fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static const char *ext4_decode_error(struct super_block *sb, int errno,
  73                                      char nbuf[16]);
  74 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  75 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  76 static int ext4_unfreeze(struct super_block *sb);
  77 static int ext4_freeze(struct super_block *sb);
  78 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  79                        const char *dev_name, void *data);
  80 static inline int ext2_feature_set_ok(struct super_block *sb);
  81 static inline int ext3_feature_set_ok(struct super_block *sb);
  82 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  83 static void ext4_destroy_lazyinit_thread(void);
  84 static void ext4_unregister_li_request(struct super_block *sb);
  85 static void ext4_clear_request_list(void);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  96 #else
  97 #define IS_EXT2_SB(sb) (0)
  98 #endif
  99
 100
 101 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 102 static struct file_system_type ext3_fs_type = {
 103         .owner          = THIS_MODULE,
 104         .name           = "ext3",
 105         .mount          = ext4_mount,
 106         .kill_sb        = kill_block_super,
 107         .fs_flags       = FS_REQUIRES_DEV,
 108 };
 109 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 110 #else
 111 #define IS_EXT3_SB(sb) (0)
 112 #endif
 113
 114 static int ext4_verify_csum_type(struct super_block *sb,
 115                                  struct ext4_super_block *es)
 116 {
 117         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 118                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 119                 return 1;
 120
 121         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 122 }
 123
 124 static __le32 ext4_superblock_csum(struct super_block *sb,
 125                                    struct ext4_super_block *es)
 126 {
 127         struct ext4_sb_info *sbi = EXT4_SB(sb);
 128         int offset = offsetof(struct ext4_super_block, s_checksum);
 129         __u32 csum;
 130
 131         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 132
 133         return cpu_to_le32(csum);
 134 }
 135
 136 int ext4_superblock_csum_verify(struct super_block *sb,
 137                                 struct ext4_super_block *es)
 138 {
 139         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 140                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 141                 return 1;
 142
 143         return es->s_checksum == ext4_superblock_csum(sb, es);
 144 }
 145
 146 void ext4_superblock_csum_set(struct super_block *sb)
 147 {
 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         ei->vfs_inode.i_data.writeback_index = 0;
 943         memset(&ei->i_cached_extent, 0, sizeof(struct ext4_ext_cache));
 944         INIT_LIST_HEAD(&ei->i_prealloc_list);
 945         spin_lock_init(&ei->i_prealloc_lock);
 946         ei->i_reserved_data_blocks = 0;
 947         ei->i_reserved_meta_blocks = 0;
 948         ei->i_allocated_meta_blocks = 0;
 949         ei->i_da_metadata_calc_len = 0;
 950         ei->i_da_metadata_calc_last_lblock = 0;
 951         spin_lock_init(&(ei->i_block_reservation_lock));
 952 #ifdef CONFIG_QUOTA
 953         ei->i_reserved_quota = 0;
 954 #endif
 955         ei->jinode = NULL;
 956         INIT_LIST_HEAD(&ei->i_completed_io_list);
 957         spin_lock_init(&ei->i_completed_io_lock);
 958         ei->i_sync_tid = 0;
 959         ei->i_datasync_tid = 0;
 960         atomic_set(&ei->i_ioend_count, 0);
 961         atomic_set(&ei->i_unwritten, 0);
 962
 963         return &ei->vfs_inode;
 964 }
 965
 966 static int ext4_drop_inode(struct inode *inode)
 967 {
 968         int drop = generic_drop_inode(inode);
 969
 970         trace_ext4_drop_inode(inode, drop);
 971         return drop;
 972 }
 973
 974 static void ext4_i_callback(struct rcu_head *head)
 975 {
 976         struct inode *inode = container_of(head, struct inode, i_rcu);
 977         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 978 }
 979
 980 static void ext4_destroy_inode(struct inode *inode)
 981 {
 982         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 983                 ext4_msg(inode->i_sb, KERN_ERR,
 984                          "Inode %lu (%p): orphan list check failed!",
 985                          inode->i_ino, EXT4_I(inode));
 986                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 987                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 988                                 true);
 989                 dump_stack();
 990         }
 991         call_rcu(&inode->i_rcu, ext4_i_callback);
 992 }
 993
 994 static void init_once(void *foo)
 995 {
 996         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 997
 998         INIT_LIST_HEAD(&ei->i_orphan);
 999 #ifdef CONFIG_EXT4_FS_XATTR
1000         init_rwsem(&ei->xattr_sem);
1001 #endif
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         if (EXT4_I(inode)->jinode) {
1035                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
1036                                                EXT4_I(inode)->jinode);
1037                 jbd2_free_inode(EXT4_I(inode)->jinode);
1038                 EXT4_I(inode)->jinode = NULL;
1039         }
1040 }
1041
1042 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1043                                         u64 ino, u32 generation)
1044 {
1045         struct inode *inode;
1046
1047         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1048                 return ERR_PTR(-ESTALE);
1049         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1050                 return ERR_PTR(-ESTALE);
1051
1052         /* iget isn't really right if the inode is currently unallocated!!
1053          *
1054          * ext4_read_inode will return a bad_inode if the inode had been
1055          * deleted, so we should be safe.
1056          *
1057          * Currently we don't know the generation for parent directory, so
1058          * a generation of 0 means "accept any"
1059          */
1060         inode = ext4_iget(sb, ino);
1061         if (IS_ERR(inode))
1062                 return ERR_CAST(inode);
1063         if (generation && inode->i_generation != generation) {
1064                 iput(inode);
1065                 return ERR_PTR(-ESTALE);
1066         }
1067
1068         return inode;
1069 }
1070
1071 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1072                                         int fh_len, int fh_type)
1073 {
1074         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1075                                     ext4_nfs_get_inode);
1076 }
1077
1078 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1079                                         int fh_len, int fh_type)
1080 {
1081         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1082                                     ext4_nfs_get_inode);
1083 }
1084
1085 /*
1086  * Try to release metadata pages (indirect blocks, directories) which are
1087  * mapped via the block device.  Since these pages could have journal heads
1088  * which would prevent try_to_free_buffers() from freeing them, we must use
1089  * jbd2 layer's try_to_free_buffers() function to release them.
1090  */
1091 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1092                                  gfp_t wait)
1093 {
1094         journal_t *journal = EXT4_SB(sb)->s_journal;
1095
1096         WARN_ON(PageChecked(page));
1097         if (!page_has_buffers(page))
1098                 return 0;
1099         if (journal)
1100                 return jbd2_journal_try_to_free_buffers(journal, page,
1101                                                         wait & ~__GFP_WAIT);
1102         return try_to_free_buffers(page);
1103 }
1104
1105 #ifdef CONFIG_QUOTA
1106 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1107 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1108
1109 static int ext4_write_dquot(struct dquot *dquot);
1110 static int ext4_acquire_dquot(struct dquot *dquot);
1111 static int ext4_release_dquot(struct dquot *dquot);
1112 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1113 static int ext4_write_info(struct super_block *sb, int type);
1114 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1115                          struct path *path);
1116 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1117                                  int format_id);
1118 static int ext4_quota_off(struct super_block *sb, int type);
1119 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1120 static int ext4_quota_on_mount(struct super_block *sb, int type);
1121 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1122                                size_t len, loff_t off);
1123 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1124                                 const char *data, size_t len, loff_t off);
1125 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1126                              unsigned int flags);
1127 static int ext4_enable_quotas(struct super_block *sb);
1128
1129 static const struct dquot_operations ext4_quota_operations = {
1130         .get_reserved_space = ext4_get_reserved_space,
1131         .write_dquot    = ext4_write_dquot,
1132         .acquire_dquot  = ext4_acquire_dquot,
1133         .release_dquot  = ext4_release_dquot,
1134         .mark_dirty     = ext4_mark_dquot_dirty,
1135         .write_info     = ext4_write_info,
1136         .alloc_dquot    = dquot_alloc,
1137         .destroy_dquot  = dquot_destroy,
1138 };
1139
1140 static const struct quotactl_ops ext4_qctl_operations = {
1141         .quota_on       = ext4_quota_on,
1142         .quota_off      = ext4_quota_off,
1143         .quota_sync     = dquot_quota_sync,
1144         .get_info       = dquot_get_dqinfo,
1145         .set_info       = dquot_set_dqinfo,
1146         .get_dqblk      = dquot_get_dqblk,
1147         .set_dqblk      = dquot_set_dqblk
1148 };
1149
1150 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1151         .quota_on_meta  = ext4_quota_on_sysfile,
1152         .quota_off      = ext4_quota_off_sysfile,
1153         .quota_sync     = dquot_quota_sync,
1154         .get_info       = dquot_get_dqinfo,
1155         .set_info       = dquot_set_dqinfo,
1156         .get_dqblk      = dquot_get_dqblk,
1157         .set_dqblk      = dquot_set_dqblk
1158 };
1159 #endif
1160
1161 static const struct super_operations ext4_sops = {
1162         .alloc_inode    = ext4_alloc_inode,
1163         .destroy_inode  = ext4_destroy_inode,
1164         .write_inode    = ext4_write_inode,
1165         .dirty_inode    = ext4_dirty_inode,
1166         .drop_inode     = ext4_drop_inode,
1167         .evict_inode    = ext4_evict_inode,
1168         .put_super      = ext4_put_super,
1169         .sync_fs        = ext4_sync_fs,
1170         .freeze_fs      = ext4_freeze,
1171         .unfreeze_fs    = ext4_unfreeze,
1172         .statfs         = ext4_statfs,
1173         .remount_fs     = ext4_remount,
1174         .show_options   = ext4_show_options,
1175 #ifdef CONFIG_QUOTA
1176         .quota_read     = ext4_quota_read,
1177         .quota_write    = ext4_quota_write,
1178 #endif
1179         .bdev_try_to_free_page = bdev_try_to_free_page,
1180 };
1181
1182 static const struct super_operations ext4_nojournal_sops = {
1183         .alloc_inode    = ext4_alloc_inode,
1184         .destroy_inode  = ext4_destroy_inode,
1185         .write_inode    = ext4_write_inode,
1186         .dirty_inode    = ext4_dirty_inode,
1187         .drop_inode     = ext4_drop_inode,
1188         .evict_inode    = ext4_evict_inode,
1189         .put_super      = ext4_put_super,
1190         .statfs         = ext4_statfs,
1191         .remount_fs     = ext4_remount,
1192         .show_options   = ext4_show_options,
1193 #ifdef CONFIG_QUOTA
1194         .quota_read     = ext4_quota_read,
1195         .quota_write    = ext4_quota_write,
1196 #endif
1197         .bdev_try_to_free_page = bdev_try_to_free_page,
1198 };
1199
1200 static const struct export_operations ext4_export_ops = {
1201         .fh_to_dentry = ext4_fh_to_dentry,
1202         .fh_to_parent = ext4_fh_to_parent,
1203         .get_parent = ext4_get_parent,
1204 };
1205
1206 enum {
1207         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1208         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1209         Opt_nouid32, Opt_debug, Opt_removed,
1210         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1211         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1212         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1213         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1214         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1215         Opt_data_err_abort, Opt_data_err_ignore,
1216         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1217         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1218         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1219         Opt_usrquota, Opt_grpquota, Opt_i_version,
1220         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1221         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1222         Opt_inode_readahead_blks, Opt_journal_ioprio,
1223         Opt_dioread_nolock, Opt_dioread_lock,
1224         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1225         Opt_max_dir_size_kb,
1226 };
1227
1228 static const match_table_t tokens = {
1229         {Opt_bsd_df, "bsddf"},
1230         {Opt_minix_df, "minixdf"},
1231         {Opt_grpid, "grpid"},
1232         {Opt_grpid, "bsdgroups"},
1233         {Opt_nogrpid, "nogrpid"},
1234         {Opt_nogrpid, "sysvgroups"},
1235         {Opt_resgid, "resgid=%u"},
1236         {Opt_resuid, "resuid=%u"},
1237         {Opt_sb, "sb=%u"},
1238         {Opt_err_cont, "errors=continue"},
1239         {Opt_err_panic, "errors=panic"},
1240         {Opt_err_ro, "errors=remount-ro"},
1241         {Opt_nouid32, "nouid32"},
1242         {Opt_debug, "debug"},
1243         {Opt_removed, "oldalloc"},
1244         {Opt_removed, "orlov"},
1245         {Opt_user_xattr, "user_xattr"},
1246         {Opt_nouser_xattr, "nouser_xattr"},
1247         {Opt_acl, "acl"},
1248         {Opt_noacl, "noacl"},
1249         {Opt_noload, "norecovery"},
1250         {Opt_noload, "noload"},
1251         {Opt_removed, "nobh"},
1252         {Opt_removed, "bh"},
1253         {Opt_commit, "commit=%u"},
1254         {Opt_min_batch_time, "min_batch_time=%u"},
1255         {Opt_max_batch_time, "max_batch_time=%u"},
1256         {Opt_journal_dev, "journal_dev=%u"},
1257         {Opt_journal_checksum, "journal_checksum"},
1258         {Opt_journal_async_commit, "journal_async_commit"},
1259         {Opt_abort, "abort"},
1260         {Opt_data_journal, "data=journal"},
1261         {Opt_data_ordered, "data=ordered"},
1262         {Opt_data_writeback, "data=writeback"},
1263         {Opt_data_err_abort, "data_err=abort"},
1264         {Opt_data_err_ignore, "data_err=ignore"},
1265         {Opt_offusrjquota, "usrjquota="},
1266         {Opt_usrjquota, "usrjquota=%s"},
1267         {Opt_offgrpjquota, "grpjquota="},
1268         {Opt_grpjquota, "grpjquota=%s"},
1269         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1270         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1271         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1272         {Opt_grpquota, "grpquota"},
1273         {Opt_noquota, "noquota"},
1274         {Opt_quota, "quota"},
1275         {Opt_usrquota, "usrquota"},
1276         {Opt_barrier, "barrier=%u"},
1277         {Opt_barrier, "barrier"},
1278         {Opt_nobarrier, "nobarrier"},
1279         {Opt_i_version, "i_version"},
1280         {Opt_stripe, "stripe=%u"},
1281         {Opt_delalloc, "delalloc"},
1282         {Opt_nodelalloc, "nodelalloc"},
1283         {Opt_mblk_io_submit, "mblk_io_submit"},
1284         {Opt_nomblk_io_submit, "nomblk_io_submit"},
1285         {Opt_block_validity, "block_validity"},
1286         {Opt_noblock_validity, "noblock_validity"},
1287         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1288         {Opt_journal_ioprio, "journal_ioprio=%u"},
1289         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1290         {Opt_auto_da_alloc, "auto_da_alloc"},
1291         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1292         {Opt_dioread_nolock, "dioread_nolock"},
1293         {Opt_dioread_lock, "dioread_lock"},
1294         {Opt_discard, "discard"},
1295         {Opt_nodiscard, "nodiscard"},
1296         {Opt_init_itable, "init_itable=%u"},
1297         {Opt_init_itable, "init_itable"},
1298         {Opt_noinit_itable, "noinit_itable"},
