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"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static const char *ext4_decode_error(struct super_block *sb, int errno,
  73                                      char nbuf[16]);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  96 #else
  97 #define IS_EXT2_SB(sb) (0)
  98 #endif
  99
 100
 101 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 102 static struct file_system_type ext3_fs_type = {
 103         .owner          = THIS_MODULE,
 104         .name           = "ext3",
 105         .mount          = ext4_mount,
 106         .kill_sb        = kill_block_super,
 107         .fs_flags       = FS_REQUIRES_DEV,
 108 };
 109 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 110 #else
 111 #define IS_EXT3_SB(sb) (0)
 112 #endif
 113
 114 static int ext4_verify_csum_type(struct super_block *sb,
 115                                  struct ext4_super_block *es)
 116 {
 117         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 118                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 119                 return 1;
 120
 121         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 122 }
 123
 124 static __le32 ext4_superblock_csum(struct super_block *sb,
 125                                    struct ext4_super_block *es)
 126 {
 127         struct ext4_sb_info *sbi = EXT4_SB(sb);
 128         int offset = offsetof(struct ext4_super_block, s_checksum);
 129         __u32 csum;
 130
 131         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 132
 133         return cpu_to_le32(csum);
 134 }
 135
 136 int ext4_superblock_csum_verify(struct super_block *sb,
 137                                 struct ext4_super_block *es)
 138 {
 139         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 140                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 141                 return 1;
 142
 143         return es->s_checksum == ext4_superblock_csum(sb, es);
 144 }
 145
 146 void ext4_superblock_csum_set(struct super_block *sb)
 147 {
 148         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 149
 150         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 151                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 152                 return;
 153
 154         es->s_checksum = ext4_superblock_csum(sb, es);
 155 }
 156
 157 void *ext4_kvmalloc(size_t size, gfp_t flags)
 158 {
 159         void *ret;
 160
 161         ret = kmalloc(size, flags);
 162         if (!ret)
 163                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 164         return ret;
 165 }
 166
 167 void *ext4_kvzalloc(size_t size, gfp_t flags)
 168 {
 169         void *ret;
 170
 171         ret = kzalloc(size, flags);
 172         if (!ret)
 173                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 174         return ret;
 175 }
 176
 177 void ext4_kvfree(void *ptr)
 178 {
 179         if (is_vmalloc_addr(ptr))
 180                 vfree(ptr);
 181         else
 182                 kfree(ptr);
 183
 184 }
 185
 186 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 187                                struct ext4_group_desc *bg)
 188 {
 189         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 190                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 191                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 192 }
 193
 194 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 195                                struct ext4_group_desc *bg)
 196 {
 197         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 198                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 199                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 200 }
 201
 202 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 203                               struct ext4_group_desc *bg)
 204 {
 205         return le32_to_cpu(bg->bg_inode_table_lo) |
 206                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 207                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 208 }
 209
 210 __u32 ext4_free_group_clusters(struct super_block *sb,
 211                                struct ext4_group_desc *bg)
 212 {
 213         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 214                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 215                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 216 }
 217
 218 __u32 ext4_free_inodes_count(struct super_block *sb,
 219                               struct ext4_group_desc *bg)
 220 {
 221         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 222                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 223                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 224 }
 225
 226 __u32 ext4_used_dirs_count(struct super_block *sb,
 227                               struct ext4_group_desc *bg)
 228 {
 229         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 230                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 231                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 232 }
 233
 234 __u32 ext4_itable_unused_count(struct super_block *sb,
 235                               struct ext4_group_desc *bg)
 236 {
 237         return le16_to_cpu(bg->bg_itable_unused_lo) |
 238                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 239                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 240 }
 241
 242 void ext4_block_bitmap_set(struct super_block *sb,
 243                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 244 {
 245         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 246         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 247                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 248 }
 249
 250 void ext4_inode_bitmap_set(struct super_block *sb,
 251                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 252 {
 253         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 254         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 255                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 256 }
 257
 258 void ext4_inode_table_set(struct super_block *sb,
 259                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 260 {
 261         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 262         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 263                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 264 }
 265
 266 void ext4_free_group_clusters_set(struct super_block *sb,
 267                                   struct ext4_group_desc *bg, __u32 count)
 268 {
 269         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 270         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 271                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 272 }
 273
 274 void ext4_free_inodes_set(struct super_block *sb,
 275                           struct ext4_group_desc *bg, __u32 count)
 276 {
 277         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 278         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 279                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 280 }
 281
 282 void ext4_used_dirs_set(struct super_block *sb,
 283                           struct ext4_group_desc *bg, __u32 count)
 284 {
 285         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 286         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 287                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 288 }
 289
 290 void ext4_itable_unused_set(struct super_block *sb,
 291                           struct ext4_group_desc *bg, __u32 count)
 292 {
 293         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 294         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 295                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 296 }
 297
 298
 299 /* Just increment the non-pointer handle value */
 300 static handle_t *ext4_get_nojournal(void)
 301 {
 302         handle_t *handle = current->journal_info;
 303         unsigned long ref_cnt = (unsigned long)handle;
 304
 305         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 306
 307         ref_cnt++;
 308         handle = (handle_t *)ref_cnt;
 309
 310         current->journal_info = handle;
 311         return handle;
 312 }
 313
 314
 315 /* Decrement the non-pointer handle value */
 316 static void ext4_put_nojournal(handle_t *handle)
 317 {
 318         unsigned long ref_cnt = (unsigned long)handle;
 319
 320         BUG_ON(ref_cnt == 0);
 321
 322         ref_cnt--;
 323         handle = (handle_t *)ref_cnt;
 324
 325         current->journal_info = handle;
 326 }
 327
 328 /*
 329  * Wrappers for jbd2_journal_start/end.
 330  */
 331 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 332 {
 333         journal_t *journal;
 334
 335         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 336         if (sb->s_flags & MS_RDONLY)
 337                 return ERR_PTR(-EROFS);
 338
 339         WARN_ON(sb->s_writers.frozen == SB_FREEZE_COMPLETE);
 340         journal = EXT4_SB(sb)->s_journal;
 341         if (!journal)
 342                 return ext4_get_nojournal();
 343         /*
 344          * Special case here: if the journal has aborted behind our
 345          * backs (eg. EIO in the commit thread), then we still need to
 346          * take the FS itself readonly cleanly.
 347          */
 348         if (is_journal_aborted(journal)) {
 349                 ext4_abort(sb, "Detected aborted journal");
 350                 return ERR_PTR(-EROFS);
 351         }
 352         return jbd2_journal_start(journal, nblocks);
 353 }
 354
 355 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 356 {
 357         struct super_block *sb;
 358         int err;
 359         int rc;
 360
 361         if (!ext4_handle_valid(handle)) {
 362                 ext4_put_nojournal(handle);
 363                 return 0;
 364         }
 365         sb = handle->h_transaction->t_journal->j_private;
 366         err = handle->h_err;
 367         rc = jbd2_journal_stop(handle);
 368
 369         if (!err)
 370                 err = rc;
 371         if (err)
 372                 __ext4_std_error(sb, where, line, err);
 373         return err;
 374 }
 375
 376 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 377                                const char *err_fn, struct buffer_head *bh,
 378                                handle_t *handle, int err)
 379 {
 380         char nbuf[16];
 381         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 382
 383         BUG_ON(!ext4_handle_valid(handle));
 384
 385         if (bh)
 386                 BUFFER_TRACE(bh, "abort");
 387
 388         if (!handle->h_err)
 389                 handle->h_err = err;
 390
 391         if (is_handle_aborted(handle))
 392                 return;
 393
 394         printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
 395                caller, line, errstr, err_fn);
 396
 397         jbd2_journal_abort_handle(handle);
 398 }
 399
 400 static void __save_error_info(struct super_block *sb, const char *func,
 401                             unsigned int line)
 402 {
 403         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 404
 405         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 406         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 407         es->s_last_error_time = cpu_to_le32(get_seconds());
 408         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 409         es->s_last_error_line = cpu_to_le32(line);
 410         if (!es->s_first_error_time) {
 411                 es->s_first_error_time = es->s_last_error_time;
 412                 strncpy(es->s_first_error_func, func,
 413                         sizeof(es->s_first_error_func));
 414                 es->s_first_error_line = cpu_to_le32(line);
 415                 es->s_first_error_ino = es->s_last_error_ino;
 416                 es->s_first_error_block = es->s_last_error_block;
 417         }
 418         /*
 419          * Start the daily error reporting function if it hasn't been
 420          * started already
 421          */
 422         if (!es->s_error_count)
 423                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 424         le32_add_cpu(&es->s_error_count, 1);
 425 }
 426
 427 static void save_error_info(struct super_block *sb, const char *func,
 428                             unsigned int line)
 429 {
 430         __save_error_info(sb, func, line);
 431         ext4_commit_super(sb, 1);
 432 }
 433
 434 /*
 435  * The del_gendisk() function uninitializes the disk-specific data
 436  * structures, including the bdi structure, without telling anyone
 437  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 438  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 439  * This is a kludge to prevent these oops until we can put in a proper
 440  * hook in del_gendisk() to inform the VFS and file system layers.
 441  */
 442 static int block_device_ejected(struct super_block *sb)
 443 {
 444         struct inode *bd_inode = sb->s_bdev->bd_inode;
 445         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 446
 447         return bdi->dev == NULL;
 448 }
 449
 450 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 451 {
 452         struct super_block              *sb = journal->j_private;
 453         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 454         int                             error = is_journal_aborted(journal);
 455         struct ext4_journal_cb_entry    *jce, *tmp;
 456
 457         spin_lock(&sbi->s_md_lock);
 458         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 459                 list_del_init(&jce->jce_list);
 460                 spin_unlock(&sbi->s_md_lock);
 461                 jce->jce_func(sb, jce, error);
 462                 spin_lock(&sbi->s_md_lock);
 463         }
 464         spin_unlock(&sbi->s_md_lock);
 465 }
 466
 467 /* Deal with the reporting of failure conditions on a filesystem such as
 468  * inconsistencies detected or read IO failures.
 469  *
 470  * On ext2, we can store the error state of the filesystem in the
 471  * superblock.  That is not possible on ext4, because we may have other
 472  * write ordering constraints on the superblock which prevent us from
 473  * writing it out straight away; and given that the journal is about to
 474  * be aborted, we can't rely on the current, or future, transactions to
 475  * write out the superblock safely.
 476  *
 477  * We'll just use the jbd2_journal_abort() error code to record an error in
 478  * the journal instead.  On recovery, the journal will complain about
 479  * that error until we've noted it down and cleared it.
 480  */
 481
 482 static void ext4_handle_error(struct super_block *sb)
 483 {
 484         if (sb->s_flags & MS_RDONLY)
 485                 return;
 486
 487         if (!test_opt(sb, ERRORS_CONT)) {
 488                 journal_t *journal = EXT4_SB(sb)->s_journal;
 489
 490                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 491                 if (journal)
 492                         jbd2_journal_abort(journal, -EIO);
 493         }
 494         if (test_opt(sb, ERRORS_RO)) {
 495                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 496                 sb->s_flags |= MS_RDONLY;
 497         }
 498         if (test_opt(sb, ERRORS_PANIC))
 499                 panic("EXT4-fs (device %s): panic forced after error\n",
 500                         sb->s_id);
 501 }
 502
 503 void __ext4_error(struct super_block *sb, const char *function,
 504                   unsigned int line, const char *fmt, ...)
 505 {
 506         struct va_format vaf;
 507         va_list args;
 508
 509         va_start(args, fmt);
 510         vaf.fmt = fmt;
 511         vaf.va = &args;
 512         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 513                sb->s_id, function, line, current->comm, &vaf);
 514         va_end(args);
 515         save_error_info(sb, function, line);
 516
 517         ext4_handle_error(sb);
 518 }
 519
 520 void ext4_error_inode(struct inode *inode, const char *function,
 521                       unsigned int line, ext4_fsblk_t block,
 522                       const char *fmt, ...)
 523 {
 524         va_list args;
 525         struct va_format vaf;
 526         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 527
 528         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 529         es->s_last_error_block = cpu_to_le64(block);
 530         save_error_info(inode->i_sb, function, line);
 531         va_start(args, fmt);
 532         vaf.fmt = fmt;
 533         vaf.va = &args;
 534         if (block)
 535                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 536                        "inode #%lu: block %llu: comm %s: %pV\n",
 537                        inode->i_sb->s_id, function, line, inode->i_ino,
 538                        block, current->comm, &vaf);
 539         else
 540                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 541                        "inode #%lu: comm %s: %pV\n",
 542                        inode->i_sb->s_id, function, line, inode->i_ino,
 543                        current->comm, &vaf);
 544         va_end(args);
 545
 546         ext4_handle_error(inode->i_sb);
 547 }
 548
 549 void ext4_error_file(struct file *file, const char *function,
 550                      unsigned int line, ext4_fsblk_t block,
 551                      const char *fmt, ...)
 552 {
 553         va_list args;
 554         struct va_format vaf;
 555         struct ext4_super_block *es;
 556         struct inode *inode = file->f_dentry->d_inode;
 557         char pathname[80], *path;
 558
 559         es = EXT4_SB(inode->i_sb)->s_es;
 560         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 561         save_error_info(inode->i_sb, function, line);
 562         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 563         if (IS_ERR(path))
 564                 path = "(unknown)";
 565         va_start(args, fmt);
 566         vaf.fmt = fmt;
 567         vaf.va = &args;
 568         if (block)
 569                 printk(KERN_CRIT
 570                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 571                        "block %llu: comm %s: path %s: %pV\n",
 572                        inode->i_sb->s_id, function, line, inode->i_ino,
 573                        block, current->comm, path, &vaf);
 574         else
 575                 printk(KERN_CRIT
 576                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 577                        "comm %s: path %s: %pV\n",
 578                        inode->i_sb->s_id, function, line, inode->i_ino,
 579                        current->comm, path, &vaf);
 580         va_end(args);
 581
 582         ext4_handle_error(inode->i_sb);
 583 }
 584
 585 static const char *ext4_decode_error(struct super_block *sb, int errno,
 586                                      char nbuf[16])
 587 {
 588         char *errstr = NULL;
 589
 590         switch (errno) {
 591         case -EIO:
 592                 errstr = "IO failure";
 593                 break;
 594         case -ENOMEM:
 595                 errstr = "Out of memory";
 596                 break;
 597         case -EROFS:
 598                 if (!sb || (EXT4_SB(sb)->s_journal &&
 599                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 600                         errstr = "Journal has aborted";
 601                 else
 602                         errstr = "Readonly filesystem";
 603                 break;
 604         default:
 605                 /* If the caller passed in an extra buffer for unknown
 606                  * errors, textualise them now.  Else we just return
 607                  * NULL. */
 608                 if (nbuf) {
 609                         /* Check for truncated error codes... */
 610                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 611                                 errstr = nbuf;
 612                 }
 613                 break;
 614         }
 615
 616         return errstr;
 617 }
 618
 619 /* __ext4_std_error decodes expected errors from journaling functions
 620  * automatically and invokes the appropriate error response.  */
 621
 622 void __ext4_std_error(struct super_block *sb, const char *function,
 623                       unsigned int line, int errno)
 624 {
 625         char nbuf[16];
 626         const char *errstr;
 627
 628         /* Special case: if the error is EROFS, and we're not already
 629          * inside a transaction, then there's really no point in logging
 630          * an error. */
 631         if (errno == -EROFS && journal_current_handle() == NULL &&
 632             (sb->s_flags & MS_RDONLY))
 633                 return;
 634
 635         errstr = ext4_decode_error(sb, errno, nbuf);
 636         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 637                sb->s_id, function, line, errstr);
 638         save_error_info(sb, function, line);
 639
 640         ext4_handle_error(sb);
 641 }
 642
 643 /*
 644  * ext4_abort is a much stronger failure handler than ext4_error.  The
 645  * abort function may be used to deal with unrecoverable failures such
 646  * as journal IO errors or ENOMEM at a critical moment in log management.
 647  *
 648  * We unconditionally force the filesystem into an ABORT|READONLY state,
 649  * unless the error response on the fs has been set to panic in which
 650  * case we take the easy way out and panic immediately.
 651  */
 652
 653 void __ext4_abort(struct super_block *sb, const char *function,
 654                 unsigned int line, const char *fmt, ...)
 655 {
 656         va_list args;
 657
 658         save_error_info(sb, function, line);
 659         va_start(args, fmt);
 660         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 661                function, line);
 662         vprintk(fmt, args);
 663         printk("\n");
 664         va_end(args);
 665
 666         if ((sb->s_flags & MS_RDONLY) == 0) {
 667                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 668                 sb->s_flags |= MS_RDONLY;
 669                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 670                 if (EXT4_SB(sb)->s_journal)
 671                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 672                 save_error_info(sb, function, line);
 673         }
 674         if (test_opt(sb, ERRORS_PANIC))
 675                 panic("EXT4-fs panic from previous error\n");
 676 }
 677
 678 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 679 {
 680         struct va_format vaf;
 681         va_list args;
 682
 683         va_start(args, fmt);
 684         vaf.fmt = fmt;
 685         vaf.va = &args;
 686         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 687         va_end(args);
 688 }
 689
 690 void __ext4_warning(struct super_block *sb, const char *function,
 691                     unsigned int line, const char *fmt, ...)
 692 {
 693         struct va_format vaf;
 694         va_list args;
 695
 696         va_start(args, fmt);
 697         vaf.fmt = fmt;
 698         vaf.va = &args;
 699         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 700                sb->s_id, function, line, &vaf);
 701         va_end(args);
 702 }
 703
 704 void __ext4_grp_locked_error(const char *function, unsigned int line,
 705                              struct super_block *sb, ext4_group_t grp,
 706                              unsigned long ino, ext4_fsblk_t block,
 707                              const char *fmt, ...)
 708 __releases(bitlock)
 709 __acquires(bitlock)
 710 {
 711         struct va_format vaf;
 712         va_list args;
 713         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 714
 715         es->s_last_error_ino = cpu_to_le32(ino);
 716         es->s_last_error_block = cpu_to_le64(block);
 717         __save_error_info(sb, function, line);
 718
 719         va_start(args, fmt);
 720
 721         vaf.fmt = fmt;
 722         vaf.va = &args;
 723         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 724                sb->s_id, function, line, grp);
 725         if (ino)
 726                 printk(KERN_CONT "inode %lu: ", ino);
 727         if (block)
 728                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 729         printk(KERN_CONT "%pV\n", &vaf);
 730         va_end(args);
 731
 732         if (test_opt(sb, ERRORS_CONT)) {
 733                 ext4_commit_super(sb, 0);
 734                 return;
 735         }
 736
 737         ext4_unlock_group(sb, grp);
 738         ext4_handle_error(sb);
 739         /*
 740          * We only get here in the ERRORS_RO case; relocking the group
 741          * may be dangerous, but nothing bad will happen since the
 742          * filesystem will have already been marked read/only and the
 743          * journal has been aborted.  We return 1 as a hint to callers
 744          * who might what to use the return value from
 745          * ext4_grp_locked_error() to distinguish between the
 746          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 747          * aggressively from the ext4 function in question, with a
 748          * more appropriate error code.
 749          */
 750         ext4_lock_group(sb, grp);
 751         return;
 752 }
 753
 754 void ext4_update_dynamic_rev(struct super_block *sb)
 755 {
 756         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 757
 758         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 759                 return;
 760
 761         ext4_warning(sb,
 762                      "updating to rev %d because of new feature flag, "
 763                      "running e2fsck is recommended",
 764                      EXT4_DYNAMIC_REV);
 765
 766         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 767         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 768         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 769         /* leave es->s_feature_*compat flags alone */
 770         /* es->s_uuid will be set by e2fsck if empty */
 771
 772         /*
 773          * The rest of the superblock fields should be zero, and if not it
 774          * means they are likely already in use, so leave them alone.  We
 775          * can leave it up to e2fsck to clean up any inconsistencies there.
