fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller ([email protected]), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* 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_removed, "mblk_io_submit"},
1284         {Opt_removed, "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 [email protected] 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         int ret = -1;
1341
1342         if (sb_any_quota_loaded(sb) &&
1343                 !sbi->s_qf_names[qtype]) {
1344                 ext4_msg(sb, KERN_ERR,
1345                         "Cannot change journaled "
1346                         "quota options when quota turned on");
1347                 return -1;
1348         }
1349         qname = match_strdup(args);
1350         if (!qname) {
1351                 ext4_msg(sb, KERN_ERR,
1352                         "Not enough memory for storing quotafile name");
1353                 return -1;
1354         }
1355         if (sbi->s_qf_names[qtype]) {
1356                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1357                         ret = 1;
1358                 else
1359                         ext4_msg(sb, KERN_ERR,
1360                                  "%s quota file already specified",
1361                                  QTYPE2NAME(qtype));
1362                 goto errout;
1363         }
1364         if (strchr(qname, '/')) {
1365                 ext4_msg(sb, KERN_ERR,
1366                         "quotafile must be on filesystem root");
1367                 goto errout;
1368         }
1369         sbi->s_qf_names[qtype] = qname;
1370         set_opt(sb, QUOTA);
1371         return 1;
1372 errout:
1373         kfree(qname);
1374         return ret;
1375 }
1376
1377 static int clear_qf_name(struct super_block *sb, int qtype)
1378 {
1379
1380         struct ext4_sb_info *sbi = EXT4_SB(sb);
1381
1382         if (sb_any_quota_loaded(sb) &&
1383                 sbi->s_qf_names[qtype]) {
1384                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1385                         " when quota turned on");
1386                 return -1;
1387         }
1388         kfree(sbi->s_qf_names[qtype]);
1389         sbi->s_qf_names[qtype] = NULL;
1390         return 1;
1391 }
1392 #endif
1393
1394 #define MOPT_SET        0x0001
1395 #define MOPT_CLEAR      0x0002
1396 #define MOPT_NOSUPPORT  0x0004
1397 #define MOPT_EXPLICIT   0x0008
1398 #define MOPT_CLEAR_ERR  0x0010
1399 #define MOPT_GTE0       0x0020
1400 #ifdef CONFIG_QUOTA
1401 #define MOPT_Q          0
1402 #define MOPT_QFMT       0x0040
1403 #else
1404 #define MOPT_Q          MOPT_NOSUPPORT
1405 #define MOPT_QFMT       MOPT_NOSUPPORT
1406 #endif
1407 #define MOPT_DATAJ      0x0080
1408
1409 static const struct mount_opts {
1410         int     token;
1411         int     mount_opt;
1412         int     flags;
1413 } ext4_mount_opts[] = {
1414         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1415         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1416         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1417         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1418         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1419         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1420         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1421         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1422         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1423         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1424         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1425         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1426         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1427         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1428                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1429         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1430         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1431         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1432         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1433         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1434         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1435         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1436         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1437         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1438         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1439         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1440         {Opt_commit, 0, MOPT_GTE0},
1441         {Opt_max_batch_time, 0, MOPT_GTE0},
1442         {Opt_min_batch_time, 0, MOPT_GTE0},
1443         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1444         {Opt_init_itable, 0, MOPT_GTE0},
1445         {Opt_stripe, 0, MOPT_GTE0},
1446         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1447         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1448         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1449         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1450         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1451 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1452         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1453         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1454 #else
1455         {Opt_acl, 0, MOPT_NOSUPPORT},
1456         {Opt_noacl, 0, MOPT_NOSUPPORT},
1457 #endif
1458         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1459         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1460         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1461         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1462                                                         MOPT_SET | MOPT_Q},
1463         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1464                                                         MOPT_SET | MOPT_Q},
1465         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1466                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1467         {Opt_usrjquota, 0, MOPT_Q},
1468         {Opt_grpjquota, 0, MOPT_Q},
1469         {Opt_offusrjquota, 0, MOPT_Q},
1470         {Opt_offgrpjquota, 0, MOPT_Q},
1471         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1472         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1473         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1474         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1475         {Opt_err, 0, 0}
1476 };
1477
1478 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1479                             substring_t *args, unsigned long *journal_devnum,
1480                             unsigned int *journal_ioprio, int is_remount)
1481 {
1482         struct ext4_sb_info *sbi = EXT4_SB(sb);
1483         const struct mount_opts *m;
1484         kuid_t uid;
1485         kgid_t gid;
1486         int arg = 0;
1487
1488 #ifdef CONFIG_QUOTA
1489         if (token == Opt_usrjquota)
1490                 return set_qf_name(sb, USRQUOTA, &args[0]);
1491         else if (token == Opt_grpjquota)
1492                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1493         else if (token == Opt_offusrjquota)
1494                 return clear_qf_name(sb, USRQUOTA);
1495         else if (token == Opt_offgrpjquota)
1496                 return clear_qf_name(sb, GRPQUOTA);
1497 #endif
1498         if (args->from && match_int(args, &arg))
1499                 return -1;
1500         switch (token) {
1501         case Opt_noacl:
1502         case Opt_nouser_xattr:
1503                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1504                 break;
1505         case Opt_sb:
1506                 return 1;       /* handled by get_sb_block() */
1507         case Opt_removed:
1508                 ext4_msg(sb, KERN_WARNING,
1509                          "Ignoring removed %s option", opt);
1510                 return 1;
1511         case Opt_resuid:
1512                 uid = make_kuid(current_user_ns(), arg);
1513                 if (!uid_valid(uid)) {
1514                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1515                         return -1;
1516                 }
1517                 sbi->s_resuid = uid;
1518                 return 1;
1519         case Opt_resgid:
1520                 gid = make_kgid(current_user_ns(), arg);
1521                 if (!gid_valid(gid)) {
1522                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1523                         return -1;
1524                 }
1525                 sbi->s_resgid = gid;
1526                 return 1;
1527         case Opt_abort:
1528                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1529                 return 1;
1530         case Opt_i_version:
1531                 sb->s_flags |= MS_I_VERSION;
1532                 return 1;
1533         case Opt_journal_dev:
1534                 if (is_remount) {
1535                         ext4_msg(sb, KERN_ERR,
1536                                  "Cannot specify journal on remount");
1537                         return -1;
1538                 }
1539                 *journal_devnum = arg;
1540                 return 1;
1541         case Opt_journal_ioprio:
1542                 if (arg < 0 || arg > 7)
1543                         return -1;
1544                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1545                 return 1;
1546         }
1547
1548         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1549                 if (token != m->token)
1550                         continue;
1551                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1552                         return -1;
1553                 if (m->flags & MOPT_EXPLICIT)
1554                         set_opt2(sb, EXPLICIT_DELALLOC);
1555                 if (m->flags & MOPT_CLEAR_ERR)
1556                         clear_opt(sb, ERRORS_MASK);
1557                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1558                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1559                                  "options when quota turned on");
1560                         return -1;
1561                 }
1562
1563                 if (m->flags & MOPT_NOSUPPORT) {
1564                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1565                 } else if (token == Opt_commit) {
1566                         if (arg == 0)
1567                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1568                         sbi->s_commit_interval = HZ * arg;
1569                 } else if (token == Opt_max_batch_time) {
1570                         if (arg == 0)
1571                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1572                         sbi->s_max_batch_time = arg;
1573                 } else if (token == Opt_min_batch_time) {
1574                         sbi->s_min_batch_time = arg;
1575                 } else if (token == Opt_inode_readahead_blks) {
1576                         if (arg > (1 << 30))
1577                                 return -1;
1578                         if (arg && !is_power_of_2(arg)) {
1579                                 ext4_msg(sb, KERN_ERR,
1580                                          "EXT4-fs: inode_readahead_blks"
1581                                          " must be a power of 2");
1582                                 return -1;
1583                         }
1584                         sbi->s_inode_readahead_blks = arg;
1585                 } else if (token == Opt_init_itable) {
1586                         set_opt(sb, INIT_INODE_TABLE);
1587                         if (!args->from)
1588                                 arg = EXT4_DEF_LI_WAIT_MULT;
1589                         sbi->s_li_wait_mult = arg;
1590                 } else if (token == Opt_max_dir_size_kb) {
1591                         sbi->s_max_dir_size_kb = arg;
1592                 } else if (token == Opt_stripe) {
1593                         sbi->s_stripe = arg;
1594                 } else if (m->flags & MOPT_DATAJ) {
1595                         if (is_remount) {
1596                                 if (!sbi->s_journal)
1597                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1598                                 else if (test_opt(sb, DATA_FLAGS) !=
1599                                          m->mount_opt) {
1600                                         ext4_msg(sb, KERN_ERR,
1601                                          "Cannot change data mode on remount");
1602                                         return -1;
1603                                 }
1604                         } else {
1605                                 clear_opt(sb, DATA_FLAGS);
1606                                 sbi->s_mount_opt |= m->mount_opt;
1607                         }
1608 #ifdef CONFIG_QUOTA
1609                 } else if (m->flags & MOPT_QFMT) {
1610                         if (sb_any_quota_loaded(sb) &&
1611                             sbi->s_jquota_fmt != m->mount_opt) {
1612                                 ext4_msg(sb, KERN_ERR, "Cannot "
1613                                          "change journaled quota options "
1614                                          "when quota turned on");
1615                                 return -1;
1616                         }
1617                         sbi->s_jquota_fmt = m->mount_opt;
1618 #endif
1619                 } else {
1620                         if (!args->from)
1621                                 arg = 1;
1622                         if (m->flags & MOPT_CLEAR)
1623                                 arg = !arg;
1624                         else if (unlikely(!(m->flags & MOPT_SET))) {
1625                                 ext4_msg(sb, KERN_WARNING,
1626                                          "buggy handling of option %s", opt);
1627                                 WARN_ON(1);
1628                                 return -1;
1629                         }
1630                         if (arg != 0)
1631                                 sbi->s_mount_opt |= m->mount_opt;
1632                         else
1633                                 sbi->s_mount_opt &= ~m->mount_opt;
1634                 }
1635                 return 1;
1636         }
1637         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1638                  "or missing value", opt);
1639         return -1;
1640 }
1641
1642 static int parse_options(char *options, struct super_block *sb,
1643                          unsigned long *journal_devnum,
1644                          unsigned int *journal_ioprio,
1645                          int is_remount)
1646 {
1647         struct ext4_sb_info *sbi = EXT4_SB(sb);
1648         char *p;
1649         substring_t args[MAX_OPT_ARGS];
1650         int token;
1651
1652         if (!options)
1653                 return 1;
1654
1655         while ((p = strsep(&options, ",")) != NULL) {
1656                 if (!*p)
1657                         continue;
1658                 /*
1659                  * Initialize args struct so we know whether arg was
1660                  * found; some options take optional arguments.
1661                  */
1662                 args[0].to = args[0].from = NULL;
1663                 token = match_token(p, tokens, args);
1664                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1665                                      journal_ioprio, is_remount) < 0)
1666                         return 0;
1667         }
1668 #ifdef CONFIG_QUOTA
1669         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1670                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1671                         clear_opt(sb, USRQUOTA);
1672
1673                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1674                         clear_opt(sb, GRPQUOTA);
1675
1676                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1677                         ext4_msg(sb, KERN_ERR, "old and new quota "
1678                                         "format mixing");
1679                         return 0;
1680                 }
1681
1682                 if (!sbi->s_jquota_fmt) {
1683                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1684                                         "not specified");
1685                         return 0;
1686                 }
1687         } else {
1688                 if (sbi->s_jquota_fmt) {
1689                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1690                                         "specified with no journaling "
1691                                         "enabled");
1692                         return 0;
1693                 }
1694         }
1695 #endif
1696         if (test_opt(sb, DIOREAD_NOLOCK)) {
1697                 int blocksize =
1698                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1699
1700                 if (blocksize < PAGE_CACHE_SIZE) {
1701                         ext4_msg(sb, KERN_ERR, "can't mount with "
1702                                  "dioread_nolock if block size != PAGE_SIZE");
1703                         return 0;
1704                 }
1705         }
1706         return 1;
1707 }
1708
1709 static inline void ext4_show_quota_options(struct seq_file *seq,
1710                                            struct super_block *sb)
1711 {
1712 #if defined(CONFIG_QUOTA)
1713         struct ext4_sb_info *sbi = EXT4_SB(sb);
1714
1715         if (sbi->s_jquota_fmt) {
1716                 char *fmtname = "";
1717
1718                 switch (sbi->s_jquota_fmt) {
1719                 case QFMT_VFS_OLD:
1720                         fmtname = "vfsold";
1721                         break;
1722                 case QFMT_VFS_V0:
1723                         fmtname = "vfsv0";
1724                         break;
1725                 case QFMT_VFS_V1:
1726                         fmtname = "vfsv1";
1727                         break;
1728                 }
1729                 seq_printf(seq, ",jqfmt=%s", fmtname);
1730         }
1731
1732         if (sbi->s_qf_names[USRQUOTA])
1733                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1734
1735         if (sbi->s_qf_names[GRPQUOTA])
1736                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1737
1738         if (test_opt(sb, USRQUOTA))
1739                 seq_puts(seq, ",usrquota");
1740
1741         if (test_opt(sb, GRPQUOTA))
1742                 seq_puts(seq, ",grpquota");
1743 #endif
1744 }
1745
1746 static const char *token2str(int token)
1747 {
1748         const struct match_token *t;
1749
1750         for (t = tokens; t->token != Opt_err; t++)
1751                 if (t->token == token && !strchr(t->pattern, '='))
1752                         break;
1753         return t->pattern;
1754 }
1755
1756 /*
1757  * Show an option if
1758  *  - it's set to a non-default value OR
1759  *  - if the per-sb default is different from the global default
1760  */
1761 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1762                               int nodefs)
1763 {
1764         struct ext4_sb_info *sbi = EXT4_SB(sb);
1765         struct ext4_super_block *es = sbi->s_es;
1766         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1767         const struct mount_opts *m;
1768         char sep = nodefs ? '\n' : ',';
1769
1770 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1771 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1772
1773         if (sbi->s_sb_block != 1)
1774                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1775
1776         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1777                 int want_set = m->flags & MOPT_SET;
1778                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1779                     (m->flags & MOPT_CLEAR_ERR))
1780                         continue;
1781                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1782                         continue; /* skip if same as the default */
1783                 if ((want_set &&
1784                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1785                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1786                         continue; /* select Opt_noFoo vs Opt_Foo */
1787                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1788         }
1789
1790         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1791             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1792                 SEQ_OPTS_PRINT("resuid=%u",
1793                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1794         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1795             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1796                 SEQ_OPTS_PRINT("resgid=%u",
1797                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1798         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1799         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1800                 SEQ_OPTS_PUTS("errors=remount-ro");
1801         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1802                 SEQ_OPTS_PUTS("errors=continue");
1803         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1804                 SEQ_OPTS_PUTS("errors=panic");
1805         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1806                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1807         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1808                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1809         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1810                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1811         if (sb->s_flags & MS_I_VERSION)
1812                 SEQ_OPTS_PUTS("i_version");
1813         if (nodefs || sbi->s_stripe)