1299         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1300         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1301         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1302         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1303         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1304         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1305         {Opt_err, NULL},
1306 };
1307
1308 static ext4_fsblk_t get_sb_block(void **data)
1309 {
1310         ext4_fsblk_t    sb_block;
1311         char            *options = (char *) *data;
1312
1313         if (!options || strncmp(options, "sb=", 3) != 0)
1314                 return 1;       /* Default location */
1315
1316         options += 3;
1317         /* TODO: use simple_strtoll with >32bit ext4 */
1318         sb_block = simple_strtoul(options, &options, 0);
1319         if (*options && *options != ',') {
1320                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1321                        (char *) *data);
1322                 return 1;
1323         }
1324         if (*options == ',')
1325                 options++;
1326         *data = (void *) options;
1327
1328         return sb_block;
1329 }
1330
1331 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1332 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1333         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1334
1335 #ifdef CONFIG_QUOTA
1336 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1337 {
1338         struct ext4_sb_info *sbi = EXT4_SB(sb);
1339         char *qname;
1340
1341         if (sb_any_quota_loaded(sb) &&
1342                 !sbi->s_qf_names[qtype]) {
1343                 ext4_msg(sb, KERN_ERR,
1344                         "Cannot change journaled "
1345                         "quota options when quota turned on");
1346                 return -1;
1347         }
1348         qname = match_strdup(args);
1349         if (!qname) {
1350                 ext4_msg(sb, KERN_ERR,
1351                         "Not enough memory for storing quotafile name");
1352                 return -1;
1353         }
1354         if (sbi->s_qf_names[qtype] &&
1355                 strcmp(sbi->s_qf_names[qtype], qname)) {
1356                 ext4_msg(sb, KERN_ERR,
1357                         "%s quota file already specified", QTYPE2NAME(qtype));
1358                 kfree(qname);
1359                 return -1;
1360         }
1361         sbi->s_qf_names[qtype] = qname;
1362         if (strchr(sbi->s_qf_names[qtype], '/')) {
1363                 ext4_msg(sb, KERN_ERR,
1364                         "quotafile must be on filesystem root");
1365                 kfree(sbi->s_qf_names[qtype]);
1366                 sbi->s_qf_names[qtype] = NULL;
1367                 return -1;
1368         }
1369         set_opt(sb, QUOTA);
1370         return 1;
1371 }
1372
1373 static int clear_qf_name(struct super_block *sb, int qtype)
1374 {
1375
1376         struct ext4_sb_info *sbi = EXT4_SB(sb);
1377
1378         if (sb_any_quota_loaded(sb) &&
1379                 sbi->s_qf_names[qtype]) {
1380                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1381                         " when quota turned on");
1382                 return -1;
1383         }
1384         /*
1385          * The space will be released later when all options are confirmed
1386          * to be correct
1387          */
1388         sbi->s_qf_names[qtype] = NULL;
1389         return 1;
1390 }
1391 #endif
1392
1393 #define MOPT_SET        0x0001
1394 #define MOPT_CLEAR      0x0002
1395 #define MOPT_NOSUPPORT  0x0004
1396 #define MOPT_EXPLICIT   0x0008
1397 #define MOPT_CLEAR_ERR  0x0010
1398 #define MOPT_GTE0       0x0020
1399 #ifdef CONFIG_QUOTA
1400 #define MOPT_Q          0
1401 #define MOPT_QFMT       0x0040
1402 #else
1403 #define MOPT_Q          MOPT_NOSUPPORT
1404 #define MOPT_QFMT       MOPT_NOSUPPORT
1405 #endif
1406 #define MOPT_DATAJ      0x0080
1407
1408 static const struct mount_opts {
1409         int     token;
1410         int     mount_opt;
1411         int     flags;
1412 } ext4_mount_opts[] = {
1413         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1414         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1415         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1416         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1417         {Opt_mblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_SET},
1418         {Opt_nomblk_io_submit, EXT4_MOUNT_MBLK_IO_SUBMIT, MOPT_CLEAR},
1419         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1420         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1421         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_SET},
1422         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK, MOPT_CLEAR},
1423         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1424         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1425         {Opt_delalloc, EXT4_MOUNT_DELALLOC, MOPT_SET | MOPT_EXPLICIT},
1426         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC, MOPT_CLEAR | MOPT_EXPLICIT},
1427         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM, MOPT_SET},
1428         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1429                                     EXT4_MOUNT_JOURNAL_CHECKSUM), MOPT_SET},
1430         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_SET},
1431         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1432         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1433         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1434         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_SET},
1435         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT, MOPT_CLEAR},
1436         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1437         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1438         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1439         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1440         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1441         {Opt_commit, 0, MOPT_GTE0},
1442         {Opt_max_batch_time, 0, MOPT_GTE0},
1443         {Opt_min_batch_time, 0, MOPT_GTE0},
1444         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1445         {Opt_init_itable, 0, MOPT_GTE0},
1446         {Opt_stripe, 0, MOPT_GTE0},
1447         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_DATAJ},
1448         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_DATAJ},
1449         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA, MOPT_DATAJ},
1450 #ifdef CONFIG_EXT4_FS_XATTR
1451         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1452         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1453 #else
1454         {Opt_user_xattr, 0, MOPT_NOSUPPORT},
1455         {Opt_nouser_xattr, 0, MOPT_NOSUPPORT},
1456 #endif
1457 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1458         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1459         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1460 #else
1461         {Opt_acl, 0, MOPT_NOSUPPORT},
1462         {Opt_noacl, 0, MOPT_NOSUPPORT},
1463 #endif
1464         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1465         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1466         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1467         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1468                                                         MOPT_SET | MOPT_Q},
1469         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1470                                                         MOPT_SET | MOPT_Q},
1471         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1472                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1473         {Opt_usrjquota, 0, MOPT_Q},
1474         {Opt_grpjquota, 0, MOPT_Q},
1475         {Opt_offusrjquota, 0, MOPT_Q},
1476         {Opt_offgrpjquota, 0, MOPT_Q},
1477         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1478         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1479         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1480         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1481         {Opt_err, 0, 0}
1482 };
1483
1484 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1485                             substring_t *args, unsigned long *journal_devnum,
1486                             unsigned int *journal_ioprio, int is_remount)
1487 {
1488         struct ext4_sb_info *sbi = EXT4_SB(sb);
1489         const struct mount_opts *m;
1490         kuid_t uid;
1491         kgid_t gid;
1492         int arg = 0;
1493
1494 #ifdef CONFIG_QUOTA
1495         if (token == Opt_usrjquota)
1496                 return set_qf_name(sb, USRQUOTA, &args[0]);
1497         else if (token == Opt_grpjquota)
1498                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1499         else if (token == Opt_offusrjquota)
1500                 return clear_qf_name(sb, USRQUOTA);
1501         else if (token == Opt_offgrpjquota)
1502                 return clear_qf_name(sb, GRPQUOTA);
1503 #endif
1504         if (args->from && match_int(args, &arg))
1505                 return -1;
1506         switch (token) {
1507         case Opt_noacl:
1508         case Opt_nouser_xattr:
1509                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1510                 break;
1511         case Opt_sb:
1512                 return 1;       /* handled by get_sb_block() */
1513         case Opt_removed:
1514                 ext4_msg(sb, KERN_WARNING,
1515                          "Ignoring removed %s option", opt);
1516                 return 1;
1517         case Opt_resuid:
1518                 uid = make_kuid(current_user_ns(), arg);
1519                 if (!uid_valid(uid)) {
1520                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1521                         return -1;
1522                 }
1523                 sbi->s_resuid = uid;
1524                 return 1;
1525         case Opt_resgid:
1526                 gid = make_kgid(current_user_ns(), arg);
1527                 if (!gid_valid(gid)) {
1528                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1529                         return -1;
1530                 }
1531                 sbi->s_resgid = gid;
1532                 return 1;
1533         case Opt_abort:
1534                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1535                 return 1;
1536         case Opt_i_version:
1537                 sb->s_flags |= MS_I_VERSION;
1538                 return 1;
1539         case Opt_journal_dev:
1540                 if (is_remount) {
1541                         ext4_msg(sb, KERN_ERR,
1542                                  "Cannot specify journal on remount");
1543                         return -1;
1544                 }
1545                 *journal_devnum = arg;
1546                 return 1;
1547         case Opt_journal_ioprio:
1548                 if (arg < 0 || arg > 7)
1549                         return -1;
1550                 *journal_ioprio = IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1551                 return 1;
1552         }
1553
1554         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1555                 if (token != m->token)
1556                         continue;
1557                 if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1558                         return -1;
1559                 if (m->flags & MOPT_EXPLICIT)
1560                         set_opt2(sb, EXPLICIT_DELALLOC);
1561                 if (m->flags & MOPT_CLEAR_ERR)
1562                         clear_opt(sb, ERRORS_MASK);
1563                 if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1564                         ext4_msg(sb, KERN_ERR, "Cannot change quota "
1565                                  "options when quota turned on");
1566                         return -1;
1567                 }
1568
1569                 if (m->flags & MOPT_NOSUPPORT) {
1570                         ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1571                 } else if (token == Opt_commit) {
1572                         if (arg == 0)
1573                                 arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1574                         sbi->s_commit_interval = HZ * arg;
1575                 } else if (token == Opt_max_batch_time) {
1576                         if (arg == 0)
1577                                 arg = EXT4_DEF_MAX_BATCH_TIME;
1578                         sbi->s_max_batch_time = arg;
1579                 } else if (token == Opt_min_batch_time) {
1580                         sbi->s_min_batch_time = arg;
1581                 } else if (token == Opt_inode_readahead_blks) {
1582                         if (arg > (1 << 30))
1583                                 return -1;
1584                         if (arg && !is_power_of_2(arg)) {
1585                                 ext4_msg(sb, KERN_ERR,
1586                                          "EXT4-fs: inode_readahead_blks"
1587                                          " must be a power of 2");
1588                                 return -1;
1589                         }
1590                         sbi->s_inode_readahead_blks = arg;
1591                 } else if (token == Opt_init_itable) {
1592                         set_opt(sb, INIT_INODE_TABLE);
1593                         if (!args->from)
1594                                 arg = EXT4_DEF_LI_WAIT_MULT;
1595                         sbi->s_li_wait_mult = arg;
1596                 } else if (token == Opt_max_dir_size_kb) {
1597                         sbi->s_max_dir_size_kb = arg;
1598                 } else if (token == Opt_stripe) {
1599                         sbi->s_stripe = arg;
1600                 } else if (m->flags & MOPT_DATAJ) {
1601                         if (is_remount) {
1602                                 if (!sbi->s_journal)
1603                                         ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1604                                 else if (test_opt(sb, DATA_FLAGS) !=
1605                                          m->mount_opt) {
1606                                         ext4_msg(sb, KERN_ERR,
1607                                          "Cannot change data mode on remount");
1608                                         return -1;
1609                                 }
1610                         } else {
1611                                 clear_opt(sb, DATA_FLAGS);
1612                                 sbi->s_mount_opt |= m->mount_opt;
1613                         }
1614 #ifdef CONFIG_QUOTA
1615                 } else if (m->flags & MOPT_QFMT) {
1616                         if (sb_any_quota_loaded(sb) &&
1617                             sbi->s_jquota_fmt != m->mount_opt) {
1618                                 ext4_msg(sb, KERN_ERR, "Cannot "
1619                                          "change journaled quota options "
1620                                          "when quota turned on");
1621                                 return -1;
1622                         }
1623                         sbi->s_jquota_fmt = m->mount_opt;
1624 #endif
1625                 } else {
1626                         if (!args->from)
1627                                 arg = 1;
1628                         if (m->flags & MOPT_CLEAR)
1629                                 arg = !arg;
1630                         else if (unlikely(!(m->flags & MOPT_SET))) {
1631                                 ext4_msg(sb, KERN_WARNING,
1632                                          "buggy handling of option %s", opt);
1633                                 WARN_ON(1);
1634                                 return -1;
1635                         }
1636                         if (arg != 0)
1637                                 sbi->s_mount_opt |= m->mount_opt;
1638                         else
1639                                 sbi->s_mount_opt &= ~m->mount_opt;
1640                 }
1641                 return 1;
1642         }
1643         ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1644                  "or missing value", opt);
1645         return -1;
1646 }
1647
1648 static int parse_options(char *options, struct super_block *sb,
1649                          unsigned long *journal_devnum,
1650                          unsigned int *journal_ioprio,
1651                          int is_remount)
1652 {
1653 #ifdef CONFIG_QUOTA
1654         struct ext4_sb_info *sbi = EXT4_SB(sb);
1655 #endif
1656         char *p;
1657         substring_t args[MAX_OPT_ARGS];
1658         int token;
1659
1660         if (!options)
1661                 return 1;
1662
1663         while ((p = strsep(&options, ",")) != NULL) {
1664                 if (!*p)
1665                         continue;
1666                 /*
1667                  * Initialize args struct so we know whether arg was
1668                  * found; some options take optional arguments.