 776          */
 777 }
 778
 779 /*
 780  * Open the external journal device
 781  */
 782 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 783 {
 784         struct block_device *bdev;
 785         char b[BDEVNAME_SIZE];
 786
 787         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 788         if (IS_ERR(bdev))
 789                 goto fail;
 790         return bdev;
 791
 792 fail:
 793         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 794                         __bdevname(dev, b), PTR_ERR(bdev));
 795         return NULL;
 796 }
 797
 798 /*
 799  * Release the journal device
 800  */
 801 static int ext4_blkdev_put(struct block_device *bdev)
 802 {
 803         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 804 }
 805
 806 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 807 {
 808         struct block_device *bdev;
 809         int ret = -ENODEV;
 810
 811         bdev = sbi->journal_bdev;
 812         if (bdev) {
 813                 ret = ext4_blkdev_put(bdev);
 814                 sbi->journal_bdev = NULL;
 815         }
 816         return ret;
 817 }
 818
 819 static inline struct inode *orphan_list_entry(struct list_head *l)
 820 {
 821         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 822 }
 823
 824 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 825 {
 826         struct list_head *l;
 827
 828         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 829                  le32_to_cpu(sbi->s_es->s_last_orphan));
 830
 831         printk(KERN_ERR "sb_info orphan list:\n");
 832         list_for_each(l, &sbi->s_orphan) {
 833                 struct inode *inode = orphan_list_entry(l);
 834                 printk(KERN_ERR "  "
 835                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 836                        inode->i_sb->s_id, inode->i_ino, inode,
 837                        inode->i_mode, inode->i_nlink,
 838                        NEXT_ORPHAN(inode));
 839         }
 840 }
 841
 842 static void ext4_put_super(struct super_block *sb)
 843 {
 844         struct ext4_sb_info *sbi = EXT4_SB(sb);
 845         struct ext4_super_block *es = sbi->s_es;
 846         int i, err;
 847
 848         ext4_unregister_li_request(sb);
 849         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 850
 851         flush_workqueue(sbi->dio_unwritten_wq);
 852         destroy_workqueue(sbi->dio_unwritten_wq);
 853
 854         if (sbi->s_journal) {
 855                 err = jbd2_journal_destroy(sbi->s_journal);
 856                 sbi->s_journal = NULL;
 857                 if (err < 0)
 858                         ext4_abort(sb, "Couldn't clean up the journal");
 859         }
 860
 861         del_timer(&sbi->s_err_report);
 862         ext4_release_system_zone(sb);
 863         ext4_mb_release(sb);
 864         ext4_ext_release(sb);
 865         ext4_xattr_put_super(sb);
 866
 867         if (!(sb->s_flags & MS_RDONLY)) {
 868                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 869                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 870         }
 871         if (!(sb->s_flags & MS_RDONLY))
 872                 ext4_commit_super(sb, 1);
 873
 874         if (sbi->s_proc) {
 875                 remove_proc_entry("options", sbi->s_proc);
 876                 remove_proc_entry(sb->s_id, ext4_proc_root);
 877         }
 878         kobject_del(&sbi->s_kobj);
 879
 880         for (i = 0; i < sbi->s_gdb_count; i++)
 881                 brelse(sbi->s_group_desc[i]);
 882         ext4_kvfree(sbi->s_group_desc);
 883         ext4_kvfree(sbi->s_flex_groups);
 884         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 885         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 886         percpu_counter_destroy(&sbi->s_dirs_counter);
 887         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 888         brelse(sbi->s_sbh);
 889 #ifdef CONFIG_QUOTA
 890         for (i = 0; i < MAXQUOTAS; i++)
 891                 kfree(sbi->s_qf_names[i]);
 892 #endif
 893
 894         /* Debugging code just in case the in-memory inode orphan list
 895          * isn't empty.  The on-disk one can be non-empty if we've
 896          * detected an error and taken the fs readonly, but the
 897          * in-memory list had better be clean by this point. */
 898         if (!list_empty(&sbi->s_orphan))
 899                 dump_orphan_list(sb, sbi);
 900         J_ASSERT(list_empty(&sbi->s_orphan));
 901
 902         invalidate_bdev(sb->s_bdev);
 903         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 904                 /*
 905                  * Invalidate the journal device's buffers.  We don't want them
 906                  * floating about in memory - the physical journal device may
 907                  * hotswapped, and it breaks the `ro-after' testing code.
 908                  */
 909                 sync_blockdev(sbi->journal_bdev);
 910                 invalidate_bdev(sbi->journal_bdev);
 911                 ext4_blkdev_remove(sbi);
 912         }
 913         if (sbi->s_mmp_tsk)
 914                 kthread_stop(sbi->s_mmp_tsk);
 915         sb->s_fs_info = NULL;
 916         /*
 917          * Now that we are completely done shutting down the
 918          * superblock, we need to actually destroy the kobject.
 919          */
 920         kobject_put(&sbi->s_kobj);
 921         wait_for_completion(&sbi->s_kobj_unregister);
 922         if (sbi->s_chksum_driver)
 923                 crypto_free_shash(sbi->s_chksum_driver);
 924         kfree(sbi->s_blockgroup_lock);
 925         kfree(sbi);
 926 }
 927
 928 static struct kmem_cache *ext4_inode_cachep;
 929
 930 /*
 931  * Called inside transaction, so use GFP_NOFS
 932  */
 933 static struct inode *ext4_alloc_inode(struct super_block *sb)
 934 {
 935         struct ext4_inode_info *ei;
 936
 937         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 938         if (!ei)
 939                 return NULL;
 940
 941         ei->vfs_inode.i_version = 1;
 942         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 943         INIT_LIST_HEAD(&ei->i_prealloc_list);
 944         spin_lock_init(&ei->i_prealloc_lock);
 945         ext4_es_init_tree(&ei->i_es_tree);
 946         rwlock_init(&ei->i_es_lock);
 947         ei->i_reserved_data_blocks = 0;
 948         ei->i_reserved_meta_blocks = 0;
 949         ei->i_allocated_meta_blocks = 0;
 950         ei->i_da_metadata_calc_len = 0;
 951         ei->i_da_metadata_calc_last_lblock = 0;
 952         spin_lock_init(&(ei->i_block_reservation_lock));
 953 #ifdef CONFIG_QUOTA
 954         ei->i_reserved_quota = 0;
 955 #endif
 956         ei->jinode = NULL;
 957         INIT_LIST_HEAD(&ei->i_completed_io_list);
 958         spin_lock_init(&ei->i_completed_io_lock);
 959         ei->i_sync_tid = 0;
 960         ei->i_datasync_tid = 0;
 961         atomic_set(&ei->i_ioend_count, 0);
 962         atomic_set(&ei->i_unwritten, 0);
 963
 964         return &ei->vfs_inode;
 965 }
 966
 967 static int ext4_drop_inode(struct inode *inode)
 968 {
 969         int drop = generic_drop_inode(inode);
 970
 971         trace_ext4_drop_inode(inode, drop);
 972         return drop;
 973 }
 974
 975 static void ext4_i_callback(struct rcu_head *head)
 976 {
 977         struct inode *inode = container_of(head, struct inode, i_rcu);
 978         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 979 }
 980
 981 static void ext4_destroy_inode(struct inode *inode)
 982 {
 983         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 984                 ext4_msg(inode->i_sb, KERN_ERR,
 985                          "Inode %lu (%p): orphan list check failed!",
 986                          inode->i_ino, EXT4_I(inode));
 987                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 988                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 989                                 true);
 990                 dump_stack();
 991         }
 992         call_rcu(&inode->i_rcu, ext4_i_callback);
 993 }
 994
 995 static void init_once(void *foo)
 996 {
 997         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 998
 999         INIT_LIST_HEAD(&ei->i_orphan);
1000         init_rwsem(&ei->xattr_sem);
1001         init_rwsem(&ei->i_data_sem);
1002         inode_init_once(&ei->vfs_inode);
1003 }
1004
1005 static int init_inodecache(void)
1006 {
1007         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1008                                              sizeof(struct ext4_inode_info),
1009                                              0, (SLAB_RECLAIM_ACCOUNT|
1010                                                 SLAB_MEM_SPREAD),
1011                                              init_once);
1012         if (ext4_inode_cachep == NULL)
1013                 return -ENOMEM;
1014         return 0;
1015 }
1016
1017 static void destroy_inodecache(void)
1018 {
1019         /*
1020          * Make sure all delayed rcu free inodes are flushed before we
1021          * destroy cache.
1022          */
1023         rcu_barrier();
1024         kmem_cache_destroy(ext4_inode_cachep);
1025 }
1026
1027 void ext4_clear_inode(struct inode *inode)
1028 {
1029         invalidate_inode_buffers(inode);
1030         clear_inode(inode);
1031         dquot_drop(inode);
1032         ext4_discard_preallocations(inode);
1033         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1034         if (EXT4_I(inode)->jinode) {
1035                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1036                                                EXT4_I(inode)->jinode);
1037                 jbd2_free_inode(EXT4_I(inode)->jinode);
1038                 EXT4_I(inode)->jinode = NULL;
1039         }
1040 }
1041
1042 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1043                                         u64 ino, u32 generation)
1044 {
1045         struct inode *inode;
1046
1047         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1048                 return ERR_PTR(-ESTALE);
1049         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1050                 return ERR_PTR(-ESTALE);
1051
1052         /* iget isn't really right if the inode is currently unallocated!!
1053          *
1054          * ext4_read_inode will return a bad_inode if the inode had been
1055          * deleted, so we should be safe.
1056          *
1057          * Currently we don't know the generation for parent directory, so
1058          * a generation of 0 means "accept any"
1059          */
1060         inode = ext4_iget(sb, ino);
1061         if (IS_ERR(inode))
1062                 return ERR_CAST(inode);
1063         if (generation && inode->i_generation != generation) {
1064                 iput(inode);
1065                 return ERR_PTR(-ESTALE);
1066         }
1067
1068         return inode;
1069 }
1070
1071 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1072                                         int fh_len, int fh_type)
1073 {
1074         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1075                                     ext4_nfs_get_inode);
1076 }
1077
1078 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1079                                         int fh_len, int fh_type)
1080 {
1081         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1082                                     ext4_nfs_get_inode);
1083 }
1084
1085 /*
1086  * Try to release metadata pages (indirect blocks, directories) which are
1087  * mapped via the block device.  Since these pages could have journal heads
1088  * which would prevent try_to_free_buffers() from freeing them, we must use
1089  * jbd2 layer's try_to_free_buffers() function to release them.
1090  */
1091 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1092                                  gfp_t wait)
1093 {
1094         journal_t *journal = EXT4_SB(sb)->s_journal;
1095
1096         WARN_ON(PageChecked(page));
1097         if (!page_has_buffers(page))
1098                 return 0;
1099         if (journal)
1100                 return jbd2_journal_try_to_free_buffers(journal, page,
1101                                                         wait & ~__GFP_WAIT);
1102         return try_to_free_buffers(page);
1103 }
1104
1105 #ifdef CONFIG_QUOTA
1106 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1107 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1108
1109 static int ext4_write_dquot(struct dquot *dquot);
1110 static int ext4_acquire_dquot(struct dquot *dquot);
1111 static int ext4_release_dquot(struct dquot *dquot);
1112 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1113 static int ext4_write_info(struct super_block *sb, int type);
1114 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1115                          struct path *path);
1116 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1117                                  int format_id);
1118 static int ext4_quota_off(struct super_block *sb, int type);
1119 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1120 static int ext4_quota_on_mount(struct super_block *sb, int type);
1121 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1122                                size_t len, loff_t off);
1123 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1124                                 const char *data, size_t len, loff_t off);
1125 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1126                              unsigned int flags);
1127 static int ext4_enable_quotas(struct super_block *sb);
1128
1129 static const struct dquot_operations ext4_quota_operations = {
1130         .get_reserved_space = ext4_get_reserved_space,
1131         .write_dquot    = ext4_write_dquot,
1132         .acquire_dquot  = ext4_acquire_dquot,
1133         .release_dquot  = ext4_release_dquot,
1134         .mark_dirty     = ext4_mark_dquot_dirty,
1135         .write_info     = ext4_write_info,
1136         .alloc_dquot    = dquot_alloc,
1137         .destroy_dquot  = dquot_destroy,
1138 };
1139
1140 static const struct quotactl_ops ext4_qctl_operations = {
1141         .quota_on       = ext4_quota_on,
1142         .quota_off      = ext4_quota_off,
1143         .quota_sync     = dquot_quota_sync,
1144         .get_info       = dquot_get_dqinfo,
1145         .set_info       = dquot_set_dqinfo,
1146         .get_dqblk      = dquot_get_dqblk,
1147         .set_dqblk      = dquot_set_dqblk
1148 };
1149
1150 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1151         .quota_on_meta  = ext4_quota_on_sysfile,
1152         .quota_off      = ext4_quota_off_sysfile,
1153         .quota_sync     = dquot_quota_sync,
1154         .get_info       = dquot_get_dqinfo,
1155         .set_info       = dquot_set_dqinfo,
1156         .get_dqblk      = dquot_get_dqblk,
1157         .set_dqblk      = dquot_set_dqblk
1158 };
1159 #endif
1160
1161 static const struct super_operations ext4_sops = {
1162         .alloc_inode    = ext4_alloc_inode,
1163         .destroy_inode  = ext4_destroy_inode,
1164         .write_inode    = ext4_write_inode,
1165         .dirty_inode    = ext4_dirty_inode,
1166         .drop_inode     = ext4_drop_inode,
1167         .evict_inode    = ext4_evict_inode,
1168         .put_super      = ext4_put_super,
1169         .sync_fs        = ext4_sync_fs,
1170         .freeze_fs      = ext4_freeze,
1171         .unfreeze_fs    = ext4_unfreeze,
1172         .statfs         = ext4_statfs,
1173         .remount_fs     = ext4_remount,
1174         .show_options   = ext4_show_options,
1175 #ifdef CONFIG_QUOTA
1176         .quota_read     = ext4_quota_read,
1177         .quota_write    = ext4_quota_write,
1178 #endif
1179         .bdev_try_to_free_page = bdev_try_to_free_page,
1180 };
1181
1182 static const struct super_operations ext4_nojournal_sops = {
1183         .alloc_inode    = ext4_alloc_inode,
1184         .destroy_inode  = ext4_destroy_inode,
1185         .write_inode    = ext4_write_inode,
1186         .dirty_inode    = ext4_dirty_inode,
1187         .drop_inode     = ext4_drop_inode,
1188         .evict_inode    = ext4_evict_inode,
1189         .put_super      = ext4_put_super,
1190         .statfs         = ext4_statfs,
1191         .remount_fs     = ext4_remount,
1192         .show_options   = ext4_show_options,
1193 #ifdef CONFIG_QUOTA
1194         .quota_read     = ext4_quota_read,
1195         .quota_write    = ext4_quota_write,
1196 #endif
1197         .bdev_try_to_free_page = bdev_try_to_free_page,
1198 };
1199
1200 static const struct export_operations ext4_export_ops = {
1201         .fh_to_dentry = ext4_fh_to_dentry,
1202         .fh_to_parent = ext4_fh_to_parent,
1203         .get_parent = ext4_get_parent,
1204 };
1205
1206 enum {
1207         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1208         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1209         Opt_nouid32, Opt_debug, Opt_removed,
1210         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1211         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1212         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1213         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1214         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1215         Opt_data_err_abort, Opt_data_err_ignore,
1216         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1217         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1218         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1219         Opt_usrquota, Opt_grpquota, Opt_i_version,
1220         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1221         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1222         Opt_inode_readahead_blks, Opt_journal_ioprio,
1223         Opt_dioread_nolock, Opt_dioread_lock,
1224         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1225         Opt_max_dir_size_kb,
1226 };
1227
1228 static const match_table_t tokens = {
1229         {Opt_bsd_df, "bsddf"},
1230         {Opt_minix_df, "minixdf"},
1231         {Opt_grpid, "grpid"},
1232         {Opt_grpid, "bsdgroups"},
1233         {Opt_nogrpid, "nogrpid"},
1234         {Opt_nogrpid, "sysvgroups"},
1235         {Opt_resgid, "resgid=%u"},
1236         {Opt_resuid, "resuid=%u"},
1237         {Opt_sb, "sb=%u"},
1238         {Opt_err_cont, "errors=continue"},
1239         {Opt_err_panic, "errors=panic"},
1240         {Opt_err_ro, "errors=remount-ro"},
1241         {Opt_nouid32, "nouid32"},
1242         {Opt_debug, "debug"},
1243         {Opt_removed, "oldalloc"},
1244         {Opt_removed, "orlov"},
1245         {Opt_user_xattr, "user_xattr"},
1246         {Opt_nouser_xattr, "nouser_xattr"},
1247         {Opt_acl, "acl"},
1248         {Opt_noacl, "noacl"},
1249         {Opt_noload, "norecovery"},
1250         {Opt_noload, "noload"},
1251         {Opt_removed, "nobh"},
1252         {Opt_removed, "bh"},
1253         {Opt_commit, "commit=%u"},
1254         {Opt_min_batch_time, "min_batch_time=%u"},
1255         {Opt_max_batch_time, "max_batch_time=%u"},
1256         {Opt_journal_dev, "journal_dev=%u"},
1257         {Opt_journal_checksum, "journal_checksum"},
1258         {Opt_journal_async_commit, "journal_async_commit"},
1259         {Opt_abort, "abort"},
1260         {Opt_data_journal, "data=journal"},
1261         {Opt_data_ordered, "data=ordered"},
1262         {Opt_data_writeback, "data=writeback"},
1263         {Opt_data_err_abort, "data_err=abort"},
1264         {Opt_data_err_ignore, "data_err=ignore"},
1265         {Opt_offusrjquota, "usrjquota="},
1266         {Opt_usrjquota, "usrjquota=%s"},
1267         {Opt_offgrpjquota, "grpjquota="},
1268         {Opt_grpjquota, "grpjquota=%s"},
1269         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1270         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1271         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1272         {Opt_grpquota, "grpquota"},
1273         {Opt_noquota, "noquota"},
1274         {Opt_quota, "quota"},
1275         {Opt_usrquota, "usrquota"},
1276         {Opt_barrier, "barrier=%u"},
1277         {Opt_barrier, "barrier"},
1278         {Opt_nobarrier, "nobarrier"},
1279         {Opt_i_version, "i_version"},
1280         {Opt_stripe, "stripe=%u"},
1281         {Opt_delalloc, "delalloc"},
1282         {Opt_nodelalloc, "nodelalloc"},
1283         {Opt_mblk_io_submit, "mblk_io_submit"},
1284         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1285         {Opt_block_validity, "block_validity"},
1286         {Opt_noblock_validity, "noblock_validity"},
1287         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1288         {Opt_journal_ioprio, "journal_ioprio=%u"},
1289         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1290         {Opt_auto_da_alloc, "auto_da_alloc"},
1291         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1292         {Opt_dioread_nolock, "dioread_nolock"},
1293         {Opt_dioread_lock, "dioread_lock"},
1294         {Opt_discard, "discard"},
1295         {Opt_nodiscard, "nodiscard"},
1296         {Opt_init_itable, "init_itable=%u"},
1297         {Opt_init_itable, "init_itable"},
1298         {Opt_noinit_itable, "noinit_itable"},
1299         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1300         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1301         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1302         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1303         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1304         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1305         {Opt_err, NULL},
1306 };
1307
1308 static ext4_fsblk_t get_sb_block(void **data)
1309 {
1310         ext4_fsblk_t    sb_block;
1311         char            *options = (char *) *data;
1312
1313         if (!options || strncmp(options, "sb=", 3) != 0)
1314                 return 1;       /* Default location */
1315
1316         options += 3;
1317         /* TODO: use simple_strtoll with >32bit ext4 */
1318         sb_block = simple_strtoul(options, &options, 0);
1319         if (*options && *options != ',') {
1320                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1321                        (char *) *data);
1322                 return 1;
1323         }
1324         if (*options == ',')
1325                 options++;
1326         *data = (void *) options;
1327
1328         return sb_block;
1329 }
1330
1331 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1332 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1333         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1334
1335 #ifdef CONFIG_QUOTA
1336 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1337 {
1338         struct ext4_sb_info *sbi = EXT4_SB(sb);
1339         char *qname;
1340
1341         if (sb_any_quota_loaded(sb) &&
1342                 !sbi->s_qf_names[qtype]) {
1343                 ext4_msg(sb, KERN_ERR,
1344                         "Cannot change journaled "
1345                         "quota options when quota turned on");
1346                 return -1;
1347         }
1348         qname = match_strdup(args);
1349         if (!qname) {
1350                 ext4_msg(sb, KERN_ERR,
1351                         "Not enough memory for storing quotafile name");
1352                 return -1;
1353         }
1354         if (sbi->s_qf_names[qtype] &&
1355                 strcmp(sbi->s_qf_names[qtype], qname)) {
1356                 ext4_msg(sb, KERN_ERR,
1357                         "%s quota file already specified", QTYPE2NAME(qtype));
1358                 kfree(qname);
1359                 return -1;
1360         }
1361         sbi->s_qf_names[qtype] = qname;
1362         if (strchr(sbi->s_qf_names[qtype], '/')) {
1363                 ext4_msg(sb, KERN_ERR,
1364                         "quotafile must be on filesystem root");
1365                 kfree(sbi->s_qf_names[qtype]);
1366                 sbi->s_qf_names[qtype] = NULL;
1367                 return -1;
1368         }
1369         set_opt(sb, QUOTA);
1370         return 1;
1371 }
1372
1373 static int clear_qf_name(struct super_block *sb, int qtype)
1374 {
1375
1376         struct ext4_sb_info *sbi = EXT4_SB(sb);
1377
1378         if (sb_any_quota_loaded(sb) &&
1379                 sbi->s_qf_names[qtype]) {
1380                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1381                         " when quota turned on");
1382                 return -1;
1383         }
1384         /*
1385          * The space will be released later when all options are confirmed
1386          * to be correct
1387          */
1388         sbi->s_qf_names[qtype] = NULL;
1389         return 1;
1390 }
1391 #endif
1392
1393 #define MOPT_SET        0x0001
1394 #define MOPT_CLEAR      0x0002
1395 #define MOPT_NOSUPPORT  0x0004
1396 #define MOPT_EXPLICIT   0x0008
1397 #define MOPT_CLEAR_ERR  0x0010
1398 #define MOPT_GTE0       0x0020
1399 #ifdef CONFIG_QUOTA
1400 #define MOPT_Q          0
1401 #define MOPT_QFMT       0x0040
1402 #else
1403 #define MOPT_Q          MOPT_NOSUPPORT
1404 #define MOPT_QFMT       MOPT_NOSUPPORT
1405 #endif
1406 #define MOPT_DATAJ      0x0080
1407
1408 static const struct mount_opts {
1409         int     token;
1410         int     mount_opt;
1411         int     flags;
1412 } ext4_mount_opts[] = {
1413         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1414         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1415         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1416         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1417         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1418         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1419         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1420         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1421         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1422         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1423         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1424         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1425         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1426         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1427         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1428         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1429                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1430         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1431         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1432         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1433         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1434         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1435         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1436         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1437         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1438         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1439         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1440         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1441         {Opt_commit, 0, MOPT_GTE0},
1442         {Opt_max_batch_time, 0, MOPT_GTE0},
1443         {Opt_min_batch_time, 0, MOPT_GTE0},
1444         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1445         {Opt_init_itable, 0, MOPT_GTE0},
1446         {Opt_stripe, 0, MOPT_GTE0},
1447         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1448         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1449         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1450         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1451         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1452 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1453         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1454         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1455 #else
1456         {Opt_acl, 0, MOPT_NOSUPPORT},
1457         {Opt_noacl, 0, MOPT_NOSUPPORT},
1458 #endif
1459         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1460         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1461         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1462         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1463                                                         MOPT_SET | MOPT_Q},
1464         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1465                                                         MOPT_SET | MOPT_Q},
1466         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1467                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1468         {Opt_usrjquota, 0, MOPT_Q},
1469         {Opt_grpjquota, 0, MOPT_Q},
1470         {Opt_offusrjquota, 0, MOPT_Q},
1471         {Opt_offgrpjquota, 0, MOPT_Q},
1472         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1473         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1474         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1475         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1476         {Opt_err, 0, 0}