1814                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1815         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1816                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1817                         SEQ_OPTS_PUTS("data=journal");
1818                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1819                         SEQ_OPTS_PUTS("data=ordered");
1820                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1821                         SEQ_OPTS_PUTS("data=writeback");
1822         }
1823         if (nodefs ||
1824             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1825                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1826                                sbi->s_inode_readahead_blks);
1827
1828         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1829                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1830                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1831         if (nodefs || sbi->s_max_dir_size_kb)
1832                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1833
1834         ext4_show_quota_options(seq, sb);
1835         return 0;
1836 }
1837
1838 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1839 {
1840         return _ext4_show_options(seq, root->d_sb, 0);
1841 }
1842
1843 static int options_seq_show(struct seq_file *seq, void *offset)
1844 {
1845         struct super_block *sb = seq->private;
1846         int rc;
1847
1848         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1849         rc = _ext4_show_options(seq, sb, 1);
1850         seq_puts(seq, "\n");
1851         return rc;
1852 }
1853
1854 static int options_open_fs(struct inode *inode, struct file *file)
1855 {
1856         return single_open(file, options_seq_show, PDE(inode)->data);
1857 }
1858
1859 static const struct file_operations ext4_seq_options_fops = {
1860         .owner = THIS_MODULE,
1861         .open = options_open_fs,
1862         .read = seq_read,
1863         .llseek = seq_lseek,
1864         .release = single_release,
1865 };
1866
1867 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1868                             int read_only)
1869 {
1870         struct ext4_sb_info *sbi = EXT4_SB(sb);
1871         int res = 0;
1872
1873         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1874                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1875                          "forcing read-only mode");
1876                 res = MS_RDONLY;
1877         }
1878         if (read_only)
1879                 goto done;
1880         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1881                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1882                          "running e2fsck is recommended");
1883         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1884                 ext4_msg(sb, KERN_WARNING,
1885                          "warning: mounting fs with errors, "
1886                          "running e2fsck is recommended");
1887         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1888                  le16_to_cpu(es->s_mnt_count) >=
1889                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1890                 ext4_msg(sb, KERN_WARNING,
1891                          "warning: maximal mount count reached, "
1892                          "running e2fsck is recommended");
1893         else if (le32_to_cpu(es->s_checkinterval) &&
1894                 (le32_to_cpu(es->s_lastcheck) +
1895                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1896                 ext4_msg(sb, KERN_WARNING,
1897                          "warning: checktime reached, "
1898                          "running e2fsck is recommended");
1899         if (!sbi->s_journal)
1900                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1901         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1902                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1903         le16_add_cpu(&es->s_mnt_count, 1);
1904         es->s_mtime = cpu_to_le32(get_seconds());
1905         ext4_update_dynamic_rev(sb);
1906         if (sbi->s_journal)
1907                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1908
1909         ext4_commit_super(sb, 1);
1910 done:
1911         if (test_opt(sb, DEBUG))
1912                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1913                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1914                         sb->s_blocksize,
1915                         sbi->s_groups_count,
1916                         EXT4_BLOCKS_PER_GROUP(sb),
1917                         EXT4_INODES_PER_GROUP(sb),
1918                         sbi->s_mount_opt, sbi->s_mount_opt2);
1919
1920         cleancache_init_fs(sb);
1921         return res;
1922 }
1923
1924 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1925 {
1926         struct ext4_sb_info *sbi = EXT4_SB(sb);
1927         struct flex_groups *new_groups;
1928         int size;
1929
1930         if (!sbi->s_log_groups_per_flex)
1931                 return 0;
1932
1933         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1934         if (size <= sbi->s_flex_groups_allocated)
1935                 return 0;
1936
1937         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1938         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1939         if (!new_groups) {
1940                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1941                          size / (int) sizeof(struct flex_groups));
1942                 return -ENOMEM;
1943         }
1944
1945         if (sbi->s_flex_groups) {
1946                 memcpy(new_groups, sbi->s_flex_groups,
1947                        (sbi->s_flex_groups_allocated *
1948                         sizeof(struct flex_groups)));
1949                 ext4_kvfree(sbi->s_flex_groups);
1950         }
1951         sbi->s_flex_groups = new_groups;
1952         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1953         return 0;
1954 }
1955
1956 static int ext4_fill_flex_info(struct super_block *sb)
1957 {
1958         struct ext4_sb_info *sbi = EXT4_SB(sb);
1959         struct ext4_group_desc *gdp = NULL;
1960         ext4_group_t flex_group;
1961         unsigned int groups_per_flex = 0;
1962         int i, err;
1963
1964         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1965         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1966                 sbi->s_log_groups_per_flex = 0;
1967                 return 1;
1968         }
1969         groups_per_flex = 1U << sbi->s_log_groups_per_flex;
1970
1971         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1972         if (err)
1973                 goto failed;
1974
1975         for (i = 0; i < sbi->s_groups_count; i++) {
1976                 gdp = ext4_get_group_desc(sb, i, NULL);
1977
1978                 flex_group = ext4_flex_group(sbi, i);
1979                 atomic_add(ext4_free_inodes_count(sb, gdp),
1980                            &sbi->s_flex_groups[flex_group].free_inodes);
1981                 atomic_add(ext4_free_group_clusters(sb, gdp),
1982                            &sbi->s_flex_groups[flex_group].free_clusters);
1983                 atomic_add(ext4_used_dirs_count(sb, gdp),
1984                            &sbi->s_flex_groups[flex_group].used_dirs);
1985         }
1986
1987         return 1;
1988 failed:
1989         return 0;
1990 }
1991
1992 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1993                                    struct ext4_group_desc *gdp)
1994 {
1995         int offset;
1996         __u16 crc = 0;
1997         __le32 le_group = cpu_to_le32(block_group);
1998
1999         if ((sbi->s_es->s_feature_ro_compat &
2000              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
2001                 /* Use new metadata_csum algorithm */
2002                 __u16 old_csum;
2003                 __u32 csum32;
2004
2005                 old_csum = gdp->bg_checksum;
2006                 gdp->bg_checksum = 0;
2007                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2008                                      sizeof(le_group));
2009                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2010                                      sbi->s_desc_size);
2011                 gdp->bg_checksum = old_csum;
2012
2013                 crc = csum32 & 0xFFFF;
2014                 goto out;
2015         }
2016
2017         /* old crc16 code */
2018         offset = offsetof(struct ext4_group_desc, bg_checksum);
2019
2020         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2021         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2022         crc = crc16(crc, (__u8 *)gdp, offset);
2023         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2024         /* for checksum of struct ext4_group_desc do the rest...*/
2025         if ((sbi->s_es->s_feature_incompat &
2026              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2027             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2028                 crc = crc16(crc, (__u8 *)gdp + offset,
2029                             le16_to_cpu(sbi->s_es->s_desc_size) -
2030                                 offset);
2031
2032 out:
2033         return cpu_to_le16(crc);
2034 }
2035
2036 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2037                                 struct ext4_group_desc *gdp)
2038 {
2039         if (ext4_has_group_desc_csum(sb) &&
2040             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2041                                                       block_group, gdp)))
2042                 return 0;
2043
2044         return 1;
2045 }
2046
2047 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2048                               struct ext4_group_desc *gdp)
2049 {
2050         if (!ext4_has_group_desc_csum(sb))
2051                 return;
2052         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2053 }
2054
2055 /* Called at mount-time, super-block is locked */
2056 static int ext4_check_descriptors(struct super_block *sb,
2057                                   ext4_group_t *first_not_zeroed)
2058 {
2059         struct ext4_sb_info *sbi = EXT4_SB(sb);
2060         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2061         ext4_fsblk_t last_block;
2062         ext4_fsblk_t block_bitmap;
2063         ext4_fsblk_t inode_bitmap;
2064         ext4_fsblk_t inode_table;
2065         int flexbg_flag = 0;
2066         ext4_group_t i, grp = sbi->s_groups_count;
2067
2068         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2069                 flexbg_flag = 1;
2070
2071         ext4_debug("Checking group descriptors");
2072
2073         for (i = 0; i < sbi->s_groups_count; i++) {
2074                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2075
2076                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2077                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2078                 else
2079                         last_block = first_block +
2080                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2081
2082                 if ((grp == sbi->s_groups_count) &&
2083                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2084                         grp = i;
2085
2086                 block_bitmap = ext4_block_bitmap(sb, gdp);
2087                 if (block_bitmap < first_block || block_bitmap > last_block) {
2088                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2089                                "Block bitmap for group %u not in group "
2090                                "(block %llu)!", i, block_bitmap);
2091                         return 0;
2092                 }
2093                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2094                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2095                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2096                                "Inode bitmap for group %u not in group "
2097                                "(block %llu)!", i, inode_bitmap);
2098                         return 0;
2099                 }
2100                 inode_table = ext4_inode_table(sb, gdp);
2101                 if (inode_table < first_block ||
2102                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2103                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2104                                "Inode table for group %u not in group "
2105                                "(block %llu)!", i, inode_table);
2106                         return 0;
2107                 }
2108                 ext4_lock_group(sb, i);
2109                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2110                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2111                                  "Checksum for group %u failed (%u!=%u)",
2112                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2113                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2114                         if (!(sb->s_flags & MS_RDONLY)) {
2115                                 ext4_unlock_group(sb, i);
2116                                 return 0;
2117                         }
2118                 }
2119                 ext4_unlock_group(sb, i);
2120                 if (!flexbg_flag)
2121                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2122         }
2123         if (NULL != first_not_zeroed)
2124                 *first_not_zeroed = grp;
2125
2126         ext4_free_blocks_count_set(sbi->s_es,
2127                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2128         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2129         return 1;
2130 }
2131
2132 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2133  * the superblock) which were deleted from all directories, but held open by
2134  * a process at the time of a crash.  We walk the list and try to delete these
2135  * inodes at recovery time (only with a read-write filesystem).
2136  *
2137  * In order to keep the orphan inode chain consistent during traversal (in
2138  * case of crash during recovery), we link each inode into the superblock
2139  * orphan list_head and handle it the same way as an inode deletion during
2140  * normal operation (which journals the operations for us).
2141  *
2142  * We only do an iget() and an iput() on each inode, which is very safe if we
2143  * accidentally point at an in-use or already deleted inode.  The worst that
2144  * can happen in this case is that we get a "bit already cleared" message from
2145  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2146  * e2fsck was run on this filesystem, and it must have already done the orphan
2147  * inode cleanup for us, so we can safely abort without any further action.
2148  */
2149 static void ext4_orphan_cleanup(struct super_block *sb,
2150                                 struct ext4_super_block *es)
2151 {
2152         unsigned int s_flags = sb->s_flags;
2153         int nr_orphans = 0, nr_truncates = 0;
2154 #ifdef CONFIG_QUOTA
2155         int i;
2156 #endif
2157         if (!es->s_last_orphan) {
2158                 jbd_debug(4, "no orphan inodes to clean up\n");
2159                 return;
2160         }
2161
2162         if (bdev_read_only(sb->s_bdev)) {
2163                 ext4_msg(sb, KERN_ERR, "write access "
2164                         "unavailable, skipping orphan cleanup");
2165                 return;
2166         }
2167
2168         /* Check if feature set would not allow a r/w mount */
2169         if (!ext4_feature_set_ok(sb, 0)) {
2170                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2171                          "unknown ROCOMPAT features");
2172                 return;
2173         }
2174
2175         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2176                 /* don't clear list on RO mount w/ errors */
2177                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2178                         jbd_debug(1, "Errors on filesystem, "
2179                                   "clearing orphan list.\n");
2180                         es->s_last_orphan = 0;
2181                 }
2182                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2183                 return;
2184         }
2185
2186         if (s_flags & MS_RDONLY) {
2187                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2188                 sb->s_flags &= ~MS_RDONLY;
2189         }
2190 #ifdef CONFIG_QUOTA
2191         /* Needed for iput() to work correctly and not trash data */
2192         sb->s_flags |= MS_ACTIVE;
2193         /* Turn on quotas so that they are updated correctly */
2194         for (i = 0; i < MAXQUOTAS; i++) {
2195                 if (EXT4_SB(sb)->s_qf_names[i]) {
2196                         int ret = ext4_quota_on_mount(sb, i);
2197                         if (ret < 0)
2198                                 ext4_msg(sb, KERN_ERR,
2199                                         "Cannot turn on journaled "
2200                                         "quota: error %d", ret);
2201                 }
2202         }
2203 #endif
2204
2205         while (es->s_last_orphan) {
2206                 struct inode *inode;
2207
2208                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2209                 if (IS_ERR(inode)) {
2210                         es->s_last_orphan = 0;
2211                         break;
2212                 }
2213
2214                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2215                 dquot_initialize(inode);
2216                 if (inode->i_nlink) {
2217                         ext4_msg(sb, KERN_DEBUG,
2218                                 "%s: truncating inode %lu to %lld bytes",
2219                                 __func__, inode->i_ino, inode->i_size);
2220                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2221                                   inode->i_ino, inode->i_size);
2222                         mutex_lock(&inode->i_mutex);
2223                         ext4_truncate(inode);
2224                         mutex_unlock(&inode->i_mutex);
2225                         nr_truncates++;
2226                 } else {
2227                         ext4_msg(sb, KERN_DEBUG,
2228                                 "%s: deleting unreferenced inode %lu",
2229                                 __func__, inode->i_ino);
2230                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2231                                   inode->i_ino);
2232                         nr_orphans++;
2233                 }
2234                 iput(inode);  /* The delete magic happens here! */
2235         }
2236
2237 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2238
2239         if (nr_orphans)
2240                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2241                        PLURAL(nr_orphans));
2242         if (nr_truncates)
2243                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2244                        PLURAL(nr_truncates));
2245 #ifdef CONFIG_QUOTA
2246         /* Turn quotas off */
2247         for (i = 0; i < MAXQUOTAS; i++) {
2248                 if (sb_dqopt(sb)->files[i])
2249                         dquot_quota_off(sb, i);
2250         }
2251 #endif
2252         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2253 }
2254
2255 /*
2256  * Maximal extent format file size.
2257  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2258  * extent format containers, within a sector_t, and within i_blocks
2259  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2260  * so that won't be a limiting factor.
2261  *
2262  * However there is other limiting factor. We do store extents in the form
2263  * of starting block and length, hence the resulting length of the extent
2264  * covering maximum file size must fit into on-disk format containers as
2265  * well. Given that length is always by 1 unit bigger than max unit (because
2266  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2267  *
2268  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2269  */
2270 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2271 {
2272         loff_t res;
2273         loff_t upper_limit = MAX_LFS_FILESIZE;
2274
2275         /* small i_blocks in vfs inode? */
2276         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2277                 /*
2278                  * CONFIG_LBDAF is not enabled implies the inode
2279                  * i_block represent total blocks in 512 bytes
2280                  * 32 == size of vfs inode i_blocks * 8
2281                  */
2282                 upper_limit = (1LL << 32) - 1;
2283
2284                 /* total blocks in file system block size */
2285                 upper_limit >>= (blkbits - 9);
2286                 upper_limit <<= blkbits;
2287         }
2288
2289         /*
2290          * 32-bit extent-start container, ee_block. We lower the maxbytes
2291          * by one fs block, so ee_len can cover the extent of maximum file
2292          * size
2293          */
2294         res = (1LL << 32) - 1;
2295         res <<= blkbits;
2296
2297         /* Sanity check against vm- & vfs- imposed limits */
2298         if (res > upper_limit)
2299                 res = upper_limit;
2300
2301         return res;
2302 }
2303
2304 /*
2305  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2306  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2307  * We need to be 1 filesystem block less than the 2^48 sector limit.
2308  */
2309 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2310 {
2311         loff_t res = EXT4_NDIR_BLOCKS;
2312         int meta_blocks;
2313         loff_t upper_limit;
2314         /* This is calculated to be the largest file size for a dense, block
2315          * mapped file such that the file's total number of 512-byte sectors,
2316          * including data and all indirect blocks, does not exceed (2^48 - 1).