1669                  */
1670                 args[0].to = args[0].from = NULL;
1671                 token = match_token(p, tokens, args);
1672                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1673                                      journal_ioprio, is_remount) < 0)
1674                         return 0;
1675         }
1676 #ifdef CONFIG_QUOTA
1677         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1678                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1679                         clear_opt(sb, USRQUOTA);
1680
1681                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1682                         clear_opt(sb, GRPQUOTA);
1683
1684                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1685                         ext4_msg(sb, KERN_ERR, "old and new quota "
1686                                         "format mixing");
1687                         return 0;
1688                 }
1689
1690                 if (!sbi->s_jquota_fmt) {
1691                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1692                                         "not specified");
1693                         return 0;
1694                 }
1695         } else {
1696                 if (sbi->s_jquota_fmt) {
1697                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1698                                         "specified with no journaling "
1699                                         "enabled");
1700                         return 0;
1701                 }
1702         }
1703 #endif
1704         return 1;
1705 }
1706
1707 static inline void ext4_show_quota_options(struct seq_file *seq,
1708                                            struct super_block *sb)
1709 {
1710 #if defined(CONFIG_QUOTA)
1711         struct ext4_sb_info *sbi = EXT4_SB(sb);
1712
1713         if (sbi->s_jquota_fmt) {
1714                 char *fmtname = "";
1715
1716                 switch (sbi->s_jquota_fmt) {
1717                 case QFMT_VFS_OLD:
1718                         fmtname = "vfsold";
1719                         break;
1720                 case QFMT_VFS_V0:
1721                         fmtname = "vfsv0";
1722                         break;
1723                 case QFMT_VFS_V1:
1724                         fmtname = "vfsv1";
1725                         break;
1726                 }
1727                 seq_printf(seq, ",jqfmt=%s", fmtname);
1728         }
1729
1730         if (sbi->s_qf_names[USRQUOTA])
1731                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1732
1733         if (sbi->s_qf_names[GRPQUOTA])
1734                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1735
1736         if (test_opt(sb, USRQUOTA))
1737                 seq_puts(seq, ",usrquota");
1738
1739         if (test_opt(sb, GRPQUOTA))
1740                 seq_puts(seq, ",grpquota");
1741 #endif
1742 }
1743
1744 static const char *token2str(int token)
1745 {
1746         const struct match_token *t;
1747
1748         for (t = tokens; t->token != Opt_err; t++)
1749                 if (t->token == token && !strchr(t->pattern, '='))
1750                         break;
1751         return t->pattern;
1752 }
1753
1754 /*
1755  * Show an option if
1756  *  - it's set to a non-default value OR
1757  *  - if the per-sb default is different from the global default
1758  */
1759 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1760                               int nodefs)
1761 {
1762         struct ext4_sb_info *sbi = EXT4_SB(sb);
1763         struct ext4_super_block *es = sbi->s_es;
1764         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1765         const struct mount_opts *m;
1766         char sep = nodefs ? '\n' : ',';
1767
1768 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1769 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1770
1771         if (sbi->s_sb_block != 1)
1772                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1773
1774         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1775                 int want_set = m->flags & MOPT_SET;
1776                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1777                     (m->flags & MOPT_CLEAR_ERR))
1778                         continue;
1779                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1780                         continue; /* skip if same as the default */
1781                 if ((want_set &&
1782                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1783                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1784                         continue; /* select Opt_noFoo vs Opt_Foo */
1785                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1786         }
1787
1788         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1789             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1790                 SEQ_OPTS_PRINT("resuid=%u",
1791                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1792         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1793             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1794                 SEQ_OPTS_PRINT("resgid=%u",
1795                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1796         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1797         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1798                 SEQ_OPTS_PUTS("errors=remount-ro");
1799         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1800                 SEQ_OPTS_PUTS("errors=continue");
1801         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1802                 SEQ_OPTS_PUTS("errors=panic");
1803         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1804                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1805         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1806                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1807         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1808                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1809         if (sb->s_flags & MS_I_VERSION)
1810                 SEQ_OPTS_PUTS("i_version");
1811         if (nodefs || sbi->s_stripe)
1812                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1813         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1814                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1815                         SEQ_OPTS_PUTS("data=journal");
1816                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1817                         SEQ_OPTS_PUTS("data=ordered");
1818                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1819                         SEQ_OPTS_PUTS("data=writeback");
1820         }
1821         if (nodefs ||
1822             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1823                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1824                                sbi->s_inode_readahead_blks);
1825
1826         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1827                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1828                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1829         if (nodefs || sbi->s_max_dir_size_kb)
1830                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1831
1832         ext4_show_quota_options(seq, sb);
1833         return 0;
1834 }
1835
1836 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1837 {
1838         return _ext4_show_options(seq, root->d_sb, 0);
1839 }
1840
1841 static int options_seq_show(struct seq_file *seq, void *offset)
1842 {
1843         struct super_block *sb = seq->private;
1844         int rc;
1845
1846         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1847         rc = _ext4_show_options(seq, sb, 1);
1848         seq_puts(seq, "\n");
1849         return rc;
1850 }
1851
1852 static int options_open_fs(struct inode *inode, struct file *file)
1853 {
1854         return single_open(file, options_seq_show, PDE(inode)->data);
1855 }
1856
1857 static const struct file_operations ext4_seq_options_fops = {
1858         .owner = THIS_MODULE,
1859         .open = options_open_fs,
1860         .read = seq_read,
1861         .llseek = seq_lseek,
1862         .release = single_release,
1863 };
1864
1865 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1866                             int read_only)
1867 {
1868         struct ext4_sb_info *sbi = EXT4_SB(sb);
1869         int res = 0;
1870
1871         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1872                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1873                          "forcing read-only mode");
1874                 res = MS_RDONLY;
1875         }
1876         if (read_only)
1877                 goto done;
1878         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1879                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1880                          "running e2fsck is recommended");
1881         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1882                 ext4_msg(sb, KERN_WARNING,
1883                          "warning: mounting fs with errors, "
1884                          "running e2fsck is recommended");
1885         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1886                  le16_to_cpu(es->s_mnt_count) >=
1887                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1888                 ext4_msg(sb, KERN_WARNING,
1889                          "warning: maximal mount count reached, "
1890                          "running e2fsck is recommended");
1891         else if (le32_to_cpu(es->s_checkinterval) &&
1892                 (le32_to_cpu(es->s_lastcheck) +
1893                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1894                 ext4_msg(sb, KERN_WARNING,
1895                          "warning: checktime reached, "
1896                          "running e2fsck is recommended");
1897         if (!sbi->s_journal)
1898                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1899         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1900                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1901         le16_add_cpu(&es->s_mnt_count, 1);
1902         es->s_mtime = cpu_to_le32(get_seconds());
1903         ext4_update_dynamic_rev(sb);
1904         if (sbi->s_journal)
1905                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1906
1907         ext4_commit_super(sb, 1);
1908 done:
1909         if (test_opt(sb, DEBUG))
1910                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1911                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1912                         sb->s_blocksize,
1913                         sbi->s_groups_count,
1914                         EXT4_BLOCKS_PER_GROUP(sb),
1915                         EXT4_INODES_PER_GROUP(sb),
1916                         sbi->s_mount_opt, sbi->s_mount_opt2);
1917
1918         cleancache_init_fs(sb);
1919         return res;
1920 }
1921
1922 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1923 {
1924         struct ext4_sb_info *sbi = EXT4_SB(sb);
1925         struct flex_groups *new_groups;
1926         int size;
1927
1928         if (!sbi->s_log_groups_per_flex)
1929                 return 0;
1930
1931         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1932         if (size <= sbi->s_flex_groups_allocated)
1933                 return 0;
1934
1935         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1936         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1937         if (!new_groups) {
1938                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1939                          size / (int) sizeof(struct flex_groups));
1940                 return -ENOMEM;
1941         }
1942
1943         if (sbi->s_flex_groups) {
1944                 memcpy(new_groups, sbi->s_flex_groups,
1945                        (sbi->s_flex_groups_allocated *
1946                         sizeof(struct flex_groups)));
1947                 ext4_kvfree(sbi->s_flex_groups);
1948         }
1949         sbi->s_flex_groups = new_groups;
1950         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1951         return 0;
1952 }
1953
1954 static int ext4_fill_flex_info(struct super_block *sb)
1955 {
1956         struct ext4_sb_info *sbi = EXT4_SB(sb);
1957         struct ext4_group_desc *gdp = NULL;
1958         ext4_group_t flex_group;
1959         unsigned int groups_per_flex = 0;
1960         int i, err;
1961
1962         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1963         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1964                 sbi->s_log_groups_per_flex = 0;
1965                 return 1;
1966         }
1967         groups_per_flex = 1U << sbi->s_log_groups_per_flex;
1968
1969         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1970         if (err)
1971                 goto failed;
1972
1973         for (i = 0; i < sbi->s_groups_count; i++) {
1974                 gdp = ext4_get_group_desc(sb, i, NULL);
1975
1976                 flex_group = ext4_flex_group(sbi, i);
1977                 atomic_add(ext4_free_inodes_count(sb, gdp),
1978                            &sbi->s_flex_groups[flex_group].free_inodes);
1979                 atomic_add(ext4_free_group_clusters(sb, gdp),
1980                            &sbi->s_flex_groups[flex_group].free_clusters);
1981                 atomic_add(ext4_used_dirs_count(sb, gdp),
1982                            &sbi->s_flex_groups[flex_group].used_dirs);
1983         }
1984
1985         return 1;
1986 failed:
1987         return 0;
1988 }
1989
1990 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1991                                    struct ext4_group_desc *gdp)
1992 {
1993         int offset;
1994         __u16 crc = 0;
1995         __le32 le_group = cpu_to_le32(block_group);
1996
1997         if ((sbi->s_es->s_feature_ro_compat &
1998              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1999                 /* Use new metadata_csum algorithm */
2000                 __u16 old_csum;
2001                 __u32 csum32;
2002
2003                 old_csum = gdp->bg_checksum;
2004                 gdp->bg_checksum = 0;
2005                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2006                                      sizeof(le_group));
2007                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2008                                      sbi->s_desc_size);
2009                 gdp->bg_checksum = old_csum;
2010
2011                 crc = csum32 & 0xFFFF;
2012                 goto out;
2013         }
2014
2015         /* old crc16 code */
2016         offset = offsetof(struct ext4_group_desc, bg_checksum);
2017
2018         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2019         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2020         crc = crc16(crc, (__u8 *)gdp, offset);
2021         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2022         /* for checksum of struct ext4_group_desc do the rest...*/
2023         if ((sbi->s_es->s_feature_incompat &
2024              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
2025             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2026                 crc = crc16(crc, (__u8 *)gdp + offset,
2027                             le16_to_cpu(sbi->s_es->s_desc_size) -
2028                                 offset);
2029
2030 out:
2031         return cpu_to_le16(crc);
2032 }
2033
2034 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2035                                 struct ext4_group_desc *gdp)
2036 {
2037         if (ext4_has_group_desc_csum(sb) &&
2038             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2039                                                       block_group, gdp)))
2040                 return 0;
2041
2042         return 1;
2043 }
2044
2045 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2046                               struct ext4_group_desc *gdp)
2047 {
2048         if (!ext4_has_group_desc_csum(sb))
2049                 return;
2050         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2051 }
2052
2053 /* Called at mount-time, super-block is locked */
2054 static int ext4_check_descriptors(struct super_block *sb,
2055                                   ext4_group_t *first_not_zeroed)
2056 {
2057         struct ext4_sb_info *sbi = EXT4_SB(sb);
2058         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2059         ext4_fsblk_t last_block;
2060         ext4_fsblk_t block_bitmap;
2061         ext4_fsblk_t inode_bitmap;
2062         ext4_fsblk_t inode_table;
2063         int flexbg_flag = 0;
2064         ext4_group_t i, grp = sbi->s_groups_count;
2065
2066         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2067                 flexbg_flag = 1;
2068
2069         ext4_debug("Checking group descriptors");
2070
2071         for (i = 0; i < sbi->s_groups_count; i++) {
2072                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2073
2074                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2075                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2076                 else
2077                         last_block = first_block +
2078                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2079
2080                 if ((grp == sbi->s_groups_count) &&
2081                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2082                         grp = i;
2083
2084                 block_bitmap = ext4_block_bitmap(sb, gdp);
2085                 if (block_bitmap < first_block || block_bitmap > last_block) {
2086                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2087                                "Block bitmap for group %u not in group "
2088                                "(block %llu)!", i, block_bitmap);
2089                         return 0;
2090                 }
2091                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2092                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2093                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2094                                "Inode bitmap for group %u not in group "
2095                                "(block %llu)!", i, inode_bitmap);
2096                         return 0;
2097                 }
2098                 inode_table = ext4_inode_table(sb, gdp);
2099                 if (inode_table < first_block ||
2100                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2101                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2102                                "Inode table for group %u not in group "
2103                                "(block %llu)!", i, inode_table);
2104                         return 0;
2105                 }
2106                 ext4_lock_group(sb, i);
2107                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2108                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2109                                  "Checksum for group %u failed (%u!=%u)",
2110                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2111                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2112                         if (!(sb->s_flags & MS_RDONLY)) {
2113                                 ext4_unlock_group(sb, i);
2114                                 return 0;
2115                         }
2116                 }
2117                 ext4_unlock_group(sb, i);
2118                 if (!flexbg_flag)
2119                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2120         }
2121         if (NULL != first_not_zeroed)
2122                 *first_not_zeroed = grp;
2123
2124         ext4_free_blocks_count_set(sbi->s_es,
2125                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2126         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2127         return 1;
2128 }
2129
2130 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2131  * the superblock) which were deleted from all directories, but held open by
2132  * a process at the time of a crash.  We walk the list and try to delete these
2133  * inodes at recovery time (only with a read-write filesystem).
2134  *
2135  * In order to keep the orphan inode chain consistent during traversal (in
2136  * case of crash during recovery), we link each inode into the superblock
2137  * orphan list_head and handle it the same way as an inode deletion during
2138  * normal operation (which journals the operations for us).
2139  *
2140  * We only do an iget() and an iput() on each inode, which is very safe if we
2141  * accidentally point at an in-use or already deleted inode.  The worst that
2142  * can happen in this case is that we get a "bit already cleared" message from
2143  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2144  * e2fsck was run on this filesystem, and it must have already done the orphan
2145  * inode cleanup for us, so we can safely abort without any further action.
2146  */
2147 static void ext4_orphan_cleanup(struct super_block *sb,
2148                                 struct ext4_super_block *es)
2149 {
2150         unsigned int s_flags = sb->s_flags;
2151         int nr_orphans = 0, nr_truncates = 0;
2152 #ifdef CONFIG_QUOTA
2153         int i;
2154 #endif
2155         if (!es->s_last_orphan) {
2156                 jbd_debug(4, "no orphan inodes to clean up\n");
2157                 return;
2158         }
2159
2160         if (bdev_read_only(sb->s_bdev)) {
2161                 ext4_msg(sb, KERN_ERR, "write access "
2162                         "unavailable, skipping orphan cleanup");
2163                 return;
2164         }
2165
2166         /* Check if feature set would not allow a r/w mount */
2167         if (!ext4_feature_set_ok(sb, 0)) {
2168                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2169                          "unknown ROCOMPAT features");
2170                 return;
2171         }
2172
2173         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2174                 /* don't clear list on RO mount w/ errors */
2175                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2176                         jbd_debug(1, "Errors on filesystem, "
2177                                   "clearing orphan list.\n");
2178                         es->s_last_orphan = 0;
2179                 }
2180                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2181                 return;
2182         }
2183
2184         if (s_flags & MS_RDONLY) {
2185                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2186                 sb->s_flags &= ~MS_RDONLY;
2187         }
2188 #ifdef CONFIG_QUOTA
2189         /* Needed for iput() to work correctly and not trash data */
2190         sb->s_flags |= MS_ACTIVE;
2191         /* Turn on quotas so that they are updated correctly */
2192         for (i = 0; i < MAXQUOTAS; i++) {
2193                 if (EXT4_SB(sb)->s_qf_names[i]) {
2194                         int ret = ext4_quota_on_mount(sb, i);
2195                         if (ret < 0)
2196                                 ext4_msg(sb, KERN_ERR,
2197                                         "Cannot turn on journaled "
2198                                         "quota: error %d", ret);
2199                 }
2200         }
2201 #endif
2202
2203         while (es->s_last_orphan) {
2204                 struct inode *inode;
2205
2206                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2207                 if (IS_ERR(inode)) {
2208                         es->s_last_orphan = 0;
2209                         break;
2210                 }
2211
2212                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2213                 dquot_initialize(inode);
2214                 if (inode->i_nlink) {
2215                         ext4_msg(sb, KERN_DEBUG,
2216                                 "%s: truncating inode %lu to %lld bytes",
2217                                 __func__, inode->i_ino, inode->i_size);
2218                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2219                                   inode->i_ino, inode->i_size);
2220                         ext4_truncate(inode);
2221                         nr_truncates++;
2222                 } else {
2223                         ext4_msg(sb, KERN_DEBUG,
2224                                 "%s: deleting unreferenced inode %lu",
2225                                 __func__, inode->i_ino);
2226                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2227                                   inode->i_ino);
2228                         nr_orphans++;
2229                 }
2230                 iput(inode);  /* The delete magic happens here! */
2231         }
2232
2233 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2234
2235         if (nr_orphans)
2236                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2237                        PLURAL(nr_orphans));
2238         if (nr_truncates)
2239                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2240                        PLURAL(nr_truncates));
2241 #ifdef CONFIG_QUOTA
2242         /* Turn quotas off */
2243         for (i = 0; i < MAXQUOTAS; i++) {
2244                 if (sb_dqopt(sb)->files[i])
2245                         dquot_quota_off(sb, i);
2246         }
2247 #endif
2248         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2249 }
2250
2251 /*
2252  * Maximal extent format file size.
2253  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2254  * extent format containers, within a sector_t, and within i_blocks
2255  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2256  * so that won't be a limiting factor.