1477 };
1478
1479 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1480                             substring_t *args, unsigned long *journal_devnum,
1481                             unsigned int *journal_ioprio, int is_remount)
1482 {
1483         struct ext4_sb_info *sbi = EXT4_SB(sb);
1484         const struct mount_opts *m;
1485         kuid_t uid;
1486         kgid_t gid;
1487         int arg = 0;
1488
1489 #ifdef CONFIG_QUOTA
1490         if (token == Opt_usrjquota)
1491                 return set_qf_name(sb, USRQUOTA, &args[0]);
1492         else if (token == Opt_grpjquota)
1493                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1494         else if (token == Opt_offusrjquota)
1495                 return clear_qf_name(sb, USRQUOTA);
1496         else if (token == Opt_offgrpjquota)
1497                 return clear_qf_name(sb, GRPQUOTA);
1498 #endif
1499         if (args->from && match_int(args, &arg))
1500                 return -1;
1501         switch (token) {
1502         case Opt_noacl:
1503         case Opt_nouser_xattr:
1504                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1505                 break;
1506         case Opt_sb:
1507                 return 1;       /* handled by get_sb_block() */
1508         case Opt_removed:
1509                 ext4_msg(sb, KERN_WARNING,
1510                          "Ignoring removed %s option", opt);
1511                 return 1;
1512         case Opt_resuid:
1513                 uid = make_kuid(current_user_ns(), arg);
1514                 if (!uid_valid(uid)) {
1515                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1516                         return -1;
1517                 }
1518                 sbi->s_resuid = uid;
1519                 return 1;
1520         case Opt_resgid:
1521                 gid = make_kgid(current_user_ns(), arg);
1522                 if (!gid_valid(gid)) {
1523                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1524                         return -1;
1525                 }
1526                 sbi->s_resgid = gid;
1527                 return 1;
1528         case Opt_abort:
1529                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1530                 return 1;
1531         case Opt_i_version:
1532                 sb->s_flags |= MS_I_VERSION;
1533                 return 1;
1534         case Opt_journal_dev:
1535                 if (is_remount) {
1536                         ext4_msg(sb, KERN_ERR,
1537                                  "Cannot specify journal on remount");
1538                         return -1;
1539                 }
1540                 *journal_devnum = arg;
1541                 return 1;
1542         case Opt_journal_ioprio:
1543                 if (arg < 0 || arg > 7)
1544                         return -1;
1545                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1546                 return 1;
1547         }
1548
1549         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1550                 if (token != m->token)
1551                         continue;
1552                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1553                         return -1;
1554                 if (m->flags & MOPT_EXPLICIT)
1555                         set_opt2(sb, EXPLICIT_DELALLOC);
1556                 if (m->flags & MOPT_CLEAR_ERR)
1557                         clear_opt(sb, ERRORS_MASK);
1558                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1559                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1560                                  "options when quota turned on");
1561                         return -1;
1562                 }
1563
1564                 if (m->flags & MOPT_NOSUPPORT) {
1565                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1566                 } else if (token == Opt_commit) {
1567                         if (arg == 0)
1568                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1569                         sbi->s_commit_interval = HZ * arg;
1570                 } else if (token == Opt_max_batch_time) {
1571                         if (arg == 0)
1572                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1573                         sbi->s_max_batch_time = arg;
1574                 } else if (token == Opt_min_batch_time) {
1575                         sbi->s_min_batch_time = arg;
1576                 } else if (token == Opt_inode_readahead_blks) {
1577                         if (arg > (1 << 30))
1578                                 return -1;
1579                         if (arg && !is_power_of_2(arg)) {
1580                                 ext4_msg(sb, KERN_ERR,
1581                                          "EXT4-fs: inode_readahead_blks"
1582                                          " must be a power of 2");
1583                                 return -1;
1584                         }
1585                         sbi->s_inode_readahead_blks = arg;
1586                 } else if (token == Opt_init_itable) {
1587                         set_opt(sb, INIT_INODE_TABLE);
1588                         if (!args->from)
1589                                 arg = EXT4_DEF_LI_WAIT_MULT;
1590                         sbi->s_li_wait_mult = arg;
1591                 } else if (token == Opt_max_dir_size_kb) {
1592                         sbi->s_max_dir_size_kb = arg;
1593                 } else if (token == Opt_stripe) {
1594                         sbi->s_stripe = arg;
1595                 } else if (m->flags & MOPT_DATAJ) {
1596                         if (is_remount) {
1597                                 if (!sbi->s_journal)
1598                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1599                                 else if (test_opt(sb, DATA_FLAGS) !=
1600                                          m->mount_opt) {
1601                                         ext4_msg(sb, KERN_ERR,
1602                                          "Cannot change data mode on remount");
1603                                         return -1;
1604                                 }
1605                         } else {
1606                                 clear_opt(sb, DATA_FLAGS);
1607                                 sbi->s_mount_opt |= m->mount_opt;
1608                         }
1609 #ifdef CONFIG_QUOTA
1610                 } else if (m->flags & MOPT_QFMT) {
1611                         if (sb_any_quota_loaded(sb) &&
1612                             sbi->s_jquota_fmt != m->mount_opt) {
1613                                 ext4_msg(sb, KERN_ERR, "Cannot "
1614                                          "change journaled quota options "
1615                                          "when quota turned on");
1616                                 return -1;
1617                         }
1618                         sbi->s_jquota_fmt = m->mount_opt;
1619 #endif
1620                 } else {
1621                         if (!args->from)
1622                                 arg = 1;
1623                         if (m->flags & MOPT_CLEAR)
1624                                 arg = !arg;
1625                         else if (unlikely(!(m->flags & MOPT_SET))) {
1626                                 ext4_msg(sb, KERN_WARNING,
1627                                          "buggy handling of option %s", opt);
1628                                 WARN_ON(1);
1629                                 return -1;
1630                         }
1631                         if (arg != 0)
1632                                 sbi->s_mount_opt |= m->mount_opt;
1633                         else
1634                                 sbi->s_mount_opt &= ~m->mount_opt;
1635                 }
1636                 return 1;
1637         }
1638         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1639                  "or missing value", opt);
1640         return -1;
1641 }
1642
1643 static int parse_options(char *options, struct super_block *sb,
1644                          unsigned long *journal_devnum,
1645                          unsigned int *journal_ioprio,
1646                          int is_remount)
1647 {
1648 #ifdef CONFIG_QUOTA
1649         struct ext4_sb_info *sbi = EXT4_SB(sb);
1650 #endif
1651         char *p;
1652         substring_t args[MAX_OPT_ARGS];
1653         int token;
1654
1655         if (!options)
1656                 return 1;
1657
1658         while ((p = strsep(&options, ",")) != NULL) {
1659                 if (!*p)
1660                         continue;
1661                 /*
1662                  * Initialize args struct so we know whether arg was
1663                  * found; some options take optional arguments.
1664                  */
1665                 args[0].to = args[0].from = NULL;
1666                 token = match_token(p, tokens, args);
1667                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1668                                      journal_ioprio, is_remount) < 0)
1669                         return 0;
1670         }
1671 #ifdef CONFIG_QUOTA
1672         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1673                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1674                         clear_opt(sb, USRQUOTA);
1675
1676                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1677                         clear_opt(sb, GRPQUOTA);
1678
1679                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1680                         ext4_msg(sb, KERN_ERR, "old and new quota "
1681                                         "format mixing");
1682                         return 0;
1683                 }
1684
1685                 if (!sbi->s_jquota_fmt) {
1686                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1687                                         "not specified");
1688                         return 0;
1689                 }
1690         } else {
1691                 if (sbi->s_jquota_fmt) {
1692                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1693                                         "specified with no journaling "
1694                                         "enabled");
1695                         return 0;
1696                 }
1697         }
1698 #endif
1699         return 1;
1700 }
1701
1702 static inline void ext4_show_quota_options(struct seq_file *seq,
1703                                            struct super_block *sb)
1704 {
1705 #if defined(CONFIG_QUOTA)
1706         struct ext4_sb_info *sbi = EXT4_SB(sb);
1707
1708         if (sbi->s_jquota_fmt) {
1709                 char *fmtname = "";
1710
1711                 switch (sbi->s_jquota_fmt) {
1712                 case QFMT_VFS_OLD:
1713                         fmtname = "vfsold";
1714                         break;
1715                 case QFMT_VFS_V0:
1716                         fmtname = "vfsv0";
1717                         break;
1718                 case QFMT_VFS_V1:
1719                         fmtname = "vfsv1";
1720                         break;
1721                 }
1722                 seq_printf(seq, ",jqfmt=%s", fmtname);
1723         }
1724
1725         if (sbi->s_qf_names[USRQUOTA])
1726                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1727
1728         if (sbi->s_qf_names[GRPQUOTA])
1729                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1730
1731         if (test_opt(sb, USRQUOTA))
1732                 seq_puts(seq, ",usrquota");
1733
1734         if (test_opt(sb, GRPQUOTA))
1735                 seq_puts(seq, ",grpquota");
1736 #endif
1737 }
1738
1739 static const char *token2str(int token)
1740 {
1741         const struct match_token *t;
1742
1743         for (t = tokens; t->token != Opt_err; t++)
1744                 if (t->token == token && !strchr(t->pattern, '='))
1745                         break;
1746         return t->pattern;
1747 }
1748
1749 /*
1750  * Show an option if
1751  *  - it's set to a non-default value OR
1752  *  - if the per-sb default is different from the global default
1753  */
1754 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1755                               int nodefs)
1756 {
1757         struct ext4_sb_info *sbi = EXT4_SB(sb);
1758         struct ext4_super_block *es = sbi->s_es;
1759         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1760         const struct mount_opts *m;
1761         char sep = nodefs ? '\n' : ',';
1762
1763 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1764 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1765
1766         if (sbi->s_sb_block != 1)
1767                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1768
1769         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1770                 int want_set = m->flags & MOPT_SET;
1771                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1772                     (m->flags & MOPT_CLEAR_ERR))
1773                         continue;
1774                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1775                         continue; /* skip if same as the default */
1776                 if ((want_set &&
1777                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1778                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1779                         continue; /* select Opt_noFoo vs Opt_Foo */
1780                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1781         }
1782
1783         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1784             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1785                 SEQ_OPTS_PRINT("resuid=%u",
1786                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1787         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1788             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1789                 SEQ_OPTS_PRINT("resgid=%u",
1790                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1791         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1792         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1793                 SEQ_OPTS_PUTS("errors=remount-ro");
1794         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1795                 SEQ_OPTS_PUTS("errors=continue");
1796         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1797                 SEQ_OPTS_PUTS("errors=panic");
1798         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1799                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1800         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1801                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1802         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1803                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1804         if (sb->s_flags & MS_I_VERSION)
1805                 SEQ_OPTS_PUTS("i_version");
1806         if (nodefs || sbi->s_stripe)
1807                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1808         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1809                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1810                         SEQ_OPTS_PUTS("data=journal");
1811                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1812                         SEQ_OPTS_PUTS("data=ordered");
1813                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1814                         SEQ_OPTS_PUTS("data=writeback");
1815         }
1816         if (nodefs ||
1817             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1818                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1819                                sbi->s_inode_readahead_blks);
1820
1821         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1822                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1823                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1824         if (nodefs || sbi->s_max_dir_size_kb)
1825                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1826
1827         ext4_show_quota_options(seq, sb);
1828         return 0;
1829 }
1830
1831 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1832 {
1833         return _ext4_show_options(seq, root->d_sb, 0);
1834 }
1835
1836 static int options_seq_show(struct seq_file *seq, void *offset)
1837 {
1838         struct super_block *sb = seq->private;
1839         int rc;
1840
1841         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1842         rc = _ext4_show_options(seq, sb, 1);
1843         seq_puts(seq, "\n");
1844         return rc;
1845 }
1846
1847 static int options_open_fs(struct inode *inode, struct file *file)
1848 {
1849         return single_open(file, options_seq_show, PDE(inode)->data);
1850 }
1851
1852 static const struct file_operations ext4_seq_options_fops = {
1853         .owner = THIS_MODULE,
1854         .open = options_open_fs,
1855         .read = seq_read,
1856         .llseek = seq_lseek,
1857         .release = single_release,
1858 };
1859
1860 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1861                             int read_only)
1862 {
1863         struct ext4_sb_info *sbi = EXT4_SB(sb);
1864         int res = 0;
1865
1866         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1867                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1868                          "forcing read-only mode");
1869                 res = MS_RDONLY;
1870         }
1871         if (read_only)
1872                 goto done;
1873         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1874                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1875                          "running e2fsck is recommended");
1876         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1877                 ext4_msg(sb, KERN_WARNING,
1878                          "warning: mounting fs with errors, "
1879                          "running e2fsck is recommended");
1880         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1881                  le16_to_cpu(es->s_mnt_count) >=
1882                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1883                 ext4_msg(sb, KERN_WARNING,
1884                          "warning: maximal mount count reached, "
1885                          "running e2fsck is recommended");
1886         else if (le32_to_cpu(es->s_checkinterval) &&
1887                 (le32_to_cpu(es->s_lastcheck) +
1888                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1889                 ext4_msg(sb, KERN_WARNING,
1890                          "warning: checktime reached, "
1891                          "running e2fsck is recommended");
1892         if (!sbi->s_journal)
1893                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1894         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1895                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1896         le16_add_cpu(&es->s_mnt_count, 1);
1897         es->s_mtime = cpu_to_le32(get_seconds());
1898         ext4_update_dynamic_rev(sb);
1899         if (sbi->s_journal)
1900                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1901
1902         ext4_commit_super(sb, 1);
1903 done:
1904         if (test_opt(sb, DEBUG))
1905                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1906                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1907                         sb->s_blocksize,
1908                         sbi->s_groups_count,
1909                         EXT4_BLOCKS_PER_GROUP(sb),
1910                         EXT4_INODES_PER_GROUP(sb),
1911                         sbi->s_mount_opt, sbi->s_mount_opt2);
1912
1913         cleancache_init_fs(sb);
1914         return res;
1915 }
1916
1917 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1918 {
1919         struct ext4_sb_info *sbi = EXT4_SB(sb);
1920         struct flex_groups *new_groups;
1921         int size;
1922
1923         if (!sbi->s_log_groups_per_flex)
1924                 return 0;
1925
1926         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1927         if (size <= sbi->s_flex_groups_allocated)
1928                 return 0;
1929
1930         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1931         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1932         if (!new_groups) {
1933                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1934                          size / (int) sizeof(struct flex_groups));
1935                 return -ENOMEM;
1936         }
1937
1938         if (sbi->s_flex_groups) {
1939                 memcpy(new_groups, sbi->s_flex_groups,
1940                        (sbi->s_flex_groups_allocated *
1941                         sizeof(struct flex_groups)));
1942                 ext4_kvfree(sbi->s_flex_groups);
1943         }
1944         sbi->s_flex_groups = new_groups;
1945         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1946         return 0;
1947 }
1948
1949 static int ext4_fill_flex_info(struct super_block *sb)
1950 {
1951         struct ext4_sb_info *sbi = EXT4_SB(sb);
1952         struct ext4_group_desc *gdp = NULL;
1953         ext4_group_t flex_group;
1954         unsigned int groups_per_flex = 0;
1955         int i, err;
1956
1957         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1958         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1959                 sbi->s_log_groups_per_flex = 0;
1960                 return 1;
1961         }
1962         groups_per_flex = 1U << sbi->s_log_groups_per_flex;
1963
1964         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1965         if (err)
1966                 goto failed;
1967
1968         for (i = 0; i < sbi->s_groups_count; i++) {
1969                 gdp = ext4_get_group_desc(sb, i, NULL);
1970
1971                 flex_group = ext4_flex_group(sbi, i);
1972                 atomic_add(ext4_free_inodes_count(sb, gdp),
1973                            &sbi->s_flex_groups[flex_group].free_inodes);
1974                 atomic_add(ext4_free_group_clusters(sb, gdp),
1975                            &sbi->s_flex_groups[flex_group].free_clusters);
1976                 atomic_add(ext4_used_dirs_count(sb, gdp),
1977                            &sbi->s_flex_groups[flex_group].used_dirs);
1978         }
1979
1980         return 1;
1981 failed:
1982         return 0;
1983 }
1984
1985 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1986                                    struct ext4_group_desc *gdp)
1987 {
1988         int offset;
1989         __u16 crc = 0;
1990         __le32 le_group = cpu_to_le32(block_group);
1991
1992         if ((sbi->s_es->s_feature_ro_compat &
1993              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1994                 /* Use new metadata_csum algorithm */
1995                 __u16 old_csum;
1996                 __u32 csum32;
1997
1998                 old_csum = gdp->bg_checksum;
1999                 gdp->bg_checksum = 0;
2000                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2001                                      sizeof(le_group));
2002                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2003                                      sbi->s_desc_size);
2004                 gdp->bg_checksum = old_csum;
2005
2006                 crc = csum32 & 0xFFFF;
2007                 goto out;
2008         }
2009
2010         /* old crc16 code */
2011         offset = offsetof(struct ext4_group_desc, bg_checksum);
2012
2013         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2014         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2015         crc = crc16(crc, (__u8 *)gdp, offset);
2016         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2017         /* for checksum of struct ext4_group_desc do the rest...*/
2018         if ((sbi->s_es->s_feature_incompat &
2019              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2020             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2021                 crc = crc16(crc, (__u8 *)gdp + offset,
2022                             le16_to_cpu(sbi->s_es->s_desc_size) -
2023                                 offset);
2024
2025 out:
2026         return cpu_to_le16(crc);
2027 }
2028
2029 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2030                                 struct ext4_group_desc *gdp)
2031 {
2032         if (ext4_has_group_desc_csum(sb) &&
2033             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2034                                                       block_group, gdp)))
2035                 return 0;
2036
2037         return 1;
2038 }
2039
2040 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2041                               struct ext4_group_desc *gdp)
2042 {
2043         if (!ext4_has_group_desc_csum(sb))
2044                 return;
2045         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2046 }
2047
2048 /* Called at mount-time, super-block is locked */
2049 static int ext4_check_descriptors(struct super_block *sb,
2050                                   ext4_group_t *first_not_zeroed)
2051 {
2052         struct ext4_sb_info *sbi = EXT4_SB(sb);
2053         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2054         ext4_fsblk_t last_block;
2055         ext4_fsblk_t block_bitmap;
2056         ext4_fsblk_t inode_bitmap;
2057         ext4_fsblk_t inode_table;
2058         int flexbg_flag = 0;
2059         ext4_group_t i, grp = sbi->s_groups_count;
2060
2061         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2062                 flexbg_flag = 1;
2063
2064         ext4_debug("Checking group descriptors");
2065
2066         for (i = 0; i < sbi->s_groups_count; i++) {
2067                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2068
2069                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2070                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2071                 else
2072                         last_block = first_block +
2073                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2074
2075                 if ((grp == sbi->s_groups_count) &&
2076                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2077                         grp = i;
2078
2079                 block_bitmap = ext4_block_bitmap(sb, gdp);
2080                 if (block_bitmap < first_block || block_bitmap > last_block) {
2081                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2082                                "Block bitmap for group %u not in group "
2083                                "(block %llu)!", i, block_bitmap);
2084                         return 0;
2085                 }
2086                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2087                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2088                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2089                                "Inode bitmap for group %u not in group "
2090                                "(block %llu)!", i, inode_bitmap);
2091                         return 0;
2092                 }
2093                 inode_table = ext4_inode_table(sb, gdp);
2094                 if (inode_table < first_block ||
2095                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2096                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2097                                "Inode table for group %u not in group "
2098                                "(block %llu)!", i, inode_table);
2099                         return 0;
2100                 }
2101                 ext4_lock_group(sb, i);
2102                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2103                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2104                                  "Checksum for group %u failed (%u!=%u)",
2105                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2106                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2107                         if (!(sb->s_flags & MS_RDONLY)) {
2108                                 ext4_unlock_group(sb, i);
2109                                 return 0;
2110                         }
2111                 }
2112                 ext4_unlock_group(sb, i);
2113                 if (!flexbg_flag)
2114                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2115         }
2116         if (NULL != first_not_zeroed)
2117                 *first_not_zeroed = grp;
2118
2119         ext4_free_blocks_count_set(sbi->s_es,
2120                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2121         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2122         return 1;
2123 }
2124
2125 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2126  * the superblock) which were deleted from all directories, but held open by
2127  * a process at the time of a crash.  We walk the list and try to delete these
2128  * inodes at recovery time (only with a read-write filesystem).