2317          *
2318          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2319          * number of 512-byte sectors of the file.
2320          */
2321
2322         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2323                 /*
2324                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2325                  * the inode i_block field represents total file blocks in
2326                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2327                  */
2328                 upper_limit = (1LL << 32) - 1;
2329
2330                 /* total blocks in file system block size */
2331                 upper_limit >>= (bits - 9);
2332
2333         } else {
2334                 /*
2335                  * We use 48 bit ext4_inode i_blocks
2336                  * With EXT4_HUGE_FILE_FL set the i_blocks
2337                  * represent total number of blocks in
2338                  * file system block size
2339                  */
2340                 upper_limit = (1LL << 48) - 1;
2341
2342         }
2343
2344         /* indirect blocks */
2345         meta_blocks = 1;
2346         /* double indirect blocks */
2347         meta_blocks += 1 + (1LL << (bits-2));
2348         /* tripple indirect blocks */
2349         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2350
2351         upper_limit -= meta_blocks;
2352         upper_limit <<= bits;
2353
2354         res += 1LL << (bits-2);
2355         res += 1LL << (2*(bits-2));
2356         res += 1LL << (3*(bits-2));
2357         res <<= bits;
2358         if (res > upper_limit)
2359                 res = upper_limit;
2360
2361         if (res > MAX_LFS_FILESIZE)
2362                 res = MAX_LFS_FILESIZE;
2363
2364         return res;
2365 }
2366
2367 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2368                                    ext4_fsblk_t logical_sb_block, int nr)
2369 {
2370         struct ext4_sb_info *sbi = EXT4_SB(sb);
2371         ext4_group_t bg, first_meta_bg;
2372         int has_super = 0;
2373
2374         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2375
2376         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2377             nr < first_meta_bg)
2378                 return logical_sb_block + nr + 1;
2379         bg = sbi->s_desc_per_block * nr;
2380         if (ext4_bg_has_super(sb, bg))
2381                 has_super = 1;
2382
2383         return (has_super + ext4_group_first_block_no(sb, bg));
2384 }
2385
2386 /**
2387  * ext4_get_stripe_size: Get the stripe size.
2388  * @sbi: In memory super block info
2389  *
2390  * If we have specified it via mount option, then
2391  * use the mount option value. If the value specified at mount time is
2392  * greater than the blocks per group use the super block value.
2393  * If the super block value is greater than blocks per group return 0.
2394  * Allocator needs it be less than blocks per group.
2395  *
2396  */
2397 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2398 {
2399         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2400         unsigned long stripe_width =
2401                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2402         int ret;
2403
2404         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2405                 ret = sbi->s_stripe;
2406         else if (stripe_width <= sbi->s_blocks_per_group)
2407                 ret = stripe_width;
2408         else if (stride <= sbi->s_blocks_per_group)
2409                 ret = stride;
2410         else
2411                 ret = 0;
2412
2413         /*
2414          * If the stripe width is 1, this makes no sense and
2415          * we set it to 0 to turn off stripe handling code.
2416          */
2417         if (ret <= 1)
2418                 ret = 0;
2419
2420         return ret;
2421 }
2422
2423 /* sysfs supprt */
2424
2425 struct ext4_attr {
2426         struct attribute attr;
2427         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2428         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2429                          const char *, size_t);
2430         int offset;
2431 };
2432
2433 static int parse_strtoul(const char *buf,
2434                 unsigned long max, unsigned long *value)
2435 {
2436         char *endp;
2437
2438         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2439         endp = skip_spaces(endp);
2440         if (*endp || *value > max)
2441                 return -EINVAL;
2442
2443         return 0;
2444 }
2445
2446 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2447                                               struct ext4_sb_info *sbi,
2448                                               char *buf)
2449 {
2450         return snprintf(buf, PAGE_SIZE, "%llu\n",
2451                 (s64) EXT4_C2B(sbi,
2452                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2453 }
2454
2455 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2456                                          struct ext4_sb_info *sbi, char *buf)
2457 {
2458         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2459
2460         if (!sb->s_bdev->bd_part)
2461                 return snprintf(buf, PAGE_SIZE, "0\n");
2462         return snprintf(buf, PAGE_SIZE, "%lu\n",
2463                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2464                          sbi->s_sectors_written_start) >> 1);
2465 }
2466
2467 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2468                                           struct ext4_sb_info *sbi, char *buf)
2469 {
2470         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2471
2472         if (!sb->s_bdev->bd_part)
2473                 return snprintf(buf, PAGE_SIZE, "0\n");
2474         return snprintf(buf, PAGE_SIZE, "%llu\n",
2475                         (unsigned long long)(sbi->s_kbytes_written +
2476                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2477                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2478 }
2479
2480 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2481                                           struct ext4_sb_info *sbi,
2482                                           const char *buf, size_t count)
2483 {
2484         unsigned long t;
2485
2486         if (parse_strtoul(buf, 0x40000000, &t))
2487                 return -EINVAL;
2488
2489         if (t && !is_power_of_2(t))
2490                 return -EINVAL;
2491
2492         sbi->s_inode_readahead_blks = t;
2493         return count;
2494 }
2495
2496 static ssize_t sbi_ui_show(struct ext4_attr *a,
2497                            struct ext4_sb_info *sbi, char *buf)
2498 {
2499         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2500
2501         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2502 }
2503
2504 static ssize_t sbi_ui_store(struct ext4_attr *a,
2505                             struct ext4_sb_info *sbi,
2506                             const char *buf, size_t count)
2507 {
2508         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2509         unsigned long t;
2510
2511         if (parse_strtoul(buf, 0xffffffff, &t))
2512                 return -EINVAL;
2513         *ui = t;
2514         return count;
2515 }
2516
2517 static ssize_t trigger_test_error(struct ext4_attr *a,
2518                                   struct ext4_sb_info *sbi,
2519                                   const char *buf, size_t count)
2520 {
2521         int len = count;
2522
2523         if (!capable(CAP_SYS_ADMIN))
2524                 return -EPERM;
2525
2526         if (len && buf[len-1] == '\n')
2527                 len--;
2528
2529         if (len)
2530                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2531         return count;
2532 }
2533
2534 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2535 static struct ext4_attr ext4_attr_##_name = {                   \
2536         .attr = {.name = __stringify(_name), .mode = _mode },   \
2537         .show   = _show,                                        \
2538         .store  = _store,                                       \
2539         .offset = offsetof(struct ext4_sb_info, _elname),       \
2540 }
2541 #define EXT4_ATTR(name, mode, show, store) \
2542 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2543
2544 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2545 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2546 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2547 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2548         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2549 #define ATTR_LIST(name) &ext4_attr_##name.attr
2550
2551 EXT4_RO_ATTR(delayed_allocation_blocks);
2552 EXT4_RO_ATTR(session_write_kbytes);
2553 EXT4_RO_ATTR(lifetime_write_kbytes);
2554 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2555                  inode_readahead_blks_store, s_inode_readahead_blks);
2556 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2557 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2558 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2559 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2560 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2561 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2562 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2563 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2564 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2565 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2566
2567 static struct attribute *ext4_attrs[] = {
2568         ATTR_LIST(delayed_allocation_blocks),
2569         ATTR_LIST(session_write_kbytes),
2570         ATTR_LIST(lifetime_write_kbytes),
2571         ATTR_LIST(inode_readahead_blks),
2572         ATTR_LIST(inode_goal),
2573         ATTR_LIST(mb_stats),
2574         ATTR_LIST(mb_max_to_scan),
2575         ATTR_LIST(mb_min_to_scan),
2576         ATTR_LIST(mb_order2_req),
2577         ATTR_LIST(mb_stream_req),
2578         ATTR_LIST(mb_group_prealloc),
2579         ATTR_LIST(max_writeback_mb_bump),
2580         ATTR_LIST(extent_max_zeroout_kb),
2581         ATTR_LIST(trigger_fs_error),
2582         NULL,
2583 };
2584
2585 /* Features this copy of ext4 supports */
2586 EXT4_INFO_ATTR(lazy_itable_init);
2587 EXT4_INFO_ATTR(batched_discard);
2588 EXT4_INFO_ATTR(meta_bg_resize);
2589
2590 static struct attribute *ext4_feat_attrs[] = {
2591         ATTR_LIST(lazy_itable_init),
2592         ATTR_LIST(batched_discard),
2593         ATTR_LIST(meta_bg_resize),
2594         NULL,
2595 };
2596
2597 static ssize_t ext4_attr_show(struct kobject *kobj,
2598                               struct attribute *attr, char *buf)
2599 {
2600         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2601                                                 s_kobj);
2602         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2603
2604         return a->show ? a->show(a, sbi, buf) : 0;
2605 }
2606
2607 static ssize_t ext4_attr_store(struct kobject *kobj,
2608                                struct attribute *attr,
2609                                const char *buf, size_t len)
2610 {
2611         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2612                                                 s_kobj);
2613         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2614
2615         return a->store ? a->store(a, sbi, buf, len) : 0;
2616 }
2617
2618 static void ext4_sb_release(struct kobject *kobj)
2619 {
2620         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2621                                                 s_kobj);
2622         complete(&sbi->s_kobj_unregister);
2623 }
2624
2625 static const struct sysfs_ops ext4_attr_ops = {
2626         .show   = ext4_attr_show,
2627         .store  = ext4_attr_store,
2628 };
2629
2630 static struct kobj_type ext4_ktype = {
2631         .default_attrs  = ext4_attrs,
2632         .sysfs_ops      = &ext4_attr_ops,
2633         .release        = ext4_sb_release,
2634 };
2635
2636 static void ext4_feat_release(struct kobject *kobj)
2637 {
2638         complete(&ext4_feat->f_kobj_unregister);
2639 }
2640
2641 static struct kobj_type ext4_feat_ktype = {
2642         .default_attrs  = ext4_feat_attrs,
2643         .sysfs_ops      = &ext4_attr_ops,
2644         .release        = ext4_feat_release,
2645 };
2646
2647 /*
2648  * Check whether this filesystem can be mounted based on
2649  * the features present and the RDONLY/RDWR mount requested.
2650  * Returns 1 if this filesystem can be mounted as requested,
2651  * 0 if it cannot be.
2652  */
2653 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2654 {
2655         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2656                 ext4_msg(sb, KERN_ERR,
2657                         "Couldn't mount because of "
2658                         "unsupported optional features (%x)",
2659                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2660                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2661                 return 0;
2662         }
2663
2664         if (readonly)
2665                 return 1;
2666
2667         /* Check that feature set is OK for a read-write mount */
2668         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2669                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2670                          "unsupported optional features (%x)",
2671                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2672                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2673                 return 0;
2674         }
2675         /*
2676          * Large file size enabled file system can only be mounted
2677          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2678          */
2679         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2680                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2681                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2682                                  "cannot be mounted RDWR without "
2683                                  "CONFIG_LBDAF");
2684                         return 0;
2685                 }
2686         }
2687         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2688             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2689                 ext4_msg(sb, KERN_ERR,
2690                          "Can't support bigalloc feature without "
2691                          "extents feature\n");
2692                 return 0;
2693         }
2694
2695 #ifndef CONFIG_QUOTA
2696         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2697             !readonly) {
2698                 ext4_msg(sb, KERN_ERR,
2699                          "Filesystem with quota feature cannot be mounted RDWR "
2700                          "without CONFIG_QUOTA");
2701                 return 0;
2702         }
2703 #endif  /* CONFIG_QUOTA */
2704         return 1;
2705 }
2706
2707 /*
2708  * This function is called once a day if we have errors logged
2709  * on the file system
2710  */
2711 static void print_daily_error_info(unsigned long arg)
2712 {
2713         struct super_block *sb = (struct super_block *) arg;
2714         struct ext4_sb_info *sbi;
2715         struct ext4_super_block *es;
2716
2717         sbi = EXT4_SB(sb);
2718         es = sbi->s_es;
2719
2720         if (es->s_error_count)
2721                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2722                          le32_to_cpu(es->s_error_count));
2723         if (es->s_first_error_time) {
2724                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2725                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2726                        (int) sizeof(es->s_first_error_func),
2727                        es->s_first_error_func,
2728                        le32_to_cpu(es->s_first_error_line));
2729                 if (es->s_first_error_ino)
2730                         printk(": inode %u",
2731                                le32_to_cpu(es->s_first_error_ino));
2732                 if (es->s_first_error_block)
2733                         printk(": block %llu", (unsigned long long)
2734                                le64_to_cpu(es->s_first_error_block));
2735                 printk("\n");
2736         }
2737         if (es->s_last_error_time) {
2738                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2739                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2740                        (int) sizeof(es->s_last_error_func),
2741                        es->s_last_error_func,
2742                        le32_to_cpu(es->s_last_error_line));
2743                 if (es->s_last_error_ino)
2744                         printk(": inode %u",
2745                                le32_to_cpu(es->s_last_error_ino));
2746                 if (es->s_last_error_block)
2747                         printk(": block %llu", (unsigned long long)
2748                                le64_to_cpu(es->s_last_error_block));
2749                 printk("\n");
2750         }
2751         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2752 }
2753
2754 /* Find next suitable group and run ext4_init_inode_table */
2755 static int ext4_run_li_request(struct ext4_li_request *elr)
2756 {
2757         struct ext4_group_desc *gdp = NULL;
2758         ext4_group_t group, ngroups;
2759         struct super_block *sb;
2760         unsigned long timeout = 0;
2761         int ret = 0;
2762
2763         sb = elr->lr_super;
2764         ngroups = EXT4_SB(sb)->s_groups_count;
2765
2766         sb_start_write(sb);
2767         for (group = elr->lr_next_group; group < ngroups; group++) {
2768                 gdp = ext4_get_group_desc(sb, group, NULL);
2769                 if (!gdp) {
2770                         ret = 1;
2771                         break;
2772                 }
2773
2774                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2775                         break;
2776         }
2777
2778         if (group >= ngroups)
2779                 ret = 1;
2780
2781         if (!ret) {
2782                 timeout = jiffies;
2783                 ret = ext4_init_inode_table(sb, group,
2784                                             elr->lr_timeout ? 0 : 1);
2785                 if (elr->lr_timeout == 0) {
2786                         timeout = (jiffies - timeout) *
2787                                   elr->lr_sbi->s_li_wait_mult;
2788                         elr->lr_timeout = timeout;
2789                 }
2790                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2791                 elr->lr_next_group = group + 1;
2792         }
2793         sb_end_write(sb);
2794
2795         return ret;
2796 }
2797
2798 /*
2799  * Remove lr_request from the list_request and free the
2800  * request structure. Should be called with li_list_mtx held
2801  */
2802 static void ext4_remove_li_request(struct ext4_li_request *elr)
2803 {
2804         struct ext4_sb_info *sbi;
2805
2806         if (!elr)
2807                 return;
2808
2809         sbi = elr->lr_sbi;
2810
2811         list_del(&elr->lr_request);
2812         sbi->s_li_request = NULL;
2813         kfree(elr);
2814 }
2815
2816 static void ext4_unregister_li_request(struct super_block *sb)
2817 {
2818         mutex_lock(&ext4_li_mtx);
2819         if (!ext4_li_info) {
2820                 mutex_unlock(&ext4_li_mtx);
2821                 return;
2822         }
2823
2824         mutex_lock(&ext4_li_info->li_list_mtx);
2825         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2826         mutex_unlock(&ext4_li_info->li_list_mtx);
2827         mutex_unlock(&ext4_li_mtx);
2828 }
2829
2830 static struct task_struct *ext4_lazyinit_task;
2831
2832 /*
2833  * This is the function where ext4lazyinit thread lives. It walks
2834  * through the request list searching for next scheduled filesystem.
