472         tid_t target;
4473         struct ext4_sb_info *sbi = EXT4_SB(sb);
4474
4475         trace_ext4_sync_fs(sb, wait);
4476         flush_workqueue(sbi->dio_unwritten_wq);
4477         /*
4478          * Writeback quota in non-journalled quota case - journalled quota has
4479          * no dirty dquots
4480          */
4481         dquot_writeback_dquots(sb, -1);
4482         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4483                 if (wait)
4484                         jbd2_log_wait_commit(sbi->s_journal, target);
4485         }
4486         return ret;
4487 }
4488
4489 /*
4490  * LVM calls this function before a (read-only) snapshot is created.  This
4491  * gives us a chance to flush the journal completely and mark the fs clean.
4492  *
4493  * Note that only this function cannot bring a filesystem to be in a clean
4494  * state independently. It relies on upper layer to stop all data & metadata
4495  * modifications.
4496  */
4497 static int ext4_freeze(struct super_block *sb)
4498 {
4499         int error = 0;
4500         journal_t *journal;
4501
4502         if (sb->s_flags & MS_RDONLY)
4503                 return 0;
4504
4505         journal = EXT4_SB(sb)->s_journal;
4506
4507         /* Now we set up the journal barrier. */
4508         jbd2_journal_lock_updates(journal);
4509
4510         /*
4511          * Don't clear the needs_recovery flag if we failed to flush
4512          * the journal.
4513          */
4514         error = jbd2_journal_flush(journal);
4515         if (error < 0)
4516                 goto out;
4517
4518         /* Journal blocked and flushed, clear needs_recovery flag. */
4519         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4520         error = ext4_commit_super(sb, 1);
4521 out:
4522         /* we rely on upper layer to stop further updates */
4523         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4524         return error;
4525 }
4526
4527 /*
4528  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4529  * flag here, even though the filesystem is not technically dirty yet.
4530  */
4531 static int ext4_unfreeze(struct super_block *sb)
4532 {
4533         if (sb->s_flags & MS_RDONLY)
4534                 return 0;
4535
4536         /* Reset the needs_recovery flag before the fs is unlocked. */
4537         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4538         ext4_commit_super(sb, 1);
4539         return 0;
4540 }
4541
4542 /*
4543  * Structure to save mount options for ext4_remount's benefit
4544  */
4545 struct ext4_mount_options {
4546         unsigned long s_mount_opt;
4547         unsigned long s_mount_opt2;
4548         kuid_t s_resuid;
4549         kgid_t s_resgid;
4550         unsigned long s_commit_interval;
4551         u32 s_min_batch_time, s_max_batch_time;
4552 #ifdef CONFIG_QUOTA
4553         int s_jquota_fmt;
4554         char *s_qf_names[MAXQUOTAS];
4555 #endif
4556 };
4557
4558 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4559 {
4560         struct ext4_super_block *es;
4561         struct ext4_sb_info *sbi = EXT4_SB(sb);
4562         unsigned long old_sb_flags;
4563         struct ext4_mount_options old_opts;
4564         int enable_quota = 0;
4565         ext4_group_t g;
4566         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4567         int err = 0;
4568 #ifdef CONFIG_QUOTA
4569         int i;
4570 #endif
4571         char *orig_data = kstrdup(data, GFP_KERNEL);
4572
4573         /* Store the original options */
4574         old_sb_flags = sb->s_flags;
4575         old_opts.s_mount_opt = sbi->s_mount_opt;
4576         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4577         old_opts.s_resuid = sbi->s_resuid;
4578         old_opts.s_resgid = sbi->s_resgid;
4579         old_opts.s_commit_interval = sbi->s_commit_interval;
4580         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4581         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4582 #ifdef CONFIG_QUOTA
4583         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4584         for (i = 0; i < MAXQUOTAS; i++)
4585                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4586 #endif
4587         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4588                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4589
4590         /*
4591          * Allow the "check" option to be passed as a remount option.