2257  *
2258  * However there is other limiting factor. We do store extents in the form
2259  * of starting block and length, hence the resulting length of the extent
2260  * covering maximum file size must fit into on-disk format containers as
2261  * well. Given that length is always by 1 unit bigger than max unit (because
2262  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2263  *
2264  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2265  */
2266 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2267 {
2268         loff_t res;
2269         loff_t upper_limit = MAX_LFS_FILESIZE;
2270
2271         /* small i_blocks in vfs inode? */
2272         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2273                 /*
2274                  * CONFIG_LBDAF is not enabled implies the inode
2275                  * i_block represent total blocks in 512 bytes
2276                  * 32 == size of vfs inode i_blocks * 8
2277                  */
2278                 upper_limit = (1LL << 32) - 1;
2279
2280                 /* total blocks in file system block size */
2281                 upper_limit >>= (blkbits - 9);
2282                 upper_limit <<= blkbits;
2283         }
2284
2285         /*
2286          * 32-bit extent-start container, ee_block. We lower the maxbytes
2287          * by one fs block, so ee_len can cover the extent of maximum file
2288          * size
2289          */
2290         res = (1LL << 32) - 1;
2291         res <<= blkbits;
2292
2293         /* Sanity check against vm- & vfs- imposed limits */
2294         if (res > upper_limit)
2295                 res = upper_limit;
2296
2297         return res;
2298 }
2299
2300 /*
2301  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2302  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2303  * We need to be 1 filesystem block less than the 2^48 sector limit.
2304  */
2305 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2306 {
2307         loff_t res = EXT4_NDIR_BLOCKS;
2308         int meta_blocks;
2309         loff_t upper_limit;
2310         /* This is calculated to be the largest file size for a dense, block
2311          * mapped file such that the file's total number of 512-byte sectors,
2312          * including data and all indirect blocks, does not exceed (2^48 - 1).
2313          *
2314          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2315          * number of 512-byte sectors of the file.
2316          */
2317
2318         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2319                 /*
2320                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2321                  * the inode i_block field represents total file blocks in
2322                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2323                  */
2324                 upper_limit = (1LL << 32) - 1;
2325
2326                 /* total blocks in file system block size */
2327                 upper_limit >>= (bits - 9);
2328
2329         } else {
2330                 /*
2331                  * We use 48 bit ext4_inode i_blocks
2332                  * With EXT4_HUGE_FILE_FL set the i_blocks
2333                  * represent total number of blocks in
2334                  * file system block size
2335                  */
2336                 upper_limit = (1LL << 48) - 1;
2337
2338         }
2339
2340         /* indirect blocks */
2341         meta_blocks = 1;
2342         /* double indirect blocks */
2343         meta_blocks += 1 + (1LL << (bits-2));
2344         /* tripple indirect blocks */
2345         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2346
2347         upper_limit -= meta_blocks;
2348         upper_limit <<= bits;
2349
2350         res += 1LL << (bits-2);
2351         res += 1LL << (2*(bits-2));
2352         res += 1LL << (3*(bits-2));
2353         res <<= bits;
2354         if (res > upper_limit)
2355                 res = upper_limit;
2356
2357         if (res > MAX_LFS_FILESIZE)
2358                 res = MAX_LFS_FILESIZE;
2359
2360         return res;
2361 }
2362
2363 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2364                                    ext4_fsblk_t logical_sb_block, int nr)
2365 {
2366         struct ext4_sb_info *sbi = EXT4_SB(sb);
2367         ext4_group_t bg, first_meta_bg;
2368         int has_super = 0;
2369
2370         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2371
2372         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2373             nr < first_meta_bg)
2374                 return logical_sb_block + nr + 1;
2375         bg = sbi->s_desc_per_block * nr;
2376         if (ext4_bg_has_super(sb, bg))
2377                 has_super = 1;
2378
2379         return (has_super + ext4_group_first_block_no(sb, bg));
2380 }
2381
2382 /**
2383  * ext4_get_stripe_size: Get the stripe size.
2384  * @sbi: In memory super block info
2385  *
2386  * If we have specified it via mount option, then
2387  * use the mount option value. If the value specified at mount time is
2388  * greater than the blocks per group use the super block value.
2389  * If the super block value is greater than blocks per group return 0.
2390  * Allocator needs it be less than blocks per group.
2391  *
2392  */
2393 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2394 {
2395         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2396         unsigned long stripe_width =
2397                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2398         int ret;
2399
2400         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2401                 ret = sbi->s_stripe;
2402         else if (stripe_width <= sbi->s_blocks_per_group)
2403                 ret = stripe_width;
2404         else if (stride <= sbi->s_blocks_per_group)
2405                 ret = stride;
2406         else
2407                 ret = 0;
2408
2409         /*
2410          * If the stripe width is 1, this makes no sense and
2411          * we set it to 0 to turn off stripe handling code.
2412          */
2413         if (ret <= 1)
2414                 ret = 0;
2415
2416         return ret;
2417 }
2418
2419 /* sysfs supprt */
2420
2421 struct ext4_attr {
2422         struct attribute attr;
2423         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2424         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2425                          const char *, size_t);
2426         int offset;
2427 };
2428
2429 static int parse_strtoul(const char *buf,
2430                 unsigned long max, unsigned long *value)
2431 {
2432         char *endp;
2433
2434         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2435         endp = skip_spaces(endp);
2436         if (*endp || *value > max)
2437                 return -EINVAL;
2438
2439         return 0;
2440 }
2441
2442 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2443                                               struct ext4_sb_info *sbi,
2444                                               char *buf)
2445 {
2446         return snprintf(buf, PAGE_SIZE, "%llu\n",
2447                 (s64) EXT4_C2B(sbi,
2448                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2449 }
2450
2451 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2452                                          struct ext4_sb_info *sbi, char *buf)
2453 {
2454         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2455
2456         if (!sb->s_bdev->bd_part)
2457                 return snprintf(buf, PAGE_SIZE, "0\n");
2458         return snprintf(buf, PAGE_SIZE, "%lu\n",
2459                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2460                          sbi->s_sectors_written_start) >> 1);
2461 }
2462
2463 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2464                                           struct ext4_sb_info *sbi, char *buf)
2465 {
2466         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2467
2468         if (!sb->s_bdev->bd_part)
2469                 return snprintf(buf, PAGE_SIZE, "0\n");
2470         return snprintf(buf, PAGE_SIZE, "%llu\n",
2471                         (unsigned long long)(sbi->s_kbytes_written +
2472                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2473                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2474 }
2475
2476 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2477                                           struct ext4_sb_info *sbi,
2478                                           const char *buf, size_t count)
2479 {
2480         unsigned long t;
2481
2482         if (parse_strtoul(buf, 0x40000000, &t))
2483                 return -EINVAL;
2484
2485         if (t && !is_power_of_2(t))
2486                 return -EINVAL;
2487
2488         sbi->s_inode_readahead_blks = t;
2489         return count;
2490 }
2491
2492 static ssize_t sbi_ui_show(struct ext4_attr *a,
2493                            struct ext4_sb_info *sbi, char *buf)
2494 {
2495         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2496
2497         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2498 }
2499
2500 static ssize_t sbi_ui_store(struct ext4_attr *a,
2501                             struct ext4_sb_info *sbi,
2502                             const char *buf, size_t count)
2503 {
2504         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2505         unsigned long t;
2506
2507         if (parse_strtoul(buf, 0xffffffff, &t))
2508                 return -EINVAL;
2509         *ui = t;
2510         return count;
2511 }
2512
2513 static ssize_t trigger_test_error(struct ext4_attr *a,
2514                                   struct ext4_sb_info *sbi,
2515                                   const char *buf, size_t count)
2516 {
2517         int len = count;
2518
2519         if (!capable(CAP_SYS_ADMIN))
2520                 return -EPERM;
2521
2522         if (len && buf[len-1] == '\n')
2523                 len--;
2524
2525         if (len)
2526                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2527         return count;
2528 }
2529
2530 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2531 static struct ext4_attr ext4_attr_##_name = {                   \
2532         .attr = {.name = __stringify(_name), .mode = _mode },   \
2533         .show   = _show,                                        \
2534         .store  = _store,                                       \
2535         .offset = offsetof(struct ext4_sb_info, _elname),       \
2536 }
2537 #define EXT4_ATTR(name, mode, show, store) \
2538 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2539
2540 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2541 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2542 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2543 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2544         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2545 #define ATTR_LIST(name) &ext4_attr_##name.attr
2546
2547 EXT4_RO_ATTR(delayed_allocation_blocks);
2548 EXT4_RO_ATTR(session_write_kbytes);
2549 EXT4_RO_ATTR(lifetime_write_kbytes);
2550 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2551                  inode_readahead_blks_store, s_inode_readahead_blks);
2552 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2553 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2554 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2555 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2556 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2557 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2558 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2559 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2560 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2561 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2562
2563 static struct attribute *ext4_attrs[] = {
2564         ATTR_LIST(delayed_allocation_blocks),
2565         ATTR_LIST(session_write_kbytes),
2566         ATTR_LIST(lifetime_write_kbytes),
2567         ATTR_LIST(inode_readahead_blks),
2568         ATTR_LIST(inode_goal),
2569         ATTR_LIST(mb_stats),
2570         ATTR_LIST(mb_max_to_scan),
2571         ATTR_LIST(mb_min_to_scan),
2572         ATTR_LIST(mb_order2_req),
2573         ATTR_LIST(mb_stream_req),
2574         ATTR_LIST(mb_group_prealloc),
2575         ATTR_LIST(max_writeback_mb_bump),
2576         ATTR_LIST(extent_max_zeroout_kb),
2577         ATTR_LIST(trigger_fs_error),
2578         NULL,
2579 };
2580
2581 /* Features this copy of ext4 supports */
2582 EXT4_INFO_ATTR(lazy_itable_init);
2583 EXT4_INFO_ATTR(batched_discard);
2584 EXT4_INFO_ATTR(meta_bg_resize);
2585
2586 static struct attribute *ext4_feat_attrs[] = {
2587         ATTR_LIST(lazy_itable_init),
2588         ATTR_LIST(batched_discard),
2589         ATTR_LIST(meta_bg_resize),
2590         NULL,
2591 };
2592
2593 static ssize_t ext4_attr_show(struct kobject *kobj,
2594                               struct attribute *attr, char *buf)
2595 {
2596         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2597                                                 s_kobj);
2598         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2599
2600         return a->show ? a->show(a, sbi, buf) : 0;
2601 }
2602
2603 static ssize_t ext4_attr_store(struct kobject *kobj,
2604                                struct attribute *attr,
2605                                const char *buf, size_t len)
2606 {
2607         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2608                                                 s_kobj);
2609         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2610
2611         return a->store ? a->store(a, sbi, buf, len) : 0;
2612 }
2613
2614 static void ext4_sb_release(struct kobject *kobj)
2615 {
2616         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2617                                                 s_kobj);
2618         complete(&sbi->s_kobj_unregister);
2619 }
2620
2621 static const struct sysfs_ops ext4_attr_ops = {
2622         .show   = ext4_attr_show,
2623         .store  = ext4_attr_store,
2624 };
2625
2626 static struct kobj_type ext4_ktype = {
2627         .default_attrs  = ext4_attrs,
2628         .sysfs_ops      = &ext4_attr_ops,
2629         .release        = ext4_sb_release,
2630 };
2631
2632 static void ext4_feat_release(struct kobject *kobj)
2633 {
2634         complete(&ext4_feat->f_kobj_unregister);
2635 }
2636
2637 static struct kobj_type ext4_feat_ktype = {
2638         .default_attrs  = ext4_feat_attrs,
2639         .sysfs_ops      = &ext4_attr_ops,
2640         .release        = ext4_feat_release,
2641 };
2642
2643 /*
2644  * Check whether this filesystem can be mounted based on
2645  * the features present and the RDONLY/RDWR mount requested.
2646  * Returns 1 if this filesystem can be mounted as requested,
2647  * 0 if it cannot be.
2648  */
2649 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2650 {
2651         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2652                 ext4_msg(sb, KERN_ERR,
2653                         "Couldn't mount because of "
2654                         "unsupported optional features (%x)",
2655                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2656                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2657                 return 0;
2658         }
2659
2660         if (readonly)
2661                 return 1;
2662
2663         /* Check that feature set is OK for a read-write mount */
2664         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2665                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2666                          "unsupported optional features (%x)",
2667                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2668                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2669                 return 0;
2670         }
2671         /*
2672          * Large file size enabled file system can only be mounted
2673          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2674          */
2675         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2676                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2677                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2678                                  "cannot be mounted RDWR without "
2679                                  "CONFIG_LBDAF");
2680                         return 0;
2681                 }
2682         }
2683         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2684             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2685                 ext4_msg(sb, KERN_ERR,
2686                          "Can't support bigalloc feature without "
2687                          "extents feature\n");
2688                 return 0;
2689         }
2690
2691 #ifndef CONFIG_QUOTA
2692         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2693             !readonly) {
2694                 ext4_msg(sb, KERN_ERR,
2695                          "Filesystem with quota feature cannot be mounted RDWR "
2696                          "without CONFIG_QUOTA");
2697                 return 0;
2698         }
2699 #endif  /* CONFIG_QUOTA */
2700         return 1;
2701 }
2702
2703 /*
2704  * This function is called once a day if we have errors logged
2705  * on the file system
2706  */
2707 static void print_daily_error_info(unsigned long arg)
2708 {
2709         struct super_block *sb = (struct super_block *) arg;
2710         struct ext4_sb_info *sbi;
2711         struct ext4_super_block *es;
2712
2713         sbi = EXT4_SB(sb);
2714         es = sbi->s_es;
2715
2716         if (es->s_error_count)
2717                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2718                          le32_to_cpu(es->s_error_count));
2719         if (es->s_first_error_time) {
2720                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2721                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2722                        (int) sizeof(es->s_first_error_func),
2723                        es->s_first_error_func,
2724                        le32_to_cpu(es->s_first_error_line));
2725                 if (es->s_first_error_ino)
2726                         printk(": inode %u",
2727                                le32_to_cpu(es->s_first_error_ino));
2728                 if (es->s_first_error_block)
2729                         printk(": block %llu", (unsigned long long)
2730                                le64_to_cpu(es->s_first_error_block));
2731                 printk("\n");
2732         }
2733         if (es->s_last_error_time) {
2734                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2735                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2736                        (int) sizeof(es->s_last_error_func),
2737                        es->s_last_error_func,
2738                        le32_to_cpu(es->s_last_error_line));
2739                 if (es->s_last_error_ino)
2740                         printk(": inode %u",
2741                                le32_to_cpu(es->s_last_error_ino));
2742                 if (es->s_last_error_block)
2743                         printk(": block %llu", (unsigned long long)
2744                                le64_to_cpu(es->s_last_error_block));
2745                 printk("\n");
2746         }
2747         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2748 }
2749
2750 /* Find next suitable group and run ext4_init_inode_table */
2751 static int ext4_run_li_request(struct ext4_li_request *elr)
2752 {
2753         struct ext4_group_desc *gdp = NULL;
2754         ext4_group_t group, ngroups;
2755         struct super_block *sb;
2756         unsigned long timeout = 0;
2757         int ret = 0;
2758
2759         sb = elr->lr_super;
2760         ngroups = EXT4_SB(sb)->s_groups_count;
2761
2762         sb_start_write(sb);
2763         for (group = elr->lr_next_group; group < ngroups; group++) {
2764                 gdp = ext4_get_group_desc(sb, group, NULL);
2765                 if (!gdp) {
2766                         ret = 1;
2767                         break;
2768                 }
2769
2770                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2771                         break;
2772         }
2773
2774         if (group == ngroups)
2775                 ret = 1;
2776
2777         if (!ret) {
2778                 timeout = jiffies;
2779                 ret = ext4_init_inode_table(sb, group,
2780                                             elr->lr_timeout ? 0 : 1);
2781                 if (elr->lr_timeout == 0) {
2782                         timeout = (jiffies - timeout) *
2783                                   elr->lr_sbi->s_li_wait_mult;
2784                         elr->lr_timeout = timeout;
2785                 }
2786                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2787                 elr->lr_next_group = group + 1;
2788         }
2789         sb_end_write(sb);
2790
2791         return ret;
2792 }
2793
2794 /*
2795  * Remove lr_request from the list_request and free the
2796  * request structure. Should be called with li_list_mtx held
2797  */
2798 static void ext4_remove_li_request(struct ext4_li_request *elr)
2799 {
2800         struct ext4_sb_info *sbi;
2801
2802         if (!elr)
2803                 return;
2804
2805         sbi = elr->lr_sbi;
2806
2807         list_del(&elr->lr_request);
2808         sbi->s_li_request = NULL;
2809         kfree(elr);
2810 }
2811
2812 static void ext4_unregister_li_request(struct super_block *sb)
2813 {
2814         mutex_lock(&ext4_li_mtx);
2815         if (!ext4_li_info) {
2816                 mutex_unlock(&ext4_li_mtx);
2817                 return;
2818         }
2819
2820         mutex_lock(&ext4_li_info->li_list_mtx);
2821         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2822         mutex_unlock(&ext4_li_info->li_list_mtx);
2823         mutex_unlock(&ext4_li_mtx);
2824 }
2825
2826 static struct task_struct *ext4_lazyinit_task;
2827
2828 /*
2829  * This is the function where ext4lazyinit thread lives. It walks
2830  * through the request list searching for next scheduled filesystem.