2129  *
2130  * In order to keep the orphan inode chain consistent during traversal (in
2131  * case of crash during recovery), we link each inode into the superblock
2132  * orphan list_head and handle it the same way as an inode deletion during
2133  * normal operation (which journals the operations for us).
2134  *
2135  * We only do an iget() and an iput() on each inode, which is very safe if we
2136  * accidentally point at an in-use or already deleted inode.  The worst that
2137  * can happen in this case is that we get a "bit already cleared" message from
2138  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2139  * e2fsck was run on this filesystem, and it must have already done the orphan
2140  * inode cleanup for us, so we can safely abort without any further action.
2141  */
2142 static void ext4_orphan_cleanup(struct super_block *sb,
2143                                 struct ext4_super_block *es)
2144 {
2145         unsigned int s_flags = sb->s_flags;
2146         int nr_orphans = 0, nr_truncates = 0;
2147 #ifdef CONFIG_QUOTA
2148         int i;
2149 #endif
2150         if (!es->s_last_orphan) {
2151                 jbd_debug(4, "no orphan inodes to clean up\n");
2152                 return;
2153         }
2154
2155         if (bdev_read_only(sb->s_bdev)) {
2156                 ext4_msg(sb, KERN_ERR, "write access "
2157                         "unavailable, skipping orphan cleanup");
2158                 return;
2159         }
2160
2161         /* Check if feature set would not allow a r/w mount */
2162         if (!ext4_feature_set_ok(sb, 0)) {
2163                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2164                          "unknown ROCOMPAT features");
2165                 return;
2166         }
2167
2168         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2169                 /* don't clear list on RO mount w/ errors */
2170                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2171                         jbd_debug(1, "Errors on filesystem, "
2172                                   "clearing orphan list.\n");
2173                         es->s_last_orphan = 0;
2174                 }
2175                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2176                 return;
2177         }
2178
2179         if (s_flags & MS_RDONLY) {
2180                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2181                 sb->s_flags &= ~MS_RDONLY;
2182         }
2183 #ifdef CONFIG_QUOTA
2184         /* Needed for iput() to work correctly and not trash data */
2185         sb->s_flags |= MS_ACTIVE;
2186         /* Turn on quotas so that they are updated correctly */
2187         for (i = 0; i < MAXQUOTAS; i++) {
2188                 if (EXT4_SB(sb)->s_qf_names[i]) {
2189                         int ret = ext4_quota_on_mount(sb, i);
2190                         if (ret < 0)
2191                                 ext4_msg(sb, KERN_ERR,
2192                                         "Cannot turn on journaled "
2193                                         "quota: error %d", ret);
2194                 }
2195         }
2196 #endif
2197
2198         while (es->s_last_orphan) {
2199                 struct inode *inode;
2200
2201                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2202                 if (IS_ERR(inode)) {
2203                         es->s_last_orphan = 0;
2204                         break;
2205                 }
2206
2207                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2208                 dquot_initialize(inode);
2209                 if (inode->i_nlink) {
2210                         ext4_msg(sb, KERN_DEBUG,
2211                                 "%s: truncating inode %lu to %lld bytes",
2212                                 __func__, inode->i_ino, inode->i_size);
2213                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2214                                   inode->i_ino, inode->i_size);
2215                         ext4_truncate(inode);
2216                         nr_truncates++;
2217                 } else {
2218                         ext4_msg(sb, KERN_DEBUG,
2219                                 "%s: deleting unreferenced inode %lu",
2220                                 __func__, inode->i_ino);
2221                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2222                                   inode->i_ino);
2223                         nr_orphans++;
2224                 }
2225                 iput(inode);  /* The delete magic happens here! */
2226         }
2227
2228 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2229
2230         if (nr_orphans)
2231                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2232                        PLURAL(nr_orphans));
2233         if (nr_truncates)
2234                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2235                        PLURAL(nr_truncates));
2236 #ifdef CONFIG_QUOTA
2237         /* Turn quotas off */
2238         for (i = 0; i < MAXQUOTAS; i++) {
2239                 if (sb_dqopt(sb)->files[i])
2240                         dquot_quota_off(sb, i);
2241         }
2242 #endif
2243         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2244 }
2245
2246 /*
2247  * Maximal extent format file size.
2248  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2249  * extent format containers, within a sector_t, and within i_blocks
2250  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2251  * so that won't be a limiting factor.
2252  *
2253  * However there is other limiting factor. We do store extents in the form
2254  * of starting block and length, hence the resulting length of the extent
2255  * covering maximum file size must fit into on-disk format containers as
2256  * well. Given that length is always by 1 unit bigger than max unit (because
2257  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2258  *
2259  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2260  */
2261 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2262 {
2263         loff_t res;
2264         loff_t upper_limit = MAX_LFS_FILESIZE;
2265
2266         /* small i_blocks in vfs inode? */
2267         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2268                 /*
2269                  * CONFIG_LBDAF is not enabled implies the inode
2270                  * i_block represent total blocks in 512 bytes
2271                  * 32 == size of vfs inode i_blocks * 8
2272                  */
2273                 upper_limit = (1LL << 32) - 1;
2274
2275                 /* total blocks in file system block size */
2276                 upper_limit >>= (blkbits - 9);
2277                 upper_limit <<= blkbits;
2278         }
2279
2280         /*
2281          * 32-bit extent-start container, ee_block. We lower the maxbytes
2282          * by one fs block, so ee_len can cover the extent of maximum file
2283          * size
2284          */
2285         res = (1LL << 32) - 1;
2286         res <<= blkbits;
2287
2288         /* Sanity check against vm- & vfs- imposed limits */
2289         if (res > upper_limit)
2290                 res = upper_limit;
2291
2292         return res;
2293 }
2294
2295 /*
2296  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2297  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2298  * We need to be 1 filesystem block less than the 2^48 sector limit.
2299  */
2300 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2301 {
2302         loff_t res = EXT4_NDIR_BLOCKS;
2303         int meta_blocks;
2304         loff_t upper_limit;
2305         /* This is calculated to be the largest file size for a dense, block
2306          * mapped file such that the file's total number of 512-byte sectors,
2307          * including data and all indirect blocks, does not exceed (2^48 - 1).
2308          *
2309          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2310          * number of 512-byte sectors of the file.
2311          */
2312
2313         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2314                 /*
2315                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2316                  * the inode i_block field represents total file blocks in
2317                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2318                  */
2319                 upper_limit = (1LL << 32) - 1;
2320
2321                 /* total blocks in file system block size */
2322                 upper_limit >>= (bits - 9);
2323
2324         } else {
2325                 /*
2326                  * We use 48 bit ext4_inode i_blocks
2327                  * With EXT4_HUGE_FILE_FL set the i_blocks
2328                  * represent total number of blocks in
2329                  * file system block size
2330                  */
2331                 upper_limit = (1LL << 48) - 1;
2332
2333         }
2334
2335         /* indirect blocks */
2336         meta_blocks = 1;
2337         /* double indirect blocks */
2338         meta_blocks += 1 + (1LL << (bits-2));
2339         /* tripple indirect blocks */
2340         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2341
2342         upper_limit -= meta_blocks;
2343         upper_limit <<= bits;
2344
2345         res += 1LL << (bits-2);
2346         res += 1LL << (2*(bits-2));
2347         res += 1LL << (3*(bits-2));
2348         res <<= bits;
2349         if (res > upper_limit)
2350                 res = upper_limit;
2351
2352         if (res > MAX_LFS_FILESIZE)
2353                 res = MAX_LFS_FILESIZE;
2354
2355         return res;
2356 }
2357
2358 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2359                                    ext4_fsblk_t logical_sb_block, int nr)
2360 {
2361         struct ext4_sb_info *sbi = EXT4_SB(sb);
2362         ext4_group_t bg, first_meta_bg;
2363         int has_super = 0;
2364
2365         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2366
2367         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2368             nr < first_meta_bg)
2369                 return logical_sb_block + nr + 1;
2370         bg = sbi->s_desc_per_block * nr;
2371         if (ext4_bg_has_super(sb, bg))
2372                 has_super = 1;
2373
2374         return (has_super + ext4_group_first_block_no(sb, bg));
2375 }
2376
2377 /**
2378  * ext4_get_stripe_size: Get the stripe size.
2379  * @sbi: In memory super block info
2380  *
2381  * If we have specified it via mount option, then
2382  * use the mount option value. If the value specified at mount time is
2383  * greater than the blocks per group use the super block value.
2384  * If the super block value is greater than blocks per group return 0.
2385  * Allocator needs it be less than blocks per group.
2386  *
2387  */
2388 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2389 {
2390         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2391         unsigned long stripe_width =
2392                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2393         int ret;
2394
2395         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2396                 ret = sbi->s_stripe;
2397         else if (stripe_width <= sbi->s_blocks_per_group)
2398                 ret = stripe_width;
2399         else if (stride <= sbi->s_blocks_per_group)
2400                 ret = stride;
2401         else
2402                 ret = 0;
2403
2404         /*
2405          * If the stripe width is 1, this makes no sense and
2406          * we set it to 0 to turn off stripe handling code.
2407          */
2408         if (ret <= 1)
2409                 ret = 0;
2410
2411         return ret;
2412 }
2413
2414 /* sysfs supprt */
2415
2416 struct ext4_attr {
2417         struct attribute attr;
2418         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2419         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2420                          const char *, size_t);
2421         int offset;
2422 };
2423
2424 static int parse_strtoul(const char *buf,
2425                 unsigned long max, unsigned long *value)
2426 {
2427         char *endp;
2428
2429         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2430         endp = skip_spaces(endp);
2431         if (*endp || *value > max)
2432                 return -EINVAL;
2433
2434         return 0;
2435 }
2436
2437 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2438                                               struct ext4_sb_info *sbi,
2439                                               char *buf)
2440 {
2441         return snprintf(buf, PAGE_SIZE, "%llu\n",
2442                 (s64) EXT4_C2B(sbi,
2443                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2444 }
2445
2446 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2447                                          struct ext4_sb_info *sbi, char *buf)
2448 {
2449         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2450
2451         if (!sb->s_bdev->bd_part)
2452                 return snprintf(buf, PAGE_SIZE, "0\n");
2453         return snprintf(buf, PAGE_SIZE, "%lu\n",
2454                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2455                          sbi->s_sectors_written_start) >> 1);
2456 }
2457
2458 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2459                                           struct ext4_sb_info *sbi, char *buf)
2460 {
2461         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2462
2463         if (!sb->s_bdev->bd_part)
2464                 return snprintf(buf, PAGE_SIZE, "0\n");
2465         return snprintf(buf, PAGE_SIZE, "%llu\n",
2466                         (unsigned long long)(sbi->s_kbytes_written +
2467                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2468                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2469 }
2470
2471 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2472                                           struct ext4_sb_info *sbi,
2473                                           const char *buf, size_t count)
2474 {
2475         unsigned long t;
2476
2477         if (parse_strtoul(buf, 0x40000000, &t))
2478                 return -EINVAL;
2479
2480         if (t && !is_power_of_2(t))
2481                 return -EINVAL;
2482
2483         sbi->s_inode_readahead_blks = t;
2484         return count;
2485 }
2486
2487 static ssize_t sbi_ui_show(struct ext4_attr *a,
2488                            struct ext4_sb_info *sbi, char *buf)
2489 {
2490         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2491
2492         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2493 }
2494
2495 static ssize_t sbi_ui_store(struct ext4_attr *a,
2496                             struct ext4_sb_info *sbi,
2497                             const char *buf, size_t count)
2498 {
2499         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2500         unsigned long t;
2501
2502         if (parse_strtoul(buf, 0xffffffff, &t))
2503                 return -EINVAL;
2504         *ui = t;
2505         return count;
2506 }
2507
2508 static ssize_t trigger_test_error(struct ext4_attr *a,
2509                                   struct ext4_sb_info *sbi,
2510                                   const char *buf, size_t count)
2511 {
2512         int len = count;
2513
2514         if (!capable(CAP_SYS_ADMIN))
2515                 return -EPERM;
2516
2517         if (len && buf[len-1] == '\n')
2518                 len--;
2519
2520         if (len)
2521                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2522         return count;
2523 }
2524
2525 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2526 static struct ext4_attr ext4_attr_##_name = {                   \
2527         .attr = {.name = __stringify(_name), .mode = _mode },   \
2528         .show   = _show,                                        \
2529         .store  = _store,                                       \
2530         .offset = offsetof(struct ext4_sb_info, _elname),       \
2531 }
2532 #define EXT4_ATTR(name, mode, show, store) \
2533 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2534
2535 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2536 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2537 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2538 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2539         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2540 #define ATTR_LIST(name) &ext4_attr_##name.attr
2541
2542 EXT4_RO_ATTR(delayed_allocation_blocks);
2543 EXT4_RO_ATTR(session_write_kbytes);
2544 EXT4_RO_ATTR(lifetime_write_kbytes);
2545 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2546                  inode_readahead_blks_store, s_inode_readahead_blks);
2547 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2548 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2549 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2550 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2551 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2552 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2553 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2554 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2555 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2556 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2557
2558 static struct attribute *ext4_attrs[] = {
2559         ATTR_LIST(delayed_allocation_blocks),
2560         ATTR_LIST(session_write_kbytes),
2561         ATTR_LIST(lifetime_write_kbytes),
2562         ATTR_LIST(inode_readahead_blks),
2563         ATTR_LIST(inode_goal),
2564         ATTR_LIST(mb_stats),
2565         ATTR_LIST(mb_max_to_scan),
2566         ATTR_LIST(mb_min_to_scan),
2567         ATTR_LIST(mb_order2_req),
2568         ATTR_LIST(mb_stream_req),
2569         ATTR_LIST(mb_group_prealloc),
2570         ATTR_LIST(max_writeback_mb_bump),
2571         ATTR_LIST(extent_max_zeroout_kb),
2572         ATTR_LIST(trigger_fs_error),
2573         NULL,
2574 };
2575
2576 /* Features this copy of ext4 supports */
2577 EXT4_INFO_ATTR(lazy_itable_init);
2578 EXT4_INFO_ATTR(batched_discard);
2579 EXT4_INFO_ATTR(meta_bg_resize);
2580
2581 static struct attribute *ext4_feat_attrs[] = {
2582         ATTR_LIST(lazy_itable_init),
2583         ATTR_LIST(batched_discard),
2584         ATTR_LIST(meta_bg_resize),
2585         NULL,
2586 };
2587
2588 static ssize_t ext4_attr_show(struct kobject *kobj,
2589                               struct attribute *attr, char *buf)
2590 {
2591         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2592                                                 s_kobj);
2593         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2594
2595         return a->show ? a->show(a, sbi, buf) : 0;
2596 }
2597
2598 static ssize_t ext4_attr_store(struct kobject *kobj,
2599                                struct attribute *attr,
2600                                const char *buf, size_t len)
2601 {
2602         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2603                                                 s_kobj);
2604         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2605
2606         return a->store ? a->store(a, sbi, buf, len) : 0;
2607 }
2608
2609 static void ext4_sb_release(struct kobject *kobj)
2610 {
2611         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2612                                                 s_kobj);
2613         complete(&sbi->s_kobj_unregister);
2614 }
2615
2616 static const struct sysfs_ops ext4_attr_ops = {
2617         .show   = ext4_attr_show,
2618         .store  = ext4_attr_store,
2619 };
2620
2621 static struct kobj_type ext4_ktype = {
2622         .default_attrs  = ext4_attrs,
2623         .sysfs_ops      = &ext4_attr_ops,
2624         .release        = ext4_sb_release,
2625 };
2626
2627 static void ext4_feat_release(struct kobject *kobj)
2628 {
2629         complete(&ext4_feat->f_kobj_unregister);
2630 }
2631
2632 static struct kobj_type ext4_feat_ktype = {
2633         .default_attrs  = ext4_feat_attrs,
2634         .sysfs_ops      = &ext4_attr_ops,
2635         .release        = ext4_feat_release,
2636 };
2637
2638 /*
2639  * Check whether this filesystem can be mounted based on
2640  * the features present and the RDONLY/RDWR mount requested.
2641  * Returns 1 if this filesystem can be mounted as requested,
2642  * 0 if it cannot be.
