fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static const char *ext4_decode_error(struct super_block *sb, int errno,
  73                                      char nbuf[16]);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  96 #else
  97 #define IS_EXT2_SB(sb) (0)
  98 #endif
  99
 100
 101 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 102 static struct file_system_type ext3_fs_type = {
 103         .owner          = THIS_MODULE,
 104         .name           = "ext3",
 105         .mount          = ext4_mount,
 106         .kill_sb        = kill_block_super,
 107         .fs_flags       = FS_REQUIRES_DEV,
 108 };
 109 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 110 #else
 111 #define IS_EXT3_SB(sb) (0)
 112 #endif
 113
 114 static int ext4_verify_csum_type(struct super_block *sb,
 115                                  struct ext4_super_block *es)
 116 {
 117         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 118                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 119                 return 1;
 120
 121         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 122 }
 123
 124 static __le32 ext4_superblock_csum(struct super_block *sb,
 125                                    struct ext4_super_block *es)
 126 {
 127         struct ext4_sb_info *sbi = EXT4_SB(sb);
 128         int offset = offsetof(struct ext4_super_block, s_checksum);
 129         __u32 csum;
 130
 131         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 132
 133         return cpu_to_le32(csum);
 134 }
 135
 136 int ext4_superblock_csum_verify(struct super_block *sb,
 137                                 struct ext4_super_block *es)
 138 {
 139         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 140                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 141                 return 1;
 142
 143         return es->s_checksum == ext4_superblock_csum(sb, es);
 144 }
 145
 146 void ext4_superblock_csum_set(struct super_block *sb)
 147 {
 148         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 149
 150         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 151                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 152                 return;
 153
 154         es->s_checksum = ext4_superblock_csum(sb, es);
 155 }
 156
 157 void *ext4_kvmalloc(size_t size, gfp_t flags)
 158 {
 159         void *ret;
 160
 161         ret = kmalloc(size, flags);
 162         if (!ret)
 163                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 164         return ret;
 165 }
 166
 167 void *ext4_kvzalloc(size_t size, gfp_t flags)
 168 {
 169         void *ret;
 170
 171         ret = kzalloc(size, flags);
 172         if (!ret)
 173                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 174         return ret;
 175 }
 176
 177 void ext4_kvfree(void *ptr)
 178 {
 179         if (is_vmalloc_addr(ptr))
 180                 vfree(ptr);
 181         else
 182                 kfree(ptr);
 183
 184 }
 185
 186 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 187                                struct ext4_group_desc *bg)
 188 {
 189         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 190                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 191                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 192 }
 193
 194 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 195                                struct ext4_group_desc *bg)
 196 {
 197         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 198                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 199                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 200 }
 201
 202 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 203                               struct ext4_group_desc *bg)
 204 {
 205         return le32_to_cpu(bg->bg_inode_table_lo) |
 206                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 207                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 208 }
 209
 210 __u32 ext4_free_group_clusters(struct super_block *sb,
 211                                struct ext4_group_desc *bg)
 212 {
 213         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 214                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 215                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 216 }
 217
 218 __u32 ext4_free_inodes_count(struct super_block *sb,
 219                               struct ext4_group_desc *bg)
 220 {
 221         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 222                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 223                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 224 }
 225
 226 __u32 ext4_used_dirs_count(struct super_block *sb,
 227                               struct ext4_group_desc *bg)
 228 {
 229         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 230                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 231                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 232 }
 233
 234 __u32 ext4_itable_unused_count(struct super_block *sb,
 235                               struct ext4_group_desc *bg)
 236 {
 237         return le16_to_cpu(bg->bg_itable_unused_lo) |
 238                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 239                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 240 }
 241
 242 void ext4_block_bitmap_set(struct super_block *sb,
 243                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 244 {
 245         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 246         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 247                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 248 }
 249
 250 void ext4_inode_bitmap_set(struct super_block *sb,
 251                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 252 {
 253         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 254         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 255                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 256 }
 257
 258 void ext4_inode_table_set(struct super_block *sb,
 259                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 260 {
 261         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 262         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 263                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 264 }
 265
 266 void ext4_free_group_clusters_set(struct super_block *sb,
 267                                   struct ext4_group_desc *bg, __u32 count)
 268 {
 269         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 270         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 271                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 272 }
 273
 274 void ext4_free_inodes_set(struct super_block *sb,
 275                           struct ext4_group_desc *bg, __u32 count)
 276 {
 277         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 278         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 279                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 280 }
 281
 282 void ext4_used_dirs_set(struct super_block *sb,
 283                           struct ext4_group_desc *bg, __u32 count)
 284 {
 285         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 286         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 287                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 288 }
 289
 290 void ext4_itable_unused_set(struct super_block *sb,
 291                           struct ext4_group_desc *bg, __u32 count)
 292 {
 293         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 294         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 295                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 296 }
 297
 298
 299 /* Just increment the non-pointer handle value */
 300 static handle_t *ext4_get_nojournal(void)
 301 {
 302         handle_t *handle = current->journal_info;
 303         unsigned long ref_cnt = (unsigned long)handle;
 304
 305         BUG_ON(ref_cnt >= EXT4_NOJOURNAL_MAX_REF_COUNT);
 306
 307         ref_cnt++;
 308         handle = (handle_t *)ref_cnt;
 309
 310         current->journal_info = handle;
 311         return handle;
 312 }
 313
 314
 315 /* Decrement the non-pointer handle value */
 316 static void ext4_put_nojournal(handle_t *handle)
 317 {
 318         unsigned long ref_cnt = (unsigned long)handle;
 319
 320         BUG_ON(ref_cnt == 0);
 321
 322         ref_cnt--;
 323         handle = (handle_t *)ref_cnt;
 324
 325         current->journal_info = handle;
 326 }
 327
 328 /*
 329  * Wrappers for jbd2_journal_start/end.
 330  */
 331 handle_t *ext4_journal_start_sb(struct super_block *sb, int nblocks)
 332 {
 333         journal_t *journal;
 334
 335         trace_ext4_journal_start(sb, nblocks, _RET_IP_);
 336         if (sb->s_flags & MS_RDONLY)
 337                 return ERR_PTR(-EROFS);
 338
 339         WARN_ON(sb->s_writers.frozen == SB_FREEZE_COMPLETE);
 340         journal = EXT4_SB(sb)->s_journal;
 341         if (!journal)
 342                 return ext4_get_nojournal();
 343         /*
 344          * Special case here: if the journal has aborted behind our
 345          * backs (eg. EIO in the commit thread), then we still need to
 346          * take the FS itself readonly cleanly.
 347          */
 348         if (is_journal_aborted(journal)) {
 349                 ext4_abort(sb, "Detected aborted journal");
 350                 return ERR_PTR(-EROFS);
 351         }
 352         return jbd2_journal_start(journal, nblocks);
 353 }
 354
 355 int __ext4_journal_stop(const char *where, unsigned int line, handle_t *handle)
 356 {
 357         struct super_block *sb;
 358         int err;
 359         int rc;
 360
 361         if (!ext4_handle_valid(handle)) {
 362                 ext4_put_nojournal(handle);
 363                 return 0;
 364         }
 365         sb = handle->h_transaction->t_journal->j_private;
 366         err = handle->h_err;
 367         rc = jbd2_journal_stop(handle);
 368
 369         if (!err)
 370                 err = rc;
 371         if (err)
 372                 __ext4_std_error(sb, where, line, err);
 373         return err;
 374 }
 375
 376 void ext4_journal_abort_handle(const char *caller, unsigned int line,
 377                                const char *err_fn, struct buffer_head *bh,
 378                                handle_t *handle, int err)
 379 {
 380         char nbuf[16];
 381         const char *errstr = ext4_decode_error(NULL, err, nbuf);
 382
 383         BUG_ON(!ext4_handle_valid(handle));
 384
 385         if (bh)
 386                 BUFFER_TRACE(bh, "abort");
 387
 388         if (!handle->h_err)
 389                 handle->h_err = err;
 390
 391         if (is_handle_aborted(handle))
 392                 return;
 393
 394         printk(KERN_ERR "EXT4-fs: %s:%d: aborting transaction: %s in %s\n",
 395                caller, line, errstr, err_fn);
 396
 397         jbd2_journal_abort_handle(handle);
 398 }
 399
 400 static void __save_error_info(struct super_block *sb, const char *func,
 401                             unsigned int line)
 402 {
 403         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 404
 405         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 406         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 407         es->s_last_error_time = cpu_to_le32(get_seconds());
 408         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 409         es->s_last_error_line = cpu_to_le32(line);
 410         if (!es->s_first_error_time) {
 411                 es->s_first_error_time = es->s_last_error_time;
 412                 strncpy(es->s_first_error_func, func,
 413                         sizeof(es->s_first_error_func));
 414                 es->s_first_error_line = cpu_to_le32(line);
 415                 es->s_first_error_ino = es->s_last_error_ino;
 416                 es->s_first_error_block = es->s_last_error_block;
 417         }
 418         /*
 419          * Start the daily error reporting function if it hasn't been
 420          * started already
 421          */
 422         if (!es->s_error_count)
 423                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 424         le32_add_cpu(&es->s_error_count, 1);
 425 }
 426
 427 static void save_error_info(struct super_block *sb, const char *func,
 428                             unsigned int line)
 429 {
 430         __save_error_info(sb, func, line);
 431         ext4_commit_super(sb, 1);
 432 }
 433
 434 /*
 435  * The del_gendisk() function uninitializes the disk-specific data
 436  * structures, including the bdi structure, without telling anyone
 437  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 438  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 439  * This is a kludge to prevent these oops until we can put in a proper
 440  * hook in del_gendisk() to inform the VFS and file system layers.
 441  */
 442 static int block_device_ejected(struct super_block *sb)
 443 {
 444         struct inode *bd_inode = sb->s_bdev->bd_inode;
 445         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 446
 447         return bdi->dev == NULL;
 448 }
 449
 450 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 451 {
 452         struct super_block              *sb = journal->j_private;
 453         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 454         int                             error = is_journal_aborted(journal);
 455         struct ext4_journal_cb_entry    *jce, *tmp;
 456
 457         spin_lock(&sbi->s_md_lock);
 458         list_for_each_entry_safe(jce, tmp, &txn->t_private_list, jce_list) {
 459                 list_del_init(&jce->jce_list);
 460                 spin_unlock(&sbi->s_md_lock);
 461                 jce->jce_func(sb, jce, error);
 462                 spin_lock(&sbi->s_md_lock);
 463         }
 464         spin_unlock(&sbi->s_md_lock);
 465 }
 466
 467 /* Deal with the reporting of failure conditions on a filesystem such as
 468  * inconsistencies detected or read IO failures.
 469  *
 470  * On ext2, we can store the error state of the filesystem in the
 471  * superblock.  That is not possible on ext4, because we may have other
 472  * write ordering constraints on the superblock which prevent us from
 473  * writing it out straight away; and given that the journal is about to
 474  * be aborted, we can't rely on the current, or future, transactions to
 475  * write out the superblock safely.
 476  *
 477  * We'll just use the jbd2_journal_abort() error code to record an error in
 478  * the journal instead.  On recovery, the journal will complain about
 479  * that error until we've noted it down and cleared it.
 480  */
 481
 482 static void ext4_handle_error(struct super_block *sb)
 483 {
 484         if (sb->s_flags & MS_RDONLY)
 485                 return;
 486
 487         if (!test_opt(sb, ERRORS_CONT)) {
 488                 journal_t *journal = EXT4_SB(sb)->s_journal;
 489
 490                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 491                 if (journal)
 492                         jbd2_journal_abort(journal, -EIO);
 493         }
 494         if (test_opt(sb, ERRORS_RO)) {
 495                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 496                 sb->s_flags |= MS_RDONLY;
 497         }
 498         if (test_opt(sb, ERRORS_PANIC))
 499                 panic("EXT4-fs (device %s): panic forced after error\n",
 500                         sb->s_id);
 501 }
 502
 503 void __ext4_error(struct super_block *sb, const char *function,
 504                   unsigned int line, const char *fmt, ...)
 505 {
 506         struct va_format vaf;
 507         va_list args;
 508
 509         va_start(args, fmt);
 510         vaf.fmt = fmt;
 511         vaf.va = &args;
 512         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 513                sb->s_id, function, line, current->comm, &vaf);
 514         va_end(args);
 515         save_error_info(sb, function, line);
 516
 517         ext4_handle_error(sb);
 518 }
 519
 520 void ext4_error_inode(struct inode *inode, const char *function,
 521                       unsigned int line, ext4_fsblk_t block,
 522                       const char *fmt, ...)
 523 {
 524         va_list args;
 525         struct va_format vaf;
 526         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 527
 528         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 529         es->s_last_error_block = cpu_to_le64(block);
 530         save_error_info(inode->i_sb, function, line);
 531         va_start(args, fmt);
 532         vaf.fmt = fmt;
 533         vaf.va = &args;
 534         if (block)
 535                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 536                        "inode #%lu: block %llu: comm %s: %pV\n",
 537                        inode->i_sb->s_id, function, line, inode->i_ino,
 538                        block, current->comm, &vaf);
 539         else
 540                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 541                        "inode #%lu: comm %s: %pV\n",
 542                        inode->i_sb->s_id, function, line, inode->i_ino,
 543                        current->comm, &vaf);
 544         va_end(args);
 545
 546         ext4_handle_error(inode->i_sb);
 547 }
 548
 549 void ext4_error_file(struct file *file, const char *function,
 550                      unsigned int line, ext4_fsblk_t block,
 551                      const char *fmt, ...)
 552 {
 553         va_list args;
 554         struct va_format vaf;
 555         struct ext4_super_block *es;
 556         struct inode *inode = file->f_dentry->d_inode;
 557         char pathname[80], *path;
 558
 559         es = EXT4_SB(inode->i_sb)->s_es;
 560         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 561         save_error_info(inode->i_sb, function, line);
 562         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 563         if (IS_ERR(path))
 564                 path = "(unknown)";
 565         va_start(args, fmt);
 566         vaf.fmt = fmt;
 567         vaf.va = &args;
 568         if (block)
 569                 printk(KERN_CRIT
 570                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 571                        "block %llu: comm %s: path %s: %pV\n",
 572                        inode->i_sb->s_id, function, line, inode->i_ino,
 573                        block, current->comm, path, &vaf);
 574         else
 575                 printk(KERN_CRIT
 576                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 577                        "comm %s: path %s: %pV\n",
 578                        inode->i_sb->s_id, function, line, inode->i_ino,
 579                        current->comm, path, &vaf);
 580         va_end(args);
 581
 582         ext4_handle_error(inode->i_sb);
 583 }
 584
 585 static const char *ext4_decode_error(struct super_block *sb, int errno,
 586                                      char nbuf[16])
 587 {
 588         char *errstr = NULL;
 589
 590         switch (errno) {
 591         case -EIO:
 592                 errstr = "IO failure";
 593                 break;
 594         case -ENOMEM:
 595                 errstr = "Out of memory";
 596                 break;
 597         case -EROFS:
 598                 if (!sb || (EXT4_SB(sb)->s_journal &&
 599                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 600                         errstr = "Journal has aborted";
 601                 else
 602                         errstr = "Readonly filesystem";
 603                 break;
 604         default:
 605                 /* If the caller passed in an extra buffer for unknown
 606                  * errors, textualise them now.  Else we just return
 607                  * NULL. */
 608                 if (nbuf) {
 609                         /* Check for truncated error codes... */
 610                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 611                                 errstr = nbuf;
 612                 }
 613                 break;
 614         }
 615
 616         return errstr;
 617 }
 618
 619 /* __ext4_std_error decodes expected errors from journaling functions
 620  * automatically and invokes the appropriate error response.  */
 621
 622 void __ext4_std_error(struct super_block *sb, const char *function,
 623                       unsigned int line, int errno)
 624 {
 625         char nbuf[16];
 626         const char *errstr;
 627
 628         /* Special case: if the error is EROFS, and we're not already
 629          * inside a transaction, then there's really no point in logging
 630          * an error. */
 631         if (errno == -EROFS && journal_current_handle() == NULL &&
 632             (sb->s_flags & MS_RDONLY))
 633                 return;
 634
 635         errstr = ext4_decode_error(sb, errno, nbuf);
 636         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 637                sb->s_id, function, line, errstr);
 638         save_error_info(sb, function, line);
 639
 640         ext4_handle_error(sb);
 641 }
 642
 643 /*
 644  * ext4_abort is a much stronger failure handler than ext4_error.  The
 645  * abort function may be used to deal with unrecoverable failures such
 646  * as journal IO errors or ENOMEM at a critical moment in log management.
 647  *
 648  * We unconditionally force the filesystem into an ABORT|READONLY state,
 649  * unless the error response on the fs has been set to panic in which
 650  * case we take the easy way out and panic immediately.
 651  */
 652
 653 void __ext4_abort(struct super_block *sb, const char *function,
 654                 unsigned int line, const char *fmt, ...)