2835  * When such a fs is found, run the lazy initialization request
2836  * (ext4_rn_li_request) and keep track of the time spend in this
2837  * function. Based on that time we compute next schedule time of
2838  * the request. When walking through the list is complete, compute
2839  * next waking time and put itself into sleep.
2840  */
2841 static int ext4_lazyinit_thread(void *arg)
2842 {
2843         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2844         struct list_head *pos, *n;
2845         struct ext4_li_request *elr;
2846         unsigned long next_wakeup, cur;
2847
2848         BUG_ON(NULL == eli);
2849
2850 cont_thread:
2851         while (true) {
2852                 next_wakeup = MAX_JIFFY_OFFSET;
2853
2854                 mutex_lock(&eli->li_list_mtx);
2855                 if (list_empty(&eli->li_request_list)) {
2856                         mutex_unlock(&eli->li_list_mtx);
2857                         goto exit_thread;
2858                 }
2859
2860                 list_for_each_safe(pos, n, &eli->li_request_list) {
2861                         elr = list_entry(pos, struct ext4_li_request,
2862                                          lr_request);
2863
2864                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2865                                 if (ext4_run_li_request(elr) != 0) {
2866                                         /* error, remove the lazy_init job */
2867                                         ext4_remove_li_request(elr);
2868                                         continue;
2869                                 }
2870                         }
2871
2872                         if (time_before(elr->lr_next_sched, next_wakeup))
2873                                 next_wakeup = elr->lr_next_sched;
2874                 }
2875                 mutex_unlock(&eli->li_list_mtx);
2876
2877                 try_to_freeze();
2878
2879                 cur = jiffies;
2880                 if ((time_after_eq(cur, next_wakeup)) ||
2881                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2882                         cond_resched();
2883                         continue;
2884                 }
2885
2886                 schedule_timeout_interruptible(next_wakeup - cur);
2887
2888                 if (kthread_should_stop()) {
2889                         ext4_clear_request_list();
2890                         goto exit_thread;
2891                 }
2892         }
2893
2894 exit_thread:
2895         /*
2896          * It looks like the request list is empty, but we need
2897          * to check it under the li_list_mtx lock, to prevent any
2898          * additions into it, and of course we should lock ext4_li_mtx
2899          * to atomically free the list and ext4_li_info, because at
2900          * this point another ext4 filesystem could be registering
2901          * new one.
2902          */
2903         mutex_lock(&ext4_li_mtx);
2904         mutex_lock(&eli->li_list_mtx);
2905         if (!list_empty(&eli->li_request_list)) {
2906                 mutex_unlock(&eli->li_list_mtx);
2907                 mutex_unlock(&ext4_li_mtx);
2908                 goto cont_thread;
2909         }
2910         mutex_unlock(&eli->li_list_mtx);
2911         kfree(ext4_li_info);
2912         ext4_li_info = NULL;
2913         mutex_unlock(&ext4_li_mtx);
2914
2915         return 0;
2916 }
2917
2918 static void ext4_clear_request_list(void)
2919 {
2920         struct list_head *pos, *n;
2921         struct ext4_li_request *elr;
2922
2923         mutex_lock(&ext4_li_info->li_list_mtx);
2924         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2925                 elr = list_entry(pos, struct ext4_li_request,
2926                                  lr_request);
2927                 ext4_remove_li_request(elr);
2928         }
2929         mutex_unlock(&ext4_li_info->li_list_mtx);
2930 }
2931
2932 static int ext4_run_lazyinit_thread(void)
2933 {
2934         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2935                                          ext4_li_info, "ext4lazyinit");
2936         if (IS_ERR(ext4_lazyinit_task)) {
2937                 int err = PTR_ERR(ext4_lazyinit_task);
2938                 ext4_clear_request_list();
2939                 kfree(ext4_li_info);
2940                 ext4_li_info = NULL;
2941                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2942                                  "initialization thread\n",
2943                                  err);
2944                 return err;
2945         }
2946         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2947         return 0;
2948 }
2949
2950 /*
2951  * Check whether it make sense to run itable init. thread or not.
2952  * If there is at least one uninitialized inode table, return
2953  * corresponding group number, else the loop goes through all
2954  * groups and return total number of groups.
2955  */
2956 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2957 {
2958         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2959         struct ext4_group_desc *gdp = NULL;
2960
2961         for (group = 0; group < ngroups; group++) {
2962                 gdp = ext4_get_group_desc(sb, group, NULL);
2963                 if (!gdp)
2964                         continue;
2965
2966                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2967                         break;
2968         }
2969
2970         return group;
2971 }
2972
2973 static int ext4_li_info_new(void)
2974 {
2975         struct ext4_lazy_init *eli = NULL;
2976
2977         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2978         if (!eli)
2979                 return -ENOMEM;
2980
2981         INIT_LIST_HEAD(&eli->li_request_list);
2982         mutex_init(&eli->li_list_mtx);
2983
2984         eli->li_state |= EXT4_LAZYINIT_QUIT;
2985
2986         ext4_li_info = eli;
2987
2988         return 0;
2989 }
2990
2991 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2992                                             ext4_group_t start)
2993 {
2994         struct ext4_sb_info *sbi = EXT4_SB(sb);
2995         struct ext4_li_request *elr;
2996         unsigned long rnd;
2997
2998         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2999         if (!elr)
3000                 return NULL;
3001
3002         elr->lr_super = sb;
3003         elr->lr_sbi = sbi;
3004         elr->lr_next_group = start;
3005
3006         /*
3007          * Randomize first schedule time of the request to
3008          * spread the inode table initialization requests
3009          * better.
3010          */
3011         get_random_bytes(&rnd, sizeof(rnd));
3012         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3013                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3014
3015         return elr;
3016 }
3017
3018 int ext4_register_li_request(struct super_block *sb,
3019                              ext4_group_t first_not_zeroed)
3020 {
3021         struct ext4_sb_info *sbi = EXT4_SB(sb);
3022         struct ext4_li_request *elr = NULL;
3023         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3024         int ret = 0;
3025
3026         mutex_lock(&ext4_li_mtx);
3027         if (sbi->s_li_request != NULL) {
3028                 /*
3029                  * Reset timeout so it can be computed again, because
3030                  * s_li_wait_mult might have changed.
3031                  */
3032                 sbi->s_li_request->lr_timeout = 0;
3033                 goto out;
3034         }
3035
3036         if (first_not_zeroed == ngroups ||
3037             (sb->s_flags & MS_RDONLY) ||
3038             !test_opt(sb, INIT_INODE_TABLE))
3039                 goto out;
3040
3041         elr = ext4_li_request_new(sb, first_not_zeroed);
3042         if (!elr) {
3043                 ret = -ENOMEM;
3044                 goto out;
3045         }
3046
3047         if (NULL == ext4_li_info) {
3048                 ret = ext4_li_info_new();
3049                 if (ret)
3050                         goto out;
3051         }
3052
3053         mutex_lock(&ext4_li_info->li_list_mtx);
3054         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3055         mutex_unlock(&ext4_li_info->li_list_mtx);
3056
3057         sbi->s_li_request = elr;
3058         /*
3059          * set elr to NULL here since it has been inserted to
3060          * the request_list and the removal and free of it is
3061          * handled by ext4_clear_request_list from now on.
3062          */
3063         elr = NULL;
3064
3065         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3066                 ret = ext4_run_lazyinit_thread();
3067                 if (ret)
3068                         goto out;
3069         }
3070 out:
3071         mutex_unlock(&ext4_li_mtx);
3072         if (ret)
3073                 kfree(elr);
3074         return ret;
3075 }
3076
3077 /*
3078  * We do not need to lock anything since this is called on
3079  * module unload.
3080  */
3081 static void ext4_destroy_lazyinit_thread(void)
3082 {
3083         /*
3084          * If thread exited earlier
3085          * there's nothing to be done.
3086          */
3087         if (!ext4_li_info || !ext4_lazyinit_task)
3088                 return;
3089
3090         kthread_stop(ext4_lazyinit_task);
3091 }
3092
3093 static int set_journal_csum_feature_set(struct super_block *sb)
3094 {
3095         int ret = 1;
3096         int compat, incompat;
3097         struct ext4_sb_info *sbi = EXT4_SB(sb);
3098
3099         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3100                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3101                 /* journal checksum v2 */
3102                 compat = 0;
3103                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3104         } else {
3105                 /* journal checksum v1 */
3106                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3107                 incompat = 0;
3108         }
3109
3110         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3111                 ret = jbd2_journal_set_features(sbi->s_journal,
3112                                 compat, 0,
3113                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3114                                 incompat);
3115         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3116                 ret = jbd2_journal_set_features(sbi->s_journal,
3117                                 compat, 0,
3118                                 incompat);
3119                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3120                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3121         } else {
3122                 jbd2_journal_clear_features(sbi->s_journal,
3123                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3124                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3125                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3126         }
3127
3128         return ret;
3129 }
3130
3131 /*
3132  * Note: calculating the overhead so we can be compatible with
3133  * historical BSD practice is quite difficult in the face of
3134  * clusters/bigalloc.  This is because multiple metadata blocks from
3135  * different block group can end up in the same allocation cluster.
3136  * Calculating the exact overhead in the face of clustered allocation
3137  * requires either O(all block bitmaps) in memory or O(number of block
3138  * groups**2) in time.  We will still calculate the superblock for
3139  * older file systems --- and if we come across with a bigalloc file
3140  * system with zero in s_overhead_clusters the estimate will be close to
3141  * correct especially for very large cluster sizes --- but for newer
3142  * file systems, it's better to calculate this figure once at mkfs
3143  * time, and store it in the superblock.  If the superblock value is
3144  * present (even for non-bigalloc file systems), we will use it.
3145  */
3146 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3147                           char *buf)
3148 {
3149         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3150         struct ext4_group_desc  *gdp;
3151         ext4_fsblk_t            first_block, last_block, b;
3152         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3153         int                     s, j, count = 0;
3154
3155         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3156                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3157                         sbi->s_itb_per_group + 2);
3158
3159         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3160                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3161         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3162         for (i = 0; i < ngroups; i++) {
3163                 gdp = ext4_get_group_desc(sb, i, NULL);
3164                 b = ext4_block_bitmap(sb, gdp);
3165                 if (b >= first_block && b <= last_block) {
3166                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3167                         count++;
3168                 }
3169                 b = ext4_inode_bitmap(sb, gdp);
3170                 if (b >= first_block && b <= last_block) {
3171                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3172                         count++;
3173                 }
3174                 b = ext4_inode_table(sb, gdp);
3175                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3176                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3177                                 int c = EXT4_B2C(sbi, b - first_block);
3178                                 ext4_set_bit(c, buf);
3179                                 count++;
3180                         }
3181                 if (i != grp)
3182                         continue;
3183                 s = 0;
3184                 if (ext4_bg_has_super(sb, grp)) {
3185                         ext4_set_bit(s++, buf);
3186                         count++;
3187                 }
3188                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3189                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3190                         count++;
3191                 }
3192         }
3193         if (!count)
3194                 return 0;
3195         return EXT4_CLUSTERS_PER_GROUP(sb) -
3196                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3197 }
3198
3199 /*
3200  * Compute the overhead and stash it in sbi->s_overhead
3201  */
3202 int ext4_calculate_overhead(struct super_block *sb)
3203 {
3204         struct ext4_sb_info *sbi = EXT4_SB(sb);
3205         struct ext4_super_block *es = sbi->s_es;
3206         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3207         ext4_fsblk_t overhead = 0;
3208         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3209
3210         if (!buf)
3211                 return -ENOMEM;
3212
3213         /*
3214          * Compute the overhead (FS structures).  This is constant
3215          * for a given filesystem unless the number of block groups
3216          * changes so we cache the previous value until it does.
3217          */
3218
3219         /*
3220          * All of the blocks before first_data_block are overhead
3221          */
3222         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3223
3224         /*
3225          * Add the overhead found in each block group
3226          */
3227         for (i = 0; i < ngroups; i++) {
3228                 int blks;
3229
3230                 blks = count_overhead(sb, i, buf);
3231                 overhead += blks;
3232                 if (blks)
3233                         memset(buf, 0, PAGE_SIZE);
3234                 cond_resched();
3235         }
3236         /* Add the journal blocks as well */
3237         if (sbi->s_journal)
3238                 overhead += EXT4_B2C(sbi, sbi->s_journal->j_maxlen);
3239
3240         sbi->s_overhead = overhead;
3241         smp_wmb();
3242         free_page((unsigned long) buf);
3243         return 0;
3244 }
3245
3246 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3247 {
3248         char *orig_data = kstrdup(data, GFP_KERNEL);
3249         struct buffer_head *bh;
3250         struct ext4_super_block *es = NULL;
3251         struct ext4_sb_info *sbi;
3252         ext4_fsblk_t block;
3253         ext4_fsblk_t sb_block = get_sb_block(&data);
3254         ext4_fsblk_t logical_sb_block;
3255         unsigned long offset = 0;
3256         unsigned long journal_devnum = 0;
3257         unsigned long def_mount_opts;
3258         struct inode *root;
3259         char *cp;
3260         const char *descr;
3261         int ret = -ENOMEM;
3262         int blocksize, clustersize;
3263         unsigned int db_count;
3264         unsigned int i;
3265         int needs_recovery, has_huge_files, has_bigalloc;
3266         __u64 blocks_count;
3267         int err = 0;
3268         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3269         ext4_group_t first_not_zeroed;
3270
3271         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3272         if (!sbi)
3273                 goto out_free_orig;
3274
3275         sbi->s_blockgroup_lock =
3276                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3277         if (!sbi->s_blockgroup_lock) {
3278                 kfree(sbi);
3279                 goto out_free_orig;
3280         }
3281         sb->s_fs_info = sbi;
3282         sbi->s_sb = sb;
3283         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3284         sbi->s_sb_block = sb_block;
3285         if (sb->s_bdev->bd_part)
3286                 sbi->s_sectors_written_start =
3287                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3288
3289         /* Cleanup superblock name */
3290         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3291                 *cp = '!';
3292
3293         /* -EINVAL is default */
3294         ret = -EINVAL;
3295         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3296         if (!blocksize) {
3297                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3298                 goto out_fail;
3299         }
3300
3301         /*
3302          * The ext4 superblock will not be buffer aligned for other than 1kB
3303          * block sizes.  We need to calculate the offset from buffer start.