2643  */
2644 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2645 {
2646         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2647                 ext4_msg(sb, KERN_ERR,
2648                         "Couldn't mount because of "
2649                         "unsupported optional features (%x)",
2650                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2651                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2652                 return 0;
2653         }
2654
2655         if (readonly)
2656                 return 1;
2657
2658         /* Check that feature set is OK for a read-write mount */
2659         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2660                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2661                          "unsupported optional features (%x)",
2662                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2663                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2664                 return 0;
2665         }
2666         /*
2667          * Large file size enabled file system can only be mounted
2668          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2669          */
2670         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2671                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2672                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2673                                  "cannot be mounted RDWR without "
2674                                  "CONFIG_LBDAF");
2675                         return 0;
2676                 }
2677         }
2678         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2679             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2680                 ext4_msg(sb, KERN_ERR,
2681                          "Can't support bigalloc feature without "
2682                          "extents feature\n");
2683                 return 0;
2684         }
2685
2686 #ifndef CONFIG_QUOTA
2687         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2688             !readonly) {
2689                 ext4_msg(sb, KERN_ERR,
2690                          "Filesystem with quota feature cannot be mounted RDWR "
2691                          "without CONFIG_QUOTA");
2692                 return 0;
2693         }
2694 #endif  /* CONFIG_QUOTA */
2695         return 1;
2696 }
2697
2698 /*
2699  * This function is called once a day if we have errors logged
2700  * on the file system
2701  */
2702 static void print_daily_error_info(unsigned long arg)
2703 {
2704         struct super_block *sb = (struct super_block *) arg;
2705         struct ext4_sb_info *sbi;
2706         struct ext4_super_block *es;
2707
2708         sbi = EXT4_SB(sb);
2709         es = sbi->s_es;
2710
2711         if (es->s_error_count)
2712                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2713                          le32_to_cpu(es->s_error_count));
2714         if (es->s_first_error_time) {
2715                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2716                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2717                        (int) sizeof(es->s_first_error_func),
2718                        es->s_first_error_func,
2719                        le32_to_cpu(es->s_first_error_line));
2720                 if (es->s_first_error_ino)
2721                         printk(": inode %u",
2722                                le32_to_cpu(es->s_first_error_ino));
2723                 if (es->s_first_error_block)
2724                         printk(": block %llu", (unsigned long long)
2725                                le64_to_cpu(es->s_first_error_block));
2726                 printk("\n");
2727         }
2728         if (es->s_last_error_time) {
2729                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2730                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2731                        (int) sizeof(es->s_last_error_func),
2732                        es->s_last_error_func,
2733                        le32_to_cpu(es->s_last_error_line));
2734                 if (es->s_last_error_ino)
2735                         printk(": inode %u",
2736                                le32_to_cpu(es->s_last_error_ino));
2737                 if (es->s_last_error_block)
2738                         printk(": block %llu", (unsigned long long)
2739                                le64_to_cpu(es->s_last_error_block));
2740                 printk("\n");
2741         }
2742         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2743 }
2744
2745 /* Find next suitable group and run ext4_init_inode_table */
2746 static int ext4_run_li_request(struct ext4_li_request *elr)
2747 {
2748         struct ext4_group_desc *gdp = NULL;
2749         ext4_group_t group, ngroups;
2750         struct super_block *sb;
2751         unsigned long timeout = 0;
2752         int ret = 0;
2753
2754         sb = elr->lr_super;
2755         ngroups = EXT4_SB(sb)->s_groups_count;
2756
2757         sb_start_write(sb);
2758         for (group = elr->lr_next_group; group < ngroups; group++) {
2759                 gdp = ext4_get_group_desc(sb, group, NULL);
2760                 if (!gdp) {
2761                         ret = 1;
2762                         break;
2763                 }
2764
2765                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2766                         break;
2767         }
2768
2769         if (group == ngroups)
2770                 ret = 1;
2771
2772         if (!ret) {
2773                 timeout = jiffies;
2774                 ret = ext4_init_inode_table(sb, group,
2775                                             elr->lr_timeout ? 0 : 1);
2776                 if (elr->lr_timeout == 0) {
2777                         timeout = (jiffies - timeout) *
2778                                   elr->lr_sbi->s_li_wait_mult;
2779                         elr->lr_timeout = timeout;
2780                 }
2781                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2782                 elr->lr_next_group = group + 1;
2783         }
2784         sb_end_write(sb);
2785
2786         return ret;
2787 }
2788
2789 /*
2790  * Remove lr_request from the list_request and free the
2791  * request structure. Should be called with li_list_mtx held
2792  */
2793 static void ext4_remove_li_request(struct ext4_li_request *elr)
2794 {
2795         struct ext4_sb_info *sbi;
2796
2797         if (!elr)
2798                 return;
2799
2800         sbi = elr->lr_sbi;
2801
2802         list_del(&elr->lr_request);
2803         sbi->s_li_request = NULL;
2804         kfree(elr);
2805 }
2806
2807 static void ext4_unregister_li_request(struct super_block *sb)
2808 {
2809         mutex_lock(&ext4_li_mtx);
2810         if (!ext4_li_info) {
2811                 mutex_unlock(&ext4_li_mtx);
2812                 return;
2813         }
2814
2815         mutex_lock(&ext4_li_info->li_list_mtx);
2816         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2817         mutex_unlock(&ext4_li_info->li_list_mtx);
2818         mutex_unlock(&ext4_li_mtx);
2819 }
2820
2821 static struct task_struct *ext4_lazyinit_task;
2822
2823 /*
2824  * This is the function where ext4lazyinit thread lives. It walks
2825  * through the request list searching for next scheduled filesystem.
2826  * When such a fs is found, run the lazy initialization request
2827  * (ext4_rn_li_request) and keep track of the time spend in this
2828  * function. Based on that time we compute next schedule time of
2829  * the request. When walking through the list is complete, compute
2830  * next waking time and put itself into sleep.
2831  */
2832 static int ext4_lazyinit_thread(void *arg)
2833 {
2834         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2835         struct list_head *pos, *n;
2836         struct ext4_li_request *elr;
2837         unsigned long next_wakeup, cur;
2838
2839         BUG_ON(NULL == eli);
2840
2841 cont_thread:
2842         while (true) {
2843                 next_wakeup = MAX_JIFFY_OFFSET;
2844
2845                 mutex_lock(&eli->li_list_mtx);
2846                 if (list_empty(&eli->li_request_list)) {
2847                         mutex_unlock(&eli->li_list_mtx);
2848                         goto exit_thread;
2849                 }
2850
2851                 list_for_each_safe(pos, n, &eli->li_request_list) {
2852                         elr = list_entry(pos, struct ext4_li_request,
2853                                          lr_request);
2854
2855                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2856                                 if (ext4_run_li_request(elr) != 0) {
2857                                         /* error, remove the lazy_init job */
2858                                         ext4_remove_li_request(elr);
2859                                         continue;
2860                                 }
2861                         }
2862
2863                         if (time_before(elr->lr_next_sched, next_wakeup))
2864                                 next_wakeup = elr->lr_next_sched;
2865                 }
2866                 mutex_unlock(&eli->li_list_mtx);
2867
2868                 try_to_freeze();
2869
2870                 cur = jiffies;
2871                 if ((time_after_eq(cur, next_wakeup)) ||
2872                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2873                         cond_resched();
2874                         continue;
2875                 }
2876
2877                 schedule_timeout_interruptible(next_wakeup - cur);
2878
2879                 if (kthread_should_stop()) {
2880                         ext4_clear_request_list();
2881                         goto exit_thread;
2882                 }
2883         }
2884
2885 exit_thread:
2886         /*
2887          * It looks like the request list is empty, but we need
2888          * to check it under the li_list_mtx lock, to prevent any
2889          * additions into it, and of course we should lock ext4_li_mtx
2890          * to atomically free the list and ext4_li_info, because at
2891          * this point another ext4 filesystem could be registering
2892          * new one.
2893          */
2894         mutex_lock(&ext4_li_mtx);
2895         mutex_lock(&eli->li_list_mtx);
2896         if (!list_empty(&eli->li_request_list)) {
2897                 mutex_unlock(&eli->li_list_mtx);
2898                 mutex_unlock(&ext4_li_mtx);
2899                 goto cont_thread;
2900         }
2901         mutex_unlock(&eli->li_list_mtx);
2902         kfree(ext4_li_info);
2903         ext4_li_info = NULL;
2904         mutex_unlock(&ext4_li_mtx);
2905
2906         return 0;
2907 }
2908
2909 static void ext4_clear_request_list(void)
2910 {
2911         struct list_head *pos, *n;
2912         struct ext4_li_request *elr;
2913
2914         mutex_lock(&ext4_li_info->li_list_mtx);
2915         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2916                 elr = list_entry(pos, struct ext4_li_request,
2917                                  lr_request);
2918                 ext4_remove_li_request(elr);
2919         }
2920         mutex_unlock(&ext4_li_info->li_list_mtx);
2921 }
2922
2923 static int ext4_run_lazyinit_thread(void)
2924 {
2925         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2926                                          ext4_li_info, "ext4lazyinit");
2927         if (IS_ERR(ext4_lazyinit_task)) {
2928                 int err = PTR_ERR(ext4_lazyinit_task);
2929                 ext4_clear_request_list();
2930                 kfree(ext4_li_info);
2931                 ext4_li_info = NULL;
2932                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2933                                  "initialization thread\n",
2934                                  err);
2935                 return err;
2936         }
2937         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2938         return 0;
2939 }
2940
2941 /*
2942  * Check whether it make sense to run itable init. thread or not.
2943  * If there is at least one uninitialized inode table, return
2944  * corresponding group number, else the loop goes through all
2945  * groups and return total number of groups.
2946  */
2947 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2948 {
2949         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2950         struct ext4_group_desc *gdp = NULL;
2951
2952         for (group = 0; group < ngroups; group++) {
2953                 gdp = ext4_get_group_desc(sb, group, NULL);
2954                 if (!gdp)
2955                         continue;
2956
2957                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2958                         break;
2959         }
2960
2961         return group;
2962 }
2963
2964 static int ext4_li_info_new(void)
2965 {
2966         struct ext4_lazy_init *eli = NULL;
2967
2968         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2969         if (!eli)
2970                 return -ENOMEM;
2971
2972         INIT_LIST_HEAD(&eli->li_request_list);
2973         mutex_init(&eli->li_list_mtx);
2974
2975         eli->li_state |= EXT4_LAZYINIT_QUIT;
2976
2977         ext4_li_info = eli;
2978
2979         return 0;
2980 }
2981
2982 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2983                                             ext4_group_t start)
2984 {
2985         struct ext4_sb_info *sbi = EXT4_SB(sb);
2986         struct ext4_li_request *elr;
2987         unsigned long rnd;
2988
2989         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2990         if (!elr)
2991                 return NULL;
2992
2993         elr->lr_super = sb;
2994         elr->lr_sbi = sbi;
2995         elr->lr_next_group = start;
2996
2997         /*
2998          * Randomize first schedule time of the request to
2999          * spread the inode table initialization requests
3000          * better.
3001          */
3002         get_random_bytes(&rnd, sizeof(rnd));
3003         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3004                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3005
3006         return elr;
3007 }
3008
3009 static int ext4_register_li_request(struct super_block *sb,
3010                                     ext4_group_t first_not_zeroed)
3011 {
3012         struct ext4_sb_info *sbi = EXT4_SB(sb);
3013         struct ext4_li_request *elr;
3014         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3015         int ret = 0;
3016
3017         if (sbi->s_li_request != NULL) {
3018                 /*
3019                  * Reset timeout so it can be computed again, because
3020                  * s_li_wait_mult might have changed.
3021                  */
3022                 sbi->s_li_request->lr_timeout = 0;
3023                 return 0;
3024         }
3025
3026         if (first_not_zeroed == ngroups ||
3027             (sb->s_flags & MS_RDONLY) ||
3028             !test_opt(sb, INIT_INODE_TABLE))
3029                 return 0;
3030
3031         elr = ext4_li_request_new(sb, first_not_zeroed);
3032         if (!elr)
3033                 return -ENOMEM;
3034
3035         mutex_lock(&ext4_li_mtx);
3036
3037         if (NULL == ext4_li_info) {
3038                 ret = ext4_li_info_new();
3039                 if (ret)
3040                         goto out;
3041         }
3042
3043         mutex_lock(&ext4_li_info->li_list_mtx);
3044         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3045         mutex_unlock(&ext4_li_info->li_list_mtx);
3046
3047         sbi->s_li_request = elr;
3048         /*
3049          * set elr to NULL here since it has been inserted to
3050          * the request_list and the removal and free of it is
3051          * handled by ext4_clear_request_list from now on.
3052          */
3053         elr = NULL;
3054
3055         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3056                 ret = ext4_run_lazyinit_thread();
3057                 if (ret)
3058                         goto out;
3059         }
3060 out:
3061         mutex_unlock(&ext4_li_mtx);
3062         if (ret)
3063                 kfree(elr);
3064         return ret;
3065 }
3066
3067 /*
3068  * We do not need to lock anything since this is called on
3069  * module unload.
3070  */
3071 static void ext4_destroy_lazyinit_thread(void)
3072 {
3073         /*
3074          * If thread exited earlier
3075          * there's nothing to be done.
3076          */
3077         if (!ext4_li_info || !ext4_lazyinit_task)
3078                 return;
3079
3080         kthread_stop(ext4_lazyinit_task);
3081 }
3082
3083 static int set_journal_csum_feature_set(struct super_block *sb)
3084 {
3085         int ret = 1;
3086         int compat, incompat;
3087         struct ext4_sb_info *sbi = EXT4_SB(sb);
3088
3089         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3090                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3091                 /* journal checksum v2 */
3092                 compat = 0;
3093                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3094         } else {
3095                 /* journal checksum v1 */
3096                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3097                 incompat = 0;
3098         }
3099
3100         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3101                 ret = jbd2_journal_set_features(sbi->s_journal,
3102                                 compat, 0,
3103                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3104                                 incompat);
3105         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3106                 ret = jbd2_journal_set_features(sbi->s_journal,
3107                                 compat, 0,
3108                                 incompat);
3109                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3110                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3111         } else {
3112                 jbd2_journal_clear_features(sbi->s_journal,
3113                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3114                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3115                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3116         }
3117
3118         return ret;
3119 }
3120
3121 /*
3122  * Note: calculating the overhead so we can be compatible with
3123  * historical BSD practice is quite difficult in the face of
3124  * clusters/bigalloc.  This is because multiple metadata blocks from
3125  * different block group can end up in the same allocation cluster.
3126  * Calculating the exact overhead in the face of clustered allocation
3127  * requires either O(all block bitmaps) in memory or O(number of block
3128  * groups**2) in time.  We will still calculate the superblock for
3129  * older file systems --- and if we come across with a bigalloc file
3130  * system with zero in s_overhead_clusters the estimate will be close to
3131  * correct especially for very large cluster sizes --- but for newer
3132  * file systems, it's better to calculate this figure once at mkfs
3133  * time, and store it in the superblock.  If the superblock value is
3134  * present (even for non-bigalloc file systems), we will use it.
3135  */
3136 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3137                           char *buf)
3138 {
3139         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3140         struct ext4_group_desc  *gdp;
3141         ext4_fsblk_t            first_block, last_block, b;
3142         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3143         int                     s, j, count = 0;
3144
3145         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3146                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3147                         sbi->s_itb_per_group + 2);
3148
3149         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3150                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3151         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3152         for (i = 0; i < ngroups; i++) {
3153                 gdp = ext4_get_group_desc(sb, i, NULL);
3154                 b = ext4_block_bitmap(sb, gdp);
3155                 if (b >= first_block && b <= last_block) {
3156                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3157                         count++;
3158                 }
3159                 b = ext4_inode_bitmap(sb, gdp);
3160                 if (b >= first_block && b <= last_block) {
3161                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3162                         count++;
3163                 }
3164                 b = ext4_inode_table(sb, gdp);
3165                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3166                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3167                                 int c = EXT4_B2C(sbi, b - first_block);
3168                                 ext4_set_bit(c, buf);
3169                                 count++;
3170                         }
3171                 if (i != grp)
3172                         continue;
3173                 s = 0;
3174                 if (ext4_bg_has_super(sb, grp)) {
3175                         ext4_set_bit(s++, buf);
3176                         count++;
3177                 }
3178                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3179                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3180                         count++;
3181                 }
3182         }
3183         if (!count)
3184                 return 0;
3185         return EXT4_CLUSTERS_PER_GROUP(sb) -
3186                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3187 }
3188
3189 /*
3190  * Compute the overhead and stash it in sbi->s_overhead
3191  */
3192 int ext4_calculate_overhead(struct super_block *sb)
3193 {
3194         struct ext4_sb_info *sbi = EXT4_SB(sb);
3195         struct ext4_super_block *es = sbi->s_es;
3196         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3197         ext4_fsblk_t overhead = 0;
3198         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3199
3200         if (!buf)
3201                 return -ENOMEM;
3202
3203         /*
3204          * Compute the overhead (FS structures).  This is constant
3205          * for a given filesystem unless the number of block groups
3206          * changes so we cache the previous value until it does.
3207          */
3208
3209         /*
3210          * All of the blocks before first_data_block are overhead
3211          */
3212         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3213
3214         /*
3215          * Add the overhead found in each block group
3216          */
3217         for (i = 0; i < ngroups; i++) {
3218                 int blks;
3219
3220                 blks = count_overhead(sb, i, buf);
3221                 overhead += blks;
3222                 if (blks)
3223                         memset(buf, 0, PAGE_SIZE);
3224                 cond_resched();
3225         }
3226         sbi->s_overhead = overhead;
3227         smp_wmb();
3228         free_page((unsigned long) buf);
3229         return 0;
3230 }
3231
3232 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3233 {
3234         char *orig_data = kstrdup(data, GFP_KERNEL);
3235         struct buffer_head *bh;
3236         struct ext4_super_block *es = NULL;
3237         struct ext4_sb_info *sbi;
3238         ext4_fsblk_t block;
3239         ext4_fsblk_t sb_block = get_sb_block(&data);
3240         ext4_fsblk_t logical_sb_block;
3241         unsigned long offset = 0;
3242         unsigned long journal_devnum = 0;
3243         unsigned long def_mount_opts;
3244         struct inode *root;
3245         char *cp;
3246         const char *descr;
3247         int ret = -ENOMEM;
3248         int blocksize, clustersize;
3249         unsigned int db_count;
3250         unsigned int i;
3251         int needs_recovery, has_huge_files, has_bigalloc;
3252         __u64 blocks_count;
3253         int err = 0;
3254         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3255         ext4_group_t first_not_zeroed;
3256
3257         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3258         if (!sbi)
3259                 goto out_free_orig;
3260
3261         sbi->s_blockgroup_lock =
3262                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3263         if (!sbi->s_blockgroup_lock) {
3264                 kfree(sbi);
3265                 goto out_free_orig;
3266         }
3267         sb->s_fs_info = sbi;
3268         sbi->s_sb = sb;
3269         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3270         sbi->s_sb_block = sb_block;
3271         if (sb->s_bdev->bd_part)
3272                 sbi->s_sectors_written_start =
3273                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3274
3275         /* Cleanup superblock name */
3276         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3277                 *cp = '!';
3278
3279         /* -EINVAL is default */
3280         ret = -EINVAL;
3281         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3282         if (!blocksize) {
3283                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3284                 goto out_fail;
3285         }
3286
3287         /*
3288          * The ext4 superblock will not be buffer aligned for other than 1kB
3289          * block sizes.  We need to calculate the offset from buffer start.