 655 {
 656         va_list args;
 657
 658         save_error_info(sb, function, line);
 659         va_start(args, fmt);
 660         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 661                function, line);
 662         vprintk(fmt, args);
 663         printk("\n");
 664         va_end(args);
 665
 666         if ((sb->s_flags & MS_RDONLY) == 0) {
 667                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 668                 sb->s_flags |= MS_RDONLY;
 669                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 670                 if (EXT4_SB(sb)->s_journal)
 671                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 672                 save_error_info(sb, function, line);
 673         }
 674         if (test_opt(sb, ERRORS_PANIC))
 675                 panic("EXT4-fs panic from previous error\n");
 676 }
 677
 678 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 679 {
 680         struct va_format vaf;
 681         va_list args;
 682
 683         va_start(args, fmt);
 684         vaf.fmt = fmt;
 685         vaf.va = &args;
 686         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 687         va_end(args);
 688 }
 689
 690 void __ext4_warning(struct super_block *sb, const char *function,
 691                     unsigned int line, const char *fmt, ...)
 692 {
 693         struct va_format vaf;
 694         va_list args;
 695
 696         va_start(args, fmt);
 697         vaf.fmt = fmt;
 698         vaf.va = &args;
 699         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 700                sb->s_id, function, line, &vaf);
 701         va_end(args);
 702 }
 703
 704 void __ext4_grp_locked_error(const char *function, unsigned int line,
 705                              struct super_block *sb, ext4_group_t grp,
 706                              unsigned long ino, ext4_fsblk_t block,
 707                              const char *fmt, ...)
 708 __releases(bitlock)
 709 __acquires(bitlock)
 710 {
 711         struct va_format vaf;
 712         va_list args;
 713         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 714
 715         es->s_last_error_ino = cpu_to_le32(ino);
 716         es->s_last_error_block = cpu_to_le64(block);
 717         __save_error_info(sb, function, line);
 718
 719         va_start(args, fmt);
 720
 721         vaf.fmt = fmt;
 722         vaf.va = &args;
 723         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 724                sb->s_id, function, line, grp);
 725         if (ino)
 726                 printk(KERN_CONT "inode %lu: ", ino);
 727         if (block)
 728                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 729         printk(KERN_CONT "%pV\n", &vaf);
 730         va_end(args);
 731
 732         if (test_opt(sb, ERRORS_CONT)) {
 733                 ext4_commit_super(sb, 0);
 734                 return;
 735         }
 736
 737         ext4_unlock_group(sb, grp);
 738         ext4_handle_error(sb);
 739         /*
 740          * We only get here in the ERRORS_RO case; relocking the group
 741          * may be dangerous, but nothing bad will happen since the
 742          * filesystem will have already been marked read/only and the
 743          * journal has been aborted.  We return 1 as a hint to callers
 744          * who might what to use the return value from
 745          * ext4_grp_locked_error() to distinguish between the
 746          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 747          * aggressively from the ext4 function in question, with a
 748          * more appropriate error code.
 749          */
 750         ext4_lock_group(sb, grp);
 751         return;
 752 }
 753
 754 void ext4_update_dynamic_rev(struct super_block *sb)
 755 {
 756         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 757
 758         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 759                 return;
 760
 761         ext4_warning(sb,
 762                      "updating to rev %d because of new feature flag, "
 763                      "running e2fsck is recommended",
 764                      EXT4_DYNAMIC_REV);
 765
 766         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 767         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 768         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 769         /* leave es->s_feature_*compat flags alone */
 770         /* es->s_uuid will be set by e2fsck if empty */
 771
 772         /*
 773          * The rest of the superblock fields should be zero, and if not it
 774          * means they are likely already in use, so leave them alone.  We
 775          * can leave it up to e2fsck to clean up any inconsistencies there.
 776          */
 777 }
 778
 779 /*
 780  * Open the external journal device
 781  */
 782 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 783 {
 784         struct block_device *bdev;
 785         char b[BDEVNAME_SIZE];
 786
 787         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 788         if (IS_ERR(bdev))
 789                 goto fail;
 790         return bdev;
 791
 792 fail:
 793         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 794                         __bdevname(dev, b), PTR_ERR(bdev));
 795         return NULL;
 796 }
 797
 798 /*
 799  * Release the journal device
 800  */
 801 static int ext4_blkdev_put(struct block_device *bdev)
 802 {
 803         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 804 }
 805
 806 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 807 {
 808         struct block_device *bdev;
 809         int ret = -ENODEV;
 810
 811         bdev = sbi->journal_bdev;
 812         if (bdev) {
 813                 ret = ext4_blkdev_put(bdev);
 814                 sbi->journal_bdev = NULL;
 815         }
 816         return ret;
 817 }
 818
 819 static inline struct inode *orphan_list_entry(struct list_head *l)
 820 {
 821         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 822 }
 823
 824 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 825 {
 826         struct list_head *l;
 827
 828         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 829                  le32_to_cpu(sbi->s_es->s_last_orphan));
 830
 831         printk(KERN_ERR "sb_info orphan list:\n");
 832         list_for_each(l, &sbi->s_orphan) {
 833                 struct inode *inode = orphan_list_entry(l);
 834                 printk(KERN_ERR "  "
 835                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 836                        inode->i_sb->s_id, inode->i_ino, inode,
 837                        inode->i_mode, inode->i_nlink,
 838                        NEXT_ORPHAN(inode));
 839         }
 840 }
 841
 842 static void ext4_put_super(struct super_block *sb)
 843 {
 844         struct ext4_sb_info *sbi = EXT4_SB(sb);
 845         struct ext4_super_block *es = sbi->s_es;
 846         int i, err;
 847
 848         ext4_unregister_li_request(sb);
 849         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 850
 851         flush_workqueue(sbi->dio_unwritten_wq);
 852         destroy_workqueue(sbi->dio_unwritten_wq);
 853
 854         if (sbi->s_journal) {
 855                 err = jbd2_journal_destroy(sbi->s_journal);
 856                 sbi->s_journal = NULL;
 857                 if (err < 0)
 858                         ext4_abort(sb, "Couldn't clean up the journal");
 859         }
 860
 861         del_timer(&sbi->s_err_report);
 862         ext4_release_system_zone(sb);
 863         ext4_mb_release(sb);
 864         ext4_ext_release(sb);
 865         ext4_xattr_put_super(sb);
 866
 867         if (!(sb->s_flags & MS_RDONLY)) {
 868                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 869                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 870         }
 871         if (!(sb->s_flags & MS_RDONLY))
 872                 ext4_commit_super(sb, 1);
 873
 874         if (sbi->s_proc) {
 875                 remove_proc_entry("options", sbi->s_proc);
 876                 remove_proc_entry(sb->s_id, ext4_proc_root);
 877         }
 878         kobject_del(&sbi->s_kobj);
 879
 880         for (i = 0; i < sbi->s_gdb_count; i++)
 881                 brelse(sbi->s_group_desc[i]);
 882         ext4_kvfree(sbi->s_group_desc);
 883         ext4_kvfree(sbi->s_flex_groups);
 884         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 885         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 886         percpu_counter_destroy(&sbi->s_dirs_counter);
 887         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 888         brelse(sbi->s_sbh);
 889 #ifdef CONFIG_QUOTA
 890         for (i = 0; i < MAXQUOTAS; i++)
 891                 kfree(sbi->s_qf_names[i]);
 892 #endif
 893
 894         /* Debugging code just in case the in-memory inode orphan list
 895          * isn't empty.  The on-disk one can be non-empty if we've
 896          * detected an error and taken the fs readonly, but the
 897          * in-memory list had better be clean by this point. */
 898         if (!list_empty(&sbi->s_orphan))
 899                 dump_orphan_list(sb, sbi);
 900         J_ASSERT(list_empty(&sbi->s_orphan));
 901
 902         invalidate_bdev(sb->s_bdev);
 903         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 904                 /*
 905                  * Invalidate the journal device's buffers.  We don't want them
 906                  * floating about in memory - the physical journal device may
 907                  * hotswapped, and it breaks the `ro-after' testing code.
 908                  */
 909                 sync_blockdev(sbi->journal_bdev);
 910                 invalidate_bdev(sbi->journal_bdev);
 911                 ext4_blkdev_remove(sbi);
 912         }
 913         if (sbi->s_mmp_tsk)
 914                 kthread_stop(sbi->s_mmp_tsk);
 915         sb->s_fs_info = NULL;
 916         /*
 917          * Now that we are completely done shutting down the
 918          * superblock, we need to actually destroy the kobject.
 919          */
 920         kobject_put(&sbi->s_kobj);
 921         wait_for_completion(&sbi->s_kobj_unregister);
 922         if (sbi->s_chksum_driver)
 923                 crypto_free_shash(sbi->s_chksum_driver);
 924         kfree(sbi->s_blockgroup_lock);
 925         kfree(sbi);
 926 }
 927
 928 static struct kmem_cache *ext4_inode_cachep;
 929
 930 /*
 931  * Called inside transaction, so use GFP_NOFS
 932  */
 933 static struct inode *ext4_alloc_inode(struct super_block *sb)
 934 {
 935         struct ext4_inode_info *ei;
 936
 937         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 938         if (!ei)
 939                 return NULL;
 940
 941         ei->vfs_inode.i_version = 1;
 942         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 943         INIT_LIST_HEAD(&ei->i_prealloc_list);
 944         spin_lock_init(&ei->i_prealloc_lock);
 945         ext4_es_init_tree(&ei->i_es_tree);
 946         rwlock_init(&ei->i_es_lock);
 947         ei->i_reserved_data_blocks = 0;
 948         ei->i_reserved_meta_blocks = 0;
 949         ei->i_allocated_meta_blocks = 0;
 950         ei->i_da_metadata_calc_len = 0;
 951         ei->i_da_metadata_calc_last_lblock = 0;
 952         spin_lock_init(&(ei->i_block_reservation_lock));
 953 #ifdef CONFIG_QUOTA
 954         ei->i_reserved_quota = 0;
 955 #endif
 956         ei->jinode = NULL;
 957         INIT_LIST_HEAD(&ei->i_completed_io_list);
 958         spin_lock_init(&ei->i_completed_io_lock);
 959         ei->i_sync_tid = 0;
 960         ei->i_datasync_tid = 0;
 961         atomic_set(&ei->i_ioend_count, 0);
 962         atomic_set(&ei->i_unwritten, 0);
 963         INIT_WORK(&ei->i_unwritten_work, ext4_end_io_work);
 964
 965         return &ei->vfs_inode;
 966 }
 967
 968 static int ext4_drop_inode(struct inode *inode)
 969 {
 970         int drop = generic_drop_inode(inode);
 971
 972         trace_ext4_drop_inode(inode, drop);
 973         return drop;
 974 }
 975
 976 static void ext4_i_callback(struct rcu_head *head)
 977 {
 978         struct inode *inode = container_of(head, struct inode, i_rcu);
 979         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 980 }
 981
 982 static void ext4_destroy_inode(struct inode *inode)
 983 {
 984         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 985                 ext4_msg(inode->i_sb, KERN_ERR,
 986                          "Inode %lu (%p): orphan list check failed!",
 987                          inode->i_ino, EXT4_I(inode));
 988                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 989                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 990                                 true);
 991                 dump_stack();
 992         }
 993         call_rcu(&inode->i_rcu, ext4_i_callback);
 994 }
 995
 996 static void init_once(void *foo)
 997 {
 998         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 999
1000         INIT_LIST_HEAD(&ei->i_orphan);
1001         init_rwsem(&ei->xattr_sem);
1002         init_rwsem(&ei->i_data_sem);
1003         inode_init_once(&ei->vfs_inode);
1004 }
1005
1006 static int init_inodecache(void)
1007 {
1008         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
1009                                              sizeof(struct ext4_inode_info),
1010                                              0, (SLAB_RECLAIM_ACCOUNT|
1011                                                 SLAB_MEM_SPREAD),
1012                                              init_once);
1013         if (ext4_inode_cachep == NULL)
1014                 return -ENOMEM;
1015         return 0;
1016 }
1017
1018 static void destroy_inodecache(void)
1019 {
1020         /*
1021          * Make sure all delayed rcu free inodes are flushed before we
1022          * destroy cache.
1023          */
1024         rcu_barrier();
1025         kmem_cache_destroy(ext4_inode_cachep);
1026 }
1027
1028 void ext4_clear_inode(struct inode *inode)
1029 {
1030         invalidate_inode_buffers(inode);
1031         clear_inode(inode);
1032         dquot_drop(inode);
1033         ext4_discard_preallocations(inode);
1034         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
1035         if (EXT4_I(inode)->jinode) {
1036                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1037                                                EXT4_I(inode)->jinode);
1038                 jbd2_free_inode(EXT4_I(inode)->jinode);
1039                 EXT4_I(inode)->jinode = NULL;
1040         }
1041 }
1042
1043 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1044                                         u64 ino, u32 generation)
1045 {
1046         struct inode *inode;
1047
1048         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1049                 return ERR_PTR(-ESTALE);
1050         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1051                 return ERR_PTR(-ESTALE);
1052
1053         /* iget isn't really right if the inode is currently unallocated!!
1054          *
1055          * ext4_read_inode will return a bad_inode if the inode had been
1056          * deleted, so we should be safe.
1057          *
1058          * Currently we don't know the generation for parent directory, so
1059          * a generation of 0 means "accept any"
1060          */
1061         inode = ext4_iget(sb, ino);
1062         if (IS_ERR(inode))
1063                 return ERR_CAST(inode);
1064         if (generation && inode->i_generation != generation) {
1065                 iput(inode);
1066                 return ERR_PTR(-ESTALE);
1067         }
1068
1069         return inode;
1070 }
1071
1072 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1073                                         int fh_len, int fh_type)
1074 {
1075         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1076                                     ext4_nfs_get_inode);
1077 }
1078
1079 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1080                                         int fh_len, int fh_type)
1081 {
1082         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1083                                     ext4_nfs_get_inode);
1084 }
1085
1086 /*
1087  * Try to release metadata pages (indirect blocks, directories) which are
1088  * mapped via the block device.  Since these pages could have journal heads
1089  * which would prevent try_to_free_buffers() from freeing them, we must use
1090  * jbd2 layer's try_to_free_buffers() function to release them.