3304          */
3305         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3306                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3307                 offset = do_div(logical_sb_block, blocksize);
3308         } else {
3309                 logical_sb_block = sb_block;
3310         }
3311
3312         if (!(bh = sb_bread(sb, logical_sb_block))) {
3313                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3314                 goto out_fail;
3315         }
3316         /*
3317          * Note: s_es must be initialized as soon as possible because
3318          *       some ext4 macro-instructions depend on its value
3319          */
3320         es = (struct ext4_super_block *) (bh->b_data + offset);
3321         sbi->s_es = es;
3322         sb->s_magic = le16_to_cpu(es->s_magic);
3323         if (sb->s_magic != EXT4_SUPER_MAGIC)
3324                 goto cantfind_ext4;
3325         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3326
3327         /* Warn if metadata_csum and gdt_csum are both set. */
3328         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3329                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3330             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3331                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3332                              "redundant flags; please run fsck.");
3333
3334         /* Check for a known checksum algorithm */
3335         if (!ext4_verify_csum_type(sb, es)) {
3336                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3337                          "unknown checksum algorithm.");
3338                 silent = 1;
3339                 goto cantfind_ext4;
3340         }
3341
3342         /* Load the checksum driver */
3343         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3344                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3345                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3346                 if (IS_ERR(sbi->s_chksum_driver)) {
3347                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3348                         ret = PTR_ERR(sbi->s_chksum_driver);
3349                         sbi->s_chksum_driver = NULL;
3350                         goto failed_mount;
3351                 }
3352         }
3353
3354         /* Check superblock checksum */
3355         if (!ext4_superblock_csum_verify(sb, es)) {
3356                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3357                          "invalid superblock checksum.  Run e2fsck?");
3358                 silent = 1;
3359                 goto cantfind_ext4;
3360         }
3361
3362         /* Precompute checksum seed for all metadata */
3363         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3364                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3365                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3366                                                sizeof(es->s_uuid));
3367
3368         /* Set defaults before we parse the mount options */
3369         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3370         set_opt(sb, INIT_INODE_TABLE);
3371         if (def_mount_opts & EXT4_DEFM_DEBUG)
3372                 set_opt(sb, DEBUG);
3373         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3374                 set_opt(sb, GRPID);
3375         if (def_mount_opts & EXT4_DEFM_UID16)
3376                 set_opt(sb, NO_UID32);
3377         /* xattr user namespace & acls are now defaulted on */
3378         set_opt(sb, XATTR_USER);
3379 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3380         set_opt(sb, POSIX_ACL);
3381 #endif
3382         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3383                 set_opt(sb, JOURNAL_DATA);
3384         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3385                 set_opt(sb, ORDERED_DATA);
3386         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3387                 set_opt(sb, WRITEBACK_DATA);
3388
3389         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3390                 set_opt(sb, ERRORS_PANIC);
3391         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3392                 set_opt(sb, ERRORS_CONT);
3393         else
3394                 set_opt(sb, ERRORS_RO);
3395         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3396                 set_opt(sb, BLOCK_VALIDITY);
3397         if (def_mount_opts & EXT4_DEFM_DISCARD)
3398                 set_opt(sb, DISCARD);
3399
3400         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3401         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3402         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3403         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3404         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3405
3406         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3407                 set_opt(sb, BARRIER);
3408
3409         /*
3410          * enable delayed allocation by default
3411          * Use -o nodelalloc to turn it off
3412          */
3413         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3414             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3415                 set_opt(sb, DELALLOC);
3416
3417         /*
3418          * set default s_li_wait_mult for lazyinit, for the case there is
3419          * no mount option specified.
3420          */
3421         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3422
3423         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3424                            &journal_devnum, &journal_ioprio, 0)) {
3425                 ext4_msg(sb, KERN_WARNING,
3426                          "failed to parse options in superblock: %s",
3427                          sbi->s_es->s_mount_opts);
3428         }
3429         sbi->s_def_mount_opt = sbi->s_mount_opt;
3430         if (!parse_options((char *) data, sb, &journal_devnum,
3431                            &journal_ioprio, 0))
3432                 goto failed_mount;
3433
3434         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3435                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3436                             "with data=journal disables delayed "
3437                             "allocation and O_DIRECT support!\n");
3438                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3439                         ext4_msg(sb, KERN_ERR, "can't mount with "
3440                                  "both data=journal and delalloc");
3441                         goto failed_mount;
3442                 }
3443                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3444                         ext4_msg(sb, KERN_ERR, "can't mount with "
3445                                  "both data=journal and delalloc");
3446                         goto failed_mount;
3447                 }
3448                 if (test_opt(sb, DELALLOC))
3449                         clear_opt(sb, DELALLOC);
3450         }
3451
3452         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3453                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3454
3455         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3456             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3457              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3458              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3459                 ext4_msg(sb, KERN_WARNING,
3460                        "feature flags set on rev 0 fs, "
3461                        "running e2fsck is recommended");
3462
3463         if (IS_EXT2_SB(sb)) {
3464                 if (ext2_feature_set_ok(sb))
3465                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3466                                  "using the ext4 subsystem");
3467                 else {
3468                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3469                                  "to feature incompatibilities");
3470                         goto failed_mount;
3471                 }
3472         }
3473
3474         if (IS_EXT3_SB(sb)) {
3475                 if (ext3_feature_set_ok(sb))
3476                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3477                                  "using the ext4 subsystem");
3478                 else {
3479                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3480                                  "to feature incompatibilities");
3481                         goto failed_mount;
3482                 }
3483         }
3484
3485         /*
3486          * Check feature flags regardless of the revision level, since we
3487          * previously didn't change the revision level when setting the flags,
3488          * so there is a chance incompat flags are set on a rev 0 filesystem.
3489          */
3490         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3491                 goto failed_mount;
3492
3493         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3494         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3495             blocksize > EXT4_MAX_BLOCK_SIZE) {
3496                 ext4_msg(sb, KERN_ERR,
3497                        "Unsupported filesystem blocksize %d", blocksize);
3498                 goto failed_mount;
3499         }
3500
3501         if (sb->s_blocksize != blocksize) {
3502                 /* Validate the filesystem blocksize */
3503                 if (!sb_set_blocksize(sb, blocksize)) {
3504                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3505                                         blocksize);
3506                         goto failed_mount;
3507                 }
3508
3509                 brelse(bh);
3510                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3511                 offset = do_div(logical_sb_block, blocksize);
3512                 bh = sb_bread(sb, logical_sb_block);
3513                 if (!bh) {
3514                         ext4_msg(sb, KERN_ERR,
3515                                "Can't read superblock on 2nd try");
3516                         goto failed_mount;
3517                 }
3518                 es = (struct ext4_super_block *)(bh->b_data + offset);
3519                 sbi->s_es = es;
3520                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3521                         ext4_msg(sb, KERN_ERR,
3522                                "Magic mismatch, very weird!");
3523                         goto failed_mount;
3524                 }
3525         }
3526
3527         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3528                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3529         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3530                                                       has_huge_files);
3531         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3532
3533         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3534                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3535                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3536         } else {
3537                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3538                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3539                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3540                     (!is_power_of_2(sbi->s_inode_size)) ||
3541                     (sbi->s_inode_size > blocksize)) {
3542                         ext4_msg(sb, KERN_ERR,
3543                                "unsupported inode size: %d",
3544                                sbi->s_inode_size);
3545                         goto failed_mount;
3546                 }
3547                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3548                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3549         }
3550
3551         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3552         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3553                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3554                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3555                     !is_power_of_2(sbi->s_desc_size)) {
3556                         ext4_msg(sb, KERN_ERR,
3557                                "unsupported descriptor size %lu",
3558                                sbi->s_desc_size);
3559                         goto failed_mount;
3560                 }
3561         } else
3562                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3563
3564         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3565         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3566         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3567                 goto cantfind_ext4;
3568
3569         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3570         if (sbi->s_inodes_per_block == 0)
3571                 goto cantfind_ext4;
3572         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3573                                         sbi->s_inodes_per_block;
3574         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3575         sbi->s_sbh = bh;
3576         sbi->s_mount_state = le16_to_cpu(es->s_state);
3577         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3578         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3579
3580         for (i = 0; i < 4; i++)
3581                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3582         sbi->s_def_hash_version = es->s_def_hash_version;
3583         i = le32_to_cpu(es->s_flags);
3584         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3585                 sbi->s_hash_unsigned = 3;
3586         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3587 #ifdef __CHAR_UNSIGNED__
3588                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3589                 sbi->s_hash_unsigned = 3;
3590 #else
3591                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3592 #endif
3593         }
3594
3595         /* Handle clustersize */
3596         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3597         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3598                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3599         if (has_bigalloc) {
3600                 if (clustersize < blocksize) {
3601                         ext4_msg(sb, KERN_ERR,
3602                                  "cluster size (%d) smaller than "
3603                                  "block size (%d)", clustersize, blocksize);
3604                         goto failed_mount;
3605                 }
3606                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3607                         le32_to_cpu(es->s_log_block_size);
3608                 sbi->s_clusters_per_group =
3609                         le32_to_cpu(es->s_clusters_per_group);
3610                 if (sbi->s_clusters_per_group > blocksize * 8) {
3611                         ext4_msg(sb, KERN_ERR,
3612                                  "#clusters per group too big: %lu",
3613                                  sbi->s_clusters_per_group);
3614                         goto failed_mount;
3615                 }
3616                 if (sbi->s_blocks_per_group !=
3617                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3618                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3619                                  "clusters per group (%lu) inconsistent",
3620                                  sbi->s_blocks_per_group,
3621                                  sbi->s_clusters_per_group);
3622                         goto failed_mount;
3623                 }
3624         } else {
3625                 if (clustersize != blocksize) {
3626                         ext4_warning(sb, "fragment/cluster size (%d) != "
3627                                      "block size (%d)", clustersize,
3628                                      blocksize);
3629                         clustersize = blocksize;
3630                 }
3631                 if (sbi->s_blocks_per_group > blocksize * 8) {
3632                         ext4_msg(sb, KERN_ERR,
3633                                  "#blocks per group too big: %lu",
3634                                  sbi->s_blocks_per_group);
3635                         goto failed_mount;
3636                 }
3637                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3638                 sbi->s_cluster_bits = 0;
3639         }
3640         sbi->s_cluster_ratio = clustersize / blocksize;
3641
3642         if (sbi->s_inodes_per_group > blocksize * 8) {
3643                 ext4_msg(sb, KERN_ERR,
3644                        "#inodes per group too big: %lu",
3645                        sbi->s_inodes_per_group);
3646                 goto failed_mount;
3647         }
3648
3649         /*
3650          * Test whether we have more sectors than will fit in sector_t,
3651          * and whether the max offset is addressable by the page cache.
3652          */
3653         err = generic_check_addressable(sb->s_blocksize_bits,
3654                                         ext4_blocks_count(es));
3655         if (err) {
3656                 ext4_msg(sb, KERN_ERR, "filesystem"
3657                          " too large to mount safely on this system");
3658                 if (sizeof(sector_t) < 8)
3659                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3660                 goto failed_mount;
3661         }
3662
3663         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3664                 goto cantfind_ext4;
3665
3666         /* check blocks count against device size */
3667         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3668         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3669                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3670                        "exceeds size of device (%llu blocks)",
3671                        ext4_blocks_count(es), blocks_count);
3672                 goto failed_mount;
3673         }
3674
3675         /*
3676          * It makes no sense for the first data block to be beyond the end
3677          * of the filesystem.
3678          */
3679         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3680                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3681                          "block %u is beyond end of filesystem (%llu)",
3682                          le32_to_cpu(es->s_first_data_block),
3683                          ext4_blocks_count(es));
3684                 goto failed_mount;
3685         }
3686         blocks_count = (ext4_blocks_count(es) -
3687                         le32_to_cpu(es->s_first_data_block) +
3688                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3689         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3690         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3691                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3692                        "(block count %llu, first data block %u, "
3693                        "blocks per group %lu)", sbi->s_groups_count,
3694                        ext4_blocks_count(es),
3695                        le32_to_cpu(es->s_first_data_block),
3696                        EXT4_BLOCKS_PER_GROUP(sb));
3697                 goto failed_mount;
3698         }
3699         sbi->s_groups_count = blocks_count;
3700         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3701                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3702         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3703                    EXT4_DESC_PER_BLOCK(sb);
3704         sbi->s_group_desc = ext4_kvmalloc(db_count *
3705                                           sizeof(struct buffer_head *),
3706                                           GFP_KERNEL);
3707         if (sbi->s_group_desc == NULL) {
3708                 ext4_msg(sb, KERN_ERR, "not enough memory");
3709                 ret = -ENOMEM;
3710                 goto failed_mount;
3711         }
3712
3713         if (ext4_proc_root)
3714                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3715
3716         if (sbi->s_proc)
3717                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3718                                  &ext4_seq_options_fops, sb);
3719
3720         bgl_lock_init(sbi->s_blockgroup_lock);
3721
3722         for (i = 0; i < db_count; i++) {
3723                 block = descriptor_loc(sb, logical_sb_block, i);
3724                 sbi->s_group_desc[i] = sb_bread(sb, block);
3725                 if (!sbi->s_group_desc[i]) {
3726                         ext4_msg(sb, KERN_ERR,
3727                                "can't read group descriptor %d", i);
3728                         db_count = i;
3729                         goto failed_mount2;
3730                 }
3731         }
3732         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3733                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3734                 goto failed_mount2;
3735         }
3736         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3737                 if (!ext4_fill_flex_info(sb)) {
3738                         ext4_msg(sb, KERN_ERR,
3739                                "unable to initialize "
3740                                "flex_bg meta info!");
3741                         goto failed_mount2;
3742                 }
3743
3744         sbi->s_gdb_count = db_count;
3745         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3746         spin_lock_init(&sbi->s_next_gen_lock);
3747
3748         init_timer(&sbi->s_err_report);
3749         sbi->s_err_report.function = print_daily_error_info;
3750         sbi->s_err_report.data = (unsigned long) sb;
3751
3752         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3753                         ext4_count_free_clusters(sb));
3754         if (!err) {
3755                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3756                                 ext4_count_free_inodes(sb));
3757         }
3758         if (!err) {
3759                 err = percpu_counter_init(&sbi->s_dirs_counter,
3760                                 ext4_count_dirs(sb));
3761         }
3762         if (!err) {
3763                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3764         }
3765         if (err) {
3766                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3767                 goto failed_mount3;
3768         }
3769
3770         sbi->s_stripe = ext4_get_stripe_size(sbi);
3771         sbi->s_max_writeback_mb_bump = 128;
3772         sbi->s_extent_max_zeroout_kb = 32;
3773
3774         /*
3775          * set up enough so that it can read an inode
3776          */
3777         if (!test_opt(sb, NOLOAD) &&
3778             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3779                 sb->s_op = &ext4_sops;
3780         else
3781                 sb->s_op = &ext4_nojournal_sops;
3782         sb->s_export_op = &ext4_export_ops;
3783         sb->s_xattr = ext4_xattr_handlers;
3784 #ifdef CONFIG_QUOTA
3785         sb->s_qcop = &ext4_qctl_operations;
3786         sb->dq_op = &ext4_quota_operations;
3787
3788         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
3789                 /* Use qctl operations for hidden quota files. */
3790                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3791         }
3792 #endif
3793         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3794
3795         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3796         mutex_init(&sbi->s_orphan_lock);
3797
3798         sb->s_root = NULL;
3799
3800         needs_recovery = (es->s_last_orphan != 0 ||
3801                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3802                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3803
3804         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3805             !(sb->s_flags & MS_RDONLY))
3806                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3807                         goto failed_mount3;
3808
3809         /*
3810          * The first inode we look at is the journal inode.  Don't try
3811          * root first: it may be modified in the journal!