4592          */
4593         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4594                 err = -EINVAL;
4595                 goto restore_opts;
4596         }
4597
4598         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4599                 ext4_abort(sb, "Abort forced by user");
4600
4601         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4602                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4603
4604         es = sbi->s_es;
4605
4606         if (sbi->s_journal) {
4607                 ext4_init_journal_params(sb, sbi->s_journal);
4608                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4609         }
4610
4611         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4612                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4613                         err = -EROFS;
4614                         goto restore_opts;
4615                 }
4616
4617                 if (*flags & MS_RDONLY) {
4618                         err = dquot_suspend(sb, -1);
4619                         if (err < 0)
4620                                 goto restore_opts;
4621
4622                         /*
4623                          * First of all, the unconditional stuff we have to do
4624                          * to disable replay of the journal when we next remount
4625                          */
4626                         sb->s_flags |= MS_RDONLY;
4627
4628                         /*
4629                          * OK, test if we are remounting a valid rw partition
4630                          * readonly, and if so set the rdonly flag and then
4631                          * mark the partition as valid again.
4632                          */
4633                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4634                             (sbi->s_mount_state & EXT4_VALID_FS))
4635                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4636
4637                         if (sbi->s_journal)
4638                                 ext4_mark_recovery_complete(sb, es);
4639                 } else {
4640                         /* Make sure we can mount this feature set readwrite */
4641                         if (!ext4_feature_set_ok(sb, 0)) {
4642                                 err = -EROFS;
4643                                 goto restore_opts;
4644                         }
4645                         /*
4646                          * Make sure the group descriptor checksums
4647                          * are sane.  If they aren't, refuse to remount r/w.
4648                          */
4649                         for (g = 0; g < sbi->s_groups_count; g++) {
4650                                 struct ext4_group_desc *gdp =
4651                                         ext4_get_group_desc(sb, g, NULL);
4652
4653                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4654                                         ext4_msg(sb, KERN_ERR,
4655                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4656                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4657                                                le16_to_cpu(gdp->bg_checksum));
4658                                         err = -EINVAL;
4659                                         goto restore_opts;
4660                                 }
4661                         }
4662
4663                         /*
4664                          * If we have an unprocessed orphan list hanging
4665                          * around from a previously readonly bdev mount,
4666                          * require a full umount/remount for now.
4667                          */
4668                         if (es->s_last_orphan) {
4669                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4670                                        "remount RDWR because of unprocessed "
4671                                        "orphan inode list.  Please "
4672                                        "umount/remount instead");
4673                                 err = -EINVAL;
4674                                 goto restore_opts;
4675                         }
4676
4677                         /*
4678                          * Mounting a RDONLY partition read-write, so reread
4679                          * and store the current valid flag.  (It may have
4680                          * been changed by e2fsck since we originally mounted
4681                          * the partition.)
4682                          */
4683                         if (sbi->s_journal)
4684                                 ext4_clear_journal_err(sb, es);
4685                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4686                         if (!ext4_setup_super(sb, es, 0))
4687                                 sb->s_flags &= ~MS_RDONLY;
4688                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4689                                                      EXT4_FEATURE_INCOMPAT_MMP))
4690                                 if (ext4_multi_mount_protect(sb,
4691                                                 le64_to_cpu(es->s_mmp_block))) {
4692                                         err = -EROFS;
4693                                         goto restore_opts;
4694                                 }
4695                         enable_quota = 1;
4696                 }
4697         }
4698
4699         /*
4700          * Reinitialize lazy itable initialization thread based on
4701          * current settings
4702          */
4703         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4704                 ext4_unregister_li_request(sb);
4705         else {
4706                 ext4_group_t first_not_zeroed;
4707                 first_not_zeroed = ext4_has_uninit_itable(sb);
4708                 ext4_register_li_request(sb, first_not_zeroed);
4709         }
4710
4711         ext4_setup_system_zone(sb);
4712         if (sbi->s_journal == NULL)
4713                 ext4_commit_super(sb, 1);
4714
4715 #ifdef CONFIG_QUOTA
4716         /* Release old quota file names */
4717         for (i = 0; i < MAXQUOTAS; i++)
4718                 if (old_opts.s_qf_names[i] &&
4719                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4720                         kfree(old_opts.s_qf_names[i]);
4721         if (enable_quota) {
4722                 if (sb_any_quota_suspended(sb))