1091  */
1092 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1093                                  gfp_t wait)
1094 {
1095         journal_t *journal = EXT4_SB(sb)->s_journal;
1096
1097         WARN_ON(PageChecked(page));
1098         if (!page_has_buffers(page))
1099                 return 0;
1100         if (journal)
1101                 return jbd2_journal_try_to_free_buffers(journal, page,
1102                                                         wait & ~__GFP_WAIT);
1103         return try_to_free_buffers(page);
1104 }
1105
1106 #ifdef CONFIG_QUOTA
1107 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1108 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1109
1110 static int ext4_write_dquot(struct dquot *dquot);
1111 static int ext4_acquire_dquot(struct dquot *dquot);
1112 static int ext4_release_dquot(struct dquot *dquot);
1113 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1114 static int ext4_write_info(struct super_block *sb, int type);
1115 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1116                          struct path *path);
1117 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1118                                  int format_id);
1119 static int ext4_quota_off(struct super_block *sb, int type);
1120 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1121 static int ext4_quota_on_mount(struct super_block *sb, int type);
1122 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1123                                size_t len, loff_t off);
1124 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1125                                 const char *data, size_t len, loff_t off);
1126 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1127                              unsigned int flags);
1128 static int ext4_enable_quotas(struct super_block *sb);
1129
1130 static const struct dquot_operations ext4_quota_operations = {
1131         .get_reserved_space = ext4_get_reserved_space,
1132         .write_dquot    = ext4_write_dquot,
1133         .acquire_dquot  = ext4_acquire_dquot,
1134         .release_dquot  = ext4_release_dquot,
1135         .mark_dirty     = ext4_mark_dquot_dirty,
1136         .write_info     = ext4_write_info,
1137         .alloc_dquot    = dquot_alloc,
1138         .destroy_dquot  = dquot_destroy,
1139 };
1140
1141 static const struct quotactl_ops ext4_qctl_operations = {
1142         .quota_on       = ext4_quota_on,
1143         .quota_off      = ext4_quota_off,
1144         .quota_sync     = dquot_quota_sync,
1145         .get_info       = dquot_get_dqinfo,
1146         .set_info       = dquot_set_dqinfo,
1147         .get_dqblk      = dquot_get_dqblk,
1148         .set_dqblk      = dquot_set_dqblk
1149 };
1150
1151 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1152         .quota_on_meta  = ext4_quota_on_sysfile,
1153         .quota_off      = ext4_quota_off_sysfile,
1154         .quota_sync     = dquot_quota_sync,
1155         .get_info       = dquot_get_dqinfo,
1156         .set_info       = dquot_set_dqinfo,
1157         .get_dqblk      = dquot_get_dqblk,
1158         .set_dqblk      = dquot_set_dqblk
1159 };
1160 #endif
1161
1162 static const struct super_operations ext4_sops = {
1163         .alloc_inode    = ext4_alloc_inode,
1164         .destroy_inode  = ext4_destroy_inode,
1165         .write_inode    = ext4_write_inode,
1166         .dirty_inode    = ext4_dirty_inode,
1167         .drop_inode     = ext4_drop_inode,
1168         .evict_inode    = ext4_evict_inode,
1169         .put_super      = ext4_put_super,
1170         .sync_fs        = ext4_sync_fs,
1171         .freeze_fs      = ext4_freeze,
1172         .unfreeze_fs    = ext4_unfreeze,
1173         .statfs         = ext4_statfs,
1174         .remount_fs     = ext4_remount,
1175         .show_options   = ext4_show_options,
1176 #ifdef CONFIG_QUOTA
1177         .quota_read     = ext4_quota_read,
1178         .quota_write    = ext4_quota_write,
1179 #endif
1180         .bdev_try_to_free_page = bdev_try_to_free_page,
1181 };
1182
1183 static const struct super_operations ext4_nojournal_sops = {
1184         .alloc_inode    = ext4_alloc_inode,
1185         .destroy_inode  = ext4_destroy_inode,
1186         .write_inode    = ext4_write_inode,
1187         .dirty_inode    = ext4_dirty_inode,
1188         .drop_inode     = ext4_drop_inode,
1189         .evict_inode    = ext4_evict_inode,
1190         .put_super      = ext4_put_super,
1191         .statfs         = ext4_statfs,
1192         .remount_fs     = ext4_remount,
1193         .show_options   = ext4_show_options,
1194 #ifdef CONFIG_QUOTA
1195         .quota_read     = ext4_quota_read,
1196         .quota_write    = ext4_quota_write,
1197 #endif
1198         .bdev_try_to_free_page = bdev_try_to_free_page,
1199 };
1200
1201 static const struct export_operations ext4_export_ops = {
1202         .fh_to_dentry = ext4_fh_to_dentry,
1203         .fh_to_parent = ext4_fh_to_parent,
1204         .get_parent = ext4_get_parent,
1205 };
1206
1207 enum {
1208         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1209         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1210         Opt_nouid32, Opt_debug, Opt_removed,
1211         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1212         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1213         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1214         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1215         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1216         Opt_data_err_abort, Opt_data_err_ignore,
1217         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1218         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1219         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1220         Opt_usrquota, Opt_grpquota, Opt_i_version,
1221         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1222         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1223         Opt_inode_readahead_blks, Opt_journal_ioprio,
1224         Opt_dioread_nolock, Opt_dioread_lock,
1225         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1226         Opt_max_dir_size_kb,
1227 };
1228
1229 static const match_table_t tokens = {
1230         {Opt_bsd_df, "bsddf"},
1231         {Opt_minix_df, "minixdf"},
1232         {Opt_grpid, "grpid"},
1233         {Opt_grpid, "bsdgroups"},
1234         {Opt_nogrpid, "nogrpid"},
1235         {Opt_nogrpid, "sysvgroups"},
1236         {Opt_resgid, "resgid=%u"},
1237         {Opt_resuid, "resuid=%u"},
1238         {Opt_sb, "sb=%u"},
1239         {Opt_err_cont, "errors=continue"},
1240         {Opt_err_panic, "errors=panic"},
1241         {Opt_err_ro, "errors=remount-ro"},
1242         {Opt_nouid32, "nouid32"},
1243         {Opt_debug, "debug"},
1244         {Opt_removed, "oldalloc"},
1245         {Opt_removed, "orlov"},
1246         {Opt_user_xattr, "user_xattr"},
1247         {Opt_nouser_xattr, "nouser_xattr"},
1248         {Opt_acl, "acl"},
1249         {Opt_noacl, "noacl"},
1250         {Opt_noload, "norecovery"},
1251         {Opt_noload, "noload"},
1252         {Opt_removed, "nobh"},
1253         {Opt_removed, "bh"},
1254         {Opt_commit, "commit=%u"},
1255         {Opt_min_batch_time, "min_batch_time=%u"},
1256         {Opt_max_batch_time, "max_batch_time=%u"},
1257         {Opt_journal_dev, "journal_dev=%u"},
1258         {Opt_journal_checksum, "journal_checksum"},
1259         {Opt_journal_async_commit, "journal_async_commit"},
1260         {Opt_abort, "abort"},
1261         {Opt_data_journal, "data=journal"},
1262         {Opt_data_ordered, "data=ordered"},
1263         {Opt_data_writeback, "data=writeback"},
1264         {Opt_data_err_abort, "data_err=abort"},
1265         {Opt_data_err_ignore, "data_err=ignore"},
1266         {Opt_offusrjquota, "usrjquota="},
1267         {Opt_usrjquota, "usrjquota=%s"},
1268         {Opt_offgrpjquota, "grpjquota="},
1269         {Opt_grpjquota, "grpjquota=%s"},
1270         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1271         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1272         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1273         {Opt_grpquota, "grpquota"},
1274         {Opt_noquota, "noquota"},
1275         {Opt_quota, "quota"},
1276         {Opt_usrquota, "usrquota"},
1277         {Opt_barrier, "barrier=%u"},
1278         {Opt_barrier, "barrier"},
1279         {Opt_nobarrier, "nobarrier"},
1280         {Opt_i_version, "i_version"},
1281         {Opt_stripe, "stripe=%u"},
1282         {Opt_delalloc, "delalloc"},
1283         {Opt_nodelalloc, "nodelalloc"},
1284         {Opt_removed, "mblk_io_submit"},
1285         {Opt_removed, "nomblk_io_submit"},
1286         {Opt_block_validity, "block_validity"},
1287         {Opt_noblock_validity, "noblock_validity"},
1288         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1289         {Opt_journal_ioprio, "journal_ioprio=%u"},
1290         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1291         {Opt_auto_da_alloc, "auto_da_alloc"},
1292         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1293         {Opt_dioread_nolock, "dioread_nolock"},
1294         {Opt_dioread_lock, "dioread_lock"},
1295         {Opt_discard, "discard"},
1296         {Opt_nodiscard, "nodiscard"},
1297         {Opt_init_itable, "init_itable=%u"},
1298         {Opt_init_itable, "init_itable"},
1299         {Opt_noinit_itable, "noinit_itable"},
1300         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1301         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1302         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1303         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1304         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1305         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1306         {Opt_err, NULL},
1307 };
1308
1309 static ext4_fsblk_t get_sb_block(void **data)
1310 {
1311         ext4_fsblk_t    sb_block;
1312         char            *options = (char *) *data;
1313
1314         if (!options || strncmp(options, "sb=", 3) != 0)
1315                 return 1;       /* Default location */
1316
1317         options += 3;
1318         /* TODO: use simple_strtoll with >32bit ext4 */
1319         sb_block = simple_strtoul(options, &options, 0);
1320         if (*options && *options != ',') {
1321                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1322                        (char *) *data);
1323                 return 1;
1324         }
1325         if (*options == ',')
1326                 options++;
1327         *data = (void *) options;
1328
1329         return sb_block;
1330 }
1331
1332 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1333 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1334         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1335
1336 #ifdef CONFIG_QUOTA
1337 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1338 {
1339         struct ext4_sb_info *sbi = EXT4_SB(sb);
1340         char *qname;
1341         int ret = -1;
1342
1343         if (sb_any_quota_loaded(sb) &&
1344                 !sbi->s_qf_names[qtype]) {
1345                 ext4_msg(sb, KERN_ERR,
1346                         "Cannot change journaled "
1347                         "quota options when quota turned on");
1348                 return -1;
1349         }
1350         qname = match_strdup(args);
1351         if (!qname) {
1352                 ext4_msg(sb, KERN_ERR,
1353                         "Not enough memory for storing quotafile name");
1354                 return -1;
1355         }
1356         if (sbi->s_qf_names[qtype]) {
1357                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1358                         ret = 1;
1359                 else
1360                         ext4_msg(sb, KERN_ERR,
1361                                  "%s quota file already specified",
1362                                  QTYPE2NAME(qtype));
1363                 goto errout;
1364         }
1365         if (strchr(qname, '/')) {
1366                 ext4_msg(sb, KERN_ERR,
1367                         "quotafile must be on filesystem root");
1368                 goto errout;
1369         }
1370         sbi->s_qf_names[qtype] = qname;
1371         set_opt(sb, QUOTA);
1372         return 1;
1373 errout:
1374         kfree(qname);
1375         return ret;
1376 }
1377
1378 static int clear_qf_name(struct super_block *sb, int qtype)
1379 {
1380
1381         struct ext4_sb_info *sbi = EXT4_SB(sb);
1382
1383         if (sb_any_quota_loaded(sb) &&
1384                 sbi->s_qf_names[qtype]) {
1385                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1386                         " when quota turned on");
1387                 return -1;
1388         }
1389         kfree(sbi->s_qf_names[qtype]);
1390         sbi->s_qf_names[qtype] = NULL;
1391         return 1;
1392 }
1393 #endif
1394
1395 #define MOPT_SET        0x0001
1396 #define MOPT_CLEAR      0x0002
1397 #define MOPT_NOSUPPORT  0x0004
1398 #define MOPT_EXPLICIT   0x0008
1399 #define MOPT_CLEAR_ERR  0x0010
1400 #define MOPT_GTE0       0x0020
1401 #ifdef CONFIG_QUOTA
1402 #define MOPT_Q          0
1403 #define MOPT_QFMT       0x0040
1404 #else
1405 #define MOPT_Q          MOPT_NOSUPPORT
1406 #define MOPT_QFMT       MOPT_NOSUPPORT
1407 #endif
1408 #define MOPT_DATAJ      0x0080
1409
1410 static const struct mount_opts {
1411         int     token;
1412         int     mount_opt;
1413         int     flags;
1414 } ext4_mount_opts[] = {
1415         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1416         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1417         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1418         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1419         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1420         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1421         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1422         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1423         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1424         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1425         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1426         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1427         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1428         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1429                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1430         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1431         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1432         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1433         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1434         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1435         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1436         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1437         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1438         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1439         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1440         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1441         {Opt_commit, 0, MOPT_GTE0},
1442         {Opt_max_batch_time, 0, MOPT_GTE0},
1443         {Opt_min_batch_time, 0, MOPT_GTE0},
1444         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1445         {Opt_init_itable, 0, MOPT_GTE0},
1446         {Opt_stripe, 0, MOPT_GTE0},
1447         {Opt_resuid, 0, MOPT_GTE0},
1448         {Opt_resgid, 0, MOPT_GTE0},
1449         {Opt_journal_dev, 0, MOPT_GTE0},
1450         {Opt_journal_ioprio, 0, MOPT_GTE0},
1451         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1452         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1453         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1454         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1455         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1456 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1457         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1458         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1459 #else
1460         {Opt_acl, 0, MOPT_NOSUPPORT},
1461         {Opt_noacl, 0, MOPT_NOSUPPORT},
1462 #endif
1463         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1464         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1465         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1466         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1467                                                         MOPT_SET | MOPT_Q},
1468         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1469                                                         MOPT_SET | MOPT_Q},
1470         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1471                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1472         {Opt_usrjquota, 0, MOPT_Q},
1473         {Opt_grpjquota, 0, MOPT_Q},
1474         {Opt_offusrjquota, 0, MOPT_Q},
1475         {Opt_offgrpjquota, 0, MOPT_Q},
1476         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1477         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1478         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1479         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1480         {Opt_err, 0, 0}
1481 };
1482
1483 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1484                             substring_t *args, unsigned long *journal_devnum,
1485                             unsigned int *journal_ioprio, int is_remount)
1486 {
1487         struct ext4_sb_info *sbi = EXT4_SB(sb);
1488         const struct mount_opts *m;
1489         kuid_t uid;
1490         kgid_t gid;
1491         int arg = 0;
1492
1493 #ifdef CONFIG_QUOTA
1494         if (token == Opt_usrjquota)
1495                 return set_qf_name(sb, USRQUOTA, &args[0]);
1496         else if (token == Opt_grpjquota)
1497                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1498         else if (token == Opt_offusrjquota)
1499                 return clear_qf_name(sb, USRQUOTA);
1500         else if (token == Opt_offgrpjquota)
1501                 return clear_qf_name(sb, GRPQUOTA);
1502 #endif
1503         switch (token) {
1504         case Opt_noacl:
1505         case Opt_nouser_xattr:
1506                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1507                 break;
1508         case Opt_sb:
1509                 return 1;       /* handled by get_sb_block() */
1510         case Opt_removed:
1511                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1512                 return 1;
1513         case Opt_abort:
1514                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1515                 return 1;
1516         case Opt_i_version:
1517                 sb->s_flags |= MS_I_VERSION;
1518                 return 1;
1519         }
1520
1521         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1522                 if (token == m->token)
1523                         break;
1524
1525         if (m->token == Opt_err) {
1526                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1527                          "or missing value", opt);
1528                 return -1;
1529         }
1530
1531         if (args->from && match_int(args, &arg))
1532                 return -1;
1533         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1534                 return -1;
1535         if (m->flags & MOPT_EXPLICIT)
1536                 set_opt2(sb, EXPLICIT_DELALLOC);
1537         if (m->flags & MOPT_CLEAR_ERR)
1538                 clear_opt(sb, ERRORS_MASK);
1539         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1540                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1541                          "options when quota turned on");
1542                 return -1;
1543         }
1544
1545         if (m->flags & MOPT_NOSUPPORT) {
1546                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1547         } else if (token == Opt_commit) {
1548                 if (arg == 0)
1549                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1550                 sbi->s_commit_interval = HZ * arg;
1551         } else if (token == Opt_max_batch_time) {
1552                 if (arg == 0)
1553                         arg = EXT4_DEF_MAX_BATCH_TIME;
1554                 sbi->s_max_batch_time = arg;
1555         } else if (token == Opt_min_batch_time) {
1556                 sbi->s_min_batch_time = arg;
1557         } else if (token == Opt_inode_readahead_blks) {
1558                 if (arg > (1 << 30))
1559                         return -1;
1560                 if (arg && !is_power_of_2(arg)) {
1561                         ext4_msg(sb, KERN_ERR, "EXT4-fs: inode_readahead_blks"
1562                                  " must be a power of 2");
1563                         return -1;
1564                 }
1565                 sbi->s_inode_readahead_blks = arg;
1566         } else if (token == Opt_init_itable) {
1567                 set_opt(sb, INIT_INODE_TABLE);
1568                 if (!args->from)
1569                         arg = EXT4_DEF_LI_WAIT_MULT;
1570                 sbi->s_li_wait_mult = arg;
1571         } else if (token == Opt_max_dir_size_kb) {
1572                 sbi->s_max_dir_size_kb = arg;
1573         } else if (token == Opt_stripe) {
1574                 sbi->s_stripe = arg;
1575         } else if (token == Opt_resuid) {
1576                 uid = make_kuid(current_user_ns(), arg);
1577                 if (!uid_valid(uid)) {
1578                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1579                         return -1;
1580                 }
1581                 sbi->s_resuid = uid;
1582         } else if (token == Opt_resgid) {
1583                 gid = make_kgid(current_user_ns(), arg);
1584                 if (!gid_valid(gid)) {
1585                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1586                         return -1;
1587                 }
1588                 sbi->s_resgid = gid;
1589         } else if (token == Opt_journal_dev) {
1590                 if (is_remount) {
1591                         ext4_msg(sb, KERN_ERR,
1592                                  "Cannot specify journal on remount");
1593                         return -1;
1594                 }
1595                 *journal_devnum = arg;
1596         } else if (token == Opt_journal_ioprio) {
1597                 if (arg > 7) {
1598                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1599                                  " (must be 0-7)");
1600                         return -1;
1601                 }
1602                 *journal_ioprio =
1603                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1604         } else if (m->flags & MOPT_DATAJ) {
1605                 if (is_remount) {
1606                         if (!sbi->s_journal)
1607                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1608                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1609                                 ext4_msg(sb, KERN_ERR,
1610                                          "Cannot change data mode on remount");
1611                                 return -1;
1612                         }
1613                 } else {
1614                         clear_opt(sb, DATA_FLAGS);
1615                         sbi->s_mount_opt |= m->mount_opt;
1616                 }
1617 #ifdef CONFIG_QUOTA
1618         } else if (m->flags & MOPT_QFMT) {
1619                 if (sb_any_quota_loaded(sb) &&
1620                     sbi->s_jquota_fmt != m->mount_opt) {
1621                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1622                                  "quota options when quota turned on");
1623                         return -1;
1624                 }
1625                 sbi->s_jquota_fmt = m->mount_opt;
1626 #endif
1627         } else {
1628                 if (!args->from)
1629                         arg = 1;
1630                 if (m->flags & MOPT_CLEAR)
1631                         arg = !arg;
1632                 else if (unlikely(!(m->flags & MOPT_SET))) {
1633                         ext4_msg(sb, KERN_WARNING,
1634                                  "buggy handling of option %s", opt);
1635                         WARN_ON(1);
1636                         return -1;
1637                 }
1638                 if (arg != 0)
1639                         sbi->s_mount_opt |= m->mount_opt;
1640                 else
1641                         sbi->s_mount_opt &= ~m->mount_opt;
1642         }
1643         return 1;
1644 }
1645
1646 static int parse_options(char *options, struct super_block *sb,
1647                          unsigned long *journal_devnum,
1648                          unsigned int *journal_ioprio,
1649                          int is_remount)
1650 {
1651         struct ext4_sb_info *sbi = EXT4_SB(sb);
1652         char *p;
1653         substring_t args[MAX_OPT_ARGS];
1654         int token;
1655
1656         if (!options)
1657                 return 1;
1658
1659         while ((p = strsep(&options, ",")) != NULL) {
1660                 if (!*p)
1661                         continue;
1662                 /*
1663                  * Initialize args struct so we know whether arg was
1664                  * found; some options take optional arguments.