3812          */
3813         if (!test_opt(sb, NOLOAD) &&
3814             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3815                 if (ext4_load_journal(sb, es, journal_devnum))
3816                         goto failed_mount3;
3817         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3818               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3819                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3820                        "suppressed and not mounted read-only");
3821                 goto failed_mount_wq;
3822         } else {
3823                 clear_opt(sb, DATA_FLAGS);
3824                 sbi->s_journal = NULL;
3825                 needs_recovery = 0;
3826                 goto no_journal;
3827         }
3828
3829         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3830             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3831                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3832                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3833                 goto failed_mount_wq;
3834         }
3835
3836         if (!set_journal_csum_feature_set(sb)) {
3837                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3838                          "feature set");
3839                 goto failed_mount_wq;
3840         }
3841
3842         /* We have now updated the journal if required, so we can
3843          * validate the data journaling mode. */
3844         switch (test_opt(sb, DATA_FLAGS)) {
3845         case 0:
3846                 /* No mode set, assume a default based on the journal
3847                  * capabilities: ORDERED_DATA if the journal can
3848                  * cope, else JOURNAL_DATA
3849                  */
3850                 if (jbd2_journal_check_available_features
3851                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3852                         set_opt(sb, ORDERED_DATA);
3853                 else
3854                         set_opt(sb, JOURNAL_DATA);
3855                 break;
3856
3857         case EXT4_MOUNT_ORDERED_DATA:
3858         case EXT4_MOUNT_WRITEBACK_DATA:
3859                 if (!jbd2_journal_check_available_features
3860                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3861                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3862                                "requested data journaling mode");
3863                         goto failed_mount_wq;
3864                 }
3865         default:
3866                 break;
3867         }
3868         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3869
3870         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3871
3872         /*
3873          * The journal may have updated the bg summary counts, so we
3874          * need to update the global counters.
3875          */
3876         percpu_counter_set(&sbi->s_freeclusters_counter,
3877                            ext4_count_free_clusters(sb));
3878         percpu_counter_set(&sbi->s_freeinodes_counter,
3879                            ext4_count_free_inodes(sb));
3880         percpu_counter_set(&sbi->s_dirs_counter,
3881                            ext4_count_dirs(sb));
3882         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3883
3884 no_journal:
3885         /*
3886          * Get the # of file system overhead blocks from the
3887          * superblock if present.
3888          */
3889         if (es->s_overhead_clusters)
3890                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3891         else {
3892                 err = ext4_calculate_overhead(sb);
3893                 if (err)
3894                         goto failed_mount_wq;
3895         }
3896
3897         /*
3898          * The maximum number of concurrent works can be high and
3899          * concurrency isn't really necessary.  Limit it to 1.
3900          */
3901         EXT4_SB(sb)->dio_unwritten_wq =
3902                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3903         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3904                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3905                 ret = -ENOMEM;
3906                 goto failed_mount_wq;
3907         }
3908
3909         /*
3910          * The jbd2_journal_load will have done any necessary log recovery,
3911          * so we can safely mount the rest of the filesystem now.
3912          */
3913
3914         root = ext4_iget(sb, EXT4_ROOT_INO);
3915         if (IS_ERR(root)) {
3916                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3917                 ret = PTR_ERR(root);
3918                 root = NULL;
3919                 goto failed_mount4;
3920         }
3921         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3922                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3923                 iput(root);
3924                 goto failed_mount4;
3925         }
3926         sb->s_root = d_make_root(root);
3927         if (!sb->s_root) {
3928                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3929                 ret = -ENOMEM;
3930                 goto failed_mount4;
3931         }
3932
3933         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3934                 sb->s_flags |= MS_RDONLY;
3935
3936         /* determine the minimum size of new large inodes, if present */
3937         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3938                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3939                                                      EXT4_GOOD_OLD_INODE_SIZE;
3940                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3941                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3942                         if (sbi->s_want_extra_isize <
3943                             le16_to_cpu(es->s_want_extra_isize))
3944                                 sbi->s_want_extra_isize =
3945                                         le16_to_cpu(es->s_want_extra_isize);
3946                         if (sbi->s_want_extra_isize <
3947                             le16_to_cpu(es->s_min_extra_isize))
3948                                 sbi->s_want_extra_isize =
3949                                         le16_to_cpu(es->s_min_extra_isize);
3950                 }
3951         }
3952         /* Check if enough inode space is available */
3953         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3954                                                         sbi->s_inode_size) {
3955                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3956                                                        EXT4_GOOD_OLD_INODE_SIZE;
3957                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3958                          "available");
3959         }
3960
3961         err = ext4_setup_system_zone(sb);
3962         if (err) {
3963                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3964                          "zone (%d)", err);
3965                 goto failed_mount4a;
3966         }
3967
3968         ext4_ext_init(sb);
3969         err = ext4_mb_init(sb);
3970         if (err) {
3971                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3972                          err);
3973                 goto failed_mount5;
3974         }
3975
3976         err = ext4_register_li_request(sb, first_not_zeroed);
3977         if (err)
3978                 goto failed_mount6;
3979
3980         sbi->s_kobj.kset = ext4_kset;
3981         init_completion(&sbi->s_kobj_unregister);
3982         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3983                                    "%s", sb->s_id);
3984         if (err)
3985                 goto failed_mount7;
3986
3987         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3988         ext4_orphan_cleanup(sb, es);
3989         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3990         if (needs_recovery) {
3991                 ext4_msg(sb, KERN_INFO, "recovery complete");
3992                 ext4_mark_recovery_complete(sb, es);
3993         }
3994         if (EXT4_SB(sb)->s_journal) {
3995                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3996                         descr = " journalled data mode";
3997                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3998                         descr = " ordered data mode";
3999                 else
4000                         descr = " writeback data mode";
4001         } else
4002                 descr = "out journal";
4003
4004 #ifdef CONFIG_QUOTA
4005         /* Enable quota usage during mount. */
4006         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4007             !(sb->s_flags & MS_RDONLY)) {
4008                 err = ext4_enable_quotas(sb);
4009                 if (err)
4010                         goto failed_mount8;
4011         }
4012 #endif  /* CONFIG_QUOTA */
4013
4014         if (test_opt(sb, DISCARD)) {
4015                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4016                 if (!blk_queue_discard(q))
4017                         ext4_msg(sb, KERN_WARNING,
4018                                  "mounting with \"discard\" option, but "
4019                                  "the device does not support discard");
4020         }
4021
4022         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4023                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4024                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4025
4026         if (es->s_error_count)
4027                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4028
4029         kfree(orig_data);
4030         return 0;
4031
4032 cantfind_ext4:
4033         if (!silent)
4034                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4035         goto failed_mount;
4036
4037 #ifdef CONFIG_QUOTA
4038 failed_mount8:
4039         kobject_del(&sbi->s_kobj);
4040 #endif
4041 failed_mount7:
4042         ext4_unregister_li_request(sb);
4043 failed_mount6:
4044         ext4_mb_release(sb);
4045 failed_mount5:
4046         ext4_ext_release(sb);
4047         ext4_release_system_zone(sb);
4048 failed_mount4a:
4049         dput(sb->s_root);
4050         sb->s_root = NULL;
4051 failed_mount4:
4052         ext4_msg(sb, KERN_ERR, "mount failed");
4053         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4054 failed_mount_wq:
4055         if (sbi->s_journal) {
4056                 jbd2_journal_destroy(sbi->s_journal);
4057                 sbi->s_journal = NULL;
4058         }
4059 failed_mount3:
4060         del_timer(&sbi->s_err_report);
4061         if (sbi->s_flex_groups)
4062                 ext4_kvfree(sbi->s_flex_groups);
4063         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4064         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4065         percpu_counter_destroy(&sbi->s_dirs_counter);
4066         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4067         if (sbi->s_mmp_tsk)
4068                 kthread_stop(sbi->s_mmp_tsk);
4069 failed_mount2:
4070         for (i = 0; i < db_count; i++)
4071                 brelse(sbi->s_group_desc[i]);
4072         ext4_kvfree(sbi->s_group_desc);
4073 failed_mount:
4074         if (sbi->s_chksum_driver)
4075                 crypto_free_shash(sbi->s_chksum_driver);
4076         if (sbi->s_proc) {
4077                 remove_proc_entry("options", sbi->s_proc);
4078                 remove_proc_entry(sb->s_id, ext4_proc_root);
4079         }
4080 #ifdef CONFIG_QUOTA
4081         for (i = 0; i < MAXQUOTAS; i++)
4082                 kfree(sbi->s_qf_names[i]);
4083 #endif
4084         ext4_blkdev_remove(sbi);
4085         brelse(bh);
4086 out_fail:
4087         sb->s_fs_info = NULL;
4088         kfree(sbi->s_blockgroup_lock);
4089         kfree(sbi);
4090 out_free_orig:
4091         kfree(orig_data);
4092         return err ? err : ret;
4093 }
4094
4095 /*
4096  * Setup any per-fs journal parameters now.  We'll do this both on
4097  * initial mount, once the journal has been initialised but before we've
4098  * done any recovery; and again on any subsequent remount.
4099  */
4100 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4101 {
4102         struct ext4_sb_info *sbi = EXT4_SB(sb);
4103
4104         journal->j_commit_interval = sbi->s_commit_interval;
4105         journal->j_min_batch_time = sbi->s_min_batch_time;
4106         journal->j_max_batch_time = sbi->s_max_batch_time;
4107
4108         write_lock(&journal->j_state_lock);
4109         if (test_opt(sb, BARRIER))
4110                 journal->j_flags |= JBD2_BARRIER;
4111         else
4112                 journal->j_flags &= ~JBD2_BARRIER;
4113         if (test_opt(sb, DATA_ERR_ABORT))
4114                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4115         else
4116                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4117         write_unlock(&journal->j_state_lock);
4118 }
4119
4120 static journal_t *ext4_get_journal(struct super_block *sb,
4121                                    unsigned int journal_inum)
4122 {
4123         struct inode *journal_inode;
4124         journal_t *journal;
4125
4126         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4127
4128         /* First, test for the existence of a valid inode on disk.  Bad
4129          * things happen if we iget() an unused inode, as the subsequent
4130          * iput() will try to delete it. */
4131
4132         journal_inode = ext4_iget(sb, journal_inum);
4133         if (IS_ERR(journal_inode)) {
4134                 ext4_msg(sb, KERN_ERR, "no journal found");
4135                 return NULL;
4136         }
4137         if (!journal_inode->i_nlink) {
4138                 make_bad_inode(journal_inode);
4139                 iput(journal_inode);
4140                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4141                 return NULL;
4142         }
4143
4144         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4145                   journal_inode, journal_inode->i_size);
4146         if (!S_ISREG(journal_inode->i_mode)) {
4147                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4148                 iput(journal_inode);
4149                 return NULL;
4150         }
4151
4152         journal = jbd2_journal_init_inode(journal_inode);
4153         if (!journal) {
4154                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4155                 iput(journal_inode);
4156                 return NULL;
4157         }
4158         journal->j_private = sb;
4159         ext4_init_journal_params(sb, journal);
4160         return journal;
4161 }
4162
4163 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4164                                        dev_t j_dev)
4165 {
4166         struct buffer_head *bh;
4167         journal_t *journal;
4168         ext4_fsblk_t start;
4169         ext4_fsblk_t len;
4170         int hblock, blocksize;
4171         ext4_fsblk_t sb_block;
4172         unsigned long offset;
4173         struct ext4_super_block *es;
4174         struct block_device *bdev;
4175
4176         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4177
4178         bdev = ext4_blkdev_get(j_dev, sb);
4179         if (bdev == NULL)
4180                 return NULL;
4181
4182         blocksize = sb->s_blocksize;
4183         hblock = bdev_logical_block_size(bdev);
4184         if (blocksize < hblock) {
4185                 ext4_msg(sb, KERN_ERR,
4186                         "blocksize too small for journal device");
4187                 goto out_bdev;
4188         }
4189
4190         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4191         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4192         set_blocksize(bdev, blocksize);
4193         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4194                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4195                        "external journal");
4196                 goto out_bdev;
4197         }
4198
4199         es = (struct ext4_super_block *) (bh->b_data + offset);
4200         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4201             !(le32_to_cpu(es->s_feature_incompat) &
4202               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4203                 ext4_msg(sb, KERN_ERR, "external journal has "
4204                                         "bad superblock");
4205                 brelse(bh);
4206                 goto out_bdev;
4207         }
4208
4209         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4210                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4211                 brelse(bh);
4212                 goto out_bdev;
4213         }
4214
4215         len = ext4_blocks_count(es);
4216         start = sb_block + 1;
4217         brelse(bh);     /* we're done with the superblock */
4218
4219         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4220                                         start, len, blocksize);
4221         if (!journal) {
4222                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4223                 goto out_bdev;
4224         }
4225         journal->j_private = sb;
4226         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4227         wait_on_buffer(journal->j_sb_buffer);
4228         if (!buffer_uptodate(journal->j_sb_buffer)) {
4229                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4230                 goto out_journal;
4231         }
4232         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4233                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4234                                         "user (unsupported) - %d",
4235                         be32_to_cpu(journal->j_superblock->s_nr_users));
4236                 goto out_journal;
4237         }
4238         EXT4_SB(sb)->journal_bdev = bdev;
4239         ext4_init_journal_params(sb, journal);
4240         return journal;
4241
4242 out_journal:
4243         jbd2_journal_destroy(journal);
4244 out_bdev:
4245         ext4_blkdev_put(bdev);
4246         return NULL;
4247 }
4248
4249 static int ext4_load_journal(struct super_block *sb,
4250                              struct ext4_super_block *es,
4251                              unsigned long journal_devnum)
4252 {
4253         journal_t *journal;
4254         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4255         dev_t journal_dev;
4256         int err = 0;
4257         int really_read_only;
4258
4259         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4260
4261         if (journal_devnum &&
4262             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4263                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4264                         "numbers have changed");
4265                 journal_dev = new_decode_dev(journal_devnum);
4266         } else
4267                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4268
4269         really_read_only = bdev_read_only(sb->s_bdev);
4270
4271         /*
4272          * Are we loading a blank journal or performing recovery after a
4273          * crash?  For recovery, we need to check in advance whether we
4274          * can get read-write access to the device.
4275          */
4276         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4277                 if (sb->s_flags & MS_RDONLY) {
4278                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4279                                         "required on readonly filesystem");
4280                         if (really_read_only) {
4281                                 ext4_msg(sb, KERN_ERR, "write access "
4282                                         "unavailable, cannot proceed");
4283                                 return -EROFS;
4284                         }
4285                         ext4_msg(sb, KERN_INFO, "write access will "
4286                                "be enabled during recovery");
4287                 }
4288         }
4289
4290         if (journal_inum && journal_dev) {
4291                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4292                        "and inode journals!");
4293                 return -EINVAL;
4294         }
4295
4296         if (journal_inum) {
4297                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4298                         return -EINVAL;
4299         } else {
4300                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4301                         return -EINVAL;
4302         }
4303
4304         if (!(journal->j_flags & JBD2_BARRIER))
4305                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4306
4307         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4308                 err = jbd2_journal_wipe(journal, !really_read_only);
4309         if (!err) {
4310                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4311                 if (save)
4312                         memcpy(save, ((char *) es) +
4313                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4314                 err = jbd2_journal_load(journal);
4315                 if (save)
4316                         memcpy(((char *) es) + EXT4_S_ERR_START,
4317                                save, EXT4_S_ERR_LEN);
4318                 kfree(save);
4319         }
4320
4321         if (err) {
4322                 ext4_msg(sb, KERN_ERR, "error loading journal");
4323                 jbd2_journal_destroy(journal);
4324                 return err;
4325         }
4326
4327         EXT4_SB(sb)->s_journal = journal;
4328         ext4_clear_journal_err(sb, es);
4329
4330         if (!really_read_only && journal_devnum &&
4331             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4332                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4333
4334                 /* Make sure we flush the recovery flag to disk. */
4335                 ext4_commit_super(sb, 1);
4336         }
4337
4338         return 0;
4339 }
4340
4341 static int ext4_commit_super(struct super_block *sb, int sync)
4342 {
4343         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4344         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4345         int error = 0;
4346
4347         if (!sbh || block_device_ejected(sb))
4348                 return error;
4349         if (buffer_write_io_error(sbh)) {
4350                 /*
4351                  * Oh, dear.  A previous attempt to write the
4352                  * superblock failed.  This could happen because the
4353                  * USB device was yanked out.  Or it could happen to
4354                  * be a transient write error and maybe the block will
4355                  * be remapped.  Nothing we can do but to retry the
4356                  * write and hope for the best.