3290          */
3291         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3292                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3293                 offset = do_div(logical_sb_block, blocksize);
3294         } else {
3295                 logical_sb_block = sb_block;
3296         }
3297
3298         if (!(bh = sb_bread(sb, logical_sb_block))) {
3299                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3300                 goto out_fail;
3301         }
3302         /*
3303          * Note: s_es must be initialized as soon as possible because
3304          *       some ext4 macro-instructions depend on its value
3305          */
3306         es = (struct ext4_super_block *) (bh->b_data + offset);
3307         sbi->s_es = es;
3308         sb->s_magic = le16_to_cpu(es->s_magic);
3309         if (sb->s_magic != EXT4_SUPER_MAGIC)
3310                 goto cantfind_ext4;
3311         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3312
3313         /* Warn if metadata_csum and gdt_csum are both set. */
3314         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3315                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3316             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3317                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3318                              "redundant flags; please run fsck.");
3319
3320         /* Check for a known checksum algorithm */
3321         if (!ext4_verify_csum_type(sb, es)) {
3322                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3323                          "unknown checksum algorithm.");
3324                 silent = 1;
3325                 goto cantfind_ext4;
3326         }
3327
3328         /* Load the checksum driver */
3329         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3330                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3331                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3332                 if (IS_ERR(sbi->s_chksum_driver)) {
3333                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3334                         ret = PTR_ERR(sbi->s_chksum_driver);
3335                         sbi->s_chksum_driver = NULL;
3336                         goto failed_mount;
3337                 }
3338         }
3339
3340         /* Check superblock checksum */
3341         if (!ext4_superblock_csum_verify(sb, es)) {
3342                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3343                          "invalid superblock checksum.  Run e2fsck?");
3344                 silent = 1;
3345                 goto cantfind_ext4;
3346         }
3347
3348         /* Precompute checksum seed for all metadata */
3349         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3350                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3351                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3352                                                sizeof(es->s_uuid));
3353
3354         /* Set defaults before we parse the mount options */
3355         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3356         set_opt(sb, INIT_INODE_TABLE);
3357         if (def_mount_opts & EXT4_DEFM_DEBUG)
3358                 set_opt(sb, DEBUG);
3359         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3360                 set_opt(sb, GRPID);
3361         if (def_mount_opts & EXT4_DEFM_UID16)
3362                 set_opt(sb, NO_UID32);
3363         /* xattr user namespace & acls are now defaulted on */
3364         set_opt(sb, XATTR_USER);
3365 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3366         set_opt(sb, POSIX_ACL);
3367 #endif
3368         set_opt(sb, MBLK_IO_SUBMIT);
3369         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3370                 set_opt(sb, JOURNAL_DATA);
3371         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3372                 set_opt(sb, ORDERED_DATA);
3373         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3374                 set_opt(sb, WRITEBACK_DATA);
3375
3376         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3377                 set_opt(sb, ERRORS_PANIC);
3378         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3379                 set_opt(sb, ERRORS_CONT);
3380         else
3381                 set_opt(sb, ERRORS_RO);
3382         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3383                 set_opt(sb, BLOCK_VALIDITY);
3384         if (def_mount_opts & EXT4_DEFM_DISCARD)
3385                 set_opt(sb, DISCARD);
3386
3387         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3388         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3389         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3390         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3391         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3392
3393         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3394                 set_opt(sb, BARRIER);
3395
3396         /*
3397          * enable delayed allocation by default
3398          * Use -o nodelalloc to turn it off
3399          */
3400         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3401             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3402                 set_opt(sb, DELALLOC);
3403
3404         /*
3405          * set default s_li_wait_mult for lazyinit, for the case there is
3406          * no mount option specified.
3407          */
3408         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3409
3410         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3411                            &journal_devnum, &journal_ioprio, 0)) {
3412                 ext4_msg(sb, KERN_WARNING,
3413                          "failed to parse options in superblock: %s",
3414                          sbi->s_es->s_mount_opts);
3415         }
3416         sbi->s_def_mount_opt = sbi->s_mount_opt;
3417         if (!parse_options((char *) data, sb, &journal_devnum,
3418                            &journal_ioprio, 0))
3419                 goto failed_mount;
3420
3421         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3422                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3423                             "with data=journal disables delayed "
3424                             "allocation and O_DIRECT support!\n");
3425                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3426                         ext4_msg(sb, KERN_ERR, "can't mount with "
3427                                  "both data=journal and delalloc");
3428                         goto failed_mount;
3429                 }
3430                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3431                         ext4_msg(sb, KERN_ERR, "can't mount with "
3432                                  "both data=journal and delalloc");
3433                         goto failed_mount;
3434                 }
3435                 if (test_opt(sb, DELALLOC))
3436                         clear_opt(sb, DELALLOC);
3437         }
3438
3439         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3440         if (test_opt(sb, DIOREAD_NOLOCK)) {
3441                 if (blocksize < PAGE_SIZE) {
3442                         ext4_msg(sb, KERN_ERR, "can't mount with "
3443                                  "dioread_nolock if block size != PAGE_SIZE");
3444                         goto failed_mount;
3445                 }
3446         }
3447
3448         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3449                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3450
3451         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3452             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3453              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3454              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3455                 ext4_msg(sb, KERN_WARNING,
3456                        "feature flags set on rev 0 fs, "
3457                        "running e2fsck is recommended");
3458
3459         if (IS_EXT2_SB(sb)) {
3460                 if (ext2_feature_set_ok(sb))
3461                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3462                                  "using the ext4 subsystem");
3463                 else {
3464                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3465                                  "to feature incompatibilities");
3466                         goto failed_mount;
3467                 }
3468         }
3469
3470         if (IS_EXT3_SB(sb)) {
3471                 if (ext3_feature_set_ok(sb))
3472                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3473                                  "using the ext4 subsystem");
3474                 else {
3475                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3476                                  "to feature incompatibilities");
3477                         goto failed_mount;
3478                 }
3479         }
3480
3481         /*
3482          * Check feature flags regardless of the revision level, since we
3483          * previously didn't change the revision level when setting the flags,
3484          * so there is a chance incompat flags are set on a rev 0 filesystem.
3485          */
3486         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3487                 goto failed_mount;
3488
3489         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3490             blocksize > EXT4_MAX_BLOCK_SIZE) {
3491                 ext4_msg(sb, KERN_ERR,
3492                        "Unsupported filesystem blocksize %d", blocksize);
3493                 goto failed_mount;
3494         }
3495
3496         if (sb->s_blocksize != blocksize) {
3497                 /* Validate the filesystem blocksize */
3498                 if (!sb_set_blocksize(sb, blocksize)) {
3499                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3500                                         blocksize);
3501                         goto failed_mount;
3502                 }
3503
3504                 brelse(bh);
3505                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3506                 offset = do_div(logical_sb_block, blocksize);
3507                 bh = sb_bread(sb, logical_sb_block);
3508                 if (!bh) {
3509                         ext4_msg(sb, KERN_ERR,
3510                                "Can't read superblock on 2nd try");
3511                         goto failed_mount;
3512                 }
3513                 es = (struct ext4_super_block *)(bh->b_data + offset);
3514                 sbi->s_es = es;
3515                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3516                         ext4_msg(sb, KERN_ERR,
3517                                "Magic mismatch, very weird!");
3518                         goto failed_mount;
3519                 }
3520         }
3521
3522         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3523                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3524         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3525                                                       has_huge_files);
3526         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3527
3528         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3529                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3530                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3531         } else {
3532                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3533                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3534                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3535                     (!is_power_of_2(sbi->s_inode_size)) ||
3536                     (sbi->s_inode_size > blocksize)) {
3537                         ext4_msg(sb, KERN_ERR,
3538                                "unsupported inode size: %d",
3539                                sbi->s_inode_size);
3540                         goto failed_mount;
3541                 }
3542                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3543                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3544         }
3545
3546         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3547         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3548                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3549                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3550                     !is_power_of_2(sbi->s_desc_size)) {
3551                         ext4_msg(sb, KERN_ERR,
3552                                "unsupported descriptor size %lu",
3553                                sbi->s_desc_size);
3554                         goto failed_mount;
3555                 }
3556         } else
3557                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3558
3559         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3560         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3561         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3562                 goto cantfind_ext4;
3563
3564         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3565         if (sbi->s_inodes_per_block == 0)
3566                 goto cantfind_ext4;
3567         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3568                                         sbi->s_inodes_per_block;
3569         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3570         sbi->s_sbh = bh;
3571         sbi->s_mount_state = le16_to_cpu(es->s_state);
3572         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3573         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3574
3575         for (i = 0; i < 4; i++)
3576                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3577         sbi->s_def_hash_version = es->s_def_hash_version;
3578         i = le32_to_cpu(es->s_flags);
3579         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3580                 sbi->s_hash_unsigned = 3;
3581         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3582 #ifdef __CHAR_UNSIGNED__
3583                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3584                 sbi->s_hash_unsigned = 3;
3585 #else
3586                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3587 #endif
3588         }
3589
3590         /* Handle clustersize */
3591         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3592         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3593                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3594         if (has_bigalloc) {
3595                 if (clustersize < blocksize) {
3596                         ext4_msg(sb, KERN_ERR,
3597                                  "cluster size (%d) smaller than "
3598                                  "block size (%d)", clustersize, blocksize);
3599                         goto failed_mount;
3600                 }
3601                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3602                         le32_to_cpu(es->s_log_block_size);
3603                 sbi->s_clusters_per_group =
3604                         le32_to_cpu(es->s_clusters_per_group);
3605                 if (sbi->s_clusters_per_group > blocksize * 8) {
3606                         ext4_msg(sb, KERN_ERR,
3607                                  "#clusters per group too big: %lu",
3608                                  sbi->s_clusters_per_group);
3609                         goto failed_mount;
3610                 }
3611                 if (sbi->s_blocks_per_group !=
3612                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3613                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3614                                  "clusters per group (%lu) inconsistent",
3615                                  sbi->s_blocks_per_group,
3616                                  sbi->s_clusters_per_group);
3617                         goto failed_mount;
3618                 }
3619         } else {
3620                 if (clustersize != blocksize) {
3621                         ext4_warning(sb, "fragment/cluster size (%d) != "
3622                                      "block size (%d)", clustersize,
3623                                      blocksize);
3624                         clustersize = blocksize;
3625                 }
3626                 if (sbi->s_blocks_per_group > blocksize * 8) {
3627                         ext4_msg(sb, KERN_ERR,
3628                                  "#blocks per group too big: %lu",
3629                                  sbi->s_blocks_per_group);
3630                         goto failed_mount;
3631                 }
3632                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3633                 sbi->s_cluster_bits = 0;
3634         }
3635         sbi->s_cluster_ratio = clustersize / blocksize;
3636
3637         if (sbi->s_inodes_per_group > blocksize * 8) {
3638                 ext4_msg(sb, KERN_ERR,
3639                        "#inodes per group too big: %lu",
3640                        sbi->s_inodes_per_group);
3641                 goto failed_mount;
3642         }
3643
3644         /*
3645          * Test whether we have more sectors than will fit in sector_t,
3646          * and whether the max offset is addressable by the page cache.
3647          */
3648         err = generic_check_addressable(sb->s_blocksize_bits,
3649                                         ext4_blocks_count(es));
3650         if (err) {
3651                 ext4_msg(sb, KERN_ERR, "filesystem"
3652                          " too large to mount safely on this system");
3653                 if (sizeof(sector_t) < 8)
3654                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3655                 goto failed_mount;
3656         }
3657
3658         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3659                 goto cantfind_ext4;
3660
3661         /* check blocks count against device size */
3662         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3663         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3664                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3665                        "exceeds size of device (%llu blocks)",
3666                        ext4_blocks_count(es), blocks_count);
3667                 goto failed_mount;
3668         }
3669
3670         /*
3671          * It makes no sense for the first data block to be beyond the end
3672          * of the filesystem.
3673          */
3674         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3675                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3676                          "block %u is beyond end of filesystem (%llu)",
3677                          le32_to_cpu(es->s_first_data_block),
3678                          ext4_blocks_count(es));
3679                 goto failed_mount;
3680         }
3681         blocks_count = (ext4_blocks_count(es) -
3682                         le32_to_cpu(es->s_first_data_block) +
3683                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3684         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3685         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3686                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3687                        "(block count %llu, first data block %u, "
3688                        "blocks per group %lu)", sbi->s_groups_count,
3689                        ext4_blocks_count(es),
3690                        le32_to_cpu(es->s_first_data_block),
3691                        EXT4_BLOCKS_PER_GROUP(sb));
3692                 goto failed_mount;
3693         }
3694         sbi->s_groups_count = blocks_count;
3695         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3696                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3697         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3698                    EXT4_DESC_PER_BLOCK(sb);
3699         sbi->s_group_desc = ext4_kvmalloc(db_count *
3700                                           sizeof(struct buffer_head *),
3701                                           GFP_KERNEL);
3702         if (sbi->s_group_desc == NULL) {
3703                 ext4_msg(sb, KERN_ERR, "not enough memory");
3704                 ret = -ENOMEM;
3705                 goto failed_mount;
3706         }
3707
3708         if (ext4_proc_root)
3709                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3710
3711         if (sbi->s_proc)
3712                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3713                                  &ext4_seq_options_fops, sb);
3714
3715         bgl_lock_init(sbi->s_blockgroup_lock);
3716
3717         for (i = 0; i < db_count; i++) {
3718                 block = descriptor_loc(sb, logical_sb_block, i);
3719                 sbi->s_group_desc[i] = sb_bread(sb, block);
3720                 if (!sbi->s_group_desc[i]) {
3721                         ext4_msg(sb, KERN_ERR,
3722                                "can't read group descriptor %d", i);
3723                         db_count = i;
3724                         goto failed_mount2;
3725                 }
3726         }
3727         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3728                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3729                 goto failed_mount2;
3730         }
3731         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3732                 if (!ext4_fill_flex_info(sb)) {
3733                         ext4_msg(sb, KERN_ERR,
3734                                "unable to initialize "
3735                                "flex_bg meta info!");
3736                         goto failed_mount2;
3737                 }
3738
3739         sbi->s_gdb_count = db_count;
3740         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3741         spin_lock_init(&sbi->s_next_gen_lock);
3742
3743         init_timer(&sbi->s_err_report);
3744         sbi->s_err_report.function = print_daily_error_info;
3745         sbi->s_err_report.data = (unsigned long) sb;
3746
3747         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3748                         ext4_count_free_clusters(sb));
3749         if (!err) {
3750                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3751                                 ext4_count_free_inodes(sb));
3752         }
3753         if (!err) {
3754                 err = percpu_counter_init(&sbi->s_dirs_counter,
3755                                 ext4_count_dirs(sb));
3756         }
3757         if (!err) {
3758                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3759         }
3760         if (err) {
3761                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3762                 goto failed_mount3;
3763         }
3764
3765         sbi->s_stripe = ext4_get_stripe_size(sbi);
3766         sbi->s_max_writeback_mb_bump = 128;
3767         sbi->s_extent_max_zeroout_kb = 32;
3768
3769         /*
3770          * set up enough so that it can read an inode
3771          */
3772         if (!test_opt(sb, NOLOAD) &&
3773             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3774                 sb->s_op = &ext4_sops;
3775         else
3776                 sb->s_op = &ext4_nojournal_sops;
3777         sb->s_export_op = &ext4_export_ops;
3778         sb->s_xattr = ext4_xattr_handlers;
3779 #ifdef CONFIG_QUOTA
3780         sb->s_qcop = &ext4_qctl_operations;
3781         sb->dq_op = &ext4_quota_operations;
3782
3783         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
3784                 /* Use qctl operations for hidden quota files. */
3785                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3786         }
3787 #endif
3788         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3789
3790         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3791         mutex_init(&sbi->s_orphan_lock);
3792
3793         sb->s_root = NULL;
3794
3795         needs_recovery = (es->s_last_orphan != 0 ||
3796                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3797                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3798
3799         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3800             !(sb->s_flags & MS_RDONLY))
3801                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3802                         goto failed_mount3;
3803
3804         /*
3805          * The first inode we look at is the journal inode.  Don't try
3806          * root first: it may be modified in the journal!
3807          */
3808         if (!test_opt(sb, NOLOAD) &&
3809             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3810                 if (ext4_load_journal(sb, es, journal_devnum))
3811                         goto failed_mount3;
3812         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3813               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3814                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3815                        "suppressed and not mounted read-only");
3816                 goto failed_mount_wq;
3817         } else {
3818                 clear_opt(sb, DATA_FLAGS);
3819                 sbi->s_journal = NULL;
3820                 needs_recovery = 0;
3821                 goto no_journal;
3822         }
3823
3824         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3825             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3826                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3827                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3828                 goto failed_mount_wq;
3829         }
3830
3831         if (!set_journal_csum_feature_set(sb)) {
3832                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3833                          "feature set");
3834                 goto failed_mount_wq;
3835         }
3836
3837         /* We have now updated the journal if required, so we can
3838          * validate the data journaling mode. */
3839         switch (test_opt(sb, DATA_FLAGS)) {
3840         case 0:
3841                 /* No mode set, assume a default based on the journal
3842                  * capabilities: ORDERED_DATA if the journal can
3843                  * cope, else JOURNAL_DATA
3844                  */
3845                 if (jbd2_journal_check_available_features
3846                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3847                         set_opt(sb, ORDERED_DATA);
3848                 else
3849                         set_opt(sb, JOURNAL_DATA);
3850                 break;
3851
3852         case EXT4_MOUNT_ORDERED_DATA:
3853         case EXT4_MOUNT_WRITEBACK_DATA:
3854                 if (!jbd2_journal_check_available_features
3855                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3856                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3857                                "requested data journaling mode");
3858                         goto failed_mount_wq;
3859                 }
3860         default:
3861                 break;
3862         }
3863         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3864
3865         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3866
3867         /*
3868          * The journal may have updated the bg summary counts, so we
3869          * need to update the global counters.
3870          */
3871         percpu_counter_set(&sbi->s_freeclusters_counter,
3872                            ext4_count_free_clusters(sb));
3873         percpu_counter_set(&sbi->s_freeinodes_counter,
3874                            ext4_count_free_inodes(sb));
3875         percpu_counter_set(&sbi->s_dirs_counter,
3876                            ext4_count_dirs(sb));
3877         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3878
3879 no_journal:
3880         /*
3881          * Get the # of file system overhead blocks from the
3882          * superblock if present.
3883          */
3884         if (es->s_overhead_clusters)
3885                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3886         else {
3887                 err = ext4_calculate_overhead(sb);
3888                 if (err)
3889                         goto failed_mount_wq;
3890         }
3891
3892         /*
3893          * The maximum number of concurrent works can be high and
3894          * concurrency isn't really necessary.  Limit it to 1.
3895          */
3896         EXT4_SB(sb)->dio_unwritten_wq =
3897                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3898         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3899                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3900                 ret = -ENOMEM;
3901                 goto failed_mount_wq;
3902         }
3903
3904         /*
3905          * The jbd2_journal_load will have done any necessary log recovery,
3906          * so we can safely mount the rest of the filesystem now.