1665                  */
1666                 args[0].to = args[0].from = NULL;
1667                 token = match_token(p, tokens, args);
1668                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1669                                      journal_ioprio, is_remount) < 0)
1670                         return 0;
1671         }
1672 #ifdef CONFIG_QUOTA
1673         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1674                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1675                         clear_opt(sb, USRQUOTA);
1676
1677                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1678                         clear_opt(sb, GRPQUOTA);
1679
1680                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1681                         ext4_msg(sb, KERN_ERR, "old and new quota "
1682                                         "format mixing");
1683                         return 0;
1684                 }
1685
1686                 if (!sbi->s_jquota_fmt) {
1687                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1688                                         "not specified");
1689                         return 0;
1690                 }
1691         } else {
1692                 if (sbi->s_jquota_fmt) {
1693                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1694                                         "specified with no journaling "
1695                                         "enabled");
1696                         return 0;
1697                 }
1698         }
1699 #endif
1700         if (test_opt(sb, DIOREAD_NOLOCK)) {
1701                 int blocksize =
1702                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1703
1704                 if (blocksize < PAGE_CACHE_SIZE) {
1705                         ext4_msg(sb, KERN_ERR, "can't mount with "
1706                                  "dioread_nolock if block size != PAGE_SIZE");
1707                         return 0;
1708                 }
1709         }
1710         return 1;
1711 }
1712
1713 static inline void ext4_show_quota_options(struct seq_file *seq,
1714                                            struct super_block *sb)
1715 {
1716 #if defined(CONFIG_QUOTA)
1717         struct ext4_sb_info *sbi = EXT4_SB(sb);
1718
1719         if (sbi->s_jquota_fmt) {
1720                 char *fmtname = "";
1721
1722                 switch (sbi->s_jquota_fmt) {
1723                 case QFMT_VFS_OLD:
1724                         fmtname = "vfsold";
1725                         break;
1726                 case QFMT_VFS_V0:
1727                         fmtname = "vfsv0";
1728                         break;
1729                 case QFMT_VFS_V1:
1730                         fmtname = "vfsv1";
1731                         break;
1732                 }
1733                 seq_printf(seq, ",jqfmt=%s", fmtname);
1734         }
1735
1736         if (sbi->s_qf_names[USRQUOTA])
1737                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1738
1739         if (sbi->s_qf_names[GRPQUOTA])
1740                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1741
1742         if (test_opt(sb, USRQUOTA))
1743                 seq_puts(seq, ",usrquota");
1744
1745         if (test_opt(sb, GRPQUOTA))
1746                 seq_puts(seq, ",grpquota");
1747 #endif
1748 }
1749
1750 static const char *token2str(int token)
1751 {
1752         const struct match_token *t;
1753
1754         for (t = tokens; t->token != Opt_err; t++)
1755                 if (t->token == token && !strchr(t->pattern, '='))
1756                         break;
1757         return t->pattern;
1758 }
1759
1760 /*
1761  * Show an option if
1762  *  - it's set to a non-default value OR
1763  *  - if the per-sb default is different from the global default
1764  */
1765 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1766                               int nodefs)
1767 {
1768         struct ext4_sb_info *sbi = EXT4_SB(sb);
1769         struct ext4_super_block *es = sbi->s_es;
1770         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1771         const struct mount_opts *m;
1772         char sep = nodefs ? '\n' : ',';
1773
1774 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1775 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1776
1777         if (sbi->s_sb_block != 1)
1778                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1779
1780         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1781                 int want_set = m->flags & MOPT_SET;
1782                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1783                     (m->flags & MOPT_CLEAR_ERR))
1784                         continue;
1785                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1786                         continue; /* skip if same as the default */
1787                 if ((want_set &&
1788                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1789                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1790                         continue; /* select Opt_noFoo vs Opt_Foo */
1791                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1792         }
1793
1794         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1795             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1796                 SEQ_OPTS_PRINT("resuid=%u",
1797                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1798         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1799             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1800                 SEQ_OPTS_PRINT("resgid=%u",
1801                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1802         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1803         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1804                 SEQ_OPTS_PUTS("errors=remount-ro");
1805         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1806                 SEQ_OPTS_PUTS("errors=continue");
1807         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1808                 SEQ_OPTS_PUTS("errors=panic");
1809         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1810                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1811         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1812                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1813         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1814                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1815         if (sb->s_flags & MS_I_VERSION)
1816                 SEQ_OPTS_PUTS("i_version");
1817         if (nodefs || sbi->s_stripe)
1818                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1819         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1820                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1821                         SEQ_OPTS_PUTS("data=journal");
1822                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1823                         SEQ_OPTS_PUTS("data=ordered");
1824                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1825                         SEQ_OPTS_PUTS("data=writeback");
1826         }
1827         if (nodefs ||
1828             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1829                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1830                                sbi->s_inode_readahead_blks);
1831
1832         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1833                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1834                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1835         if (nodefs || sbi->s_max_dir_size_kb)
1836                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1837
1838         ext4_show_quota_options(seq, sb);
1839         return 0;
1840 }
1841
1842 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1843 {
1844         return _ext4_show_options(seq, root->d_sb, 0);
1845 }
1846
1847 static int options_seq_show(struct seq_file *seq, void *offset)
1848 {
1849         struct super_block *sb = seq->private;
1850         int rc;
1851
1852         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1853         rc = _ext4_show_options(seq, sb, 1);
1854         seq_puts(seq, "\n");
1855         return rc;
1856 }
1857
1858 static int options_open_fs(struct inode *inode, struct file *file)
1859 {
1860         return single_open(file, options_seq_show, PDE(inode)->data);
1861 }
1862
1863 static const struct file_operations ext4_seq_options_fops = {
1864         .owner = THIS_MODULE,
1865         .open = options_open_fs,
1866         .read = seq_read,
1867         .llseek = seq_lseek,
1868         .release = single_release,
1869 };
1870
1871 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1872                             int read_only)
1873 {
1874         struct ext4_sb_info *sbi = EXT4_SB(sb);
1875         int res = 0;
1876
1877         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1878                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1879                          "forcing read-only mode");
1880                 res = MS_RDONLY;
1881         }
1882         if (read_only)
1883                 goto done;
1884         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1885                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1886                          "running e2fsck is recommended");
1887         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1888                 ext4_msg(sb, KERN_WARNING,
1889                          "warning: mounting fs with errors, "
1890                          "running e2fsck is recommended");
1891         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1892                  le16_to_cpu(es->s_mnt_count) >=
1893                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1894                 ext4_msg(sb, KERN_WARNING,
1895                          "warning: maximal mount count reached, "
1896                          "running e2fsck is recommended");
1897         else if (le32_to_cpu(es->s_checkinterval) &&
1898                 (le32_to_cpu(es->s_lastcheck) +
1899                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1900                 ext4_msg(sb, KERN_WARNING,
1901                          "warning: checktime reached, "
1902                          "running e2fsck is recommended");
1903         if (!sbi->s_journal)
1904                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1905         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1906                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1907         le16_add_cpu(&es->s_mnt_count, 1);
1908         es->s_mtime = cpu_to_le32(get_seconds());
1909         ext4_update_dynamic_rev(sb);
1910         if (sbi->s_journal)
1911                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1912
1913         ext4_commit_super(sb, 1);
1914 done:
1915         if (test_opt(sb, DEBUG))
1916                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1917                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1918                         sb->s_blocksize,
1919                         sbi->s_groups_count,
1920                         EXT4_BLOCKS_PER_GROUP(sb),
1921                         EXT4_INODES_PER_GROUP(sb),
1922                         sbi->s_mount_opt, sbi->s_mount_opt2);
1923
1924         cleancache_init_fs(sb);
1925         return res;
1926 }
1927
1928 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1929 {
1930         struct ext4_sb_info *sbi = EXT4_SB(sb);
1931         struct flex_groups *new_groups;
1932         int size;
1933
1934         if (!sbi->s_log_groups_per_flex)
1935                 return 0;
1936
1937         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1938         if (size <= sbi->s_flex_groups_allocated)
1939                 return 0;
1940
1941         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1942         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1943         if (!new_groups) {
1944                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1945                          size / (int) sizeof(struct flex_groups));
1946                 return -ENOMEM;
1947         }
1948
1949         if (sbi->s_flex_groups) {
1950                 memcpy(new_groups, sbi->s_flex_groups,
1951                        (sbi->s_flex_groups_allocated *
1952                         sizeof(struct flex_groups)));
1953                 ext4_kvfree(sbi->s_flex_groups);
1954         }
1955         sbi->s_flex_groups = new_groups;
1956         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1957         return 0;
1958 }
1959
1960 static int ext4_fill_flex_info(struct super_block *sb)
1961 {
1962         struct ext4_sb_info *sbi = EXT4_SB(sb);
1963         struct ext4_group_desc *gdp = NULL;
1964         ext4_group_t flex_group;
1965         unsigned int groups_per_flex = 0;
1966         int i, err;
1967
1968         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1969         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1970                 sbi->s_log_groups_per_flex = 0;
1971                 return 1;
1972         }
1973         groups_per_flex = 1U << sbi->s_log_groups_per_flex;
1974
1975         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1976         if (err)
1977                 goto failed;
1978
1979         for (i = 0; i < sbi->s_groups_count; i++) {
1980                 gdp = ext4_get_group_desc(sb, i, NULL);
1981
1982                 flex_group = ext4_flex_group(sbi, i);
1983                 atomic_add(ext4_free_inodes_count(sb, gdp),
1984                            &sbi->s_flex_groups[flex_group].free_inodes);
1985                 atomic_add(ext4_free_group_clusters(sb, gdp),
1986                            &sbi->s_flex_groups[flex_group].free_clusters);
1987                 atomic_add(ext4_used_dirs_count(sb, gdp),
1988                            &sbi->s_flex_groups[flex_group].used_dirs);
1989         }
1990
1991         return 1;
1992 failed:
1993         return 0;
1994 }
1995
1996 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1997                                    struct ext4_group_desc *gdp)
1998 {
1999         int offset;
2000         __u16 crc = 0;
2001         __le32 le_group = cpu_to_le32(block_group);
2002
2003         if ((sbi->s_es->s_feature_ro_compat &
2004              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
2005                 /* Use new metadata_csum algorithm */
2006                 __u16 old_csum;
2007                 __u32 csum32;
2008
2009                 old_csum = gdp->bg_checksum;
2010                 gdp->bg_checksum = 0;
2011                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2012                                      sizeof(le_group));
2013                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2014                                      sbi->s_desc_size);
2015                 gdp->bg_checksum = old_csum;
2016
2017                 crc = csum32 & 0xFFFF;
2018                 goto out;
2019         }
2020
2021         /* old crc16 code */
2022         offset = offsetof(struct ext4_group_desc, bg_checksum);
2023
2024         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2025         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2026         crc = crc16(crc, (__u8 *)gdp, offset);
2027         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2028         /* for checksum of struct ext4_group_desc do the rest...*/
2029         if ((sbi->s_es->s_feature_incompat &
2030              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2031             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2032                 crc = crc16(crc, (__u8 *)gdp + offset,
2033                             le16_to_cpu(sbi->s_es->s_desc_size) -
2034                                 offset);
2035
2036 out:
2037         return cpu_to_le16(crc);
2038 }
2039
2040 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2041                                 struct ext4_group_desc *gdp)
2042 {
2043         if (ext4_has_group_desc_csum(sb) &&
2044             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2045                                                       block_group, gdp)))
2046                 return 0;
2047
2048         return 1;
2049 }
2050
2051 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2052                               struct ext4_group_desc *gdp)
2053 {
2054         if (!ext4_has_group_desc_csum(sb))
2055                 return;
2056         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2057 }
2058
2059 /* Called at mount-time, super-block is locked */
2060 static int ext4_check_descriptors(struct super_block *sb,
2061                                   ext4_group_t *first_not_zeroed)
2062 {
2063         struct ext4_sb_info *sbi = EXT4_SB(sb);
2064         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2065         ext4_fsblk_t last_block;
2066         ext4_fsblk_t block_bitmap;
2067         ext4_fsblk_t inode_bitmap;
2068         ext4_fsblk_t inode_table;
2069         int flexbg_flag = 0;
2070         ext4_group_t i, grp = sbi->s_groups_count;
2071
2072         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2073                 flexbg_flag = 1;
2074
2075         ext4_debug("Checking group descriptors");
2076
2077         for (i = 0; i < sbi->s_groups_count; i++) {
2078                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2079
2080                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2081                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2082                 else
2083                         last_block = first_block +
2084                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2085
2086                 if ((grp == sbi->s_groups_count) &&
2087                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2088                         grp = i;
2089
2090                 block_bitmap = ext4_block_bitmap(sb, gdp);
2091                 if (block_bitmap < first_block || block_bitmap > last_block) {
2092                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2093                                "Block bitmap for group %u not in group "
2094                                "(block %llu)!", i, block_bitmap);
2095                         return 0;
2096                 }
2097                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2098                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2099                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2100                                "Inode bitmap for group %u not in group "
2101                                "(block %llu)!", i, inode_bitmap);
2102                         return 0;
2103                 }
2104                 inode_table = ext4_inode_table(sb, gdp);
2105                 if (inode_table < first_block ||
2106                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2107                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2108                                "Inode table for group %u not in group "
2109                                "(block %llu)!", i, inode_table);
2110                         return 0;
2111                 }
2112                 ext4_lock_group(sb, i);
2113                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2114                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2115                                  "Checksum for group %u failed (%u!=%u)",
2116                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2117                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2118                         if (!(sb->s_flags & MS_RDONLY)) {
2119                                 ext4_unlock_group(sb, i);
2120                                 return 0;
2121                         }
2122                 }
2123                 ext4_unlock_group(sb, i);
2124                 if (!flexbg_flag)
2125                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2126         }
2127         if (NULL != first_not_zeroed)
2128                 *first_not_zeroed = grp;
2129
2130         ext4_free_blocks_count_set(sbi->s_es,
2131                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2132         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2133         return 1;
2134 }
2135
2136 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2137  * the superblock) which were deleted from all directories, but held open by
2138  * a process at the time of a crash.  We walk the list and try to delete these
2139  * inodes at recovery time (only with a read-write filesystem).
2140  *
2141  * In order to keep the orphan inode chain consistent during traversal (in
2142  * case of crash during recovery), we link each inode into the superblock
2143  * orphan list_head and handle it the same way as an inode deletion during
2144  * normal operation (which journals the operations for us).
2145  *
2146  * We only do an iget() and an iput() on each inode, which is very safe if we
2147  * accidentally point at an in-use or already deleted inode.  The worst that
2148  * can happen in this case is that we get a "bit already cleared" message from
2149  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2150  * e2fsck was run on this filesystem, and it must have already done the orphan
2151  * inode cleanup for us, so we can safely abort without any further action.
2152  */
2153 static void ext4_orphan_cleanup(struct super_block *sb,
2154                                 struct ext4_super_block *es)
2155 {
2156         unsigned int s_flags = sb->s_flags;
2157         int nr_orphans = 0, nr_truncates = 0;
2158 #ifdef CONFIG_QUOTA
2159         int i;
2160 #endif
2161         if (!es->s_last_orphan) {
2162                 jbd_debug(4, "no orphan inodes to clean up\n");
2163                 return;
2164         }
2165
2166         if (bdev_read_only(sb->s_bdev)) {
2167                 ext4_msg(sb, KERN_ERR, "write access "
2168                         "unavailable, skipping orphan cleanup");
2169                 return;
2170         }
2171
2172         /* Check if feature set would not allow a r/w mount */
2173         if (!ext4_feature_set_ok(sb, 0)) {
2174                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2175                          "unknown ROCOMPAT features");
2176                 return;
2177         }
2178
2179         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2180                 /* don't clear list on RO mount w/ errors */
2181                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2182                         jbd_debug(1, "Errors on filesystem, "
2183                                   "clearing orphan list.\n");
2184                         es->s_last_orphan = 0;
2185                 }
2186                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2187                 return;
2188         }
2189
2190         if (s_flags & MS_RDONLY) {
2191                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2192                 sb->s_flags &= ~MS_RDONLY;
2193         }
2194 #ifdef CONFIG_QUOTA
2195         /* Needed for iput() to work correctly and not trash data */
2196         sb->s_flags |= MS_ACTIVE;
2197         /* Turn on quotas so that they are updated correctly */
2198         for (i = 0; i < MAXQUOTAS; i++) {
2199                 if (EXT4_SB(sb)->s_qf_names[i]) {
2200                         int ret = ext4_quota_on_mount(sb, i);
2201                         if (ret < 0)
2202                                 ext4_msg(sb, KERN_ERR,
2203                                         "Cannot turn on journaled "
2204                                         "quota: error %d", ret);
2205                 }
2206         }
2207 #endif
2208
2209         while (es->s_last_orphan) {
2210                 struct inode *inode;
2211
2212                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2213                 if (IS_ERR(inode)) {
2214                         es->s_last_orphan = 0;
2215                         break;
2216                 }
2217
2218                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2219                 dquot_initialize(inode);
2220                 if (inode->i_nlink) {
2221                         ext4_msg(sb, KERN_DEBUG,
2222                                 "%s: truncating inode %lu to %lld bytes",
2223                                 __func__, inode->i_ino, inode->i_size);
2224                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2225                                   inode->i_ino, inode->i_size);
2226                         mutex_lock(&inode->i_mutex);
2227                         ext4_truncate(inode);
2228                         mutex_unlock(&inode->i_mutex);
2229                         nr_truncates++;
2230                 } else {
2231                         ext4_msg(sb, KERN_DEBUG,
2232                                 "%s: deleting unreferenced inode %lu",
2233                                 __func__, inode->i_ino);
2234                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2235                                   inode->i_ino);
2236                         nr_orphans++;
2237                 }
2238                 iput(inode);  /* The delete magic happens here! */
2239         }
2240
2241 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2242
2243         if (nr_orphans)
2244                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2245                        PLURAL(nr_orphans));
2246         if (nr_truncates)
2247                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2248                        PLURAL(nr_truncates));
2249 #ifdef CONFIG_QUOTA
2250         /* Turn quotas off */
2251         for (i = 0; i < MAXQUOTAS; i++) {
2252                 if (sb_dqopt(sb)->files[i])
2253                         dquot_quota_off(sb, i);
2254         }
2255 #endif
2256         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2257 }
2258
2259 /*
2260  * Maximal extent format file size.