2831  * When such a fs is found, run the lazy initialization request
2832  * (ext4_rn_li_request) and keep track of the time spend in this
2833  * function. Based on that time we compute next schedule time of
2834  * the request. When walking through the list is complete, compute
2835  * next waking time and put itself into sleep.
2836  */
2837 static int ext4_lazyinit_thread(void *arg)
2838 {
2839         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2840         struct list_head *pos, *n;
2841         struct ext4_li_request *elr;
2842         unsigned long next_wakeup, cur;
2843
2844         BUG_ON(NULL == eli);
2845
2846 cont_thread:
2847         while (true) {
2848                 next_wakeup = MAX_JIFFY_OFFSET;
2849
2850                 mutex_lock(&eli->li_list_mtx);
2851                 if (list_empty(&eli->li_request_list)) {
2852                         mutex_unlock(&eli->li_list_mtx);
2853                         goto exit_thread;
2854                 }
2855
2856                 list_for_each_safe(pos, n, &eli->li_request_list) {
2857                         elr = list_entry(pos, struct ext4_li_request,
2858                                          lr_request);
2859
2860                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2861                                 if (ext4_run_li_request(elr) != 0) {
2862                                         /* error, remove the lazy_init job */
2863                                         ext4_remove_li_request(elr);
2864                                         continue;
2865                                 }
2866                         }
2867
2868                         if (time_before(elr->lr_next_sched, next_wakeup))
2869                                 next_wakeup = elr->lr_next_sched;
2870                 }
2871                 mutex_unlock(&eli->li_list_mtx);
2872
2873                 try_to_freeze();
2874
2875                 cur = jiffies;
2876                 if ((time_after_eq(cur, next_wakeup)) ||
2877                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2878                         cond_resched();
2879                         continue;
2880                 }
2881
2882                 schedule_timeout_interruptible(next_wakeup - cur);
2883
2884                 if (kthread_should_stop()) {
2885                         ext4_clear_request_list();
2886                         goto exit_thread;
2887                 }
2888         }
2889
2890 exit_thread:
2891         /*
2892          * It looks like the request list is empty, but we need
2893          * to check it under the li_list_mtx lock, to prevent any
2894          * additions into it, and of course we should lock ext4_li_mtx
2895          * to atomically free the list and ext4_li_info, because at
2896          * this point another ext4 filesystem could be registering
2897          * new one.
2898          */
2899         mutex_lock(&ext4_li_mtx);
2900         mutex_lock(&eli->li_list_mtx);
2901         if (!list_empty(&eli->li_request_list)) {
2902                 mutex_unlock(&eli->li_list_mtx);
2903                 mutex_unlock(&ext4_li_mtx);
2904                 goto cont_thread;
2905         }
2906         mutex_unlock(&eli->li_list_mtx);
2907         kfree(ext4_li_info);
2908         ext4_li_info = NULL;
2909         mutex_unlock(&ext4_li_mtx);
2910
2911         return 0;
2912 }
2913
2914 static void ext4_clear_request_list(void)
2915 {
2916         struct list_head *pos, *n;
2917         struct ext4_li_request *elr;
2918
2919         mutex_lock(&ext4_li_info->li_list_mtx);
2920         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2921                 elr = list_entry(pos, struct ext4_li_request,
2922                                  lr_request);
2923                 ext4_remove_li_request(elr);
2924         }
2925         mutex_unlock(&ext4_li_info->li_list_mtx);
2926 }
2927
2928 static int ext4_run_lazyinit_thread(void)
2929 {
2930         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2931                                          ext4_li_info, "ext4lazyinit");
2932         if (IS_ERR(ext4_lazyinit_task)) {
2933                 int err = PTR_ERR(ext4_lazyinit_task);
2934                 ext4_clear_request_list();
2935                 kfree(ext4_li_info);
2936                 ext4_li_info = NULL;
2937                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2938                                  "initialization thread\n",
2939                                  err);
2940                 return err;
2941         }
2942         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2943         return 0;
2944 }
2945
2946 /*
2947  * Check whether it make sense to run itable init. thread or not.
2948  * If there is at least one uninitialized inode table, return
2949  * corresponding group number, else the loop goes through all
2950  * groups and return total number of groups.
2951  */
2952 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2953 {
2954         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2955         struct ext4_group_desc *gdp = NULL;
2956
2957         for (group = 0; group < ngroups; group++) {
2958                 gdp = ext4_get_group_desc(sb, group, NULL);
2959                 if (!gdp)
2960                         continue;
2961
2962                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2963                         break;
2964         }
2965
2966         return group;
2967 }
2968
2969 static int ext4_li_info_new(void)
2970 {
2971         struct ext4_lazy_init *eli = NULL;
2972
2973         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2974         if (!eli)
2975                 return -ENOMEM;
2976
2977         INIT_LIST_HEAD(&eli->li_request_list);
2978         mutex_init(&eli->li_list_mtx);
2979
2980         eli->li_state |= EXT4_LAZYINIT_QUIT;
2981
2982         ext4_li_info = eli;
2983
2984         return 0;
2985 }
2986
2987 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2988                                             ext4_group_t start)
2989 {
2990         struct ext4_sb_info *sbi = EXT4_SB(sb);
2991         struct ext4_li_request *elr;
2992         unsigned long rnd;
2993
2994         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2995         if (!elr)
2996                 return NULL;
2997
2998         elr->lr_super = sb;
2999         elr->lr_sbi = sbi;
3000         elr->lr_next_group = start;
3001
3002         /*
3003          * Randomize first schedule time of the request to
3004          * spread the inode table initialization requests
3005          * better.
3006          */
3007         get_random_bytes(&rnd, sizeof(rnd));
3008         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3009                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3010
3011         return elr;
3012 }
3013
3014 static int ext4_register_li_request(struct super_block *sb,
3015                                     ext4_group_t first_not_zeroed)
3016 {
3017         struct ext4_sb_info *sbi = EXT4_SB(sb);
3018         struct ext4_li_request *elr;
3019         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3020         int ret = 0;
3021
3022         if (sbi->s_li_request != NULL) {
3023                 /*
3024                  * Reset timeout so it can be computed again, because
3025                  * s_li_wait_mult might have changed.
3026                  */
3027                 sbi->s_li_request->lr_timeout = 0;
3028                 return 0;
3029         }
3030
3031         if (first_not_zeroed == ngroups ||
3032             (sb->s_flags & MS_RDONLY) ||
3033             !test_opt(sb, INIT_INODE_TABLE))
3034                 return 0;
3035
3036         elr = ext4_li_request_new(sb, first_not_zeroed);
3037         if (!elr)
3038                 return -ENOMEM;
3039
3040         mutex_lock(&ext4_li_mtx);
3041
3042         if (NULL == ext4_li_info) {
3043                 ret = ext4_li_info_new();
3044                 if (ret)
3045                         goto out;
3046         }
3047
3048         mutex_lock(&ext4_li_info->li_list_mtx);
3049         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3050         mutex_unlock(&ext4_li_info->li_list_mtx);
3051
3052         sbi->s_li_request = elr;
3053         /*
3054          * set elr to NULL here since it has been inserted to
3055          * the request_list and the removal and free of it is
3056          * handled by ext4_clear_request_list from now on.
3057          */
3058         elr = NULL;
3059
3060         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3061                 ret = ext4_run_lazyinit_thread();
3062                 if (ret)
3063                         goto out;
3064         }
3065 out:
3066         mutex_unlock(&ext4_li_mtx);
3067         if (ret)
3068                 kfree(elr);
3069         return ret;
3070 }
3071
3072 /*
3073  * We do not need to lock anything since this is called on
3074  * module unload.
3075  */
3076 static void ext4_destroy_lazyinit_thread(void)
3077 {
3078         /*
3079          * If thread exited earlier
3080          * there's nothing to be done.
3081          */
3082         if (!ext4_li_info || !ext4_lazyinit_task)
3083                 return;
3084
3085         kthread_stop(ext4_lazyinit_task);
3086 }
3087
3088 static int set_journal_csum_feature_set(struct super_block *sb)
3089 {
3090         int ret = 1;
3091         int compat, incompat;
3092         struct ext4_sb_info *sbi = EXT4_SB(sb);
3093
3094         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3095                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3096                 /* journal checksum v2 */
3097                 compat = 0;
3098                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3099         } else {
3100                 /* journal checksum v1 */
3101                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3102                 incompat = 0;
3103         }
3104
3105         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3106                 ret = jbd2_journal_set_features(sbi->s_journal,
3107                                 compat, 0,
3108                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3109                                 incompat);
3110         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3111                 ret = jbd2_journal_set_features(sbi->s_journal,
3112                                 compat, 0,
3113                                 incompat);
3114                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3115                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3116         } else {
3117                 jbd2_journal_clear_features(sbi->s_journal,
3118                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3119                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3120                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3121         }
3122
3123         return ret;
3124 }
3125
3126 /*
3127  * Note: calculating the overhead so we can be compatible with
3128  * historical BSD practice is quite difficult in the face of
3129  * clusters/bigalloc.  This is because multiple metadata blocks from
3130  * different block group can end up in the same allocation cluster.
3131  * Calculating the exact overhead in the face of clustered allocation
3132  * requires either O(all block bitmaps) in memory or O(number of block
3133  * groups**2) in time.  We will still calculate the superblock for
3134  * older file systems --- and if we come across with a bigalloc file
3135  * system with zero in s_overhead_clusters the estimate will be close to
3136  * correct especially for very large cluster sizes --- but for newer
3137  * file systems, it's better to calculate this figure once at mkfs
3138  * time, and store it in the superblock.  If the superblock value is
3139  * present (even for non-bigalloc file systems), we will use it.
3140  */
3141 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3142                           char *buf)
3143 {
3144         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3145         struct ext4_group_desc  *gdp;
3146         ext4_fsblk_t            first_block, last_block, b;
3147         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3148         int                     s, j, count = 0;
3149
3150         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3151                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3152                         sbi->s_itb_per_group + 2);
3153
3154         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3155                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3156         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3157         for (i = 0; i < ngroups; i++) {
3158                 gdp = ext4_get_group_desc(sb, i, NULL);
3159                 b = ext4_block_bitmap(sb, gdp);
3160                 if (b >= first_block && b <= last_block) {
3161                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3162                         count++;
3163                 }
3164                 b = ext4_inode_bitmap(sb, gdp);
3165                 if (b >= first_block && b <= last_block) {
3166                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3167                         count++;
3168                 }
3169                 b = ext4_inode_table(sb, gdp);
3170                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3171                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3172                                 int c = EXT4_B2C(sbi, b - first_block);
3173                                 ext4_set_bit(c, buf);
3174                                 count++;
3175                         }
3176                 if (i != grp)
3177                         continue;
3178                 s = 0;
3179                 if (ext4_bg_has_super(sb, grp)) {
3180                         ext4_set_bit(s++, buf);
3181                         count++;
3182                 }
3183                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3184                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3185                         count++;
3186                 }
3187         }
3188         if (!count)
3189                 return 0;
3190         return EXT4_CLUSTERS_PER_GROUP(sb) -
3191                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3192 }
3193
3194 /*
3195  * Compute the overhead and stash it in sbi->s_overhead
3196  */
3197 int ext4_calculate_overhead(struct super_block *sb)
3198 {
3199         struct ext4_sb_info *sbi = EXT4_SB(sb);
3200         struct ext4_super_block *es = sbi->s_es;
3201         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3202         ext4_fsblk_t overhead = 0;
3203         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3204
3205         memset(buf, 0, PAGE_SIZE);
3206         if (!buf)
3207                 return -ENOMEM;
3208
3209         /*
3210          * Compute the overhead (FS structures).  This is constant
3211          * for a given filesystem unless the number of block groups
3212          * changes so we cache the previous value until it does.
3213          */
3214
3215         /*
3216          * All of the blocks before first_data_block are overhead
3217          */
3218         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3219
3220         /*
3221          * Add the overhead found in each block group
3222          */
3223         for (i = 0; i < ngroups; i++) {
3224                 int blks;
3225
3226                 blks = count_overhead(sb, i, buf);
3227                 overhead += blks;
3228                 if (blks)
3229                         memset(buf, 0, PAGE_SIZE);
3230                 cond_resched();
3231         }
3232         sbi->s_overhead = overhead;
3233         smp_wmb();
3234         free_page((unsigned long) buf);
3235         return 0;
3236 }
3237
3238 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3239 {
3240         char *orig_data = kstrdup(data, GFP_KERNEL);
3241         struct buffer_head *bh;
3242         struct ext4_super_block *es = NULL;
3243         struct ext4_sb_info *sbi;
3244         ext4_fsblk_t block;
3245         ext4_fsblk_t sb_block = get_sb_block(&data);
3246         ext4_fsblk_t logical_sb_block;
3247         unsigned long offset = 0;
3248         unsigned long journal_devnum = 0;
3249         unsigned long def_mount_opts;
3250         struct inode *root;
3251         char *cp;
3252         const char *descr;
3253         int ret = -ENOMEM;
3254         int blocksize, clustersize;
3255         unsigned int db_count;
3256         unsigned int i;
3257         int needs_recovery, has_huge_files, has_bigalloc;
3258         __u64 blocks_count;
3259         int err;
3260         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3261         ext4_group_t first_not_zeroed;
3262
3263         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3264         if (!sbi)
3265                 goto out_free_orig;
3266
3267         sbi->s_blockgroup_lock =
3268                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3269         if (!sbi->s_blockgroup_lock) {
3270                 kfree(sbi);
3271                 goto out_free_orig;
3272         }
3273         sb->s_fs_info = sbi;
3274         sbi->s_sb = sb;
3275         sbi->s_mount_opt = 0;
3276         sbi->s_resuid = make_kuid(&init_user_ns, EXT4_DEF_RESUID);
3277         sbi->s_resgid = make_kgid(&init_user_ns, EXT4_DEF_RESGID);
3278         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3279         sbi->s_sb_block = sb_block;
3280         if (sb->s_bdev->bd_part)
3281                 sbi->s_sectors_written_start =
3282                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3283
3284         /* Cleanup superblock name */
3285         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3286                 *cp = '!';
3287
3288         ret = -EINVAL;
3289         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3290         if (!blocksize) {
3291                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3292                 goto out_fail;
3293         }
3294
3295         /*
3296          * The ext4 superblock will not be buffer aligned for other than 1kB
3297          * block sizes.  We need to calculate the offset from buffer start.