3907          */
3908
3909         root = ext4_iget(sb, EXT4_ROOT_INO);
3910         if (IS_ERR(root)) {
3911                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3912                 ret = PTR_ERR(root);
3913                 root = NULL;
3914                 goto failed_mount4;
3915         }
3916         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3917                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3918                 iput(root);
3919                 goto failed_mount4;
3920         }
3921         sb->s_root = d_make_root(root);
3922         if (!sb->s_root) {
3923                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3924                 ret = -ENOMEM;
3925                 goto failed_mount4;
3926         }
3927
3928         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3929                 sb->s_flags |= MS_RDONLY;
3930
3931         /* determine the minimum size of new large inodes, if present */
3932         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3933                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3934                                                      EXT4_GOOD_OLD_INODE_SIZE;
3935                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3936                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3937                         if (sbi->s_want_extra_isize <
3938                             le16_to_cpu(es->s_want_extra_isize))
3939                                 sbi->s_want_extra_isize =
3940                                         le16_to_cpu(es->s_want_extra_isize);
3941                         if (sbi->s_want_extra_isize <
3942                             le16_to_cpu(es->s_min_extra_isize))
3943                                 sbi->s_want_extra_isize =
3944                                         le16_to_cpu(es->s_min_extra_isize);
3945                 }
3946         }
3947         /* Check if enough inode space is available */
3948         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3949                                                         sbi->s_inode_size) {
3950                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3951                                                        EXT4_GOOD_OLD_INODE_SIZE;
3952                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3953                          "available");
3954         }
3955
3956         err = ext4_setup_system_zone(sb);
3957         if (err) {
3958                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3959                          "zone (%d)", err);
3960                 goto failed_mount4a;
3961         }
3962
3963         ext4_ext_init(sb);
3964         err = ext4_mb_init(sb);
3965         if (err) {
3966                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3967                          err);
3968                 goto failed_mount5;
3969         }
3970
3971         err = ext4_register_li_request(sb, first_not_zeroed);
3972         if (err)
3973                 goto failed_mount6;
3974
3975         sbi->s_kobj.kset = ext4_kset;
3976         init_completion(&sbi->s_kobj_unregister);
3977         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3978                                    "%s", sb->s_id);
3979         if (err)
3980                 goto failed_mount7;
3981
3982         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3983         ext4_orphan_cleanup(sb, es);
3984         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3985         if (needs_recovery) {
3986                 ext4_msg(sb, KERN_INFO, "recovery complete");
3987                 ext4_mark_recovery_complete(sb, es);
3988         }
3989         if (EXT4_SB(sb)->s_journal) {
3990                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3991                         descr = " journalled data mode";
3992                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3993                         descr = " ordered data mode";
3994                 else
3995                         descr = " writeback data mode";
3996         } else
3997                 descr = "out journal";
3998
3999 #ifdef CONFIG_QUOTA
4000         /* Enable quota usage during mount. */
4001         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4002             !(sb->s_flags & MS_RDONLY)) {
4003                 err = ext4_enable_quotas(sb);
4004                 if (err)
4005                         goto failed_mount7;
4006         }
4007 #endif  /* CONFIG_QUOTA */
4008
4009         if (test_opt(sb, DISCARD)) {
4010                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4011                 if (!blk_queue_discard(q))
4012                         ext4_msg(sb, KERN_WARNING,
4013                                  "mounting with \"discard\" option, but "
4014                                  "the device does not support discard");
4015         }
4016
4017         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4018                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4019                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4020
4021         if (es->s_error_count)
4022                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4023
4024         kfree(orig_data);
4025         return 0;
4026
4027 cantfind_ext4:
4028         if (!silent)
4029                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4030         goto failed_mount;
4031
4032 failed_mount7:
4033         ext4_unregister_li_request(sb);
4034 failed_mount6:
4035         ext4_mb_release(sb);
4036 failed_mount5:
4037         ext4_ext_release(sb);
4038         ext4_release_system_zone(sb);
4039 failed_mount4a:
4040         dput(sb->s_root);
4041         sb->s_root = NULL;
4042 failed_mount4:
4043         ext4_msg(sb, KERN_ERR, "mount failed");
4044         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4045 failed_mount_wq:
4046         if (sbi->s_journal) {
4047                 jbd2_journal_destroy(sbi->s_journal);
4048                 sbi->s_journal = NULL;
4049         }
4050 failed_mount3:
4051         del_timer(&sbi->s_err_report);
4052         if (sbi->s_flex_groups)
4053                 ext4_kvfree(sbi->s_flex_groups);
4054         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4055         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4056         percpu_counter_destroy(&sbi->s_dirs_counter);
4057         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4058         if (sbi->s_mmp_tsk)
4059                 kthread_stop(sbi->s_mmp_tsk);
4060 failed_mount2:
4061         for (i = 0; i < db_count; i++)
4062                 brelse(sbi->s_group_desc[i]);
4063         ext4_kvfree(sbi->s_group_desc);
4064 failed_mount:
4065         if (sbi->s_chksum_driver)
4066                 crypto_free_shash(sbi->s_chksum_driver);
4067         if (sbi->s_proc) {
4068                 remove_proc_entry("options", sbi->s_proc);
4069                 remove_proc_entry(sb->s_id, ext4_proc_root);
4070         }
4071 #ifdef CONFIG_QUOTA
4072         for (i = 0; i < MAXQUOTAS; i++)
4073                 kfree(sbi->s_qf_names[i]);
4074 #endif
4075         ext4_blkdev_remove(sbi);
4076         brelse(bh);
4077 out_fail:
4078         sb->s_fs_info = NULL;
4079         kfree(sbi->s_blockgroup_lock);
4080         kfree(sbi);
4081 out_free_orig:
4082         kfree(orig_data);
4083         return err ? err : ret;
4084 }
4085
4086 /*
4087  * Setup any per-fs journal parameters now.  We'll do this both on
4088  * initial mount, once the journal has been initialised but before we've
4089  * done any recovery; and again on any subsequent remount.
4090  */
4091 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4092 {
4093         struct ext4_sb_info *sbi = EXT4_SB(sb);
4094
4095         journal->j_commit_interval = sbi->s_commit_interval;
4096         journal->j_min_batch_time = sbi->s_min_batch_time;
4097         journal->j_max_batch_time = sbi->s_max_batch_time;
4098
4099         write_lock(&journal->j_state_lock);
4100         if (test_opt(sb, BARRIER))
4101                 journal->j_flags |= JBD2_BARRIER;
4102         else
4103                 journal->j_flags &= ~JBD2_BARRIER;
4104         if (test_opt(sb, DATA_ERR_ABORT))
4105                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4106         else
4107                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4108         write_unlock(&journal->j_state_lock);
4109 }
4110
4111 static journal_t *ext4_get_journal(struct super_block *sb,
4112                                    unsigned int journal_inum)
4113 {
4114         struct inode *journal_inode;
4115         journal_t *journal;
4116
4117         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4118
4119         /* First, test for the existence of a valid inode on disk.  Bad
4120          * things happen if we iget() an unused inode, as the subsequent
4121          * iput() will try to delete it. */
4122
4123         journal_inode = ext4_iget(sb, journal_inum);
4124         if (IS_ERR(journal_inode)) {
4125                 ext4_msg(sb, KERN_ERR, "no journal found");
4126                 return NULL;
4127         }
4128         if (!journal_inode->i_nlink) {
4129                 make_bad_inode(journal_inode);
4130                 iput(journal_inode);
4131                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4132                 return NULL;
4133         }
4134
4135         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4136                   journal_inode, journal_inode->i_size);
4137         if (!S_ISREG(journal_inode->i_mode)) {
4138                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4139                 iput(journal_inode);
4140                 return NULL;
4141         }
4142
4143         journal = jbd2_journal_init_inode(journal_inode);
4144         if (!journal) {
4145                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4146                 iput(journal_inode);
4147                 return NULL;
4148         }
4149         journal->j_private = sb;
4150         ext4_init_journal_params(sb, journal);
4151         return journal;
4152 }
4153
4154 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4155                                        dev_t j_dev)
4156 {
4157         struct buffer_head *bh;
4158         journal_t *journal;
4159         ext4_fsblk_t start;
4160         ext4_fsblk_t len;
4161         int hblock, blocksize;
4162         ext4_fsblk_t sb_block;
4163         unsigned long offset;
4164         struct ext4_super_block *es;
4165         struct block_device *bdev;
4166
4167         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4168
4169         bdev = ext4_blkdev_get(j_dev, sb);
4170         if (bdev == NULL)
4171                 return NULL;
4172
4173         blocksize = sb->s_blocksize;
4174         hblock = bdev_logical_block_size(bdev);
4175         if (blocksize < hblock) {
4176                 ext4_msg(sb, KERN_ERR,
4177                         "blocksize too small for journal device");
4178                 goto out_bdev;
4179         }
4180
4181         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4182         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4183         set_blocksize(bdev, blocksize);
4184         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4185                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4186                        "external journal");
4187                 goto out_bdev;
4188         }
4189
4190         es = (struct ext4_super_block *) (bh->b_data + offset);
4191         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4192             !(le32_to_cpu(es->s_feature_incompat) &
4193               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4194                 ext4_msg(sb, KERN_ERR, "external journal has "
4195                                         "bad superblock");
4196                 brelse(bh);
4197                 goto out_bdev;
4198         }
4199
4200         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4201                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4202                 brelse(bh);
4203                 goto out_bdev;
4204         }
4205
4206         len = ext4_blocks_count(es);
4207         start = sb_block + 1;
4208         brelse(bh);     /* we're done with the superblock */
4209
4210         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4211                                         start, len, blocksize);
4212         if (!journal) {
4213                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4214                 goto out_bdev;
4215         }
4216         journal->j_private = sb;
4217         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4218         wait_on_buffer(journal->j_sb_buffer);
4219         if (!buffer_uptodate(journal->j_sb_buffer)) {
4220                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4221                 goto out_journal;
4222         }
4223         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4224                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4225                                         "user (unsupported) - %d",
4226                         be32_to_cpu(journal->j_superblock->s_nr_users));
4227                 goto out_journal;
4228         }
4229         EXT4_SB(sb)->journal_bdev = bdev;
4230         ext4_init_journal_params(sb, journal);
4231         return journal;
4232
4233 out_journal:
4234         jbd2_journal_destroy(journal);
4235 out_bdev:
4236         ext4_blkdev_put(bdev);
4237         return NULL;
4238 }
4239
4240 static int ext4_load_journal(struct super_block *sb,
4241                              struct ext4_super_block *es,
4242                              unsigned long journal_devnum)
4243 {
4244         journal_t *journal;
4245         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4246         dev_t journal_dev;
4247         int err = 0;
4248         int really_read_only;
4249
4250         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4251
4252         if (journal_devnum &&
4253             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4254                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4255                         "numbers have changed");
4256                 journal_dev = new_decode_dev(journal_devnum);
4257         } else
4258                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4259
4260         really_read_only = bdev_read_only(sb->s_bdev);
4261
4262         /*
4263          * Are we loading a blank journal or performing recovery after a
4264          * crash?  For recovery, we need to check in advance whether we
4265          * can get read-write access to the device.
4266          */
4267         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4268                 if (sb->s_flags & MS_RDONLY) {
4269                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4270                                         "required on readonly filesystem");
4271                         if (really_read_only) {
4272                                 ext4_msg(sb, KERN_ERR, "write access "
4273                                         "unavailable, cannot proceed");
4274                                 return -EROFS;
4275                         }
4276                         ext4_msg(sb, KERN_INFO, "write access will "
4277                                "be enabled during recovery");
4278                 }
4279         }
4280
4281         if (journal_inum && journal_dev) {
4282                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4283                        "and inode journals!");
4284                 return -EINVAL;
4285         }
4286
4287         if (journal_inum) {
4288                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4289                         return -EINVAL;
4290         } else {
4291                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4292                         return -EINVAL;
4293         }
4294
4295         if (!(journal->j_flags & JBD2_BARRIER))
4296                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4297
4298         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4299                 err = jbd2_journal_wipe(journal, !really_read_only);
4300         if (!err) {
4301                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4302                 if (save)
4303                         memcpy(save, ((char *) es) +
4304                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4305                 err = jbd2_journal_load(journal);
4306                 if (save)
4307                         memcpy(((char *) es) + EXT4_S_ERR_START,
4308                                save, EXT4_S_ERR_LEN);
4309                 kfree(save);
4310         }
4311
4312         if (err) {
4313                 ext4_msg(sb, KERN_ERR, "error loading journal");
4314                 jbd2_journal_destroy(journal);
4315                 return err;
4316         }
4317
4318         EXT4_SB(sb)->s_journal = journal;
4319         ext4_clear_journal_err(sb, es);
4320
4321         if (!really_read_only && journal_devnum &&
4322             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4323                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4324
4325                 /* Make sure we flush the recovery flag to disk. */
4326                 ext4_commit_super(sb, 1);
4327         }
4328
4329         return 0;
4330 }
4331
4332 static int ext4_commit_super(struct super_block *sb, int sync)
4333 {
4334         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4335         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4336         int error = 0;
4337
4338         if (!sbh || block_device_ejected(sb))
4339                 return error;
4340         if (buffer_write_io_error(sbh)) {
4341                 /*
4342                  * Oh, dear.  A previous attempt to write the
4343                  * superblock failed.  This could happen because the
4344                  * USB device was yanked out.  Or it could happen to
4345                  * be a transient write error and maybe the block will
4346                  * be remapped.  Nothing we can do but to retry the
4347                  * write and hope for the best.
4348                  */
4349                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4350                        "superblock detected");
4351                 clear_buffer_write_io_error(sbh);
4352                 set_buffer_uptodate(sbh);
4353         }
4354         /*
4355          * If the file system is mounted read-only, don't update the
4356          * superblock write time.  This avoids updating the superblock
4357          * write time when we are mounting the root file system
4358          * read/only but we need to replay the journal; at that point,
4359          * for people who are east of GMT and who make their clock
4360          * tick in localtime for Windows bug-for-bug compatibility,
4361          * the clock is set in the future, and this will cause e2fsck
4362          * to complain and force a full file system check.
4363          */
4364         if (!(sb->s_flags & MS_RDONLY))
4365                 es->s_wtime = cpu_to_le32(get_seconds());
4366         if (sb->s_bdev->bd_part)
4367                 es->s_kbytes_written =
4368                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4369                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4370                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4371         else
4372                 es->s_kbytes_written =
4373                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4374         ext4_free_blocks_count_set(es,
4375                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4376                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4377         es->s_free_inodes_count =
4378                 cpu_to_le32(percpu_counter_sum_positive(
4379                                 &EXT4_SB(sb)->s_freeinodes_counter));
4380         BUFFER_TRACE(sbh, "marking dirty");
4381         ext4_superblock_csum_set(sb);
4382         mark_buffer_dirty(sbh);
4383         if (sync) {
4384                 error = sync_dirty_buffer(sbh);
4385                 if (error)
4386                         return error;
4387
4388                 error = buffer_write_io_error(sbh);
4389                 if (error) {
4390                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4391                                "superblock");
4392                         clear_buffer_write_io_error(sbh);
4393                         set_buffer_uptodate(sbh);
4394                 }
4395         }
4396         return error;
4397 }
4398
4399 /*
4400  * Have we just finished recovery?  If so, and if we are mounting (or
4401  * remounting) the filesystem readonly, then we will end up with a
4402  * consistent fs on disk.  Record that fact.
4403  */
4404 static void ext4_mark_recovery_complete(struct super_block *sb,
4405                                         struct ext4_super_block *es)
4406 {
4407         journal_t *journal = EXT4_SB(sb)->s_journal;
4408
4409         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4410                 BUG_ON(journal != NULL);
4411                 return;
4412         }
4413         jbd2_journal_lock_updates(journal);
4414         if (jbd2_journal_flush(journal) < 0)
4415                 goto out;
4416
4417         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4418             sb->s_flags & MS_RDONLY) {
4419                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4420                 ext4_commit_super(sb, 1);
4421         }
4422
4423 out:
4424         jbd2_journal_unlock_updates(journal);
4425 }
4426
4427 /*
4428  * If we are mounting (or read-write remounting) a filesystem whose journal
4429  * has recorded an error from a previous lifetime, move that error to the
4430  * main filesystem now.
4431  */
4432 static void ext4_clear_journal_err(struct super_block *sb,
4433                                    struct ext4_super_block *es)
4434 {
4435         journal_t *journal;
4436         int j_errno;
4437         const char *errstr;
4438
4439         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4440
4441         journal = EXT4_SB(sb)->s_journal;
4442
4443         /*
4444          * Now check for any error status which may have been recorded in the
4445          * journal by a prior ext4_error() or ext4_abort()
4446          */
4447
4448         j_errno = jbd2_journal_errno(journal);
4449         if (j_errno) {
4450                 char nbuf[16];
4451
4452                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4453                 ext4_warning(sb, "Filesystem error recorded "
4454                              "from previous mount: %s", errstr);
4455                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4456
4457                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4458                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4459                 ext4_commit_super(sb, 1);
4460
4461                 jbd2_journal_clear_err(journal);
4462                 jbd2_journal_update_sb_errno(journal);
4463         }
4464 }
4465
4466 /*
4467  * Force the running and committing transactions to commit,
4468  * and wait on the commit.
4469  */
4470 int ext4_force_commit(struct super_block *sb)
4471 {
4472         journal_t *journal;
4473         int ret = 0;
4474
4475         if (sb->s_flags & MS_RDONLY)
4476                 return 0;
4477
4478         journal = EXT4_SB(sb)->s_journal;
4479         if (journal)
4480                 ret = ext4_journal_force_commit(journal);
4481
4482         return ret;
4483 }
4484
4485 static int ext4_sync_fs(struct super_block *sb, int wait)
4486 {
4487         int ret = 0;
4488         tid_t target;
4489         struct ext4_sb_info *sbi = EXT4_SB(sb);
4490
4491         trace_ext4_sync_fs(sb, wait);
4492         flush_workqueue(sbi->dio_unwritten_wq);
4493         /*
4494          * Writeback quota in non-journalled quota case - journalled quota has
4495          * no dirty dquots
4496          */
4497         dquot_writeback_dquots(sb, -1);
4498         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4499                 if (wait)
4500                         jbd2_log_wait_commit(sbi->s_journal, target);
4501         }
4502         return ret;
4503 }
4504
4505 /*
4506  * LVM calls this function before a (read-only) snapshot is created.  This
4507  * gives us a chance to flush the journal completely and mark the fs clean.
4508  *
4509  * Note that only this function cannot bring a filesystem to be in a clean
4510  * state independently. It relies on upper layer to stop all data & metadata
4511  * modifications.
4512  */
4513 static int ext4_freeze(struct super_block *sb)
4514 {
4515         int error = 0;
4516         journal_t *journal;
4517
4518         if (sb->s_flags & MS_RDONLY)
4519                 return 0;
4520
4521         journal = EXT4_SB(sb)->s_journal;
4522
4523         /* Now we set up the journal barrier. */
4524         jbd2_journal_lock_updates(journal);
4525
4526         /*
4527          * Don't clear the needs_recovery flag if we failed to flush
4528          * the journal.
4529          */
4530         error = jbd2_journal_flush(journal);
4531         if (error < 0)
4532                 goto out;
4533
4534         /* Journal blocked and flushed, clear needs_recovery flag. */
4535         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4536         error = ext4_commit_super(sb, 1);
4537 out:
4538         /* we rely on upper layer to stop further updates */
4539         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4540         return error;
4541 }
4542
4543 /*
4544  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4545  * flag here, even though the filesystem is not technically dirty yet.
4546  */
4547 static int ext4_unfreeze(struct super_block *sb)
4548 {
4549         if (sb->s_flags & MS_RDONLY)
4550                 return 0;
4551
4552         /* Reset the needs_recovery flag before the fs is unlocked. */
4553         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4554         ext4_commit_super(sb, 1);
4555         return 0;
4556 }
4557
4558 /*
4559  * Structure to save mount options for ext4_remount's benefit
4560  */
4561 struct ext4_mount_options {
4562         unsigned long s_mount_opt;
4563         unsigned long s_mount_opt2;
4564         kuid_t s_resuid;
4565         kgid_t s_resgid;
4566         unsigned long s_commit_interval;
4567         u32 s_min_batch_time, s_max_batch_time;
4568 #ifdef CONFIG_QUOTA
4569         int s_jquota_fmt;
4570         char *s_qf_names[MAXQUOTAS];
4571 #endif
4572 };
4573
4574 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4575 {
4576         struct ext4_super_block *es;
4577         struct ext4_sb_info *sbi = EXT4_SB(sb);
4578         unsigned long old_sb_flags;
4579         struct ext4_mount_options old_opts;
4580         int enable_quota = 0;
4581         ext4_group_t g;
4582         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4583         int err = 0;
4584 #ifdef CONFIG_QUOTA
4585         int i;
4586 #endif
4587         char *orig_data = kstrdup(data, GFP_KERNEL);
4588
4589         /* Store the original options */
4590         old_sb_flags = sb->s_flags;
4591         old_opts.s_mount_opt = sbi->s_mount_opt;
4592         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4593         old_opts.s_resuid = sbi->s_resuid;
4594         old_opts.s_resgid = sbi->s_resgid;
4595         old_opts.s_commit_interval = sbi->s_commit_interval;
4596         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4597         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4598 #ifdef CONFIG_QUOTA
4599         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4600         for (i = 0; i < MAXQUOTAS; i++)
4601                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4602 #endif
4603         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4604                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4605
4606         /*
4607          * Allow the "check" option to be passed as a remount option.
4608          */
4609         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4610                 err = -EINVAL;
4611                 goto restore_opts;
4612         }
4613
4614         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4615                 ext4_abort(sb, "Abort forced by user");
4616
4617         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4618                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4619
4620         es = sbi->s_es;
4621
4622         if (sbi->s_journal) {
4623                 ext4_init_journal_params(sb, sbi->s_journal);
4624                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4625         }
4626
4627         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4628                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4629                         err = -EROFS;
4630                         goto restore_opts;
4631                 }
4632
4633                 if (*flags & MS_RDONLY) {
4634                         err = dquot_suspend(sb, -1);
4635                         if (err < 0)
4636                                 goto restore_opts;
4637
4638                         /*
4639                          * First of all, the unconditional stuff we have to do
4640                          * to disable replay of the journal when we next remount
4641                          */
4642                         sb->s_flags |= MS_RDONLY;
4643
4644                         /*
4645                          * OK, test if we are remounting a valid rw partition
4646                          * readonly, and if so set the rdonly flag and then
4647                          * mark the partition as valid again.
4648                          */
4649                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4650                             (sbi->s_mount_state & EXT4_VALID_FS))
4651                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4652
4653                         if (sbi->s_journal)
4654                                 ext4_mark_recovery_complete(sb, es);
4655                 } else {
4656                         /* Make sure we can mount this feature set readwrite */
4657                         if (!ext4_feature_set_ok(sb, 0)) {
4658                                 err = -EROFS;
4659                                 goto restore_opts;
4660                         }
4661                         /*
4662                          * Make sure the group descriptor checksums
4663                          * are sane.  If they aren't, refuse to remount r/w.