2261  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2262  * extent format containers, within a sector_t, and within i_blocks
2263  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2264  * so that won't be a limiting factor.
2265  *
2266  * However there is other limiting factor. We do store extents in the form
2267  * of starting block and length, hence the resulting length of the extent
2268  * covering maximum file size must fit into on-disk format containers as
2269  * well. Given that length is always by 1 unit bigger than max unit (because
2270  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2271  *
2272  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2273  */
2274 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2275 {
2276         loff_t res;
2277         loff_t upper_limit = MAX_LFS_FILESIZE;
2278
2279         /* small i_blocks in vfs inode? */
2280         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2281                 /*
2282                  * CONFIG_LBDAF is not enabled implies the inode
2283                  * i_block represent total blocks in 512 bytes
2284                  * 32 == size of vfs inode i_blocks * 8
2285                  */
2286                 upper_limit = (1LL << 32) - 1;
2287
2288                 /* total blocks in file system block size */
2289                 upper_limit >>= (blkbits - 9);
2290                 upper_limit <<= blkbits;
2291         }
2292
2293         /*
2294          * 32-bit extent-start container, ee_block. We lower the maxbytes
2295          * by one fs block, so ee_len can cover the extent of maximum file
2296          * size
2297          */
2298         res = (1LL << 32) - 1;
2299         res <<= blkbits;
2300
2301         /* Sanity check against vm- & vfs- imposed limits */
2302         if (res > upper_limit)
2303                 res = upper_limit;
2304
2305         return res;
2306 }
2307
2308 /*
2309  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2310  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2311  * We need to be 1 filesystem block less than the 2^48 sector limit.
2312  */
2313 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2314 {
2315         loff_t res = EXT4_NDIR_BLOCKS;
2316         int meta_blocks;
2317         loff_t upper_limit;
2318         /* This is calculated to be the largest file size for a dense, block
2319          * mapped file such that the file's total number of 512-byte sectors,
2320          * including data and all indirect blocks, does not exceed (2^48 - 1).
2321          *
2322          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2323          * number of 512-byte sectors of the file.
2324          */
2325
2326         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2327                 /*
2328                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2329                  * the inode i_block field represents total file blocks in
2330                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2331                  */
2332                 upper_limit = (1LL << 32) - 1;
2333
2334                 /* total blocks in file system block size */
2335                 upper_limit >>= (bits - 9);
2336
2337         } else {
2338                 /*
2339                  * We use 48 bit ext4_inode i_blocks
2340                  * With EXT4_HUGE_FILE_FL set the i_blocks
2341                  * represent total number of blocks in
2342                  * file system block size
2343                  */
2344                 upper_limit = (1LL << 48) - 1;
2345
2346         }
2347
2348         /* indirect blocks */
2349         meta_blocks = 1;
2350         /* double indirect blocks */
2351         meta_blocks += 1 + (1LL << (bits-2));
2352         /* tripple indirect blocks */
2353         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2354
2355         upper_limit -= meta_blocks;
2356         upper_limit <<= bits;
2357
2358         res += 1LL << (bits-2);
2359         res += 1LL << (2*(bits-2));
2360         res += 1LL << (3*(bits-2));
2361         res <<= bits;
2362         if (res > upper_limit)
2363                 res = upper_limit;
2364
2365         if (res > MAX_LFS_FILESIZE)
2366                 res = MAX_LFS_FILESIZE;
2367
2368         return res;
2369 }
2370
2371 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2372                                    ext4_fsblk_t logical_sb_block, int nr)
2373 {
2374         struct ext4_sb_info *sbi = EXT4_SB(sb);
2375         ext4_group_t bg, first_meta_bg;
2376         int has_super = 0;
2377
2378         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2379
2380         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2381             nr < first_meta_bg)
2382                 return logical_sb_block + nr + 1;
2383         bg = sbi->s_desc_per_block * nr;
2384         if (ext4_bg_has_super(sb, bg))
2385                 has_super = 1;
2386
2387         return (has_super + ext4_group_first_block_no(sb, bg));
2388 }
2389
2390 /**
2391  * ext4_get_stripe_size: Get the stripe size.
2392  * @sbi: In memory super block info
2393  *
2394  * If we have specified it via mount option, then
2395  * use the mount option value. If the value specified at mount time is
2396  * greater than the blocks per group use the super block value.
2397  * If the super block value is greater than blocks per group return 0.
2398  * Allocator needs it be less than blocks per group.
2399  *
2400  */
2401 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2402 {
2403         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2404         unsigned long stripe_width =
2405                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2406         int ret;
2407
2408         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2409                 ret = sbi->s_stripe;
2410         else if (stripe_width <= sbi->s_blocks_per_group)
2411                 ret = stripe_width;
2412         else if (stride <= sbi->s_blocks_per_group)
2413                 ret = stride;
2414         else
2415                 ret = 0;
2416
2417         /*
2418          * If the stripe width is 1, this makes no sense and
2419          * we set it to 0 to turn off stripe handling code.
2420          */
2421         if (ret <= 1)
2422                 ret = 0;
2423
2424         return ret;
2425 }
2426
2427 /* sysfs supprt */
2428
2429 struct ext4_attr {
2430         struct attribute attr;
2431         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2432         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2433                          const char *, size_t);
2434         int offset;
2435 };
2436
2437 static int parse_strtoul(const char *buf,
2438                 unsigned long max, unsigned long *value)
2439 {
2440         char *endp;
2441
2442         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2443         endp = skip_spaces(endp);
2444         if (*endp || *value > max)
2445                 return -EINVAL;
2446
2447         return 0;
2448 }
2449
2450 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2451                                               struct ext4_sb_info *sbi,
2452                                               char *buf)
2453 {
2454         return snprintf(buf, PAGE_SIZE, "%llu\n",
2455                 (s64) EXT4_C2B(sbi,
2456                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2457 }
2458
2459 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2460                                          struct ext4_sb_info *sbi, char *buf)
2461 {
2462         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2463
2464         if (!sb->s_bdev->bd_part)
2465                 return snprintf(buf, PAGE_SIZE, "0\n");
2466         return snprintf(buf, PAGE_SIZE, "%lu\n",
2467                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2468                          sbi->s_sectors_written_start) >> 1);
2469 }
2470
2471 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2472                                           struct ext4_sb_info *sbi, char *buf)
2473 {
2474         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2475
2476         if (!sb->s_bdev->bd_part)
2477                 return snprintf(buf, PAGE_SIZE, "0\n");
2478         return snprintf(buf, PAGE_SIZE, "%llu\n",
2479                         (unsigned long long)(sbi->s_kbytes_written +
2480                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2481                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2482 }
2483
2484 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2485                                           struct ext4_sb_info *sbi,
2486                                           const char *buf, size_t count)
2487 {
2488         unsigned long t;
2489
2490         if (parse_strtoul(buf, 0x40000000, &t))
2491                 return -EINVAL;
2492
2493         if (t && !is_power_of_2(t))
2494                 return -EINVAL;
2495
2496         sbi->s_inode_readahead_blks = t;
2497         return count;
2498 }
2499
2500 static ssize_t sbi_ui_show(struct ext4_attr *a,
2501                            struct ext4_sb_info *sbi, char *buf)
2502 {
2503         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2504
2505         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2506 }
2507
2508 static ssize_t sbi_ui_store(struct ext4_attr *a,
2509                             struct ext4_sb_info *sbi,
2510                             const char *buf, size_t count)
2511 {
2512         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2513         unsigned long t;
2514
2515         if (parse_strtoul(buf, 0xffffffff, &t))
2516                 return -EINVAL;
2517         *ui = t;
2518         return count;
2519 }
2520
2521 static ssize_t trigger_test_error(struct ext4_attr *a,
2522                                   struct ext4_sb_info *sbi,
2523                                   const char *buf, size_t count)
2524 {
2525         int len = count;
2526
2527         if (!capable(CAP_SYS_ADMIN))
2528                 return -EPERM;
2529
2530         if (len && buf[len-1] == '\n')
2531                 len--;
2532
2533         if (len)
2534                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2535         return count;
2536 }
2537
2538 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2539 static struct ext4_attr ext4_attr_##_name = {                   \
2540         .attr = {.name = __stringify(_name), .mode = _mode },   \
2541         .show   = _show,                                        \
2542         .store  = _store,                                       \
2543         .offset = offsetof(struct ext4_sb_info, _elname),       \
2544 }
2545 #define EXT4_ATTR(name, mode, show, store) \
2546 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2547
2548 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2549 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2550 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2551 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2552         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2553 #define ATTR_LIST(name) &ext4_attr_##name.attr
2554
2555 EXT4_RO_ATTR(delayed_allocation_blocks);
2556 EXT4_RO_ATTR(session_write_kbytes);
2557 EXT4_RO_ATTR(lifetime_write_kbytes);
2558 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2559                  inode_readahead_blks_store, s_inode_readahead_blks);
2560 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2561 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2562 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2563 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2564 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2565 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2566 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2567 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2568 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2569 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2570
2571 static struct attribute *ext4_attrs[] = {
2572         ATTR_LIST(delayed_allocation_blocks),
2573         ATTR_LIST(session_write_kbytes),
2574         ATTR_LIST(lifetime_write_kbytes),
2575         ATTR_LIST(inode_readahead_blks),
2576         ATTR_LIST(inode_goal),
2577         ATTR_LIST(mb_stats),
2578         ATTR_LIST(mb_max_to_scan),
2579         ATTR_LIST(mb_min_to_scan),
2580         ATTR_LIST(mb_order2_req),
2581         ATTR_LIST(mb_stream_req),
2582         ATTR_LIST(mb_group_prealloc),
2583         ATTR_LIST(max_writeback_mb_bump),
2584         ATTR_LIST(extent_max_zeroout_kb),
2585         ATTR_LIST(trigger_fs_error),
2586         NULL,
2587 };
2588
2589 /* Features this copy of ext4 supports */
2590 EXT4_INFO_ATTR(lazy_itable_init);
2591 EXT4_INFO_ATTR(batched_discard);
2592 EXT4_INFO_ATTR(meta_bg_resize);
2593
2594 static struct attribute *ext4_feat_attrs[] = {
2595         ATTR_LIST(lazy_itable_init),
2596         ATTR_LIST(batched_discard),
2597         ATTR_LIST(meta_bg_resize),
2598         NULL,
2599 };
2600
2601 static ssize_t ext4_attr_show(struct kobject *kobj,
2602                               struct attribute *attr, char *buf)
2603 {
2604         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2605                                                 s_kobj);
2606         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2607
2608         return a->show ? a->show(a, sbi, buf) : 0;
2609 }
2610
2611 static ssize_t ext4_attr_store(struct kobject *kobj,
2612                                struct attribute *attr,
2613                                const char *buf, size_t len)
2614 {
2615         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2616                                                 s_kobj);
2617         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2618
2619         return a->store ? a->store(a, sbi, buf, len) : 0;
2620 }
2621
2622 static void ext4_sb_release(struct kobject *kobj)
2623 {
2624         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2625                                                 s_kobj);
2626         complete(&sbi->s_kobj_unregister);
2627 }
2628
2629 static const struct sysfs_ops ext4_attr_ops = {
2630         .show   = ext4_attr_show,
2631         .store  = ext4_attr_store,
2632 };
2633
2634 static struct kobj_type ext4_ktype = {
2635         .default_attrs  = ext4_attrs,
2636         .sysfs_ops      = &ext4_attr_ops,
2637         .release        = ext4_sb_release,
2638 };
2639
2640 static void ext4_feat_release(struct kobject *kobj)
2641 {
2642         complete(&ext4_feat->f_kobj_unregister);
2643 }
2644
2645 static struct kobj_type ext4_feat_ktype = {
2646         .default_attrs  = ext4_feat_attrs,
2647         .sysfs_ops      = &ext4_attr_ops,
2648         .release        = ext4_feat_release,
2649 };
2650
2651 /*
2652  * Check whether this filesystem can be mounted based on
2653  * the features present and the RDONLY/RDWR mount requested.
2654  * Returns 1 if this filesystem can be mounted as requested,
2655  * 0 if it cannot be.
2656  */
2657 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2658 {
2659         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2660                 ext4_msg(sb, KERN_ERR,
2661                         "Couldn't mount because of "
2662                         "unsupported optional features (%x)",
2663                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2664                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2665                 return 0;
2666         }
2667
2668         if (readonly)
2669                 return 1;
2670
2671         /* Check that feature set is OK for a read-write mount */
2672         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2673                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2674                          "unsupported optional features (%x)",
2675                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2676                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2677                 return 0;
2678         }
2679         /*
2680          * Large file size enabled file system can only be mounted
2681          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2682          */
2683         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2684                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2685                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2686                                  "cannot be mounted RDWR without "
2687                                  "CONFIG_LBDAF");
2688                         return 0;
2689                 }
2690         }
2691         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2692             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2693                 ext4_msg(sb, KERN_ERR,
2694                          "Can't support bigalloc feature without "
2695                          "extents feature\n");
2696                 return 0;
2697         }
2698
2699 #ifndef CONFIG_QUOTA
2700         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2701             !readonly) {
2702                 ext4_msg(sb, KERN_ERR,
2703                          "Filesystem with quota feature cannot be mounted RDWR "
2704                          "without CONFIG_QUOTA");
2705                 return 0;
2706         }
2707 #endif  /* CONFIG_QUOTA */
2708         return 1;
2709 }
2710
2711 /*
2712  * This function is called once a day if we have errors logged
2713  * on the file system
2714  */
2715 static void print_daily_error_info(unsigned long arg)
2716 {
2717         struct super_block *sb = (struct super_block *) arg;
2718         struct ext4_sb_info *sbi;
2719         struct ext4_super_block *es;
2720
2721         sbi = EXT4_SB(sb);
2722         es = sbi->s_es;
2723
2724         if (es->s_error_count)
2725                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2726                          le32_to_cpu(es->s_error_count));
2727         if (es->s_first_error_time) {
2728                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2729                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2730                        (int) sizeof(es->s_first_error_func),
2731                        es->s_first_error_func,
2732                        le32_to_cpu(es->s_first_error_line));
2733                 if (es->s_first_error_ino)
2734                         printk(": inode %u",
2735                                le32_to_cpu(es->s_first_error_ino));
2736                 if (es->s_first_error_block)
2737                         printk(": block %llu", (unsigned long long)
2738                                le64_to_cpu(es->s_first_error_block));
2739                 printk("\n");
2740         }
2741         if (es->s_last_error_time) {
2742                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2743                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2744                        (int) sizeof(es->s_last_error_func),
2745                        es->s_last_error_func,
2746                        le32_to_cpu(es->s_last_error_line));
2747                 if (es->s_last_error_ino)
2748                         printk(": inode %u",
2749                                le32_to_cpu(es->s_last_error_ino));
2750                 if (es->s_last_error_block)
2751                         printk(": block %llu", (unsigned long long)
2752                                le64_to_cpu(es->s_last_error_block));
2753                 printk("\n");
2754         }
2755         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2756 }
2757
2758 /* Find next suitable group and run ext4_init_inode_table */
2759 static int ext4_run_li_request(struct ext4_li_request *elr)
2760 {
2761         struct ext4_group_desc *gdp = NULL;
2762         ext4_group_t group, ngroups;
2763         struct super_block *sb;
2764         unsigned long timeout = 0;
2765         int ret = 0;
2766
2767         sb = elr->lr_super;
2768         ngroups = EXT4_SB(sb)->s_groups_count;
2769
2770         sb_start_write(sb);
2771         for (group = elr->lr_next_group; group < ngroups; group++) {
2772                 gdp = ext4_get_group_desc(sb, group, NULL);
2773                 if (!gdp) {
2774                         ret = 1;
2775                         break;
2776                 }
2777
2778                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2779                         break;
2780         }
2781
2782         if (group >= ngroups)
2783                 ret = 1;
2784
2785         if (!ret) {
2786                 timeout = jiffies;
2787                 ret = ext4_init_inode_table(sb, group,
2788                                             elr->lr_timeout ? 0 : 1);
2789                 if (elr->lr_timeout == 0) {
2790                         timeout = (jiffies - timeout) *
2791                                   elr->lr_sbi->s_li_wait_mult;
2792                         elr->lr_timeout = timeout;
2793                 }
2794                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2795                 elr->lr_next_group = group + 1;
2796         }
2797         sb_end_write(sb);
2798
2799         return ret;
2800 }
2801
2802 /*
2803  * Remove lr_request from the list_request and free the
2804  * request structure. Should be called with li_list_mtx held
2805  */
2806 static void ext4_remove_li_request(struct ext4_li_request *elr)
2807 {
2808         struct ext4_sb_info *sbi;
2809
2810         if (!elr)
2811                 return;
2812
2813         sbi = elr->lr_sbi;
2814
2815         list_del(&elr->lr_request);
2816         sbi->s_li_request = NULL;
2817         kfree(elr);
2818 }
2819
2820 static void ext4_unregister_li_request(struct super_block *sb)
2821 {
2822         mutex_lock(&ext4_li_mtx);
2823         if (!ext4_li_info) {
2824                 mutex_unlock(&ext4_li_mtx);
2825                 return;
2826         }
2827
2828         mutex_lock(&ext4_li_info->li_list_mtx);
2829         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2830         mutex_unlock(&ext4_li_info->li_list_mtx);
2831         mutex_unlock(&ext4_li_mtx);
2832 }
2833
2834 static struct task_struct *ext4_lazyinit_task;
2835
2836 /*
2837  * This is the function where ext4lazyinit thread lives. It walks
2838  * through the request list searching for next scheduled filesystem.