3298          */
3299         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3300                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3301                 offset = do_div(logical_sb_block, blocksize);
3302         } else {
3303                 logical_sb_block = sb_block;
3304         }
3305
3306         if (!(bh = sb_bread(sb, logical_sb_block))) {
3307                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3308                 goto out_fail;
3309         }
3310         /*
3311          * Note: s_es must be initialized as soon as possible because
3312          *       some ext4 macro-instructions depend on its value
3313          */
3314         es = (struct ext4_super_block *) (bh->b_data + offset);
3315         sbi->s_es = es;
3316         sb->s_magic = le16_to_cpu(es->s_magic);
3317         if (sb->s_magic != EXT4_SUPER_MAGIC)
3318                 goto cantfind_ext4;
3319         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3320
3321         /* Warn if metadata_csum and gdt_csum are both set. */
3322         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3323                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3324             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3325                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3326                              "redundant flags; please run fsck.");
3327
3328         /* Check for a known checksum algorithm */
3329         if (!ext4_verify_csum_type(sb, es)) {
3330                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3331                          "unknown checksum algorithm.");
3332                 silent = 1;
3333                 goto cantfind_ext4;
3334         }
3335
3336         /* Load the checksum driver */
3337         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3338                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3339                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3340                 if (IS_ERR(sbi->s_chksum_driver)) {
3341                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3342                         ret = PTR_ERR(sbi->s_chksum_driver);
3343                         sbi->s_chksum_driver = NULL;
3344                         goto failed_mount;
3345                 }
3346         }
3347
3348         /* Check superblock checksum */
3349         if (!ext4_superblock_csum_verify(sb, es)) {
3350                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3351                          "invalid superblock checksum.  Run e2fsck?");
3352                 silent = 1;
3353                 goto cantfind_ext4;
3354         }
3355
3356         /* Precompute checksum seed for all metadata */
3357         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3358                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3359                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3360                                                sizeof(es->s_uuid));
3361
3362         /* Set defaults before we parse the mount options */
3363         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3364         set_opt(sb, INIT_INODE_TABLE);
3365         if (def_mount_opts & EXT4_DEFM_DEBUG)
3366                 set_opt(sb, DEBUG);
3367         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3368                 set_opt(sb, GRPID);
3369         if (def_mount_opts & EXT4_DEFM_UID16)
3370                 set_opt(sb, NO_UID32);
3371         /* xattr user namespace & acls are now defaulted on */
3372 #ifdef CONFIG_EXT4_FS_XATTR
3373         set_opt(sb, XATTR_USER);
3374 #endif
3375 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3376         set_opt(sb, POSIX_ACL);
3377 #endif
3378         set_opt(sb, MBLK_IO_SUBMIT);
3379         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3380                 set_opt(sb, JOURNAL_DATA);
3381         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3382                 set_opt(sb, ORDERED_DATA);
3383         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3384                 set_opt(sb, WRITEBACK_DATA);
3385
3386         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3387                 set_opt(sb, ERRORS_PANIC);
3388         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3389                 set_opt(sb, ERRORS_CONT);
3390         else
3391                 set_opt(sb, ERRORS_RO);
3392         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3393                 set_opt(sb, BLOCK_VALIDITY);
3394         if (def_mount_opts & EXT4_DEFM_DISCARD)
3395                 set_opt(sb, DISCARD);
3396
3397         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3398         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3399         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3400         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3401         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3402
3403         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3404                 set_opt(sb, BARRIER);
3405
3406         /*
3407          * enable delayed allocation by default
3408          * Use -o nodelalloc to turn it off
3409          */
3410         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3411             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3412                 set_opt(sb, DELALLOC);
3413
3414         /*
3415          * set default s_li_wait_mult for lazyinit, for the case there is
3416          * no mount option specified.
3417          */
3418         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3419
3420         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3421                            &journal_devnum, &journal_ioprio, 0)) {
3422                 ext4_msg(sb, KERN_WARNING,
3423                          "failed to parse options in superblock: %s",
3424                          sbi->s_es->s_mount_opts);
3425         }
3426         sbi->s_def_mount_opt = sbi->s_mount_opt;
3427         if (!parse_options((char *) data, sb, &journal_devnum,
3428                            &journal_ioprio, 0))
3429                 goto failed_mount;
3430
3431         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3432                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3433                             "with data=journal disables delayed "
3434                             "allocation and O_DIRECT support!\n");
3435                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3436                         ext4_msg(sb, KERN_ERR, "can't mount with "
3437                                  "both data=journal and delalloc");
3438                         goto failed_mount;
3439                 }
3440                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3441                         ext4_msg(sb, KERN_ERR, "can't mount with "
3442                                  "both data=journal and delalloc");
3443                         goto failed_mount;
3444                 }
3445                 if (test_opt(sb, DELALLOC))
3446                         clear_opt(sb, DELALLOC);
3447         }
3448
3449         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3450         if (test_opt(sb, DIOREAD_NOLOCK)) {
3451                 if (blocksize < PAGE_SIZE) {
3452                         ext4_msg(sb, KERN_ERR, "can't mount with "
3453                                  "dioread_nolock if block size != PAGE_SIZE");
3454                         goto failed_mount;
3455                 }
3456         }
3457
3458         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3459                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3460
3461         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3462             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3463              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3464              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3465                 ext4_msg(sb, KERN_WARNING,
3466                        "feature flags set on rev 0 fs, "
3467                        "running e2fsck is recommended");
3468
3469         if (IS_EXT2_SB(sb)) {
3470                 if (ext2_feature_set_ok(sb))
3471                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3472                                  "using the ext4 subsystem");
3473                 else {
3474                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3475                                  "to feature incompatibilities");
3476                         goto failed_mount;
3477                 }
3478         }
3479
3480         if (IS_EXT3_SB(sb)) {
3481                 if (ext3_feature_set_ok(sb))
3482                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3483                                  "using the ext4 subsystem");
3484                 else {
3485                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3486                                  "to feature incompatibilities");
3487                         goto failed_mount;
3488                 }
3489         }
3490
3491         /*
3492          * Check feature flags regardless of the revision level, since we
3493          * previously didn't change the revision level when setting the flags,
3494          * so there is a chance incompat flags are set on a rev 0 filesystem.
3495          */
3496         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3497                 goto failed_mount;
3498
3499         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3500             blocksize > EXT4_MAX_BLOCK_SIZE) {
3501                 ext4_msg(sb, KERN_ERR,
3502                        "Unsupported filesystem blocksize %d", blocksize);
3503                 goto failed_mount;
3504         }
3505
3506         if (sb->s_blocksize != blocksize) {
3507                 /* Validate the filesystem blocksize */
3508                 if (!sb_set_blocksize(sb, blocksize)) {
3509                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3510                                         blocksize);
3511                         goto failed_mount;
3512                 }
3513
3514                 brelse(bh);
3515                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3516                 offset = do_div(logical_sb_block, blocksize);
3517                 bh = sb_bread(sb, logical_sb_block);
3518                 if (!bh) {
3519                         ext4_msg(sb, KERN_ERR,
3520                                "Can't read superblock on 2nd try");
3521                         goto failed_mount;
3522                 }
3523                 es = (struct ext4_super_block *)(bh->b_data + offset);
3524                 sbi->s_es = es;
3525                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3526                         ext4_msg(sb, KERN_ERR,
3527                                "Magic mismatch, very weird!");
3528                         goto failed_mount;
3529                 }
3530         }
3531
3532         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3533                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3534         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3535                                                       has_huge_files);
3536         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3537
3538         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3539                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3540                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3541         } else {
3542                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3543                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3544                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3545                     (!is_power_of_2(sbi->s_inode_size)) ||
3546                     (sbi->s_inode_size > blocksize)) {
3547                         ext4_msg(sb, KERN_ERR,
3548                                "unsupported inode size: %d",
3549                                sbi->s_inode_size);
3550                         goto failed_mount;
3551                 }
3552                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3553                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3554         }
3555
3556         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3557         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3558                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3559                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3560                     !is_power_of_2(sbi->s_desc_size)) {
3561                         ext4_msg(sb, KERN_ERR,
3562                                "unsupported descriptor size %lu",
3563                                sbi->s_desc_size);
3564                         goto failed_mount;
3565                 }
3566         } else
3567                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3568
3569         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3570         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3571         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3572                 goto cantfind_ext4;
3573
3574         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3575         if (sbi->s_inodes_per_block == 0)
3576                 goto cantfind_ext4;
3577         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3578                                         sbi->s_inodes_per_block;
3579         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3580         sbi->s_sbh = bh;
3581         sbi->s_mount_state = le16_to_cpu(es->s_state);
3582         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3583         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3584
3585         for (i = 0; i < 4; i++)
3586                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3587         sbi->s_def_hash_version = es->s_def_hash_version;
3588         i = le32_to_cpu(es->s_flags);
3589         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3590                 sbi->s_hash_unsigned = 3;
3591         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3592 #ifdef __CHAR_UNSIGNED__
3593                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3594                 sbi->s_hash_unsigned = 3;
3595 #else
3596                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3597 #endif
3598         }
3599
3600         /* Handle clustersize */
3601         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3602         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3603                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3604         if (has_bigalloc) {
3605                 if (clustersize < blocksize) {
3606                         ext4_msg(sb, KERN_ERR,
3607                                  "cluster size (%d) smaller than "
3608                                  "block size (%d)", clustersize, blocksize);
3609                         goto failed_mount;
3610                 }
3611                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3612                         le32_to_cpu(es->s_log_block_size);
3613                 sbi->s_clusters_per_group =
3614                         le32_to_cpu(es->s_clusters_per_group);
3615                 if (sbi->s_clusters_per_group > blocksize * 8) {
3616                         ext4_msg(sb, KERN_ERR,
3617                                  "#clusters per group too big: %lu",
3618                                  sbi->s_clusters_per_group);
3619                         goto failed_mount;
3620                 }
3621                 if (sbi->s_blocks_per_group !=
3622                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3623                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3624                                  "clusters per group (%lu) inconsistent",
3625                                  sbi->s_blocks_per_group,
3626                                  sbi->s_clusters_per_group);
3627                         goto failed_mount;
3628                 }
3629         } else {
3630                 if (clustersize != blocksize) {
3631                         ext4_warning(sb, "fragment/cluster size (%d) != "
3632                                      "block size (%d)", clustersize,
3633                                      blocksize);
3634                         clustersize = blocksize;
3635                 }
3636                 if (sbi->s_blocks_per_group > blocksize * 8) {
3637                         ext4_msg(sb, KERN_ERR,
3638                                  "#blocks per group too big: %lu",
3639                                  sbi->s_blocks_per_group);
3640                         goto failed_mount;
3641                 }
3642                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3643                 sbi->s_cluster_bits = 0;
3644         }
3645         sbi->s_cluster_ratio = clustersize / blocksize;
3646
3647         if (sbi->s_inodes_per_group > blocksize * 8) {
3648                 ext4_msg(sb, KERN_ERR,
3649                        "#inodes per group too big: %lu",
3650                        sbi->s_inodes_per_group);
3651                 goto failed_mount;
3652         }
3653
3654         /*
3655          * Test whether we have more sectors than will fit in sector_t,
3656          * and whether the max offset is addressable by the page cache.
3657          */
3658         err = generic_check_addressable(sb->s_blocksize_bits,
3659                                         ext4_blocks_count(es));
3660         if (err) {
3661                 ext4_msg(sb, KERN_ERR, "filesystem"
3662                          " too large to mount safely on this system");
3663                 if (sizeof(sector_t) < 8)
3664                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3665                 ret = err;
3666                 goto failed_mount;
3667         }
3668
3669         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3670                 goto cantfind_ext4;
3671
3672         /* check blocks count against device size */
3673         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3674         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3675                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3676                        "exceeds size of device (%llu blocks)",
3677                        ext4_blocks_count(es), blocks_count);
3678                 goto failed_mount;
3679         }
3680
3681         /*
3682          * It makes no sense for the first data block to be beyond the end
3683          * of the filesystem.
3684          */
3685         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3686                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3687                          "block %u is beyond end of filesystem (%llu)",
3688                          le32_to_cpu(es->s_first_data_block),
3689                          ext4_blocks_count(es));
3690                 goto failed_mount;
3691         }
3692         blocks_count = (ext4_blocks_count(es) -
3693                         le32_to_cpu(es->s_first_data_block) +
3694                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3695         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3696         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3697                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3698                        "(block count %llu, first data block %u, "
3699                        "blocks per group %lu)", sbi->s_groups_count,
3700                        ext4_blocks_count(es),
3701                        le32_to_cpu(es->s_first_data_block),
3702                        EXT4_BLOCKS_PER_GROUP(sb));
3703                 goto failed_mount;
3704         }
3705         sbi->s_groups_count = blocks_count;
3706         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3707                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3708         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3709                    EXT4_DESC_PER_BLOCK(sb);
3710         sbi->s_group_desc = ext4_kvmalloc(db_count *
3711                                           sizeof(struct buffer_head *),
3712                                           GFP_KERNEL);
3713         if (sbi->s_group_desc == NULL) {
3714                 ext4_msg(sb, KERN_ERR, "not enough memory");
3715                 ret = -ENOMEM;
3716                 goto failed_mount;
3717         }
3718
3719         if (ext4_proc_root)
3720                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3721
3722         if (sbi->s_proc)
3723                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3724                                  &ext4_seq_options_fops, sb);
3725
3726         bgl_lock_init(sbi->s_blockgroup_lock);
3727
3728         for (i = 0; i < db_count; i++) {
3729                 block = descriptor_loc(sb, logical_sb_block, i);
3730                 sbi->s_group_desc[i] = sb_bread(sb, block);
3731                 if (!sbi->s_group_desc[i]) {
3732                         ext4_msg(sb, KERN_ERR,
3733                                "can't read group descriptor %d", i);
3734                         db_count = i;
3735                         goto failed_mount2;
3736                 }
3737         }
3738         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3739                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3740                 goto failed_mount2;
3741         }
3742         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3743                 if (!ext4_fill_flex_info(sb)) {
3744                         ext4_msg(sb, KERN_ERR,
3745                                "unable to initialize "
3746                                "flex_bg meta info!");
3747                         goto failed_mount2;
3748                 }
3749
3750         sbi->s_gdb_count = db_count;
3751         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3752         spin_lock_init(&sbi->s_next_gen_lock);
3753
3754         init_timer(&sbi->s_err_report);
3755         sbi->s_err_report.function = print_daily_error_info;
3756         sbi->s_err_report.data = (unsigned long) sb;
3757
3758         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3759                         ext4_count_free_clusters(sb));
3760         if (!err) {
3761                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3762                                 ext4_count_free_inodes(sb));
3763         }
3764         if (!err) {
3765                 err = percpu_counter_init(&sbi->s_dirs_counter,
3766                                 ext4_count_dirs(sb));
3767         }
3768         if (!err) {
3769                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3770         }
3771         if (err) {
3772                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3773                 ret = err;
3774                 goto failed_mount3;
3775         }
3776
3777         sbi->s_stripe = ext4_get_stripe_size(sbi);
3778         sbi->s_max_writeback_mb_bump = 128;
3779         sbi->s_extent_max_zeroout_kb = 32;
3780
3781         /*
3782          * set up enough so that it can read an inode
3783          */
3784         if (!test_opt(sb, NOLOAD) &&
3785             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3786                 sb->s_op = &ext4_sops;
3787         else
3788                 sb->s_op = &ext4_nojournal_sops;
3789         sb->s_export_op = &ext4_export_ops;
3790         sb->s_xattr = ext4_xattr_handlers;
3791 #ifdef CONFIG_QUOTA
3792         sb->s_qcop = &ext4_qctl_operations;
3793         sb->dq_op = &ext4_quota_operations;
3794
3795         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
3796                 /* Use qctl operations for hidden quota files. */
3797                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3798         }
3799 #endif
3800         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3801
3802         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3803         mutex_init(&sbi->s_orphan_lock);
3804         sbi->s_resize_flags = 0;
3805
3806         sb->s_root = NULL;
3807
3808         needs_recovery = (es->s_last_orphan != 0 ||
3809                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3810                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3811
3812         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3813             !(sb->s_flags & MS_RDONLY))
3814                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3815                         goto failed_mount3;
3816
3817         /*
3818          * The first inode we look at is the journal inode.  Don't try
3819          * root first: it may be modified in the journal!