4357                  */
4358                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4359                        "superblock detected");
4360                 clear_buffer_write_io_error(sbh);
4361                 set_buffer_uptodate(sbh);
4362         }
4363         /*
4364          * If the file system is mounted read-only, don't update the
4365          * superblock write time.  This avoids updating the superblock
4366          * write time when we are mounting the root file system
4367          * read/only but we need to replay the journal; at that point,
4368          * for people who are east of GMT and who make their clock
4369          * tick in localtime for Windows bug-for-bug compatibility,
4370          * the clock is set in the future, and this will cause e2fsck
4371          * to complain and force a full file system check.
4372          */
4373         if (!(sb->s_flags & MS_RDONLY))
4374                 es->s_wtime = cpu_to_le32(get_seconds());
4375         if (sb->s_bdev->bd_part)
4376                 es->s_kbytes_written =
4377                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4378                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4379                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4380         else
4381                 es->s_kbytes_written =
4382                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4383         ext4_free_blocks_count_set(es,
4384                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4385                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4386         es->s_free_inodes_count =
4387                 cpu_to_le32(percpu_counter_sum_positive(
4388                                 &EXT4_SB(sb)->s_freeinodes_counter));
4389         BUFFER_TRACE(sbh, "marking dirty");
4390         ext4_superblock_csum_set(sb);
4391         mark_buffer_dirty(sbh);
4392         if (sync) {
4393                 error = sync_dirty_buffer(sbh);
4394                 if (error)
4395                         return error;
4396
4397                 error = buffer_write_io_error(sbh);
4398                 if (error) {
4399                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4400                                "superblock");
4401                         clear_buffer_write_io_error(sbh);
4402                         set_buffer_uptodate(sbh);
4403                 }
4404         }
4405         return error;
4406 }
4407
4408 /*
4409  * Have we just finished recovery?  If so, and if we are mounting (or
4410  * remounting) the filesystem readonly, then we will end up with a
4411  * consistent fs on disk.  Record that fact.
4412  */
4413 static void ext4_mark_recovery_complete(struct super_block *sb,
4414                                         struct ext4_super_block *es)
4415 {
4416         journal_t *journal = EXT4_SB(sb)->s_journal;
4417
4418         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4419                 BUG_ON(journal != NULL);
4420                 return;
4421         }
4422         jbd2_journal_lock_updates(journal);
4423         if (jbd2_journal_flush(journal) < 0)
4424                 goto out;
4425
4426         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4427             sb->s_flags & MS_RDONLY) {
4428                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4429                 ext4_commit_super(sb, 1);
4430         }
4431
4432 out:
4433         jbd2_journal_unlock_updates(journal);
4434 }
4435
4436 /*
4437  * If we are mounting (or read-write remounting) a filesystem whose journal
4438  * has recorded an error from a previous lifetime, move that error to the
4439  * main filesystem now.
4440  */
4441 static void ext4_clear_journal_err(struct super_block *sb,
4442                                    struct ext4_super_block *es)
4443 {
4444         journal_t *journal;
4445         int j_errno;
4446         const char *errstr;
4447
4448         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4449
4450         journal = EXT4_SB(sb)->s_journal;
4451
4452         /*
4453          * Now check for any error status which may have been recorded in the
4454          * journal by a prior ext4_error() or ext4_abort()
4455          */
4456
4457         j_errno = jbd2_journal_errno(journal);
4458         if (j_errno) {
4459                 char nbuf[16];
4460
4461                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4462                 ext4_warning(sb, "Filesystem error recorded "
4463                              "from previous mount: %s", errstr);
4464                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4465
4466                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4467                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4468                 ext4_commit_super(sb, 1);
4469
4470                 jbd2_journal_clear_err(journal);
4471                 jbd2_journal_update_sb_errno(journal);
4472         }
4473 }
4474
4475 /*
4476  * Force the running and committing transactions to commit,
4477  * and wait on the commit.
4478  */
4479 int ext4_force_commit(struct super_block *sb)
4480 {
4481         journal_t *journal;
4482         int ret = 0;
4483
4484         if (sb->s_flags & MS_RDONLY)
4485                 return 0;
4486
4487         journal = EXT4_SB(sb)->s_journal;
4488         if (journal)
4489                 ret = ext4_journal_force_commit(journal);
4490
4491         return ret;
4492 }
4493
4494 static int ext4_sync_fs(struct super_block *sb, int wait)
4495 {
4496         int ret = 0;
4497         tid_t target;
4498         struct ext4_sb_info *sbi = EXT4_SB(sb);
4499
4500         trace_ext4_sync_fs(sb, wait);
4501         flush_workqueue(sbi->dio_unwritten_wq);
4502         /*
4503          * Writeback quota in non-journalled quota case - journalled quota has
4504          * no dirty dquots
4505          */
4506         dquot_writeback_dquots(sb, -1);
4507         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4508                 if (wait)
4509                         jbd2_log_wait_commit(sbi->s_journal, target);
4510         }
4511         return ret;
4512 }
4513
4514 /*
4515  * LVM calls this function before a (read-only) snapshot is created.  This
4516  * gives us a chance to flush the journal completely and mark the fs clean.
4517  *
4518  * Note that only this function cannot bring a filesystem to be in a clean
4519  * state independently. It relies on upper layer to stop all data & metadata
4520  * modifications.
4521  */
4522 static int ext4_freeze(struct super_block *sb)
4523 {
4524         int error = 0;
4525         journal_t *journal;
4526
4527         if (sb->s_flags & MS_RDONLY)
4528                 return 0;
4529
4530         journal = EXT4_SB(sb)->s_journal;
4531
4532         /* Now we set up the journal barrier. */
4533         jbd2_journal_lock_updates(journal);
4534
4535         /*
4536          * Don't clear the needs_recovery flag if we failed to flush
4537          * the journal.
4538          */
4539         error = jbd2_journal_flush(journal);
4540         if (error < 0)
4541                 goto out;
4542
4543         /* Journal blocked and flushed, clear needs_recovery flag. */
4544         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4545         error = ext4_commit_super(sb, 1);
4546 out:
4547         /* we rely on upper layer to stop further updates */
4548         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4549         return error;
4550 }
4551
4552 /*
4553  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4554  * flag here, even though the filesystem is not technically dirty yet.
4555  */
4556 static int ext4_unfreeze(struct super_block *sb)
4557 {
4558         if (sb->s_flags & MS_RDONLY)
4559                 return 0;
4560
4561         /* Reset the needs_recovery flag before the fs is unlocked. */
4562         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4563         ext4_commit_super(sb, 1);
4564         return 0;
4565 }
4566
4567 /*
4568  * Structure to save mount options for ext4_remount's benefit
4569  */
4570 struct ext4_mount_options {
4571         unsigned long s_mount_opt;
4572         unsigned long s_mount_opt2;
4573         kuid_t s_resuid;
4574         kgid_t s_resgid;
4575         unsigned long s_commit_interval;
4576         u32 s_min_batch_time, s_max_batch_time;
4577 #ifdef CONFIG_QUOTA
4578         int s_jquota_fmt;
4579         char *s_qf_names[MAXQUOTAS];
4580 #endif
4581 };
4582
4583 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4584 {
4585         struct ext4_super_block *es;
4586         struct ext4_sb_info *sbi = EXT4_SB(sb);
4587         unsigned long old_sb_flags;
4588         struct ext4_mount_options old_opts;
4589         int enable_quota = 0;
4590         ext4_group_t g;
4591         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4592         int err = 0;
4593 #ifdef CONFIG_QUOTA
4594         int i, j;
4595 #endif
4596         char *orig_data = kstrdup(data, GFP_KERNEL);
4597
4598         /* Store the original options */
4599         old_sb_flags = sb->s_flags;
4600         old_opts.s_mount_opt = sbi->s_mount_opt;
4601         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4602         old_opts.s_resuid = sbi->s_resuid;
4603         old_opts.s_resgid = sbi->s_resgid;
4604         old_opts.s_commit_interval = sbi->s_commit_interval;
4605         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4606         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4607 #ifdef CONFIG_QUOTA
4608         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4609         for (i = 0; i < MAXQUOTAS; i++)
4610                 if (sbi->s_qf_names[i]) {
4611                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4612                                                          GFP_KERNEL);
4613                         if (!old_opts.s_qf_names[i]) {
4614                                 for (j = 0; j < i; j++)
4615                                         kfree(old_opts.s_qf_names[j]);
4616                                 return -ENOMEM;
4617                         }
4618                 } else
4619                         old_opts.s_qf_names[i] = NULL;
4620 #endif
4621         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4622                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4623
4624         /*
4625          * Allow the "check" option to be passed as a remount option.
4626          */
4627         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4628                 err = -EINVAL;
4629                 goto restore_opts;
4630         }
4631
4632         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4633                 ext4_abort(sb, "Abort forced by user");
4634
4635         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4636                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4637
4638         es = sbi->s_es;
4639
4640         if (sbi->s_journal) {
4641                 ext4_init_journal_params(sb, sbi->s_journal);
4642                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4643         }
4644
4645         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4646                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4647                         err = -EROFS;
4648                         goto restore_opts;
4649                 }
4650
4651                 if (*flags & MS_RDONLY) {
4652                         err = dquot_suspend(sb, -1);
4653                         if (err < 0)
4654                                 goto restore_opts;
4655
4656                         /*
4657                          * First of all, the unconditional stuff we have to do
4658                          * to disable replay of the journal when we next remount
4659                          */
4660                         sb->s_flags |= MS_RDONLY;
4661
4662                         /*
4663                          * OK, test if we are remounting a valid rw partition
4664                          * readonly, and if so set the rdonly flag and then
4665                          * mark the partition as valid again.
4666                          */
4667                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4668                             (sbi->s_mount_state & EXT4_VALID_FS))
4669                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4670
4671                         if (sbi->s_journal)
4672                                 ext4_mark_recovery_complete(sb, es);
4673                 } else {
4674                         /* Make sure we can mount this feature set readwrite */
4675                         if (!ext4_feature_set_ok(sb, 0)) {
4676                                 err = -EROFS;
4677                                 goto restore_opts;
4678                         }
4679                         /*
4680                          * Make sure the group descriptor checksums
4681                          * are sane.  If they aren't, refuse to remount r/w.
4682                          */
4683                         for (g = 0; g < sbi->s_groups_count; g++) {
4684                                 struct ext4_group_desc *gdp =
4685                                         ext4_get_group_desc(sb, g, NULL);
4686
4687                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4688                                         ext4_msg(sb, KERN_ERR,
4689                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4690                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4691                                                le16_to_cpu(gdp->bg_checksum));
4692                                         err = -EINVAL;
4693                                         goto restore_opts;
4694                                 }
4695                         }
4696
4697                         /*
4698                          * If we have an unprocessed orphan list hanging
4699                          * around from a previously readonly bdev mount,
4700                          * require a full umount/remount for now.
4701                          */
4702                         if (es->s_last_orphan) {
4703                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4704                                        "remount RDWR because of unprocessed "
4705                                        "orphan inode list.  Please "
4706                                        "umount/remount instead");
4707                                 err = -EINVAL;
4708                                 goto restore_opts;
4709                         }
4710
4711                         /*
4712                          * Mounting a RDONLY partition read-write, so reread
4713                          * and store the current valid flag.  (It may have
4714                          * been changed by e2fsck since we originally mounted
4715                          * the partition.)
4716                          */
4717                         if (sbi->s_journal)
4718                                 ext4_clear_journal_err(sb, es);
4719                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4720                         if (!ext4_setup_super(sb, es, 0))
4721                                 sb->s_flags &= ~MS_RDONLY;
4722                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4723                                                      EXT4_FEATURE_INCOMPAT_MMP))
4724                                 if (ext4_multi_mount_protect(sb,
4725                                                 le64_to_cpu(es->s_mmp_block))) {
4726                                         err = -EROFS;
4727                                         goto restore_opts;
4728                                 }
4729                         enable_quota = 1;
4730                 }
4731         }
4732
4733         /*
4734          * Reinitialize lazy itable initialization thread based on
4735          * current settings
4736          */
4737         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4738                 ext4_unregister_li_request(sb);
4739         else {
4740                 ext4_group_t first_not_zeroed;
4741                 first_not_zeroed = ext4_has_uninit_itable(sb);
4742                 ext4_register_li_request(sb, first_not_zeroed);
4743         }
4744
4745         ext4_setup_system_zone(sb);
4746         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4747                 ext4_commit_super(sb, 1);
4748
4749 #ifdef CONFIG_QUOTA
4750         /* Release old quota file names */
4751         for (i = 0; i < MAXQUOTAS; i++)
4752                 kfree(old_opts.s_qf_names[i]);
4753         if (enable_quota) {
4754                 if (sb_any_quota_suspended(sb))
4755                         dquot_resume(sb, -1);
4756                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4757                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4758                         err = ext4_enable_quotas(sb);
4759                         if (err)
4760                                 goto restore_opts;
4761                 }
4762         }
4763 #endif
4764
4765         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4766         kfree(orig_data);
4767         return 0;
4768
4769 restore_opts:
4770         sb->s_flags = old_sb_flags;
4771         sbi->s_mount_opt = old_opts.s_mount_opt;
4772         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4773         sbi->s_resuid = old_opts.s_resuid;
4774         sbi->s_resgid = old_opts.s_resgid;
4775         sbi->s_commit_interval = old_opts.s_commit_interval;
4776         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4777         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4778 #ifdef CONFIG_QUOTA
4779         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4780         for (i = 0; i < MAXQUOTAS; i++) {
4781                 kfree(sbi->s_qf_names[i]);
4782                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4783         }
4784 #endif
4785         kfree(orig_data);
4786         return err;
4787 }
4788
4789 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4790 {
4791         struct super_block *sb = dentry->d_sb;
4792         struct ext4_sb_info *sbi = EXT4_SB(sb);
4793         struct ext4_super_block *es = sbi->s_es;
4794         ext4_fsblk_t overhead = 0;
4795         u64 fsid;
4796         s64 bfree;
4797
4798         if (!test_opt(sb, MINIX_DF))
4799                 overhead = sbi->s_overhead;
4800
4801         buf->f_type = EXT4_SUPER_MAGIC;
4802         buf->f_bsize = sb->s_blocksize;
4803         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4804         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4805                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4806         /* prevent underflow in case that few free space is available */
4807         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4808         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4809         if (buf->f_bfree < ext4_r_blocks_count(es))
4810                 buf->f_bavail = 0;
4811         buf->f_files = le32_to_cpu(es->s_inodes_count);
4812         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4813         buf->f_namelen = EXT4_NAME_LEN;
4814         fsid = le64_to_cpup((void *)es->s_uuid) ^
4815                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4816         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4817         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4818
4819         return 0;
4820 }
4821
4822 /* Helper function for writing quotas on sync - we need to start transaction
4823  * before quota file is locked for write. Otherwise the are possible deadlocks:
4824  * Process 1                         Process 2
4825  * ext4_create()                     quota_sync()
4826  *   jbd2_journal_start()                  write_dquot()
4827  *   dquot_initialize()                         down(dqio_mutex)
4828  *     down(dqio_mutex)                    jbd2_journal_start()
4829  *
4830  */
4831
4832 #ifdef CONFIG_QUOTA
4833
4834 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4835 {
4836         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4837 }
4838
4839 static int ext4_write_dquot(struct dquot *dquot)
4840 {
4841         int ret, err;
4842         handle_t *handle;
4843         struct inode *inode;
4844
4845         inode = dquot_to_inode(dquot);
4846         handle = ext4_journal_start(inode,
4847                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4848         if (IS_ERR(handle))
4849                 return PTR_ERR(handle);
4850         ret = dquot_commit(dquot);
4851         err = ext4_journal_stop(handle);
4852         if (!ret)
4853                 ret = err;
4854         return ret;
4855 }
4856
4857 static int ext4_acquire_dquot(struct dquot *dquot)
4858 {
4859         int ret, err;
4860         handle_t *handle;
4861
4862         handle = ext4_journal_start(dquot_to_inode(dquot),
4863                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4864         if (IS_ERR(handle))
4865                 return PTR_ERR(handle);
4866         ret = dquot_acquire(dquot);
4867         err = ext4_journal_stop(handle);
4868         if (!ret)
4869                 ret = err;
4870         return ret;
4871 }
4872
4873 static int ext4_release_dquot(struct dquot *dquot)
4874 {
4875         int ret, err;
4876         handle_t *handle;
4877
4878         handle = ext4_journal_start(dquot_to_inode(dquot),
4879                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4880         if (IS_ERR(handle)) {
4881                 /* Release dquot anyway to avoid endless cycle in dqput() */
4882                 dquot_release(dquot);
4883                 return PTR_ERR(handle);
4884         }
4885         ret = dquot_release(dquot);
4886         err = ext4_journal_stop(handle);
4887         if (!ret)
4888                 ret = err;
4889         return ret;
4890 }
4891
4892 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4893 {
4894         /* Are we journaling quotas? */
4895         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4896             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4897                 dquot_mark_dquot_dirty(dquot);
4898                 return ext4_write_dquot(dquot);
4899         } else {
4900                 return dquot_mark_dquot_dirty(dquot);
4901         }
4902 }
4903
4904 static int ext4_write_info(struct super_block *sb, int type)
4905 {
4906         int ret, err;
4907         handle_t *handle;
4908
4909         /* Data block + inode block */
4910         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4911         if (IS_ERR(handle))
4912                 return PTR_ERR(handle);
4913         ret = dquot_commit_info(sb, type);
4914         err = ext4_journal_stop(handle);
4915         if (!ret)
4916                 ret = err;
4917         return ret;
4918 }
4919
4920 /*
4921  * Turn on quotas during mount time - we need to find
4922  * the quota file and such...