4723                         dquot_resume(sb, -1);
4724                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4725                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4726                         err = ext4_enable_quotas(sb);
4727                         if (err)
4728                                 goto restore_opts;
4729                 }
4730         }
4731 #endif
4732
4733         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4734         kfree(orig_data);
4735         return 0;
4736
4737 restore_opts:
4738         sb->s_flags = old_sb_flags;
4739         sbi->s_mount_opt = old_opts.s_mount_opt;
4740         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4741         sbi->s_resuid = old_opts.s_resuid;
4742         sbi->s_resgid = old_opts.s_resgid;
4743         sbi->s_commit_interval = old_opts.s_commit_interval;
4744         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4745         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4746 #ifdef CONFIG_QUOTA
4747         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4748         for (i = 0; i < MAXQUOTAS; i++) {
4749                 if (sbi->s_qf_names[i] &&
4750                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4751                         kfree(sbi->s_qf_names[i]);
4752                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4753         }
4754 #endif
4755         kfree(orig_data);
4756         return err;
4757 }
4758
4759 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4760 {
4761         struct super_block *sb = dentry->d_sb;
4762         struct ext4_sb_info *sbi = EXT4_SB(sb);
4763         struct ext4_super_block *es = sbi->s_es;
4764         ext4_fsblk_t overhead = 0;
4765         u64 fsid;
4766         s64 bfree;
4767
4768         if (!test_opt(sb, MINIX_DF))
4769                 overhead = sbi->s_overhead;
4770
4771         buf->f_type = EXT4_SUPER_MAGIC;
4772         buf->f_bsize = sb->s_blocksize;
4773         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, sbi->s_overhead);
4774         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4775                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4776         /* prevent underflow in case that few free space is available */
4777         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4778         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4779         if (buf->f_bfree < ext4_r_blocks_count(es))
4780                 buf->f_bavail = 0;
4781         buf->f_files = le32_to_cpu(es->s_inodes_count);
4782         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4783         buf->f_namelen = EXT4_NAME_LEN;
4784         fsid = le64_to_cpup((void *)es->s_uuid) ^
4785                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4786         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4787         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4788
4789         return 0;
4790 }
4791
4792 /* Helper function for writing quotas on sync - we need to start transaction
4793  * before quota file is locked for write. Otherwise the are possible deadlocks:
4794  * Process 1                         Process 2
4795  * ext4_create()                     quota_sync()
4796  *   jbd2_journal_start()                  write_dquot()
4797  *   dquot_initialize()                         down(dqio_mutex)
4798  *     down(dqio_mutex)                    jbd2_journal_start()
4799  *
4800  */
4801
4802 #ifdef CONFIG_QUOTA
4803
4804 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4805 {
4806         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_type];
4807 }
4808
4809 static int ext4_write_dquot(struct dquot *dquot)
4810 {
4811         int ret, err;
4812         handle_t *handle;
4813         struct inode *inode;
4814
4815         inode = dquot_to_inode(dquot);
4816         handle = ext4_journal_start(inode,
4817                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4818         if (IS_ERR(handle))
4819                 return PTR_ERR(handle);
4820         ret = dquot_commit(dquot);
4821         err = ext4_journal_stop(handle);
4822         if (!ret)
4823                 ret = err;
4824         return ret;
4825 }
4826
4827 static int ext4_acquire_dquot(struct dquot *dquot)
4828 {
4829         int ret, err;
4830         handle_t *handle;
4831
4832         handle = ext4_journal_start(dquot_to_inode(dquot),
4833                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4834         if (IS_ERR(handle))
4835                 return PTR_ERR(handle);
4836         ret = dquot_acquire(dquot);
4837         err = ext4_journal_stop(handle);
4838         if (!ret)
4839                 ret = err;
4840         return ret;
4841 }
4842
4843 static int ext4_release_dquot(struct dquot *dquot)
4844 {
4845         int ret, err;
4846         handle_t *handle;
4847
4848         handle = ext4_journal_start(dquot_to_inode(dquot),
4849                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4850         if (IS_ERR(handle)) {
4851                 /* Release dquot anyway to avoid endless cycle in dqput() */
4852                 dquot_release(dquot);
4853                 return PTR_ERR(handle);
4854         }
4855         ret = dquot_release(dquot);
4856         err = ext4_journal_stop(handle);
4857         if (!ret)
4858                 ret = err;
4859         return ret;
4860 }
4861
4862 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4863 {
4864         /* Are we journaling quotas? */
4865         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4866             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4867                 dquot_mark_dquot_dirty(dquot);
4868                 return ext4_write_dquot(dquot);
4869         } else {
4870                 return dquot_mark_dquot_dirty(dquot);
4871         }
4872 }
4873
4874 static int ext4_write_info(struct super_block *sb, int type)
4875 {
4876         int ret, err;
4877         handle_t *handle;
4878
4879         /* Data block + inode block */
4880         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4881         if (IS_ERR(handle))
4882                 return PTR_ERR(handle);
4883         ret = dquot_commit_info(sb, type);
4884         err = ext4_journal_stop(handle);
4885         if (!ret)
4886                 ret = err;
4887         return ret;
4888 }
4889
4890 /*
4891  * Turn on quotas during mount time - we need to find
4892  * the quota file and such...