2839  * When such a fs is found, run the lazy initialization request
2840  * (ext4_rn_li_request) and keep track of the time spend in this
2841  * function. Based on that time we compute next schedule time of
2842  * the request. When walking through the list is complete, compute
2843  * next waking time and put itself into sleep.
2844  */
2845 static int ext4_lazyinit_thread(void *arg)
2846 {
2847         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2848         struct list_head *pos, *n;
2849         struct ext4_li_request *elr;
2850         unsigned long next_wakeup, cur;
2851
2852         BUG_ON(NULL == eli);
2853
2854 cont_thread:
2855         while (true) {
2856                 next_wakeup = MAX_JIFFY_OFFSET;
2857
2858                 mutex_lock(&eli->li_list_mtx);
2859                 if (list_empty(&eli->li_request_list)) {
2860                         mutex_unlock(&eli->li_list_mtx);
2861                         goto exit_thread;
2862                 }
2863
2864                 list_for_each_safe(pos, n, &eli->li_request_list) {
2865                         elr = list_entry(pos, struct ext4_li_request,
2866                                          lr_request);
2867
2868                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2869                                 if (ext4_run_li_request(elr) != 0) {
2870                                         /* error, remove the lazy_init job */
2871                                         ext4_remove_li_request(elr);
2872                                         continue;
2873                                 }
2874                         }
2875
2876                         if (time_before(elr->lr_next_sched, next_wakeup))
2877                                 next_wakeup = elr->lr_next_sched;
2878                 }
2879                 mutex_unlock(&eli->li_list_mtx);
2880
2881                 try_to_freeze();
2882
2883                 cur = jiffies;
2884                 if ((time_after_eq(cur, next_wakeup)) ||
2885                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2886                         cond_resched();
2887                         continue;
2888                 }
2889
2890                 schedule_timeout_interruptible(next_wakeup - cur);
2891
2892                 if (kthread_should_stop()) {
2893                         ext4_clear_request_list();
2894                         goto exit_thread;
2895                 }
2896         }
2897
2898 exit_thread:
2899         /*
2900          * It looks like the request list is empty, but we need
2901          * to check it under the li_list_mtx lock, to prevent any
2902          * additions into it, and of course we should lock ext4_li_mtx
2903          * to atomically free the list and ext4_li_info, because at
2904          * this point another ext4 filesystem could be registering
2905          * new one.
2906          */
2907         mutex_lock(&ext4_li_mtx);
2908         mutex_lock(&eli->li_list_mtx);
2909         if (!list_empty(&eli->li_request_list)) {
2910                 mutex_unlock(&eli->li_list_mtx);
2911                 mutex_unlock(&ext4_li_mtx);
2912                 goto cont_thread;
2913         }
2914         mutex_unlock(&eli->li_list_mtx);
2915         kfree(ext4_li_info);
2916         ext4_li_info = NULL;
2917         mutex_unlock(&ext4_li_mtx);
2918
2919         return 0;
2920 }
2921
2922 static void ext4_clear_request_list(void)
2923 {
2924         struct list_head *pos, *n;
2925         struct ext4_li_request *elr;
2926
2927         mutex_lock(&ext4_li_info->li_list_mtx);
2928         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2929                 elr = list_entry(pos, struct ext4_li_request,
2930                                  lr_request);
2931                 ext4_remove_li_request(elr);
2932         }
2933         mutex_unlock(&ext4_li_info->li_list_mtx);
2934 }
2935
2936 static int ext4_run_lazyinit_thread(void)
2937 {
2938         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2939                                          ext4_li_info, "ext4lazyinit");
2940         if (IS_ERR(ext4_lazyinit_task)) {
2941                 int err = PTR_ERR(ext4_lazyinit_task);
2942                 ext4_clear_request_list();
2943                 kfree(ext4_li_info);
2944                 ext4_li_info = NULL;
2945                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2946                                  "initialization thread\n",
2947                                  err);
2948                 return err;
2949         }
2950         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2951         return 0;
2952 }
2953
2954 /*
2955  * Check whether it make sense to run itable init. thread or not.
2956  * If there is at least one uninitialized inode table, return
2957  * corresponding group number, else the loop goes through all
2958  * groups and return total number of groups.
2959  */
2960 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2961 {
2962         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2963         struct ext4_group_desc *gdp = NULL;
2964
2965         for (group = 0; group < ngroups; group++) {
2966                 gdp = ext4_get_group_desc(sb, group, NULL);
2967                 if (!gdp)
2968                         continue;
2969
2970                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2971                         break;
2972         }
2973
2974         return group;
2975 }
2976
2977 static int ext4_li_info_new(void)
2978 {
2979         struct ext4_lazy_init *eli = NULL;
2980
2981         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2982         if (!eli)
2983                 return -ENOMEM;
2984
2985         INIT_LIST_HEAD(&eli->li_request_list);
2986         mutex_init(&eli->li_list_mtx);
2987
2988         eli->li_state |= EXT4_LAZYINIT_QUIT;
2989
2990         ext4_li_info = eli;
2991
2992         return 0;
2993 }
2994
2995 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2996                                             ext4_group_t start)
2997 {
2998         struct ext4_sb_info *sbi = EXT4_SB(sb);
2999         struct ext4_li_request *elr;
3000         unsigned long rnd;
3001
3002         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3003         if (!elr)
3004                 return NULL;
3005
3006         elr->lr_super = sb;
3007         elr->lr_sbi = sbi;
3008         elr->lr_next_group = start;
3009
3010         /*
3011          * Randomize first schedule time of the request to
3012          * spread the inode table initialization requests
3013          * better.
3014          */
3015         get_random_bytes(&rnd, sizeof(rnd));
3016         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3017                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3018
3019         return elr;
3020 }
3021
3022 int ext4_register_li_request(struct super_block *sb,
3023                              ext4_group_t first_not_zeroed)
3024 {
3025         struct ext4_sb_info *sbi = EXT4_SB(sb);
3026         struct ext4_li_request *elr = NULL;
3027         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3028         int ret = 0;
3029
3030         mutex_lock(&ext4_li_mtx);
3031         if (sbi->s_li_request != NULL) {
3032                 /*
3033                  * Reset timeout so it can be computed again, because
3034                  * s_li_wait_mult might have changed.
3035                  */
3036                 sbi->s_li_request->lr_timeout = 0;
3037                 goto out;
3038         }
3039
3040         if (first_not_zeroed == ngroups ||
3041             (sb->s_flags & MS_RDONLY) ||
3042             !test_opt(sb, INIT_INODE_TABLE))
3043                 goto out;
3044
3045         elr = ext4_li_request_new(sb, first_not_zeroed);
3046         if (!elr) {
3047                 ret = -ENOMEM;
3048                 goto out;
3049         }
3050
3051         if (NULL == ext4_li_info) {
3052                 ret = ext4_li_info_new();
3053                 if (ret)
3054                         goto out;
3055         }
3056
3057         mutex_lock(&ext4_li_info->li_list_mtx);
3058         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3059         mutex_unlock(&ext4_li_info->li_list_mtx);
3060
3061         sbi->s_li_request = elr;
3062         /*
3063          * set elr to NULL here since it has been inserted to
3064          * the request_list and the removal and free of it is
3065          * handled by ext4_clear_request_list from now on.
3066          */
3067         elr = NULL;
3068
3069         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3070                 ret = ext4_run_lazyinit_thread();
3071                 if (ret)
3072                         goto out;
3073         }
3074 out:
3075         mutex_unlock(&ext4_li_mtx);
3076         if (ret)
3077                 kfree(elr);
3078         return ret;
3079 }
3080
3081 /*
3082  * We do not need to lock anything since this is called on
3083  * module unload.
3084  */
3085 static void ext4_destroy_lazyinit_thread(void)
3086 {
3087         /*
3088          * If thread exited earlier
3089          * there's nothing to be done.
3090          */
3091         if (!ext4_li_info || !ext4_lazyinit_task)
3092                 return;
3093
3094         kthread_stop(ext4_lazyinit_task);
3095 }
3096
3097 static int set_journal_csum_feature_set(struct super_block *sb)
3098 {
3099         int ret = 1;
3100         int compat, incompat;
3101         struct ext4_sb_info *sbi = EXT4_SB(sb);
3102
3103         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3104                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3105                 /* journal checksum v2 */
3106                 compat = 0;
3107                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3108         } else {
3109                 /* journal checksum v1 */
3110                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3111                 incompat = 0;
3112         }
3113
3114         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3115                 ret = jbd2_journal_set_features(sbi->s_journal,
3116                                 compat, 0,
3117                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3118                                 incompat);
3119         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3120                 ret = jbd2_journal_set_features(sbi->s_journal,
3121                                 compat, 0,
3122                                 incompat);
3123                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3124                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3125         } else {
3126                 jbd2_journal_clear_features(sbi->s_journal,
3127                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3128                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3129                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3130         }
3131
3132         return ret;
3133 }
3134
3135 /*
3136  * Note: calculating the overhead so we can be compatible with
3137  * historical BSD practice is quite difficult in the face of
3138  * clusters/bigalloc.  This is because multiple metadata blocks from
3139  * different block group can end up in the same allocation cluster.
3140  * Calculating the exact overhead in the face of clustered allocation
3141  * requires either O(all block bitmaps) in memory or O(number of block
3142  * groups**2) in time.  We will still calculate the superblock for
3143  * older file systems --- and if we come across with a bigalloc file
3144  * system with zero in s_overhead_clusters the estimate will be close to
3145  * correct especially for very large cluster sizes --- but for newer
3146  * file systems, it's better to calculate this figure once at mkfs
3147  * time, and store it in the superblock.  If the superblock value is
3148  * present (even for non-bigalloc file systems), we will use it.
3149  */
3150 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3151                           char *buf)
3152 {
3153         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3154         struct ext4_group_desc  *gdp;
3155         ext4_fsblk_t            first_block, last_block, b;
3156         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3157         int                     s, j, count = 0;
3158
3159         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3160                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3161                         sbi->s_itb_per_group + 2);
3162
3163         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3164                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3165         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3166         for (i = 0; i < ngroups; i++) {
3167                 gdp = ext4_get_group_desc(sb, i, NULL);
3168                 b = ext4_block_bitmap(sb, gdp);
3169                 if (b >= first_block && b <= last_block) {
3170                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3171                         count++;
3172                 }
3173                 b = ext4_inode_bitmap(sb, gdp);
3174                 if (b >= first_block && b <= last_block) {
3175                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3176                         count++;
3177                 }
3178                 b = ext4_inode_table(sb, gdp);
3179                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3180                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3181                                 int c = EXT4_B2C(sbi, b - first_block);
3182                                 ext4_set_bit(c, buf);
3183                                 count++;
3184                         }
3185                 if (i != grp)
3186                         continue;
3187                 s = 0;
3188                 if (ext4_bg_has_super(sb, grp)) {
3189                         ext4_set_bit(s++, buf);
3190                         count++;
3191                 }
3192                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3193                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3194                         count++;
3195                 }
3196         }
3197         if (!count)
3198                 return 0;
3199         return EXT4_CLUSTERS_PER_GROUP(sb) -
3200                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3201 }
3202
3203 /*
3204  * Compute the overhead and stash it in sbi->s_overhead
3205  */
3206 int ext4_calculate_overhead(struct super_block *sb)
3207 {
3208         struct ext4_sb_info *sbi = EXT4_SB(sb);
3209         struct ext4_super_block *es = sbi->s_es;
3210         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3211         ext4_fsblk_t overhead = 0;
3212         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3213
3214         if (!buf)
3215                 return -ENOMEM;
3216
3217         /*
3218          * Compute the overhead (FS structures).  This is constant
3219          * for a given filesystem unless the number of block groups
3220          * changes so we cache the previous value until it does.
3221          */
3222
3223         /*
3224          * All of the blocks before first_data_block are overhead
3225          */
3226         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3227
3228         /*
3229          * Add the overhead found in each block group
3230          */
3231         for (i = 0; i < ngroups; i++) {
3232                 int blks;
3233
3234                 blks = count_overhead(sb, i, buf);
3235                 overhead += blks;
3236                 if (blks)
3237                         memset(buf, 0, PAGE_SIZE);
3238                 cond_resched();
3239         }
3240         /* Add the journal blocks as well */
3241         if (sbi->s_journal)
3242                 overhead += EXT4_B2C(sbi, sbi->s_journal->j_maxlen);
3243
3244         sbi->s_overhead = overhead;
3245         smp_wmb();
3246         free_page((unsigned long) buf);
3247         return 0;
3248 }
3249
3250 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3251 {
3252         char *orig_data = kstrdup(data, GFP_KERNEL);
3253         struct buffer_head *bh;
3254         struct ext4_super_block *es = NULL;
3255         struct ext4_sb_info *sbi;
3256         ext4_fsblk_t block;
3257         ext4_fsblk_t sb_block = get_sb_block(&data);
3258         ext4_fsblk_t logical_sb_block;
3259         unsigned long offset = 0;
3260         unsigned long journal_devnum = 0;
3261         unsigned long def_mount_opts;
3262         struct inode *root;
3263         char *cp;
3264         const char *descr;
3265         int ret = -ENOMEM;
3266         int blocksize, clustersize;
3267         unsigned int db_count;
3268         unsigned int i;
3269         int needs_recovery, has_huge_files, has_bigalloc;
3270         __u64 blocks_count;
3271         int err = 0;
3272         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3273         ext4_group_t first_not_zeroed;
3274
3275         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3276         if (!sbi)
3277                 goto out_free_orig;
3278
3279         sbi->s_blockgroup_lock =
3280                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3281         if (!sbi->s_blockgroup_lock) {
3282                 kfree(sbi);
3283                 goto out_free_orig;
3284         }
3285         sb->s_fs_info = sbi;
3286         sbi->s_sb = sb;
3287         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3288         sbi->s_sb_block = sb_block;
3289         if (sb->s_bdev->bd_part)
3290                 sbi->s_sectors_written_start =
3291                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3292
3293         /* Cleanup superblock name */
3294         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3295                 *cp = '!';
3296
3297         /* -EINVAL is default */
3298         ret = -EINVAL;
3299         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3300         if (!blocksize) {
3301                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3302                 goto out_fail;
3303         }
3304
3305         /*
3306          * The ext4 superblock will not be buffer aligned for other than 1kB
3307          * block sizes.  We need to calculate the offset from buffer start.
3308          */
3309         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3310                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3311                 offset = do_div(logical_sb_block, blocksize);
3312         } else {
3313                 logical_sb_block = sb_block;
3314         }
3315
3316         if (!(bh = sb_bread(sb, logical_sb_block))) {
3317                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3318                 goto out_fail;
3319         }
3320         /*
3321          * Note: s_es must be initialized as soon as possible because
3322          *       some ext4 macro-instructions depend on its value
3323          */
3324         es = (struct ext4_super_block *) (bh->b_data + offset);
3325         sbi->s_es = es;
3326         sb->s_magic = le16_to_cpu(es->s_magic);
3327         if (sb->s_magic != EXT4_SUPER_MAGIC)
3328                 goto cantfind_ext4;
3329         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3330
3331         /* Warn if metadata_csum and gdt_csum are both set. */
3332         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3333                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3334             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3335                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3336                              "redundant flags; please run fsck.");
3337
3338         /* Check for a known checksum algorithm */
3339         if (!ext4_verify_csum_type(sb, es)) {
3340                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3341                          "unknown checksum algorithm.");
3342                 silent = 1;
3343                 goto cantfind_ext4;
3344         }
3345
3346         /* Load the checksum driver */
3347         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3348                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3349                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3350                 if (IS_ERR(sbi->s_chksum_driver)) {
3351                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3352                         ret = PTR_ERR(sbi->s_chksum_driver);
3353                         sbi->s_chksum_driver = NULL;
3354                         goto failed_mount;
3355                 }
3356         }
3357
3358         /* Check superblock checksum */
3359         if (!ext4_superblock_csum_verify(sb, es)) {
3360                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3361                          "invalid superblock checksum.  Run e2fsck?");
3362                 silent = 1;
3363                 goto cantfind_ext4;
3364         }
3365
3366         /* Precompute checksum seed for all metadata */
3367         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3368                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3369                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3370                                                sizeof(es->s_uuid));
3371
3372         /* Set defaults before we parse the mount options */
3373         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3374         set_opt(sb, INIT_INODE_TABLE);
3375         if (def_mount_opts & EXT4_DEFM_DEBUG)
3376                 set_opt(sb, DEBUG);
3377         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3378                 set_opt(sb, GRPID);
3379         if (def_mount_opts & EXT4_DEFM_UID16)
3380                 set_opt(sb, NO_UID32);
3381         /* xattr user namespace & acls are now defaulted on */
3382         set_opt(sb, XATTR_USER);
3383 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3384         set_opt(sb, POSIX_ACL);
3385 #endif
3386         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3387                 set_opt(sb, JOURNAL_DATA);
3388         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3389                 set_opt(sb, ORDERED_DATA);
3390         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3391                 set_opt(sb, WRITEBACK_DATA);
3392
3393         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3394                 set_opt(sb, ERRORS_PANIC);
3395         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3396                 set_opt(sb, ERRORS_CONT);
3397         else
3398                 set_opt(sb, ERRORS_RO);
3399         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3400                 set_opt(sb, BLOCK_VALIDITY);
3401         if (def_mount_opts & EXT4_DEFM_DISCARD)
3402                 set_opt(sb, DISCARD);
3403
3404         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3405         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3406         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3407         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3408         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3409
3410         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3411                 set_opt(sb, BARRIER);
3412
3413         /*
3414          * enable delayed allocation by default
3415          * Use -o nodelalloc to turn it off
3416          */
3417         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3418             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3419                 set_opt(sb, DELALLOC);
3420
3421         /*
3422          * set default s_li_wait_mult for lazyinit, for the case there is
3423          * no mount option specified.