4923  */
4924 static int ext4_quota_on_mount(struct super_block *sb, int type)
4925 {
4926         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4927                                         EXT4_SB(sb)->s_jquota_fmt, type);
4928 }
4929
4930 /*
4931  * Standard function to be called on quota_on
4932  */
4933 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4934                          struct path *path)
4935 {
4936         int err;
4937
4938         if (!test_opt(sb, QUOTA))
4939                 return -EINVAL;
4940
4941         /* Quotafile not on the same filesystem? */
4942         if (path->dentry->d_sb != sb)
4943                 return -EXDEV;
4944         /* Journaling quota? */
4945         if (EXT4_SB(sb)->s_qf_names[type]) {
4946                 /* Quotafile not in fs root? */
4947                 if (path->dentry->d_parent != sb->s_root)
4948                         ext4_msg(sb, KERN_WARNING,
4949                                 "Quota file not on filesystem root. "
4950                                 "Journaled quota will not work");
4951         }
4952
4953         /*
4954          * When we journal data on quota file, we have to flush journal to see
4955          * all updates to the file when we bypass pagecache...
4956          */
4957         if (EXT4_SB(sb)->s_journal &&
4958             ext4_should_journal_data(path->dentry->d_inode)) {
4959                 /*
4960                  * We don't need to lock updates but journal_flush() could
4961                  * otherwise be livelocked...
4962                  */
4963                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4964                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4965                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4966                 if (err)
4967                         return err;
4968         }
4969
4970         return dquot_quota_on(sb, type, format_id, path);
4971 }
4972
4973 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
4974                              unsigned int flags)
4975 {
4976         int err;
4977         struct inode *qf_inode;
4978         unsigned long qf_inums[MAXQUOTAS] = {
4979                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4980                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4981         };
4982
4983         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
4984
4985         if (!qf_inums[type])
4986                 return -EPERM;
4987
4988         qf_inode = ext4_iget(sb, qf_inums[type]);
4989         if (IS_ERR(qf_inode)) {
4990                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
4991                 return PTR_ERR(qf_inode);
4992         }
4993
4994         err = dquot_enable(qf_inode, type, format_id, flags);
4995         iput(qf_inode);
4996
4997         return err;
4998 }
4999
5000 /* Enable usage tracking for all quota types. */
5001 static int ext4_enable_quotas(struct super_block *sb)
5002 {
5003         int type, err = 0;
5004         unsigned long qf_inums[MAXQUOTAS] = {
5005                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5006                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5007         };
5008
5009         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5010         for (type = 0; type < MAXQUOTAS; type++) {
5011                 if (qf_inums[type]) {
5012                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5013                                                 DQUOT_USAGE_ENABLED);
5014                         if (err) {
5015                                 ext4_warning(sb,
5016                                         "Failed to enable quota tracking "
5017                                         "(type=%d, err=%d). Please run "
5018                                         "e2fsck to fix.", type, err);
5019                                 return err;
5020                         }
5021                 }
5022         }
5023         return 0;
5024 }
5025
5026 /*
5027  * quota_on function that is used when QUOTA feature is set.
5028  */
5029 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5030                                  int format_id)
5031 {
5032         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5033                 return -EINVAL;
5034
5035         /*
5036          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5037          */
5038         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5039 }
5040
5041 static int ext4_quota_off(struct super_block *sb, int type)
5042 {
5043         struct inode *inode = sb_dqopt(sb)->files[type];
5044         handle_t *handle;
5045
5046         /* Force all delayed allocation blocks to be allocated.
5047          * Caller already holds s_umount sem */
5048         if (test_opt(sb, DELALLOC))
5049                 sync_filesystem(sb);
5050
5051         if (!inode)
5052                 goto out;
5053
5054         /* Update modification times of quota files when userspace can
5055          * start looking at them */
5056         handle = ext4_journal_start(inode, 1);
5057         if (IS_ERR(handle))
5058                 goto out;
5059         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5060         ext4_mark_inode_dirty(handle, inode);
5061         ext4_journal_stop(handle);
5062
5063 out:
5064         return dquot_quota_off(sb, type);
5065 }
5066
5067 /*
5068  * quota_off function that is used when QUOTA feature is set.
5069  */
5070 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5071 {
5072         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5073                 return -EINVAL;
5074
5075         /* Disable only the limits. */
5076         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5077 }
5078
5079 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5080  * acquiring the locks... As quota files are never truncated and quota code
5081  * itself serializes the operations (and no one else should touch the files)
5082  * we don't have to be afraid of races */
5083 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5084                                size_t len, loff_t off)
5085 {
5086         struct inode *inode = sb_dqopt(sb)->files[type];
5087         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5088         int err = 0;
5089         int offset = off & (sb->s_blocksize - 1);
5090         int tocopy;
5091         size_t toread;
5092         struct buffer_head *bh;
5093         loff_t i_size = i_size_read(inode);
5094
5095         if (off > i_size)
5096                 return 0;
5097         if (off+len > i_size)
5098                 len = i_size-off;
5099         toread = len;
5100         while (toread > 0) {
5101                 tocopy = sb->s_blocksize - offset < toread ?
5102                                 sb->s_blocksize - offset : toread;
5103                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5104                 if (err)
5105                         return err;
5106                 if (!bh)        /* A hole? */
5107                         memset(data, 0, tocopy);
5108                 else
5109                         memcpy(data, bh->b_data+offset, tocopy);
5110                 brelse(bh);
5111                 offset = 0;
5112                 toread -= tocopy;
5113                 data += tocopy;
5114                 blk++;
5115         }
5116         return len;
5117 }
5118
5119 /* Write to quotafile (we know the transaction is already started and has
5120  * enough credits) */
5121 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5122                                 const char *data, size_t len, loff_t off)
5123 {
5124         struct inode *inode = sb_dqopt(sb)->files[type];
5125         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5126         int err = 0;
5127         int offset = off & (sb->s_blocksize - 1);
5128         struct buffer_head *bh;
5129         handle_t *handle = journal_current_handle();
5130
5131         if (EXT4_SB(sb)->s_journal && !handle) {
5132                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5133                         " cancelled because transaction is not started",
5134                         (unsigned long long)off, (unsigned long long)len);
5135                 return -EIO;
5136         }
5137         /*
5138          * Since we account only one data block in transaction credits,
5139          * then it is impossible to cross a block boundary.
5140          */
5141         if (sb->s_blocksize - offset < len) {
5142                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5143                         " cancelled because not block aligned",
5144                         (unsigned long long)off, (unsigned long long)len);
5145                 return -EIO;
5146         }
5147
5148         bh = ext4_bread(handle, inode, blk, 1, &err);
5149         if (!bh)
5150                 goto out;
5151         err = ext4_journal_get_write_access(handle, bh);
5152         if (err) {
5153                 brelse(bh);
5154                 goto out;
5155         }
5156         lock_buffer(bh);
5157         memcpy(bh->b_data+offset, data, len);
5158         flush_dcache_page(bh->b_page);
5159         unlock_buffer(bh);
5160         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5161         brelse(bh);
5162 out:
5163         if (err)
5164                 return err;
5165         if (inode->i_size < off + len) {
5166                 i_size_write(inode, off + len);
5167                 EXT4_I(inode)->i_disksize = inode->i_size;
5168                 ext4_mark_inode_dirty(handle, inode);
5169         }
5170         return len;
5171 }
5172
5173 #endif
5174
5175 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5176                        const char *dev_name, void *data)
5177 {
5178         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5179 }
5180
5181 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5182 static inline void register_as_ext2(void)
5183 {
5184         int err = register_filesystem(&ext2_fs_type);
5185         if (err)
5186                 printk(KERN_WARNING
5187                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5188 }
5189
5190 static inline void unregister_as_ext2(void)
5191 {
5192         unregister_filesystem(&ext2_fs_type);
5193 }
5194
5195 static inline int ext2_feature_set_ok(struct super_block *sb)
5196 {
5197         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5198                 return 0;
5199         if (sb->s_flags & MS_RDONLY)
5200                 return 1;
5201         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5202                 return 0;
5203         return 1;
5204 }
5205 MODULE_ALIAS("ext2");
5206 #else
5207 static inline void register_as_ext2(void) { }
5208 static inline void unregister_as_ext2(void) { }
5209 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5210 #endif
5211
5212 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5213 static inline void register_as_ext3(void)
5214 {
5215         int err = register_filesystem(&ext3_fs_type);
5216         if (err)
5217                 printk(KERN_WARNING
5218                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5219 }
5220
5221 static inline void unregister_as_ext3(void)
5222 {
5223         unregister_filesystem(&ext3_fs_type);
5224 }
5225
5226 static inline int ext3_feature_set_ok(struct super_block *sb)
5227 {
5228         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5229                 return 0;
5230         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5231                 return 0;
5232         if (sb->s_flags & MS_RDONLY)
5233                 return 1;
5234         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5235                 return 0;
5236         return 1;
5237 }
5238 MODULE_ALIAS("ext3");
5239 #else
5240 static inline void register_as_ext3(void) { }
5241 static inline void unregister_as_ext3(void) { }
5242 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5243 #endif
5244
5245 static struct file_system_type ext4_fs_type = {
5246         .owner          = THIS_MODULE,
5247         .name           = "ext4",
5248         .mount          = ext4_mount,
5249         .kill_sb        = kill_block_super,
5250         .fs_flags       = FS_REQUIRES_DEV,
5251 };
5252
5253 static int __init ext4_init_feat_adverts(void)
5254 {
5255         struct ext4_features *ef;
5256         int ret = -ENOMEM;
5257
5258         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5259         if (!ef)
5260                 goto out;
5261
5262         ef->f_kobj.kset = ext4_kset;
5263         init_completion(&ef->f_kobj_unregister);
5264         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5265                                    "features");
5266         if (ret) {
5267                 kfree(ef);
5268                 goto out;
5269         }
5270
5271         ext4_feat = ef;
5272         ret = 0;
5273 out:
5274         return ret;
5275 }
5276
5277 static void ext4_exit_feat_adverts(void)
5278 {
5279         kobject_put(&ext4_feat->f_kobj);
5280         wait_for_completion(&ext4_feat->f_kobj_unregister);
5281         kfree(ext4_feat);
5282 }
5283
5284 /* Shared across all ext4 file systems */
5285 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5286 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5287
5288 static int __init ext4_init_fs(void)
5289 {
5290         int i, err;
5291
5292         ext4_li_info = NULL;
5293         mutex_init(&ext4_li_mtx);
5294
5295         /* Build-time check for flags consistency */
5296         ext4_check_flag_values();
5297
5298         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5299                 mutex_init(&ext4__aio_mutex[i]);
5300                 init_waitqueue_head(&ext4__ioend_wq[i]);
5301         }
5302
5303         err = ext4_init_es();
5304         if (err)
5305                 return err;
5306
5307         err = ext4_init_pageio();
5308         if (err)
5309                 goto out7;
5310
5311         err = ext4_init_system_zone();
5312         if (err)
5313                 goto out6;
5314         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5315         if (!ext4_kset) {
5316                 err = -ENOMEM;
5317                 goto out5;
5318         }
5319         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5320
5321         err = ext4_init_feat_adverts();
5322         if (err)
5323                 goto out4;
5324
5325         err = ext4_init_mballoc();
5326         if (err)
5327                 goto out3;
5328
5329         err = ext4_init_xattr();
5330         if (err)
5331                 goto out2;
5332         err = init_inodecache();
5333         if (err)
5334                 goto out1;
5335         register_as_ext3();
5336         register_as_ext2();
5337         err = register_filesystem(&ext4_fs_type);
5338         if (err)
5339                 goto out;
5340
5341         return 0;
5342 out:
5343         unregister_as_ext2();
5344         unregister_as_ext3();
5345         destroy_inodecache();
5346 out1:
5347         ext4_exit_xattr();
5348 out2:
5349         ext4_exit_mballoc();
5350 out3:
5351         ext4_exit_feat_adverts();
5352 out4:
5353         if (ext4_proc_root)
5354                 remove_proc_entry("fs/ext4", NULL);
5355         kset_unregister(ext4_kset);
5356 out5:
5357         ext4_exit_system_zone();
5358 out6:
5359         ext4_exit_pageio();
5360 out7:
5361         ext4_exit_es();
5362
5363         return err;
5364 }
5365
5366 static void __exit ext4_exit_fs(void)
5367 {
5368         ext4_destroy_lazyinit_thread();
5369         unregister_as_ext2();
5370         unregister_as_ext3();
5371         unregister_filesystem(&ext4_fs_type);
5372         destroy_inodecache();
5373         ext4_exit_xattr();
5374         ext4_exit_mballoc();
5375         ext4_exit_feat_adverts();
5376         remove_proc_entry("fs/ext4", NULL);
5377         kset_unregister(ext4_kset);
5378         ext4_exit_system_zone();
5379         ext4_exit_pageio();
5380 }
5381
5382 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5383 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5384 MODULE_LICENSE("GPL");
5385 module_init(ext4_init_fs)
5386 module_exit(ext4_exit_fs)