4893  */
4894 static int ext4_quota_on_mount(struct super_block *sb, int type)
4895 {
4896         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4897                                         EXT4_SB(sb)->s_jquota_fmt, type);
4898 }
4899
4900 /*
4901  * Standard function to be called on quota_on
4902  */
4903 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4904                          struct path *path)
4905 {
4906         int err;
4907
4908         if (!test_opt(sb, QUOTA))
4909                 return -EINVAL;
4910
4911         /* Quotafile not on the same filesystem? */
4912         if (path->dentry->d_sb != sb)
4913                 return -EXDEV;
4914         /* Journaling quota? */
4915         if (EXT4_SB(sb)->s_qf_names[type]) {
4916                 /* Quotafile not in fs root? */
4917                 if (path->dentry->d_parent != sb->s_root)
4918                         ext4_msg(sb, KERN_WARNING,
4919                                 "Quota file not on filesystem root. "
4920                                 "Journaled quota will not work");
4921         }
4922
4923         /*
4924          * When we journal data on quota file, we have to flush journal to see
4925          * all updates to the file when we bypass pagecache...
4926          */
4927         if (EXT4_SB(sb)->s_journal &&
4928             ext4_should_journal_data(path->dentry->d_inode)) {
4929                 /*
4930                  * We don't need to lock updates but journal_flush() could
4931                  * otherwise be livelocked...
4932                  */
4933                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4934                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4935                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4936                 if (err)
4937                         return err;
4938         }
4939
4940         return dquot_quota_on(sb, type, format_id, path);
4941 }
4942
4943 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
4944                              unsigned int flags)
4945 {
4946         int err;
4947         struct inode *qf_inode;
4948         unsigned long qf_inums[MAXQUOTAS] = {
4949                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4950                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4951         };
4952
4953         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
4954
4955         if (!qf_inums[type])
4956                 return -EPERM;
4957
4958         qf_inode = ext4_iget(sb, qf_inums[type]);
4959         if (IS_ERR(qf_inode)) {
4960                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
4961                 return PTR_ERR(qf_inode);
4962         }
4963
4964         err = dquot_enable(qf_inode, type, format_id, flags);
4965         iput(qf_inode);
4966
4967         return err;
4968 }
4969
4970 /* Enable usage tracking for all quota types. */
4971 static int ext4_enable_quotas(struct super_block *sb)
4972 {
4973         int type, err = 0;
4974         unsigned long qf_inums[MAXQUOTAS] = {
4975                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4976                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4977         };
4978
4979         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
4980         for (type = 0; type < MAXQUOTAS; type++) {
4981                 if (qf_inums[type]) {
4982                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
4983                                                 DQUOT_USAGE_ENABLED);
4984                         if (err) {
4985                                 ext4_warning(sb,
4986                                         "Failed to enable quota (type=%d) "
4987                                         "tracking. Please run e2fsck to fix.",
4988                                         type);
4989                                 return err;
4990                         }
4991                 }
4992         }
4993         return 0;
4994 }
4995
4996 /*
4997  * quota_on function that is used when QUOTA feature is set.
4998  */
4999 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5000                                  int format_id)
5001 {
5002         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5003                 return -EINVAL;
5004
5005         /*
5006          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5007          */
5008         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5009 }
5010
5011 static int ext4_quota_off(struct super_block *sb, int type)
5012 {
5013         struct inode *inode = sb_dqopt(sb)->files[type];
5014         handle_t *handle;
5015
5016         /* Force all delayed allocation blocks to be allocated.
5017          * Caller already holds s_umount sem */
5018         if (test_opt(sb, DELALLOC))
5019                 sync_filesystem(sb);
5020
5021         if (!inode)
5022                 goto out;
5023
5024         /* Update modification times of quota files when userspace can
5025          * start looking at them */
5026         handle = ext4_journal_start(inode, 1);
5027         if (IS_ERR(handle))
5028                 goto out;
5029         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5030         ext4_mark_inode_dirty(handle, inode);
5031         ext4_journal_stop(handle);
5032
5033 out:
5034         return dquot_quota_off(sb, type);
5035 }
5036
5037 /*
5038  * quota_off function that is used when QUOTA feature is set.