3424          */
3425         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3426
3427         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3428                            &journal_devnum, &journal_ioprio, 0)) {
3429                 ext4_msg(sb, KERN_WARNING,
3430                          "failed to parse options in superblock: %s",
3431                          sbi->s_es->s_mount_opts);
3432         }
3433         sbi->s_def_mount_opt = sbi->s_mount_opt;
3434         if (!parse_options((char *) data, sb, &journal_devnum,
3435                            &journal_ioprio, 0))
3436                 goto failed_mount;
3437
3438         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3439                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3440                             "with data=journal disables delayed "
3441                             "allocation and O_DIRECT support!\n");
3442                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3443                         ext4_msg(sb, KERN_ERR, "can't mount with "
3444                                  "both data=journal and delalloc");
3445                         goto failed_mount;
3446                 }
3447                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3448                         ext4_msg(sb, KERN_ERR, "can't mount with "
3449                                  "both data=journal and delalloc");
3450                         goto failed_mount;
3451                 }
3452                 if (test_opt(sb, DELALLOC))
3453                         clear_opt(sb, DELALLOC);
3454         }
3455
3456         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3457                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3458
3459         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3460             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3461              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3462              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3463                 ext4_msg(sb, KERN_WARNING,
3464                        "feature flags set on rev 0 fs, "
3465                        "running e2fsck is recommended");
3466
3467         if (IS_EXT2_SB(sb)) {
3468                 if (ext2_feature_set_ok(sb))
3469                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3470                                  "using the ext4 subsystem");
3471                 else {
3472                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3473                                  "to feature incompatibilities");
3474                         goto failed_mount;
3475                 }
3476         }
3477
3478         if (IS_EXT3_SB(sb)) {
3479                 if (ext3_feature_set_ok(sb))
3480                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3481                                  "using the ext4 subsystem");
3482                 else {
3483                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3484                                  "to feature incompatibilities");
3485                         goto failed_mount;
3486                 }
3487         }
3488
3489         /*
3490          * Check feature flags regardless of the revision level, since we
3491          * previously didn't change the revision level when setting the flags,
3492          * so there is a chance incompat flags are set on a rev 0 filesystem.
3493          */
3494         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3495                 goto failed_mount;
3496
3497         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3498         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3499             blocksize > EXT4_MAX_BLOCK_SIZE) {
3500                 ext4_msg(sb, KERN_ERR,
3501                        "Unsupported filesystem blocksize %d", blocksize);
3502                 goto failed_mount;
3503         }
3504
3505         if (sb->s_blocksize != blocksize) {
3506                 /* Validate the filesystem blocksize */
3507                 if (!sb_set_blocksize(sb, blocksize)) {
3508                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3509                                         blocksize);
3510                         goto failed_mount;
3511                 }
3512
3513                 brelse(bh);
3514                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3515                 offset = do_div(logical_sb_block, blocksize);
3516                 bh = sb_bread(sb, logical_sb_block);
3517                 if (!bh) {
3518                         ext4_msg(sb, KERN_ERR,
3519                                "Can't read superblock on 2nd try");
3520                         goto failed_mount;
3521                 }
3522                 es = (struct ext4_super_block *)(bh->b_data + offset);
3523                 sbi->s_es = es;
3524                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3525                         ext4_msg(sb, KERN_ERR,
3526                                "Magic mismatch, very weird!");
3527                         goto failed_mount;
3528                 }
3529         }
3530
3531         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3532                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3533         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3534                                                       has_huge_files);
3535         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3536
3537         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3538                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3539                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3540         } else {
3541                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3542                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3543                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3544                     (!is_power_of_2(sbi->s_inode_size)) ||
3545                     (sbi->s_inode_size > blocksize)) {
3546                         ext4_msg(sb, KERN_ERR,
3547                                "unsupported inode size: %d",
3548                                sbi->s_inode_size);
3549                         goto failed_mount;
3550                 }
3551                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3552                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3553         }
3554
3555         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3556         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3557                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3558                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3559                     !is_power_of_2(sbi->s_desc_size)) {
3560                         ext4_msg(sb, KERN_ERR,
3561                                "unsupported descriptor size %lu",
3562                                sbi->s_desc_size);
3563                         goto failed_mount;
3564                 }
3565         } else
3566                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3567
3568         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3569         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3570         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3571                 goto cantfind_ext4;
3572
3573         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3574         if (sbi->s_inodes_per_block == 0)
3575                 goto cantfind_ext4;
3576         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3577                                         sbi->s_inodes_per_block;
3578         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3579         sbi->s_sbh = bh;
3580         sbi->s_mount_state = le16_to_cpu(es->s_state);
3581         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3582         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3583
3584         for (i = 0; i < 4; i++)
3585                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3586         sbi->s_def_hash_version = es->s_def_hash_version;
3587         i = le32_to_cpu(es->s_flags);
3588         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3589                 sbi->s_hash_unsigned = 3;
3590         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3591 #ifdef __CHAR_UNSIGNED__
3592                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3593                 sbi->s_hash_unsigned = 3;
3594 #else
3595                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3596 #endif
3597         }
3598
3599         /* Handle clustersize */
3600         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3601         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3602                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3603         if (has_bigalloc) {
3604                 if (clustersize < blocksize) {
3605                         ext4_msg(sb, KERN_ERR,
3606                                  "cluster size (%d) smaller than "
3607                                  "block size (%d)", clustersize, blocksize);
3608                         goto failed_mount;
3609                 }
3610                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3611                         le32_to_cpu(es->s_log_block_size);
3612                 sbi->s_clusters_per_group =
3613                         le32_to_cpu(es->s_clusters_per_group);
3614                 if (sbi->s_clusters_per_group > blocksize * 8) {
3615                         ext4_msg(sb, KERN_ERR,
3616                                  "#clusters per group too big: %lu",
3617                                  sbi->s_clusters_per_group);
3618                         goto failed_mount;
3619                 }
3620                 if (sbi->s_blocks_per_group !=
3621                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3622                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3623                                  "clusters per group (%lu) inconsistent",
3624                                  sbi->s_blocks_per_group,
3625                                  sbi->s_clusters_per_group);
3626                         goto failed_mount;
3627                 }
3628         } else {
3629                 if (clustersize != blocksize) {
3630                         ext4_warning(sb, "fragment/cluster size (%d) != "
3631                                      "block size (%d)", clustersize,
3632                                      blocksize);
3633                         clustersize = blocksize;
3634                 }
3635                 if (sbi->s_blocks_per_group > blocksize * 8) {
3636                         ext4_msg(sb, KERN_ERR,
3637                                  "#blocks per group too big: %lu",
3638                                  sbi->s_blocks_per_group);
3639                         goto failed_mount;
3640                 }
3641                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3642                 sbi->s_cluster_bits = 0;
3643         }
3644         sbi->s_cluster_ratio = clustersize / blocksize;
3645
3646         if (sbi->s_inodes_per_group > blocksize * 8) {
3647                 ext4_msg(sb, KERN_ERR,
3648                        "#inodes per group too big: %lu",
3649                        sbi->s_inodes_per_group);
3650                 goto failed_mount;
3651         }
3652
3653         /*
3654          * Test whether we have more sectors than will fit in sector_t,
3655          * and whether the max offset is addressable by the page cache.
3656          */
3657         err = generic_check_addressable(sb->s_blocksize_bits,
3658                                         ext4_blocks_count(es));
3659         if (err) {
3660                 ext4_msg(sb, KERN_ERR, "filesystem"
3661                          " too large to mount safely on this system");
3662                 if (sizeof(sector_t) < 8)
3663                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3664                 goto failed_mount;
3665         }
3666
3667         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3668                 goto cantfind_ext4;
3669
3670         /* check blocks count against device size */
3671         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3672         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3673                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3674                        "exceeds size of device (%llu blocks)",
3675                        ext4_blocks_count(es), blocks_count);
3676                 goto failed_mount;
3677         }
3678
3679         /*
3680          * It makes no sense for the first data block to be beyond the end
3681          * of the filesystem.
3682          */
3683         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3684                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3685                          "block %u is beyond end of filesystem (%llu)",
3686                          le32_to_cpu(es->s_first_data_block),
3687                          ext4_blocks_count(es));
3688                 goto failed_mount;
3689         }
3690         blocks_count = (ext4_blocks_count(es) -
3691                         le32_to_cpu(es->s_first_data_block) +
3692                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3693         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3694         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3695                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3696                        "(block count %llu, first data block %u, "
3697                        "blocks per group %lu)", sbi->s_groups_count,
3698                        ext4_blocks_count(es),
3699                        le32_to_cpu(es->s_first_data_block),
3700                        EXT4_BLOCKS_PER_GROUP(sb));
3701                 goto failed_mount;
3702         }
3703         sbi->s_groups_count = blocks_count;
3704         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3705                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3706         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3707                    EXT4_DESC_PER_BLOCK(sb);
3708         sbi->s_group_desc = ext4_kvmalloc(db_count *
3709                                           sizeof(struct buffer_head *),
3710                                           GFP_KERNEL);
3711         if (sbi->s_group_desc == NULL) {
3712                 ext4_msg(sb, KERN_ERR, "not enough memory");
3713                 ret = -ENOMEM;
3714                 goto failed_mount;
3715         }
3716
3717         if (ext4_proc_root)
3718                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3719
3720         if (sbi->s_proc)
3721                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3722                                  &ext4_seq_options_fops, sb);
3723
3724         bgl_lock_init(sbi->s_blockgroup_lock);
3725
3726         for (i = 0; i < db_count; i++) {
3727                 block = descriptor_loc(sb, logical_sb_block, i);
3728                 sbi->s_group_desc[i] = sb_bread(sb, block);
3729                 if (!sbi->s_group_desc[i]) {
3730                         ext4_msg(sb, KERN_ERR,
3731                                "can't read group descriptor %d", i);
3732                         db_count = i;
3733                         goto failed_mount2;
3734                 }
3735         }
3736         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3737                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3738                 goto failed_mount2;
3739         }
3740         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3741                 if (!ext4_fill_flex_info(sb)) {
3742                         ext4_msg(sb, KERN_ERR,
3743                                "unable to initialize "
3744                                "flex_bg meta info!");
3745                         goto failed_mount2;
3746                 }
3747
3748         sbi->s_gdb_count = db_count;
3749         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3750         spin_lock_init(&sbi->s_next_gen_lock);
3751
3752         init_timer(&sbi->s_err_report);
3753         sbi->s_err_report.function = print_daily_error_info;
3754         sbi->s_err_report.data = (unsigned long) sb;
3755
3756         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3757                         ext4_count_free_clusters(sb));
3758         if (!err) {
3759                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3760                                 ext4_count_free_inodes(sb));
3761         }
3762         if (!err) {
3763                 err = percpu_counter_init(&sbi->s_dirs_counter,
3764                                 ext4_count_dirs(sb));
3765         }
3766         if (!err) {
3767                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3768         }
3769         if (err) {
3770                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3771                 goto failed_mount3;
3772         }
3773
3774         sbi->s_stripe = ext4_get_stripe_size(sbi);
3775         sbi->s_max_writeback_mb_bump = 128;
3776         sbi->s_extent_max_zeroout_kb = 32;
3777
3778         /*
3779          * set up enough so that it can read an inode
3780          */
3781         if (!test_opt(sb, NOLOAD) &&
3782             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3783                 sb->s_op = &ext4_sops;
3784         else
3785                 sb->s_op = &ext4_nojournal_sops;
3786         sb->s_export_op = &ext4_export_ops;
3787         sb->s_xattr = ext4_xattr_handlers;
3788 #ifdef CONFIG_QUOTA
3789         sb->s_qcop = &ext4_qctl_operations;
3790         sb->dq_op = &ext4_quota_operations;
3791
3792         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
3793                 /* Use qctl operations for hidden quota files. */
3794                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3795         }
3796 #endif
3797         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3798
3799         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3800         mutex_init(&sbi->s_orphan_lock);
3801
3802         sb->s_root = NULL;
3803
3804         needs_recovery = (es->s_last_orphan != 0 ||
3805                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3806                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3807
3808         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3809             !(sb->s_flags & MS_RDONLY))
3810                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3811                         goto failed_mount3;
3812
3813         /*
3814          * The first inode we look at is the journal inode.  Don't try
3815          * root first: it may be modified in the journal!
3816          */
3817         if (!test_opt(sb, NOLOAD) &&
3818             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3819                 if (ext4_load_journal(sb, es, journal_devnum))
3820                         goto failed_mount3;
3821         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3822               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3823                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3824                        "suppressed and not mounted read-only");
3825                 goto failed_mount_wq;
3826         } else {
3827                 clear_opt(sb, DATA_FLAGS);
3828                 sbi->s_journal = NULL;
3829                 needs_recovery = 0;
3830                 goto no_journal;
3831         }
3832
3833         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3834             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3835                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3836                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3837                 goto failed_mount_wq;
3838         }
3839
3840         if (!set_journal_csum_feature_set(sb)) {
3841                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3842                          "feature set");
3843                 goto failed_mount_wq;
3844         }
3845
3846         /* We have now updated the journal if required, so we can
3847          * validate the data journaling mode. */
3848         switch (test_opt(sb, DATA_FLAGS)) {
3849         case 0:
3850                 /* No mode set, assume a default based on the journal
3851                  * capabilities: ORDERED_DATA if the journal can
3852                  * cope, else JOURNAL_DATA
3853                  */
3854                 if (jbd2_journal_check_available_features
3855                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3856                         set_opt(sb, ORDERED_DATA);
3857                 else
3858                         set_opt(sb, JOURNAL_DATA);
3859                 break;
3860
3861         case EXT4_MOUNT_ORDERED_DATA:
3862         case EXT4_MOUNT_WRITEBACK_DATA:
3863                 if (!jbd2_journal_check_available_features
3864                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3865                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3866                                "requested data journaling mode");
3867                         goto failed_mount_wq;
3868                 }
3869         default:
3870                 break;
3871         }
3872         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3873
3874         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3875
3876         /*
3877          * The journal may have updated the bg summary counts, so we
3878          * need to update the global counters.
3879          */
3880         percpu_counter_set(&sbi->s_freeclusters_counter,
3881                            ext4_count_free_clusters(sb));
3882         percpu_counter_set(&sbi->s_freeinodes_counter,
3883                            ext4_count_free_inodes(sb));
3884         percpu_counter_set(&sbi->s_dirs_counter,
3885                            ext4_count_dirs(sb));
3886         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3887
3888 no_journal:
3889         /*
3890          * Get the # of file system overhead blocks from the
3891          * superblock if present.
3892          */
3893         if (es->s_overhead_clusters)
3894                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3895         else {
3896                 err = ext4_calculate_overhead(sb);
3897                 if (err)
3898                         goto failed_mount_wq;
3899         }
3900
3901         /*
3902          * The maximum number of concurrent works can be high and
3903          * concurrency isn't really necessary.  Limit it to 1.
3904          */
3905         EXT4_SB(sb)->dio_unwritten_wq =
3906                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3907         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3908                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3909                 ret = -ENOMEM;
3910                 goto failed_mount_wq;
3911         }
3912
3913         /*
3914          * The jbd2_journal_load will have done any necessary log recovery,
3915          * so we can safely mount the rest of the filesystem now.