4664                          */
4665                         for (g = 0; g < sbi->s_groups_count; g++) {
4666                                 struct ext4_group_desc *gdp =
4667                                         ext4_get_group_desc(sb, g, NULL);
4668
4669                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4670                                         ext4_msg(sb, KERN_ERR,
4671                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4672                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4673                                                le16_to_cpu(gdp->bg_checksum));
4674                                         err = -EINVAL;
4675                                         goto restore_opts;
4676                                 }
4677                         }
4678
4679                         /*
4680                          * If we have an unprocessed orphan list hanging
4681                          * around from a previously readonly bdev mount,
4682                          * require a full umount/remount for now.
4683                          */
4684                         if (es->s_last_orphan) {
4685                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4686                                        "remount RDWR because of unprocessed "
4687                                        "orphan inode list.  Please "
4688                                        "umount/remount instead");
4689                                 err = -EINVAL;
4690                                 goto restore_opts;
4691                         }
4692
4693                         /*
4694                          * Mounting a RDONLY partition read-write, so reread
4695                          * and store the current valid flag.  (It may have
4696                          * been changed by e2fsck since we originally mounted
4697                          * the partition.)
4698                          */
4699                         if (sbi->s_journal)
4700                                 ext4_clear_journal_err(sb, es);
4701                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4702                         if (!ext4_setup_super(sb, es, 0))
4703                                 sb->s_flags &= ~MS_RDONLY;
4704                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4705                                                      EXT4_FEATURE_INCOMPAT_MMP))
4706                                 if (ext4_multi_mount_protect(sb,
4707                                                 le64_to_cpu(es->s_mmp_block))) {
4708                                         err = -EROFS;
4709                                         goto restore_opts;
4710                                 }
4711                         enable_quota = 1;
4712                 }
4713         }
4714
4715         /*
4716          * Reinitialize lazy itable initialization thread based on
4717          * current settings
4718          */
4719         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4720                 ext4_unregister_li_request(sb);
4721         else {
4722                 ext4_group_t first_not_zeroed;
4723                 first_not_zeroed = ext4_has_uninit_itable(sb);
4724                 ext4_register_li_request(sb, first_not_zeroed);
4725         }
4726
4727         ext4_setup_system_zone(sb);
4728         if (sbi->s_journal == NULL)
4729                 ext4_commit_super(sb, 1);
4730
4731 #ifdef CONFIG_QUOTA
4732         /* Release old quota file names */
4733         for (i = 0; i < MAXQUOTAS; i++)
4734                 if (old_opts.s_qf_names[i] &&
4735                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4736                         kfree(old_opts.s_qf_names[i]);
4737         if (enable_quota) {
4738                 if (sb_any_quota_suspended(sb))
4739                         dquot_resume(sb, -1);
4740                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4741                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4742                         err = ext4_enable_quotas(sb);
4743                         if (err)
4744                                 goto restore_opts;
4745                 }
4746         }
4747 #endif
4748
4749         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4750         kfree(orig_data);
4751         return 0;
4752
4753 restore_opts:
4754         sb->s_flags = old_sb_flags;
4755         sbi->s_mount_opt = old_opts.s_mount_opt;
4756         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4757         sbi->s_resuid = old_opts.s_resuid;
4758         sbi->s_resgid = old_opts.s_resgid;
4759         sbi->s_commit_interval = old_opts.s_commit_interval;
4760         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4761         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4762 #ifdef CONFIG_QUOTA
4763         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4764         for (i = 0; i < MAXQUOTAS; i++) {
4765                 if (sbi->s_qf_names[i] &&
4766                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4767                         kfree(sbi->s_qf_names[i]);
4768                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4769         }
4770 #endif
4771         kfree(orig_data);
4772         return err;
4773 }
4774
4775 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4776 {
4777         struct super_block *sb = dentry->d_sb;
4778         struct ext4_sb_info *sbi = EXT4_SB(sb);
4779         struct ext4_super_block *es = sbi->s_es;
4780         ext4_fsblk_t overhead = 0;
4781         u64 fsid;
4782         s64 bfree;
4783
4784         if (!test_opt(sb, MINIX_DF))
4785                 overhead = sbi->s_overhead;
4786
4787         buf->f_type = EXT4_SUPER_MAGIC;
4788         buf->f_bsize = sb->s_blocksize;
4789         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4790         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4791                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4792         /* prevent underflow in case that few free space is available */
4793         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4794         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4795         if (buf->f_bfree < ext4_r_blocks_count(es))
4796                 buf->f_bavail = 0;
4797         buf->f_files = le32_to_cpu(es->s_inodes_count);
4798         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4799         buf->f_namelen = EXT4_NAME_LEN;
4800         fsid = le64_to_cpup((void *)es->s_uuid) ^
4801                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4802         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4803         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4804
4805         return 0;
4806 }
4807
4808 /* Helper function for writing quotas on sync - we need to start transaction
4809  * before quota file is locked for write. Otherwise the are possible deadlocks:
4810  * Process 1                         Process 2
4811  * ext4_create()                     quota_sync()
4812  *   jbd2_journal_start()                  write_dquot()
4813  *   dquot_initialize()                         down(dqio_mutex)
4814  *     down(dqio_mutex)                    jbd2_journal_start()
4815  *
4816  */
4817
4818 #ifdef CONFIG_QUOTA
4819
4820 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4821 {
4822         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4823 }
4824
4825 static int ext4_write_dquot(struct dquot *dquot)
4826 {
4827         int ret, err;
4828         handle_t *handle;
4829         struct inode *inode;
4830
4831         inode = dquot_to_inode(dquot);
4832         handle = ext4_journal_start(inode,
4833                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4834         if (IS_ERR(handle))
4835                 return PTR_ERR(handle);
4836         ret = dquot_commit(dquot);
4837         err = ext4_journal_stop(handle);
4838         if (!ret)
4839                 ret = err;
4840         return ret;
4841 }
4842
4843 static int ext4_acquire_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_INIT_BLOCKS(dquot->dq_sb));
4850         if (IS_ERR(handle))
4851                 return PTR_ERR(handle);
4852         ret = dquot_acquire(dquot);
4853         err = ext4_journal_stop(handle);
4854         if (!ret)
4855                 ret = err;
4856         return ret;
4857 }
4858
4859 static int ext4_release_dquot(struct dquot *dquot)
4860 {
4861         int ret, err;
4862         handle_t *handle;
4863
4864         handle = ext4_journal_start(dquot_to_inode(dquot),
4865                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4866         if (IS_ERR(handle)) {
4867                 /* Release dquot anyway to avoid endless cycle in dqput() */
4868                 dquot_release(dquot);
4869                 return PTR_ERR(handle);
4870         }
4871         ret = dquot_release(dquot);
4872         err = ext4_journal_stop(handle);
4873         if (!ret)
4874                 ret = err;
4875         return ret;
4876 }
4877
4878 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4879 {
4880         /* Are we journaling quotas? */
4881         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4882             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4883                 dquot_mark_dquot_dirty(dquot);
4884                 return ext4_write_dquot(dquot);
4885         } else {
4886                 return dquot_mark_dquot_dirty(dquot);
4887         }
4888 }
4889
4890 static int ext4_write_info(struct super_block *sb, int type)
4891 {
4892         int ret, err;
4893         handle_t *handle;
4894
4895         /* Data block + inode block */
4896         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4897         if (IS_ERR(handle))
4898                 return PTR_ERR(handle);
4899         ret = dquot_commit_info(sb, type);
4900         err = ext4_journal_stop(handle);
4901         if (!ret)
4902                 ret = err;
4903         return ret;
4904 }
4905
4906 /*
4907  * Turn on quotas during mount time - we need to find
4908  * the quota file and such...
4909  */
4910 static int ext4_quota_on_mount(struct super_block *sb, int type)
4911 {
4912         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4913                                         EXT4_SB(sb)->s_jquota_fmt, type);
4914 }
4915
4916 /*
4917  * Standard function to be called on quota_on
4918  */
4919 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4920                          struct path *path)
4921 {
4922         int err;
4923
4924         if (!test_opt(sb, QUOTA))
4925                 return -EINVAL;
4926
4927         /* Quotafile not on the same filesystem? */
4928         if (path->dentry->d_sb != sb)
4929                 return -EXDEV;
4930         /* Journaling quota? */
4931         if (EXT4_SB(sb)->s_qf_names[type]) {
4932                 /* Quotafile not in fs root? */
4933                 if (path->dentry->d_parent != sb->s_root)
4934                         ext4_msg(sb, KERN_WARNING,
4935                                 "Quota file not on filesystem root. "
4936                                 "Journaled quota will not work");
4937         }
4938
4939         /*
4940          * When we journal data on quota file, we have to flush journal to see
4941          * all updates to the file when we bypass pagecache...
4942          */
4943         if (EXT4_SB(sb)->s_journal &&
4944             ext4_should_journal_data(path->dentry->d_inode)) {
4945                 /*
4946                  * We don't need to lock updates but journal_flush() could
4947                  * otherwise be livelocked...
4948                  */
4949                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4950                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4951                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4952                 if (err)
4953                         return err;
4954         }
4955
4956         return dquot_quota_on(sb, type, format_id, path);
4957 }
4958
4959 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
4960                              unsigned int flags)
4961 {
4962         int err;
4963         struct inode *qf_inode;
4964         unsigned long qf_inums[MAXQUOTAS] = {
4965                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4966                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4967         };
4968
4969         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
4970
4971         if (!qf_inums[type])
4972                 return -EPERM;
4973
4974         qf_inode = ext4_iget(sb, qf_inums[type]);
4975         if (IS_ERR(qf_inode)) {
4976                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
4977                 return PTR_ERR(qf_inode);
4978         }
4979
4980         err = dquot_enable(qf_inode, type, format_id, flags);
4981         iput(qf_inode);
4982
4983         return err;
4984 }
4985
4986 /* Enable usage tracking for all quota types. */
4987 static int ext4_enable_quotas(struct super_block *sb)
4988 {
4989         int type, err = 0;
4990         unsigned long qf_inums[MAXQUOTAS] = {
4991                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4992                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4993         };
4994
4995         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
4996         for (type = 0; type < MAXQUOTAS; type++) {
4997                 if (qf_inums[type]) {
4998                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
4999                                                 DQUOT_USAGE_ENABLED);
5000                         if (err) {
5001                                 ext4_warning(sb,
5002                                         "Failed to enable quota (type=%d) "
5003                                         "tracking. Please run e2fsck to fix.",
5004                                         type);
5005                                 return err;
5006                         }
5007                 }
5008         }
5009         return 0;
5010 }
5011
5012 /*
5013  * quota_on function that is used when QUOTA feature is set.
5014  */
5015 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5016                                  int format_id)
5017 {
5018         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5019                 return -EINVAL;
5020
5021         /*
5022          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5023          */
5024         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5025 }
5026
5027 static int ext4_quota_off(struct super_block *sb, int type)
5028 {
5029         struct inode *inode = sb_dqopt(sb)->files[type];
5030         handle_t *handle;
5031
5032         /* Force all delayed allocation blocks to be allocated.
5033          * Caller already holds s_umount sem */
5034         if (test_opt(sb, DELALLOC))
5035                 sync_filesystem(sb);
5036
5037         if (!inode)
5038                 goto out;
5039
5040         /* Update modification times of quota files when userspace can
5041          * start looking at them */
5042         handle = ext4_journal_start(inode, 1);
5043         if (IS_ERR(handle))
5044                 goto out;
5045         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5046         ext4_mark_inode_dirty(handle, inode);
5047         ext4_journal_stop(handle);
5048
5049 out:
5050         return dquot_quota_off(sb, type);
5051 }
5052
5053 /*
5054  * quota_off function that is used when QUOTA feature is set.
5055  */
5056 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5057 {
5058         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5059                 return -EINVAL;
5060
5061         /* Disable only the limits. */
5062         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5063 }
5064
5065 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5066  * acquiring the locks... As quota files are never truncated and quota code
5067  * itself serializes the operations (and no one else should touch the files)
5068  * we don't have to be afraid of races */
5069 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5070                                size_t len, loff_t off)
5071 {
5072         struct inode *inode = sb_dqopt(sb)->files[type];
5073         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5074         int err = 0;
5075         int offset = off & (sb->s_blocksize - 1);
5076         int tocopy;
5077         size_t toread;
5078         struct buffer_head *bh;
5079         loff_t i_size = i_size_read(inode);
5080
5081         if (off > i_size)
5082                 return 0;
5083         if (off+len > i_size)
5084                 len = i_size-off;
5085         toread = len;
5086         while (toread > 0) {
5087                 tocopy = sb->s_blocksize - offset < toread ?
5088                                 sb->s_blocksize - offset : toread;
5089                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5090                 if (err)
5091                         return err;
5092                 if (!bh)        /* A hole? */
5093                         memset(data, 0, tocopy);
5094                 else
5095                         memcpy(data, bh->b_data+offset, tocopy);
5096                 brelse(bh);
5097                 offset = 0;
5098                 toread -= tocopy;
5099                 data += tocopy;
5100                 blk++;
5101         }
5102         return len;
5103 }
5104
5105 /* Write to quotafile (we know the transaction is already started and has
5106  * enough credits) */
5107 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5108                                 const char *data, size_t len, loff_t off)
5109 {
5110         struct inode *inode = sb_dqopt(sb)->files[type];
5111         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5112         int err = 0;
5113         int offset = off & (sb->s_blocksize - 1);
5114         struct buffer_head *bh;
5115         handle_t *handle = journal_current_handle();
5116
5117         if (EXT4_SB(sb)->s_journal && !handle) {
5118                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5119                         " cancelled because transaction is not started",
5120                         (unsigned long long)off, (unsigned long long)len);
5121                 return -EIO;
5122         }
5123         /*
5124          * Since we account only one data block in transaction credits,
5125          * then it is impossible to cross a block boundary.
5126          */
5127         if (sb->s_blocksize - offset < len) {
5128                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5129                         " cancelled because not block aligned",
5130                         (unsigned long long)off, (unsigned long long)len);
5131                 return -EIO;
5132         }
5133
5134         bh = ext4_bread(handle, inode, blk, 1, &err);
5135         if (!bh)
5136                 goto out;
5137         err = ext4_journal_get_write_access(handle, bh);
5138         if (err) {
5139                 brelse(bh);
5140                 goto out;
5141         }
5142         lock_buffer(bh);
5143         memcpy(bh->b_data+offset, data, len);
5144         flush_dcache_page(bh->b_page);
5145         unlock_buffer(bh);
5146         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5147         brelse(bh);
5148 out:
5149         if (err)
5150                 return err;
5151         if (inode->i_size < off + len) {
5152                 i_size_write(inode, off + len);
5153                 EXT4_I(inode)->i_disksize = inode->i_size;
5154                 ext4_mark_inode_dirty(handle, inode);
5155         }
5156         return len;
5157 }
5158
5159 #endif
5160
5161 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5162                        const char *dev_name, void *data)
5163 {
5164         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5165 }
5166
5167 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5168 static inline void register_as_ext2(void)
5169 {
5170         int err = register_filesystem(&ext2_fs_type);
5171         if (err)
5172                 printk(KERN_WARNING
5173                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5174 }
5175
5176 static inline void unregister_as_ext2(void)
5177 {
5178         unregister_filesystem(&ext2_fs_type);
5179 }
5180
5181 static inline int ext2_feature_set_ok(struct super_block *sb)
5182 {
5183         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5184                 return 0;
5185         if (sb->s_flags & MS_RDONLY)
5186                 return 1;
5187         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5188                 return 0;
5189         return 1;
5190 }
5191 MODULE_ALIAS("ext2");
5192 #else
5193 static inline void register_as_ext2(void) { }
5194 static inline void unregister_as_ext2(void) { }
5195 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5196 #endif
5197
5198 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5199 static inline void register_as_ext3(void)
5200 {
5201         int err = register_filesystem(&ext3_fs_type);
5202         if (err)
5203                 printk(KERN_WARNING
5204                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5205 }
5206
5207 static inline void unregister_as_ext3(void)
5208 {
5209         unregister_filesystem(&ext3_fs_type);
5210 }
5211
5212 static inline int ext3_feature_set_ok(struct super_block *sb)
5213 {
5214         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5215                 return 0;
5216         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5217                 return 0;
5218         if (sb->s_flags & MS_RDONLY)
5219                 return 1;
5220         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5221                 return 0;
5222         return 1;
5223 }
5224 MODULE_ALIAS("ext3");
5225 #else
5226 static inline void register_as_ext3(void) { }
5227 static inline void unregister_as_ext3(void) { }
5228 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5229 #endif
5230
5231 static struct file_system_type ext4_fs_type = {
5232         .owner          = THIS_MODULE,
5233         .name           = "ext4",
5234         .mount          = ext4_mount,
5235         .kill_sb        = kill_block_super,
5236         .fs_flags       = FS_REQUIRES_DEV,
5237 };
5238
5239 static int __init ext4_init_feat_adverts(void)
5240 {
5241         struct ext4_features *ef;
5242         int ret = -ENOMEM;
5243
5244         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5245         if (!ef)
5246                 goto out;
5247
5248         ef->f_kobj.kset = ext4_kset;
5249         init_completion(&ef->f_kobj_unregister);
5250         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5251                                    "features");
5252         if (ret) {
5253                 kfree(ef);
5254                 goto out;
5255         }
5256
5257         ext4_feat = ef;
5258         ret = 0;
5259 out:
5260         return ret;
5261 }
5262
5263 static void ext4_exit_feat_adverts(void)
5264 {
5265         kobject_put(&ext4_feat->f_kobj);
5266         wait_for_completion(&ext4_feat->f_kobj_unregister);
5267         kfree(ext4_feat);
5268 }
5269
5270 /* Shared across all ext4 file systems */
5271 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5272 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5273
5274 static int __init ext4_init_fs(void)
5275 {
5276         int i, err;
5277
5278         ext4_li_info = NULL;
5279         mutex_init(&ext4_li_mtx);
5280
5281         /* Build-time check for flags consistency */
5282         ext4_check_flag_values();
5283
5284         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5285                 mutex_init(&ext4__aio_mutex[i]);
5286                 init_waitqueue_head(&ext4__ioend_wq[i]);
5287         }
5288
5289         err = ext4_init_es();
5290         if (err)
5291                 return err;
5292
5293         err = ext4_init_pageio();
5294         if (err)
5295                 goto out7;
5296
5297         err = ext4_init_system_zone();
5298         if (err)
5299                 goto out6;
5300         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5301         if (!ext4_kset) {
5302                 err = -ENOMEM;
5303                 goto out5;
5304         }
5305         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5306
5307         err = ext4_init_feat_adverts();
5308         if (err)
5309                 goto out4;
5310
5311         err = ext4_init_mballoc();
5312         if (err)
5313                 goto out3;
5314
5315         err = ext4_init_xattr();
5316         if (err)
5317                 goto out2;
5318         err = init_inodecache();
5319         if (err)
5320                 goto out1;
5321         register_as_ext3();
5322         register_as_ext2();
5323         err = register_filesystem(&ext4_fs_type);
5324         if (err)
5325                 goto out;
5326
5327         return 0;
5328 out:
5329         unregister_as_ext2();
5330         unregister_as_ext3();
5331         destroy_inodecache();
5332 out1:
5333         ext4_exit_xattr();
5334 out2:
5335         ext4_exit_mballoc();
5336 out3:
5337         ext4_exit_feat_adverts();
5338 out4:
5339         if (ext4_proc_root)
5340                 remove_proc_entry("fs/ext4", NULL);
5341         kset_unregister(ext4_kset);
5342 out5:
5343         ext4_exit_system_zone();
5344 out6:
5345         ext4_exit_pageio();
5346 out7:
5347         ext4_exit_es();
5348
5349         return err;
5350 }
5351
5352 static void __exit ext4_exit_fs(void)
5353 {
5354         ext4_destroy_lazyinit_thread();
5355         unregister_as_ext2();
5356         unregister_as_ext3();
5357         unregister_filesystem(&ext4_fs_type);
5358         destroy_inodecache();
5359         ext4_exit_xattr();
5360         ext4_exit_mballoc();
5361         ext4_exit_feat_adverts();
5362         remove_proc_entry("fs/ext4", NULL);
5363         kset_unregister(ext4_kset);
5364         ext4_exit_system_zone();
5365         ext4_exit_pageio();
5366 }
5367
5368 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5369 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5370 MODULE_LICENSE("GPL");
5371 module_init(ext4_init_fs)
5372 module_exit(ext4_exit_fs)