5039  */
5040 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5041 {
5042         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5043                 return -EINVAL;
5044
5045         /* Disable only the limits. */
5046         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5047 }
5048
5049 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5050  * acquiring the locks... As quota files are never truncated and quota code
5051  * itself serializes the operations (and no one else should touch the files)
5052  * we don't have to be afraid of races */
5053 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5054                                size_t len, loff_t off)
5055 {
5056         struct inode *inode = sb_dqopt(sb)->files[type];
5057         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5058         int err = 0;
5059         int offset = off & (sb->s_blocksize - 1);
5060         int tocopy;
5061         size_t toread;
5062         struct buffer_head *bh;
5063         loff_t i_size = i_size_read(inode);
5064
5065         if (off > i_size)
5066                 return 0;
5067         if (off+len > i_size)
5068                 len = i_size-off;
5069         toread = len;
5070         while (toread > 0) {
5071                 tocopy = sb->s_blocksize - offset < toread ?
5072                                 sb->s_blocksize - offset : toread;
5073                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5074                 if (err)
5075                         return err;
5076                 if (!bh)        /* A hole? */
5077                         memset(data, 0, tocopy);
5078                 else
5079                         memcpy(data, bh->b_data+offset, tocopy);
5080                 brelse(bh);
5081                 offset = 0;
5082                 toread -= tocopy;
5083                 data += tocopy;
5084                 blk++;
5085         }
5086         return len;
5087 }
5088
5089 /* Write to quotafile (we know the transaction is already started and has
5090  * enough credits) */
5091 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5092                                 const char *data, size_t len, loff_t off)
5093 {
5094         struct inode *inode = sb_dqopt(sb)->files[type];
5095         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5096         int err = 0;
5097         int offset = off & (sb->s_blocksize - 1);
5098         struct buffer_head *bh;
5099         handle_t *handle = journal_current_handle();
5100
5101         if (EXT4_SB(sb)->s_journal && !handle) {
5102                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5103                         " cancelled because transaction is not started",
5104                         (unsigned long long)off, (unsigned long long)len);
5105                 return -EIO;
5106         }
5107         /*
5108          * Since we account only one data block in transaction credits,
5109          * then it is impossible to cross a block boundary.
5110          */
5111         if (sb->s_blocksize - offset < len) {
5112                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5113                         " cancelled because not block aligned",
5114                         (unsigned long long)off, (unsigned long long)len);
5115                 return -EIO;
5116         }
5117
5118         bh = ext4_bread(handle, inode, blk, 1, &err);
5119         if (!bh)
5120                 goto out;
5121         err = ext4_journal_get_write_access(handle, bh);
5122         if (err) {
5123                 brelse(bh);
5124                 goto out;
5125         }
5126         lock_buffer(bh);
5127         memcpy(bh->b_data+offset, data, len);
5128         flush_dcache_page(bh->b_page);
5129         unlock_buffer(bh);
5130         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5131         brelse(bh);
5132 out:
5133         if (err)
5134                 return err;
5135         if (inode->i_size < off + len) {
5136                 i_size_write(inode, off + len);
5137                 EXT4_I(inode)->i_disksize = inode->i_size;
5138                 ext4_mark_inode_dirty(handle, inode);
5139         }
5140         return len;
5141 }
5142
5143 #endif
5144
5145 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5146                        const char *dev_name, void *data)
5147 {
5148         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5149 }
5150
5151 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5152 static inline void register_as_ext2(void)
5153 {
5154         int err = register_filesystem(&ext2_fs_type);
5155         if (err)
5156                 printk(KERN_WARNING
5157                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5158 }
5159
5160 static inline void unregister_as_ext2(void)
5161 {
5162         unregister_filesystem(&ext2_fs_type);
5163 }
5164
5165 static inline int ext2_feature_set_ok(struct super_block *sb)
5166 {
5167         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5168                 return 0;
5169         if (sb->s_flags & MS_RDONLY)
5170                 return 1;
5171         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5172                 return 0;
5173         return 1;
5174 }
5175 MODULE_ALIAS("ext2");
5176 #else
5177 static inline void register_as_ext2(void) { }