3820          */
3821         if (!test_opt(sb, NOLOAD) &&
3822             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3823                 if (ext4_load_journal(sb, es, journal_devnum))
3824                         goto failed_mount3;
3825         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3826               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3827                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3828                        "suppressed and not mounted read-only");
3829                 goto failed_mount_wq;
3830         } else {
3831                 clear_opt(sb, DATA_FLAGS);
3832                 sbi->s_journal = NULL;
3833                 needs_recovery = 0;
3834                 goto no_journal;
3835         }
3836
3837         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3838             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3839                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3840                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3841                 goto failed_mount_wq;
3842         }
3843
3844         if (!set_journal_csum_feature_set(sb)) {
3845                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3846                          "feature set");
3847                 goto failed_mount_wq;
3848         }
3849
3850         /* We have now updated the journal if required, so we can
3851          * validate the data journaling mode. */
3852         switch (test_opt(sb, DATA_FLAGS)) {
3853         case 0:
3854                 /* No mode set, assume a default based on the journal
3855                  * capabilities: ORDERED_DATA if the journal can
3856                  * cope, else JOURNAL_DATA
3857                  */
3858                 if (jbd2_journal_check_available_features
3859                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3860                         set_opt(sb, ORDERED_DATA);
3861                 else
3862                         set_opt(sb, JOURNAL_DATA);
3863                 break;
3864
3865         case EXT4_MOUNT_ORDERED_DATA:
3866         case EXT4_MOUNT_WRITEBACK_DATA:
3867                 if (!jbd2_journal_check_available_features
3868                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3869                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3870                                "requested data journaling mode");
3871                         goto failed_mount_wq;
3872                 }
3873         default:
3874                 break;
3875         }
3876         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3877
3878         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3879
3880         /*
3881          * The journal may have updated the bg summary counts, so we
3882          * need to update the global counters.
3883          */
3884         percpu_counter_set(&sbi->s_freeclusters_counter,
3885                            ext4_count_free_clusters(sb));
3886         percpu_counter_set(&sbi->s_freeinodes_counter,
3887                            ext4_count_free_inodes(sb));
3888         percpu_counter_set(&sbi->s_dirs_counter,
3889                            ext4_count_dirs(sb));
3890         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3891
3892 no_journal:
3893         /*
3894          * Get the # of file system overhead blocks from the
3895          * superblock if present.
3896          */
3897         if (es->s_overhead_clusters)
3898                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3899         else {
3900                 ret = ext4_calculate_overhead(sb);
3901                 if (ret)
3902                         goto failed_mount_wq;
3903         }
3904
3905         /*
3906          * The maximum number of concurrent works can be high and
3907          * concurrency isn't really necessary.  Limit it to 1.
3908          */
3909         EXT4_SB(sb)->dio_unwritten_wq =
3910                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3911         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3912                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3913                 goto failed_mount_wq;
3914         }
3915
3916         /*
3917          * The jbd2_journal_load will have done any necessary log recovery,
3918          * so we can safely mount the rest of the filesystem now.
3919          */
3920
3921         root = ext4_iget(sb, EXT4_ROOT_INO);
3922         if (IS_ERR(root)) {
3923                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3924                 ret = PTR_ERR(root);
3925                 root = NULL;
3926                 goto failed_mount4;
3927         }
3928         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3929                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3930                 iput(root);
3931                 goto failed_mount4;
3932         }
3933         sb->s_root = d_make_root(root);
3934         if (!sb->s_root) {
3935                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3936                 ret = -ENOMEM;
3937                 goto failed_mount4;
3938         }
3939
3940         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3941                 sb->s_flags |= MS_RDONLY;
3942
3943         /* determine the minimum size of new large inodes, if present */
3944         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3945                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3946                                                      EXT4_GOOD_OLD_INODE_SIZE;
3947                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3948                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3949                         if (sbi->s_want_extra_isize <
3950                             le16_to_cpu(es->s_want_extra_isize))
3951                                 sbi->s_want_extra_isize =
3952                                         le16_to_cpu(es->s_want_extra_isize);
3953                         if (sbi->s_want_extra_isize <
3954                             le16_to_cpu(es->s_min_extra_isize))
3955                                 sbi->s_want_extra_isize =
3956                                         le16_to_cpu(es->s_min_extra_isize);
3957                 }
3958         }
3959         /* Check if enough inode space is available */
3960         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3961                                                         sbi->s_inode_size) {
3962                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3963                                                        EXT4_GOOD_OLD_INODE_SIZE;
3964                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3965                          "available");
3966         }
3967
3968         err = ext4_setup_system_zone(sb);
3969         if (err) {
3970                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3971                          "zone (%d)", err);
3972                 goto failed_mount4a;
3973         }
3974
3975         ext4_ext_init(sb);
3976         err = ext4_mb_init(sb);
3977         if (err) {
3978                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3979                          err);
3980                 goto failed_mount5;
3981         }
3982
3983         err = ext4_register_li_request(sb, first_not_zeroed);
3984         if (err)
3985                 goto failed_mount6;
3986
3987         sbi->s_kobj.kset = ext4_kset;
3988         init_completion(&sbi->s_kobj_unregister);
3989         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
3990                                    "%s", sb->s_id);
3991         if (err)
3992                 goto failed_mount7;
3993
3994         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3995         ext4_orphan_cleanup(sb, es);
3996         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3997         if (needs_recovery) {
3998                 ext4_msg(sb, KERN_INFO, "recovery complete");
3999                 ext4_mark_recovery_complete(sb, es);
4000         }
4001         if (EXT4_SB(sb)->s_journal) {
4002                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4003                         descr = " journalled data mode";
4004                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4005                         descr = " ordered data mode";
4006                 else
4007                         descr = " writeback data mode";
4008         } else
4009                 descr = "out journal";
4010
4011 #ifdef CONFIG_QUOTA
4012         /* Enable quota usage during mount. */
4013         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4014             !(sb->s_flags & MS_RDONLY)) {
4015                 ret = ext4_enable_quotas(sb);
4016                 if (ret)
4017                         goto failed_mount7;
4018         }
4019 #endif  /* CONFIG_QUOTA */
4020
4021         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4022                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4023                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4024
4025         if (es->s_error_count)
4026                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4027
4028         kfree(orig_data);
4029         return 0;
4030
4031 cantfind_ext4:
4032         if (!silent)
4033                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4034         goto failed_mount;
4035
4036 failed_mount7:
4037         ext4_unregister_li_request(sb);
4038 failed_mount6:
4039         ext4_mb_release(sb);
4040 failed_mount5:
4041         ext4_ext_release(sb);
4042         ext4_release_system_zone(sb);
4043 failed_mount4a:
4044         dput(sb->s_root);
4045         sb->s_root = NULL;
4046 failed_mount4:
4047         ext4_msg(sb, KERN_ERR, "mount failed");
4048         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4049 failed_mount_wq:
4050         if (sbi->s_journal) {
4051                 jbd2_journal_destroy(sbi->s_journal);
4052                 sbi->s_journal = NULL;
4053         }
4054 failed_mount3:
4055         del_timer(&sbi->s_err_report);
4056         if (sbi->s_flex_groups)
4057                 ext4_kvfree(sbi->s_flex_groups);
4058         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4059         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4060         percpu_counter_destroy(&sbi->s_dirs_counter);
4061         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4062         if (sbi->s_mmp_tsk)
4063                 kthread_stop(sbi->s_mmp_tsk);
4064 failed_mount2:
4065         for (i = 0; i < db_count; i++)
4066                 brelse(sbi->s_group_desc[i]);
4067         ext4_kvfree(sbi->s_group_desc);
4068 failed_mount:
4069         if (sbi->s_chksum_driver)
4070                 crypto_free_shash(sbi->s_chksum_driver);
4071         if (sbi->s_proc) {
4072                 remove_proc_entry("options", sbi->s_proc);
4073                 remove_proc_entry(sb->s_id, ext4_proc_root);
4074         }
4075 #ifdef CONFIG_QUOTA
4076         for (i = 0; i < MAXQUOTAS; i++)
4077                 kfree(sbi->s_qf_names[i]);
4078 #endif
4079         ext4_blkdev_remove(sbi);
4080         brelse(bh);
4081 out_fail:
4082         sb->s_fs_info = NULL;
4083         kfree(sbi->s_blockgroup_lock);
4084         kfree(sbi);
4085 out_free_orig:
4086         kfree(orig_data);
4087         return ret;
4088 }
4089
4090 /*
4091  * Setup any per-fs journal parameters now.  We'll do this both on
4092  * initial mount, once the journal has been initialised but before we've
4093  * done any recovery; and again on any subsequent remount.
4094  */
4095 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4096 {
4097         struct ext4_sb_info *sbi = EXT4_SB(sb);
4098
4099         journal->j_commit_interval = sbi->s_commit_interval;
4100         journal->j_min_batch_time = sbi->s_min_batch_time;
4101         journal->j_max_batch_time = sbi->s_max_batch_time;
4102
4103         write_lock(&journal->j_state_lock);
4104         if (test_opt(sb, BARRIER))
4105                 journal->j_flags |= JBD2_BARRIER;
4106         else
4107                 journal->j_flags &= ~JBD2_BARRIER;
4108         if (test_opt(sb, DATA_ERR_ABORT))
4109                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4110         else
4111                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4112         write_unlock(&journal->j_state_lock);
4113 }
4114
4115 static journal_t *ext4_get_journal(struct super_block *sb,
4116                                    unsigned int journal_inum)
4117 {
4118         struct inode *journal_inode;
4119         journal_t *journal;
4120
4121         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4122
4123         /* First, test for the existence of a valid inode on disk.  Bad
4124          * things happen if we iget() an unused inode, as the subsequent
4125          * iput() will try to delete it. */
4126
4127         journal_inode = ext4_iget(sb, journal_inum);
4128         if (IS_ERR(journal_inode)) {
4129                 ext4_msg(sb, KERN_ERR, "no journal found");
4130                 return NULL;
4131         }
4132         if (!journal_inode->i_nlink) {
4133                 make_bad_inode(journal_inode);
4134                 iput(journal_inode);
4135                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4136                 return NULL;
4137         }
4138
4139         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4140                   journal_inode, journal_inode->i_size);
4141         if (!S_ISREG(journal_inode->i_mode)) {
4142                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4143                 iput(journal_inode);
4144                 return NULL;
4145         }
4146
4147         journal = jbd2_journal_init_inode(journal_inode);
4148         if (!journal) {
4149                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4150                 iput(journal_inode);
4151                 return NULL;
4152         }
4153         journal->j_private = sb;
4154         ext4_init_journal_params(sb, journal);
4155         return journal;
4156 }
4157
4158 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4159                                        dev_t j_dev)
4160 {
4161         struct buffer_head *bh;
4162         journal_t *journal;
4163         ext4_fsblk_t start;
4164         ext4_fsblk_t len;
4165         int hblock, blocksize;
4166         ext4_fsblk_t sb_block;
4167         unsigned long offset;
4168         struct ext4_super_block *es;
4169         struct block_device *bdev;
4170
4171         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4172
4173         bdev = ext4_blkdev_get(j_dev, sb);
4174         if (bdev == NULL)
4175                 return NULL;
4176
4177         blocksize = sb->s_blocksize;
4178         hblock = bdev_logical_block_size(bdev);
4179         if (blocksize < hblock) {
4180                 ext4_msg(sb, KERN_ERR,
4181                         "blocksize too small for journal device");
4182                 goto out_bdev;
4183         }
4184
4185         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4186         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4187         set_blocksize(bdev, blocksize);
4188         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4189                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4190                        "external journal");
4191                 goto out_bdev;
4192         }
4193
4194         es = (struct ext4_super_block *) (bh->b_data + offset);
4195         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4196             !(le32_to_cpu(es->s_feature_incompat) &
4197               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4198                 ext4_msg(sb, KERN_ERR, "external journal has "
4199                                         "bad superblock");
4200                 brelse(bh);
4201                 goto out_bdev;
4202         }
4203
4204         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4205                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4206                 brelse(bh);
4207                 goto out_bdev;
4208         }
4209
4210         len = ext4_blocks_count(es);
4211         start = sb_block + 1;
4212         brelse(bh);     /* we're done with the superblock */
4213
4214         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4215                                         start, len, blocksize);
4216         if (!journal) {
4217                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4218                 goto out_bdev;
4219         }
4220         journal->j_private = sb;
4221         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4222         wait_on_buffer(journal->j_sb_buffer);
4223         if (!buffer_uptodate(journal->j_sb_buffer)) {
4224                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4225                 goto out_journal;
4226         }
4227         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4228                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4229                                         "user (unsupported) - %d",
4230                         be32_to_cpu(journal->j_superblock->s_nr_users));
4231                 goto out_journal;
4232         }
4233         EXT4_SB(sb)->journal_bdev = bdev;
4234         ext4_init_journal_params(sb, journal);
4235         return journal;
4236
4237 out_journal:
4238         jbd2_journal_destroy(journal);
4239 out_bdev:
4240         ext4_blkdev_put(bdev);
4241         return NULL;
4242 }
4243
4244 static int ext4_load_journal(struct super_block *sb,
4245                              struct ext4_super_block *es,
4246                              unsigned long journal_devnum)
4247 {
4248         journal_t *journal;
4249         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4250         dev_t journal_dev;
4251         int err = 0;
4252         int really_read_only;
4253
4254         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4255
4256         if (journal_devnum &&
4257             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4258                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4259                         "numbers have changed");
4260                 journal_dev = new_decode_dev(journal_devnum);
4261         } else
4262                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4263
4264         really_read_only = bdev_read_only(sb->s_bdev);
4265
4266         /*
4267          * Are we loading a blank journal or performing recovery after a
4268          * crash?  For recovery, we need to check in advance whether we
4269          * can get read-write access to the device.