3916          */
3917
3918         root = ext4_iget(sb, EXT4_ROOT_INO);
3919         if (IS_ERR(root)) {
3920                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3921                 ret = PTR_ERR(root);
3922                 root = NULL;
3923                 goto failed_mount4;
3924         }
3925         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3926                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3927                 iput(root);
3928                 goto failed_mount4;
3929         }
3930         sb->s_root = d_make_root(root);
3931         if (!sb->s_root) {
3932                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3933                 ret = -ENOMEM;
3934                 goto failed_mount4;
3935         }
3936
3937         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3938                 sb->s_flags |= MS_RDONLY;
3939
3940         /* determine the minimum size of new large inodes, if present */
3941         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3942                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3943                                                      EXT4_GOOD_OLD_INODE_SIZE;
3944                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3945                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3946                         if (sbi->s_want_extra_isize <
3947                             le16_to_cpu(es->s_want_extra_isize))
3948                                 sbi->s_want_extra_isize =
3949                                         le16_to_cpu(es->s_want_extra_isize);
3950                         if (sbi->s_want_extra_isize <
3951                             le16_to_cpu(es->s_min_extra_isize))
3952                                 sbi->s_want_extra_isize =
3953                                         le16_to_cpu(es->s_min_extra_isize);
3954                 }
3955         }
3956         /* Check if enough inode space is available */
3957         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3958                                                         sbi->s_inode_size) {
3959                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3960                                                        EXT4_GOOD_OLD_INODE_SIZE;
3961                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3962                          "available");
3963         }
3964
3965         err = ext4_setup_system_zone(sb);
3966         if (err) {
3967                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3968                          "zone (%d)", err);
3969                 goto failed_mount4a;
3970         }
3971
3972         ext4_ext_init(sb);
3973         err = ext4_mb_init(sb);
3974         if (err) {
3975                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3976                          err);
3977                 goto failed_mount5;
3978         }
3979
3980         err = ext4_register_li_request(sb, first_not_zeroed);
3981         if (err)
3982                 goto failed_mount6;
3983
3984         sbi->s_kobj.kset = ext4_kset;
3985         init_completion(&sbi->s_kobj_unregister);
3986         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3987                                    "%s", sb->s_id);
3988         if (err)
3989                 goto failed_mount7;
3990
3991         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3992         ext4_orphan_cleanup(sb, es);
3993         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3994         if (needs_recovery) {
3995                 ext4_msg(sb, KERN_INFO, "recovery complete");
3996                 ext4_mark_recovery_complete(sb, es);
3997         }
3998         if (EXT4_SB(sb)->s_journal) {
3999                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4000                         descr = " journalled data mode";
4001                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4002                         descr = " ordered data mode";
4003                 else
4004                         descr = " writeback data mode";
4005         } else
4006                 descr = "out journal";
4007
4008 #ifdef CONFIG_QUOTA
4009         /* Enable quota usage during mount. */
4010         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4011             !(sb->s_flags & MS_RDONLY)) {
4012                 err = ext4_enable_quotas(sb);
4013                 if (err)
4014                         goto failed_mount8;
4015         }
4016 #endif  /* CONFIG_QUOTA */
4017
4018         if (test_opt(sb, DISCARD)) {
4019                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4020                 if (!blk_queue_discard(q))
4021                         ext4_msg(sb, KERN_WARNING,
4022                                  "mounting with \"discard\" option, but "
4023                                  "the device does not support discard");
4024         }
4025
4026         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4027                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4028                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4029
4030         if (es->s_error_count)
4031                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4032
4033         kfree(orig_data);
4034         return 0;
4035
4036 cantfind_ext4:
4037         if (!silent)
4038                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4039         goto failed_mount;
4040
4041 #ifdef CONFIG_QUOTA
4042 failed_mount8:
4043         kobject_del(&sbi->s_kobj);
4044 #endif
4045 failed_mount7:
4046         ext4_unregister_li_request(sb);
4047 failed_mount6:
4048         ext4_mb_release(sb);
4049 failed_mount5:
4050         ext4_ext_release(sb);
4051         ext4_release_system_zone(sb);
4052 failed_mount4a:
4053         dput(sb->s_root);
4054         sb->s_root = NULL;
4055 failed_mount4:
4056         ext4_msg(sb, KERN_ERR, "mount failed");
4057         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4058 failed_mount_wq:
4059         if (sbi->s_journal) {
4060                 jbd2_journal_destroy(sbi->s_journal);
4061                 sbi->s_journal = NULL;
4062         }
4063 failed_mount3:
4064         del_timer(&sbi->s_err_report);
4065         if (sbi->s_flex_groups)
4066                 ext4_kvfree(sbi->s_flex_groups);
4067         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4068         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4069         percpu_counter_destroy(&sbi->s_dirs_counter);
4070         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4071         if (sbi->s_mmp_tsk)
4072                 kthread_stop(sbi->s_mmp_tsk);
4073 failed_mount2:
4074         for (i = 0; i < db_count; i++)
4075                 brelse(sbi->s_group_desc[i]);
4076         ext4_kvfree(sbi->s_group_desc);
4077 failed_mount:
4078         if (sbi->s_chksum_driver)
4079                 crypto_free_shash(sbi->s_chksum_driver);
4080         if (sbi->s_proc) {
4081                 remove_proc_entry("options", sbi->s_proc);
4082                 remove_proc_entry(sb->s_id, ext4_proc_root);
4083         }
4084 #ifdef CONFIG_QUOTA
4085         for (i = 0; i < MAXQUOTAS; i++)
4086                 kfree(sbi->s_qf_names[i]);
4087 #endif
4088         ext4_blkdev_remove(sbi);
4089         brelse(bh);
4090 out_fail:
4091         sb->s_fs_info = NULL;
4092         kfree(sbi->s_blockgroup_lock);
4093         kfree(sbi);
4094 out_free_orig:
4095         kfree(orig_data);
4096         return err ? err : ret;
4097 }
4098
4099 /*
4100  * Setup any per-fs journal parameters now.  We'll do this both on
4101  * initial mount, once the journal has been initialised but before we've
4102  * done any recovery; and again on any subsequent remount.
4103  */
4104 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4105 {
4106         struct ext4_sb_info *sbi = EXT4_SB(sb);
4107
4108         journal->j_commit_interval = sbi->s_commit_interval;
4109         journal->j_min_batch_time = sbi->s_min_batch_time;
4110         journal->j_max_batch_time = sbi->s_max_batch_time;
4111
4112         write_lock(&journal->j_state_lock);
4113         if (test_opt(sb, BARRIER))
4114                 journal->j_flags |= JBD2_BARRIER;
4115         else
4116                 journal->j_flags &= ~JBD2_BARRIER;
4117         if (test_opt(sb, DATA_ERR_ABORT))
4118                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4119         else
4120                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4121         write_unlock(&journal->j_state_lock);
4122 }
4123
4124 static journal_t *ext4_get_journal(struct super_block *sb,
4125                                    unsigned int journal_inum)
4126 {
4127         struct inode *journal_inode;
4128         journal_t *journal;
4129
4130         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4131
4132         /* First, test for the existence of a valid inode on disk.  Bad
4133          * things happen if we iget() an unused inode, as the subsequent
4134          * iput() will try to delete it. */
4135
4136         journal_inode = ext4_iget(sb, journal_inum);
4137         if (IS_ERR(journal_inode)) {
4138                 ext4_msg(sb, KERN_ERR, "no journal found");
4139                 return NULL;
4140         }
4141         if (!journal_inode->i_nlink) {
4142                 make_bad_inode(journal_inode);
4143                 iput(journal_inode);
4144                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4145                 return NULL;
4146         }
4147
4148         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4149                   journal_inode, journal_inode->i_size);
4150         if (!S_ISREG(journal_inode->i_mode)) {
4151                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4152                 iput(journal_inode);
4153                 return NULL;
4154         }
4155
4156         journal = jbd2_journal_init_inode(journal_inode);
4157         if (!journal) {
4158                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4159                 iput(journal_inode);
4160                 return NULL;
4161         }
4162         journal->j_private = sb;
4163         ext4_init_journal_params(sb, journal);
4164         return journal;
4165 }
4166
4167 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4168                                        dev_t j_dev)
4169 {
4170         struct buffer_head *bh;
4171         journal_t *journal;
4172         ext4_fsblk_t start;
4173         ext4_fsblk_t len;
4174         int hblock, blocksize;
4175         ext4_fsblk_t sb_block;
4176         unsigned long offset;
4177         struct ext4_super_block *es;
4178         struct block_device *bdev;
4179
4180         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4181
4182         bdev = ext4_blkdev_get(j_dev, sb);
4183         if (bdev == NULL)
4184                 return NULL;
4185
4186         blocksize = sb->s_blocksize;
4187         hblock = bdev_logical_block_size(bdev);
4188         if (blocksize < hblock) {
4189                 ext4_msg(sb, KERN_ERR,
4190                         "blocksize too small for journal device");
4191                 goto out_bdev;
4192         }
4193
4194         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4195         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4196         set_blocksize(bdev, blocksize);
4197         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4198                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4199                        "external journal");
4200                 goto out_bdev;
4201         }
4202
4203         es = (struct ext4_super_block *) (bh->b_data + offset);
4204         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4205             !(le32_to_cpu(es->s_feature_incompat) &
4206               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4207                 ext4_msg(sb, KERN_ERR, "external journal has "
4208                                         "bad superblock");
4209                 brelse(bh);
4210                 goto out_bdev;
4211         }
4212
4213         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4214                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4215                 brelse(bh);
4216                 goto out_bdev;
4217         }
4218
4219         len = ext4_blocks_count(es);
4220         start = sb_block + 1;
4221         brelse(bh);     /* we're done with the superblock */
4222
4223         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4224                                         start, len, blocksize);
4225         if (!journal) {
4226                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4227                 goto out_bdev;
4228         }
4229         journal->j_private = sb;
4230         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4231         wait_on_buffer(journal->j_sb_buffer);
4232         if (!buffer_uptodate(journal->j_sb_buffer)) {
4233                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4234                 goto out_journal;
4235         }
4236         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4237                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4238                                         "user (unsupported) - %d",
4239                         be32_to_cpu(journal->j_superblock->s_nr_users));
4240                 goto out_journal;
4241         }
4242         EXT4_SB(sb)->journal_bdev = bdev;
4243         ext4_init_journal_params(sb, journal);
4244         return journal;
4245
4246 out_journal:
4247         jbd2_journal_destroy(journal);
4248 out_bdev:
4249         ext4_blkdev_put(bdev);
4250         return NULL;
4251 }
4252
4253 static int ext4_load_journal(struct super_block *sb,
4254                              struct ext4_super_block *es,
4255                              unsigned long journal_devnum)
4256 {
4257         journal_t *journal;
4258         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4259         dev_t journal_dev;
4260         int err = 0;
4261         int really_read_only;
4262
4263         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4264
4265         if (journal_devnum &&
4266             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4267                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4268                         "numbers have changed");
4269                 journal_dev = new_decode_dev(journal_devnum);
4270         } else
4271                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4272
4273         really_read_only = bdev_read_only(sb->s_bdev);
4274
4275         /*
4276          * Are we loading a blank journal or performing recovery after a
4277          * crash?  For recovery, we need to check in advance whether we
4278          * can get read-write access to the device.
4279          */
4280         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4281                 if (sb->s_flags & MS_RDONLY) {
4282                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4283                                         "required on readonly filesystem");
4284                         if (really_read_only) {
4285                                 ext4_msg(sb, KERN_ERR, "write access "
4286                                         "unavailable, cannot proceed");
4287                                 return -EROFS;
4288                         }
4289                         ext4_msg(sb, KERN_INFO, "write access will "
4290                                "be enabled during recovery");
4291                 }
4292         }
4293
4294         if (journal_inum && journal_dev) {
4295                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4296                        "and inode journals!");
4297                 return -EINVAL;
4298         }
4299
4300         if (journal_inum) {
4301                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4302                         return -EINVAL;
4303         } else {
4304                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4305                         return -EINVAL;
4306         }
4307
4308         if (!(journal->j_flags & JBD2_BARRIER))
4309                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4310
4311         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4312                 err = jbd2_journal_wipe(journal, !really_read_only);
4313         if (!err) {
4314                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4315                 if (save)
4316                         memcpy(save, ((char *) es) +
4317                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4318                 err = jbd2_journal_load(journal);
4319                 if (save)
4320                         memcpy(((char *) es) + EXT4_S_ERR_START,
4321                                save, EXT4_S_ERR_LEN);
4322                 kfree(save);
4323         }
4324
4325         if (err) {
4326                 ext4_msg(sb, KERN_ERR, "error loading journal");
4327                 jbd2_journal_destroy(journal);
4328                 return err;
4329         }
4330
4331         EXT4_SB(sb)->s_journal = journal;
4332         ext4_clear_journal_err(sb, es);
4333
4334         if (!really_read_only && journal_devnum &&
4335             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4336                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4337
4338                 /* Make sure we flush the recovery flag to disk. */
4339                 ext4_commit_super(sb, 1);
4340         }
4341
4342         return 0;
4343 }
4344
4345 static int ext4_commit_super(struct super_block *sb, int sync)
4346 {
4347         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4348         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4349         int error = 0;
4350
4351         if (!sbh || block_device_ejected(sb))
4352                 return error;
4353         if (buffer_write_io_error(sbh)) {
4354                 /*
4355                  * Oh, dear.  A previous attempt to write the
4356                  * superblock failed.  This could happen because the
4357                  * USB device was yanked out.  Or it could happen to
4358                  * be a transient write error and maybe the block will
4359                  * be remapped.  Nothing we can do but to retry the
4360                  * write and hope for the best.
4361                  */
4362                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4363                        "superblock detected");
4364                 clear_buffer_write_io_error(sbh);
4365                 set_buffer_uptodate(sbh);
4366         }
4367         /*
4368          * If the file system is mounted read-only, don't update the
4369          * superblock write time.  This avoids updating the superblock
4370          * write time when we are mounting the root file system
4371          * read/only but we need to replay the journal; at that point,
4372          * for people who are east of GMT and who make their clock
4373          * tick in localtime for Windows bug-for-bug compatibility,
4374          * the clock is set in the future, and this will cause e2fsck
4375          * to complain and force a full file system check.
4376          */
4377         if (!(sb->s_flags & MS_RDONLY))
4378                 es->s_wtime = cpu_to_le32(get_seconds());
4379         if (sb->s_bdev->bd_part)
4380                 es->s_kbytes_written =
4381                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4382                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4383                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4384         else
4385                 es->s_kbytes_written =
4386                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4387         ext4_free_blocks_count_set(es,
4388                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4389                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4390         es->s_free_inodes_count =
4391                 cpu_to_le32(percpu_counter_sum_positive(
4392                                 &EXT4_SB(sb)->s_freeinodes_counter));
4393         BUFFER_TRACE(sbh, "marking dirty");
4394         ext4_superblock_csum_set(sb);
4395         mark_buffer_dirty(sbh);
4396         if (sync) {
4397                 error = sync_dirty_buffer(sbh);
4398                 if (error)
4399                         return error;
4400
4401                 error = buffer_write_io_error(sbh);
4402                 if (error) {
4403                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4404                                "superblock");
4405                         clear_buffer_write_io_error(sbh);
4406                         set_buffer_uptodate(sbh);
4407                 }
4408         }
4409         return error;
4410 }
4411
4412 /*
4413  * Have we just finished recovery?  If so, and if we are mounting (or
4414  * remounting) the filesystem readonly, then we will end up with a
4415  * consistent fs on disk.  Record that fact.
4416  */
4417 static void ext4_mark_recovery_complete(struct super_block *sb,
4418                                         struct ext4_super_block *es)
4419 {
4420         journal_t *journal = EXT4_SB(sb)->s_journal;
4421
4422         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4423                 BUG_ON(journal != NULL);
4424                 return;
4425         }
4426         jbd2_journal_lock_updates(journal);
4427         if (jbd2_journal_flush(journal) < 0)
4428                 goto out;
4429
4430         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4431             sb->s_flags & MS_RDONLY) {
4432                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4433                 ext4_commit_super(sb, 1);
4434         }
4435
4436 out:
4437         jbd2_journal_unlock_updates(journal);
4438 }
4439
4440 /*
4441  * If we are mounting (or read-write remounting) a filesystem whose journal
4442  * has recorded an error from a previous lifetime, move that error to the
4443  * main filesystem now.
4444  */
4445 static void ext4_clear_journal_err(struct super_block *sb,
4446                                    struct ext4_super_block *es)
4447 {
4448         journal_t *journal;
4449         int j_errno;
4450         const char *errstr;
4451
4452         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4453
4454         journal = EXT4_SB(sb)->s_journal;
4455
4456         /*
4457          * Now check for any error status which may have been recorded in the
4458          * journal by a prior ext4_error() or ext4_abort()
4459          */
4460
4461         j_errno = jbd2_journal_errno(journal);
4462         if (j_errno) {
4463                 char nbuf[16];
4464
4465                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4466                 ext4_warning(sb, "Filesystem error recorded "
4467                              "from previous mount: %s", errstr);
4468                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4469
4470                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4471                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4472                 ext4_commit_super(sb, 1);
4473
4474                 jbd2_journal_clear_err(journal);
4475                 jbd2_journal_update_sb_errno(journal);
4476         }
4477 }
4478
4479 /*
4480  * Force the running and committing transactions to commit,
4481  * and wait on the commit.
4482  */
4483 int ext4_force_commit(struct super_block *sb)
4484 {
4485         journal_t *journal;
4486
4487         if (sb->s_flags & MS_RDONLY)
4488                 return 0;
4489
4490         journal = EXT4_SB(sb)->s_journal;
4491         return ext4_journal_force_commit(journal);
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)