5178 static inline void unregister_as_ext2(void) { }
5179 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5180 #endif
5181
5182 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5183 static inline void register_as_ext3(void)
5184 {
5185         int err = register_filesystem(&ext3_fs_type);
5186         if (err)
5187                 printk(KERN_WARNING
5188                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5189 }
5190
5191 static inline void unregister_as_ext3(void)
5192 {
5193         unregister_filesystem(&ext3_fs_type);
5194 }
5195
5196 static inline int ext3_feature_set_ok(struct super_block *sb)
5197 {
5198         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5199                 return 0;
5200         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5201                 return 0;
5202         if (sb->s_flags & MS_RDONLY)
5203                 return 1;
5204         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5205                 return 0;
5206         return 1;
5207 }
5208 MODULE_ALIAS("ext3");
5209 #else
5210 static inline void register_as_ext3(void) { }
5211 static inline void unregister_as_ext3(void) { }
5212 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5213 #endif
5214
5215 static struct file_system_type ext4_fs_type = {
5216         .owner          = THIS_MODULE,
5217         .name           = "ext4",
5218         .mount          = ext4_mount,
5219         .kill_sb        = kill_block_super,
5220         .fs_flags       = FS_REQUIRES_DEV,
5221 };
5222
5223 static int __init ext4_init_feat_adverts(void)
5224 {
5225         struct ext4_features *ef;
5226         int ret = -ENOMEM;
5227
5228         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5229         if (!ef)
5230                 goto out;
5231
5232         ef->f_kobj.kset = ext4_kset;
5233         init_completion(&ef->f_kobj_unregister);
5234         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5235                                    "features");
5236         if (ret) {
5237                 kfree(ef);
5238                 goto out;
5239         }
5240
5241         ext4_feat = ef;
5242         ret = 0;
5243 out:
5244         return ret;
5245 }
5246
5247 static void ext4_exit_feat_adverts(void)
5248 {
5249         kobject_put(&ext4_feat->f_kobj);
5250         wait_for_completion(&ext4_feat->f_kobj_unregister);
5251         kfree(ext4_feat);
5252 }
5253
5254 /* Shared across all ext4 file systems */
5255 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5256 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5257
5258 static int __init ext4_init_fs(void)
5259 {
5260         int i, err;
5261
5262         ext4_li_info = NULL;
5263         mutex_init(&ext4_li_mtx);
5264
5265         ext4_check_flag_values();
5266
5267         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5268                 mutex_init(&ext4__aio_mutex[i]);
5269                 init_waitqueue_head(&ext4__ioend_wq[i]);
5270         }
5271
5272         err = ext4_init_pageio();
5273         if (err)
5274                 return err;
5275         err = ext4_init_system_zone();
5276         if (err)
5277                 goto out6;
5278         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5279         if (!ext4_kset) {
5280                 err = -ENOMEM;
5281                 goto out5;
5282         }
5283         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5284
5285         err = ext4_init_feat_adverts();
5286         if (err)
5287                 goto out4;
5288
5289         err = ext4_init_mballoc();
5290         if (err)
5291                 goto out3;
5292
5293         err = ext4_init_xattr();
5294         if (err)
5295                 goto out2;
5296         err = init_inodecache();
5297         if (err)
5298                 goto out1;
5299         register_as_ext3();
5300         register_as_ext2();
5301         err = register_filesystem(&ext4_fs_type);
5302         if (err)
5303                 goto out;
5304
5305         return 0;
5306 out:
5307         unregister_as_ext2();
5308         unregister_as_ext3();
5309         destroy_inodecache();
5310 out1:
5311         ext4_exit_xattr();
5312 out2:
5313         ext4_exit_mballoc();
5314 out3:
5315         ext4_exit_feat_adverts();
5316 out4:
5317         if (ext4_proc_root)
5318                 remove_proc_entry("fs/ext4", NULL);
5319         kset_unregister(ext4_kset);
5320 out5:
5321         ext4_exit_system_zone();
5322 out6:
5323         ext4_exit_pageio();
5324         return err;
5325 }
5326
5327 static void __exit ext4_exit_fs(void)
5328 {
5329         ext4_destroy_lazyinit_thread();
5330         unregister_as_ext2();
5331         unregister_as_ext3();
5332         unregister_filesystem(&ext4_fs_type);
5333         destroy_inodecache();
5334         ext4_exit_xattr();
5335         ext4_exit_mballoc();
5336         ext4_exit_feat_adverts();
5337         remove_proc_entry("fs/ext4", NULL);
5338         kset_unregister(ext4_kset);
5339         ext4_exit_system_zone();
5340         ext4_exit_pageio();
5341 }
5342
5343 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5344 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5345 MODULE_LICENSE("GPL");
5346 module_init(ext4_init_fs)
5347 module_exit(ext4_exit_fs)