4270          */
4271         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4272                 if (sb->s_flags & MS_RDONLY) {
4273                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4274                                         "required on readonly filesystem");
4275                         if (really_read_only) {
4276                                 ext4_msg(sb, KERN_ERR, "write access "
4277                                         "unavailable, cannot proceed");
4278                                 return -EROFS;
4279                         }
4280                         ext4_msg(sb, KERN_INFO, "write access will "
4281                                "be enabled during recovery");
4282                 }
4283         }
4284
4285         if (journal_inum && journal_dev) {
4286                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4287                        "and inode journals!");
4288                 return -EINVAL;
4289         }
4290
4291         if (journal_inum) {
4292                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4293                         return -EINVAL;
4294         } else {
4295                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4296                         return -EINVAL;
4297         }
4298
4299         if (!(journal->j_flags & JBD2_BARRIER))
4300                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4301
4302         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4303                 err = jbd2_journal_wipe(journal, !really_read_only);
4304         if (!err) {
4305                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4306                 if (save)
4307                         memcpy(save, ((char *) es) +
4308                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4309                 err = jbd2_journal_load(journal);
4310                 if (save)
4311                         memcpy(((char *) es) + EXT4_S_ERR_START,
4312                                save, EXT4_S_ERR_LEN);
4313                 kfree(save);
4314         }
4315
4316         if (err) {
4317                 ext4_msg(sb, KERN_ERR, "error loading journal");
4318                 jbd2_journal_destroy(journal);
4319                 return err;
4320         }
4321
4322         EXT4_SB(sb)->s_journal = journal;
4323         ext4_clear_journal_err(sb, es);
4324
4325         if (!really_read_only && journal_devnum &&
4326             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4327                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4328
4329                 /* Make sure we flush the recovery flag to disk. */
4330                 ext4_commit_super(sb, 1);
4331         }
4332
4333         return 0;
4334 }
4335
4336 static int ext4_commit_super(struct super_block *sb, int sync)
4337 {
4338         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4339         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4340         int error = 0;
4341
4342         if (!sbh || block_device_ejected(sb))
4343                 return error;
4344         if (buffer_write_io_error(sbh)) {
4345                 /*
4346                  * Oh, dear.  A previous attempt to write the
4347                  * superblock failed.  This could happen because the
4348                  * USB device was yanked out.  Or it could happen to
4349                  * be a transient write error and maybe the block will
4350                  * be remapped.  Nothing we can do but to retry the
4351                  * write and hope for the best.
4352                  */
4353                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4354                        "superblock detected");
4355                 clear_buffer_write_io_error(sbh);
4356                 set_buffer_uptodate(sbh);
4357         }
4358         /*
4359          * If the file system is mounted read-only, don't update the
4360          * superblock write time.  This avoids updating the superblock
4361          * write time when we are mounting the root file system
4362          * read/only but we need to replay the journal; at that point,
4363          * for people who are east of GMT and who make their clock
4364          * tick in localtime for Windows bug-for-bug compatibility,
4365          * the clock is set in the future, and this will cause e2fsck
4366          * to complain and force a full file system check.
4367          */
4368         if (!(sb->s_flags & MS_RDONLY))
4369                 es->s_wtime = cpu_to_le32(get_seconds());
4370         if (sb->s_bdev->bd_part)
4371                 es->s_kbytes_written =
4372                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4373                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4374                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4375         else
4376                 es->s_kbytes_written =
4377                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4378         ext4_free_blocks_count_set(es,
4379                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4380                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4381         es->s_free_inodes_count =
4382                 cpu_to_le32(percpu_counter_sum_positive(
4383                                 &EXT4_SB(sb)->s_freeinodes_counter));
4384         BUFFER_TRACE(sbh, "marking dirty");
4385         ext4_superblock_csum_set(sb);
4386         mark_buffer_dirty(sbh);
4387         if (sync) {
4388                 error = sync_dirty_buffer(sbh);
4389                 if (error)
4390                         return error;
4391
4392                 error = buffer_write_io_error(sbh);
4393                 if (error) {
4394                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4395                                "superblock");
4396                         clear_buffer_write_io_error(sbh);
4397                         set_buffer_uptodate(sbh);
4398                 }
4399         }
4400         return error;
4401 }
4402
4403 /*
4404  * Have we just finished recovery?  If so, and if we are mounting (or
4405  * remounting) the filesystem readonly, then we will end up with a
4406  * consistent fs on disk.  Record that fact.
4407  */
4408 static void ext4_mark_recovery_complete(struct super_block *sb,
4409                                         struct ext4_super_block *es)
4410 {
4411         journal_t *journal = EXT4_SB(sb)->s_journal;
4412
4413         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4414                 BUG_ON(journal != NULL);
4415                 return;
4416         }
4417         jbd2_journal_lock_updates(journal);
4418         if (jbd2_journal_flush(journal) < 0)
4419                 goto out;
4420
4421         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4422             sb->s_flags & MS_RDONLY) {
4423                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4424                 ext4_commit_super(sb, 1);
4425         }
4426
4427 out:
4428         jbd2_journal_unlock_updates(journal);
4429 }
4430
4431 /*
4432  * If we are mounting (or read-write remounting) a filesystem whose journal
4433  * has recorded an error from a previous lifetime, move that error to the
4434  * main filesystem now.
4435  */
4436 static void ext4_clear_journal_err(struct super_block *sb,
4437                                    struct ext4_super_block *es)
4438 {
4439         journal_t *journal;
4440         int j_errno;
4441         const char *errstr;
4442
4443         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4444
4445         journal = EXT4_SB(sb)->s_journal;
4446
4447         /*
4448          * Now check for any error status which may have been recorded in the
4449          * journal by a prior ext4_error() or ext4_abort()
4450          */
4451
4452         j_errno = jbd2_journal_errno(journal);
4453         if (j_errno) {
4454                 char nbuf[16];
4455
4456                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4457                 ext4_warning(sb, "Filesystem error recorded "
4458                              "from previous mount: %s", errstr);
4459                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4460
4461                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4462                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4463                 ext4_commit_super(sb, 1);
4464
4465                 jbd2_journal_clear_err(journal);
4466                 jbd2_journal_update_sb_errno(journal);
4467         }
4468 }
4469
4470 /*
4471  * Force the running and committing transactions to commit,
4472  * and wait on the commit.
4473  */
4474 int ext4_force_commit(struct super_block *sb)
4475 {
4476         journal_t *journal;
4477         int ret = 0;
4478
4479         if (sb->s_flags & MS_RDONLY)
4480                 return 0;
4481
4482         journal = EXT4_SB(sb)->s_journal;
4483         if (journal)
4484                 ret = ext4_journal_force_commit(journal);
4485
4486         return ret;
4487 }
4488
4489 static int ext4_sync_fs(struct super_block *sb, int wait)
4490 {
4491         int ret = 0;
4492         tid_t target;
4493         struct ext4_sb_info *sbi = EXT4_SB(sb);
4494
4495         trace_ext4_sync_fs(sb, wait);
4496         flush_workqueue(sbi->dio_unwritten_wq);
4497         /*
4498          * Writeback quota in non-journalled quota case - journalled quota has
4499          * no dirty dquots
4500          */
4501         dquot_writeback_dquots(sb, -1);
4502         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4503                 if (wait)
4504                         jbd2_log_wait_commit(sbi->s_journal, target);
4505         }
4506         return ret;
4507 }
4508
4509 /*
4510  * LVM calls this function before a (read-only) snapshot is created.  This
4511  * gives us a chance to flush the journal completely and mark the fs clean.
4512  *
4513  * Note that only this function cannot bring a filesystem to be in a clean
4514  * state independently. It relies on upper layer to stop all data & metadata
4515  * modifications.
4516  */
4517 static int ext4_freeze(struct super_block *sb)
4518 {
4519         int error = 0;
4520         journal_t *journal;
4521
4522         if (sb->s_flags & MS_RDONLY)
4523                 return 0;
4524
4525         journal = EXT4_SB(sb)->s_journal;
4526
4527         /* Now we set up the journal barrier. */
4528         jbd2_journal_lock_updates(journal);
4529
4530         /*
4531          * Don't clear the needs_recovery flag if we failed to flush
4532          * the journal.
4533          */
4534         error = jbd2_journal_flush(journal);
4535         if (error < 0)
4536                 goto out;
4537
4538         /* Journal blocked and flushed, clear needs_recovery flag. */
4539         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4540         error = ext4_commit_super(sb, 1);
4541 out:
4542         /* we rely on upper layer to stop further updates */
4543         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4544         return error;
4545 }
4546
4547 /*
4548  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4549  * flag here, even though the filesystem is not technically dirty yet.
4550  */
4551 static int ext4_unfreeze(struct super_block *sb)
4552 {
4553         if (sb->s_flags & MS_RDONLY)
4554                 return 0;
4555
4556         /* Reset the needs_recovery flag before the fs is unlocked. */
4557         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4558         ext4_commit_super(sb, 1);
4559         return 0;
4560 }
4561
4562 /*
4563  * Structure to save mount options for ext4_remount's benefit
4564  */
4565 struct ext4_mount_options {
4566         unsigned long s_mount_opt;
4567         unsigned long s_mount_opt2;
4568         kuid_t s_resuid;
4569         kgid_t s_resgid;
4570         unsigned long s_commit_interval;
4571         u32 s_min_batch_time, s_max_batch_time;
4572 #ifdef CONFIG_QUOTA
4573         int s_jquota_fmt;
4574         char *s_qf_names[MAXQUOTAS];
4575 #endif
4576 };
4577
4578 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4579 {
4580         struct ext4_super_block *es;
4581         struct ext4_sb_info *sbi = EXT4_SB(sb);
4582         unsigned long old_sb_flags;
4583         struct ext4_mount_options old_opts;
4584         int enable_quota = 0;
4585         ext4_group_t g;
4586         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4587         int err = 0;
4588 #ifdef CONFIG_QUOTA
4589         int i;
4590 #endif
4591         char *orig_data = kstrdup(data, GFP_KERNEL);
4592
4593         /* Store the original options */
4594         old_sb_flags = sb->s_flags;
4595         old_opts.s_mount_opt = sbi->s_mount_opt;
4596         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4597         old_opts.s_resuid = sbi->s_resuid;
4598         old_opts.s_resgid = sbi->s_resgid;
4599         old_opts.s_commit_interval = sbi->s_commit_interval;
4600         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4601         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4602 #ifdef CONFIG_QUOTA
4603         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4604         for (i = 0; i < MAXQUOTAS; i++)
4605                 old_opts.s_qf_names[i] = sbi->s_qf_names[i];
4606 #endif
4607         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4608                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4609
4610         /*
4611          * Allow the "check" option to be passed as a remount option.
4612          */
4613         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4614                 err = -EINVAL;
4615                 goto restore_opts;
4616         }
4617
4618         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4619                 ext4_abort(sb, "Abort forced by user");
4620
4621         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4622                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4623
4624         es = sbi->s_es;
4625
4626         if (sbi->s_journal) {
4627                 ext4_init_journal_params(sb, sbi->s_journal);
4628                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4629         }
4630
4631         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4632                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4633                         err = -EROFS;
4634                         goto restore_opts;
4635                 }
4636
4637                 if (*flags & MS_RDONLY) {
4638                         err = dquot_suspend(sb, -1);
4639                         if (err < 0)
4640                                 goto restore_opts;
4641
4642                         /*
4643                          * First of all, the unconditional stuff we have to do
4644                          * to disable replay of the journal when we next remount
4645                          */
4646                         sb->s_flags |= MS_RDONLY;
4647
4648                         /*
4649                          * OK, test if we are remounting a valid rw partition
4650                          * readonly, and if so set the rdonly flag and then
4651                          * mark the partition as valid again.
4652                          */
4653                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4654                             (sbi->s_mount_state & EXT4_VALID_FS))
4655                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4656
4657                         if (sbi->s_journal)
4658                                 ext4_mark_recovery_complete(sb, es);
4659                 } else {
4660                         /* Make sure we can mount this feature set readwrite */
4661                         if (!ext4_feature_set_ok(sb, 0)) {
4662                                 err = -EROFS;
4663                                 goto restore_opts;
4664                         }
4665                         /*
4666                          * Make sure the group descriptor checksums
4667                          * are sane.  If they aren't, refuse to remount r/w.
4668                          */
4669                         for (g = 0; g < sbi->s_groups_count; g++) {
4670                                 struct ext4_group_desc *gdp =
4671                                         ext4_get_group_desc(sb, g, NULL);
4672
4673                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4674                                         ext4_msg(sb, KERN_ERR,
4675                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4676                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4677                                                le16_to_cpu(gdp->bg_checksum));
4678                                         err = -EINVAL;
4679                                         goto restore_opts;
4680                                 }
4681                         }
4682
4683                         /*
4684                          * If we have an unprocessed orphan list hanging
4685                          * around from a previously readonly bdev mount,
4686                          * require a full umount/remount for now.
4687                          */
4688                         if (es->s_last_orphan) {
4689                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4690                                        "remount RDWR because of unprocessed "
4691                                        "orphan inode list.  Please "
4692                                        "umount/remount instead");
4693                                 err = -EINVAL;
4694                                 goto restore_opts;
4695                         }
4696
4697                         /*
4698                          * Mounting a RDONLY partition read-write, so reread
4699                          * and store the current valid flag.  (It may have
4700                          * been changed by e2fsck since we originally mounted
4701                          * the partition.)
4702                          */
4703                         if (sbi->s_journal)
4704                                 ext4_clear_journal_err(sb, es);
4705                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4706                         if (!ext4_setup_super(sb, es, 0))
4707                                 sb->s_flags &= ~MS_RDONLY;
4708                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4709                                                      EXT4_FEATURE_INCOMPAT_MMP))
4710                                 if (ext4_multi_mount_protect(sb,
4711                                                 le64_to_cpu(es->s_mmp_block))) {
4712                                         err = -EROFS;
4713                                         goto restore_opts;
4714                                 }
4715                         enable_quota = 1;
4716                 }
4717         }
4718
4719         /*
4720          * Reinitialize lazy itable initialization thread based on
4721          * current settings
4722          */
4723         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4724                 ext4_unregister_li_request(sb);
4725         else {
4726                 ext4_group_t first_not_zeroed;
4727                 first_not_zeroed = ext4_has_uninit_itable(sb);
4728                 ext4_register_li_request(sb, first_not_zeroed);
4729         }
4730
4731         ext4_setup_system_zone(sb);
4732         if (sbi->s_journal == NULL)
4733                 ext4_commit_super(sb, 1);
4734
4735 #ifdef CONFIG_QUOTA
4736         /* Release old quota file names */
4737         for (i = 0; i < MAXQUOTAS; i++)
4738                 if (old_opts.s_qf_names[i] &&
4739                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4740                         kfree(old_opts.s_qf_names[i]);
4741         if (enable_quota) {
4742                 if (sb_any_quota_suspended(sb))
4743                         dquot_resume(sb, -1);
4744                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4745                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4746                         err = ext4_enable_quotas(sb);
4747                         if (err)
4748                                 goto restore_opts;
4749                 }
4750         }
4751 #endif
4752
4753         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4754         kfree(orig_data);
4755         return 0;
4756
4757 restore_opts:
4758         sb->s_flags = old_sb_flags;
4759         sbi->s_mount_opt = old_opts.s_mount_opt;
4760         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4761         sbi->s_resuid = old_opts.s_resuid;
4762         sbi->s_resgid = old_opts.s_resgid;
4763         sbi->s_commit_interval = old_opts.s_commit_interval;
4764         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4765         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4766 #ifdef CONFIG_QUOTA
4767         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4768         for (i = 0; i < MAXQUOTAS; i++) {
4769                 if (sbi->s_qf_names[i] &&
4770                     old_opts.s_qf_names[i] != sbi->s_qf_names[i])
4771                         kfree(sbi->s_qf_names[i]);
4772                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4773         }
4774 #endif
4775         kfree(orig_data);
4776         return err;
4777 }
4778
4779 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4780 {
4781         struct super_block *sb = dentry->d_sb;
4782         struct ext4_sb_info *sbi = EXT4_SB(sb);
4783         struct ext4_super_block *es = sbi->s_es;
4784         ext4_fsblk_t overhead = 0;
4785         u64 fsid;
4786         s64 bfree;
4787
4788         if (!test_opt(sb, MINIX_DF))
4789                 overhead = sbi->s_overhead;
4790
4791         buf->f_type = EXT4_SUPER_MAGIC;
4792         buf->f_bsize = sb->s_blocksize;
4793         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, sbi->s_overhead);
4794         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4795                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4796         /* prevent underflow in case that few free space is available */
4797         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4798         buf->f_bavail = buf->f_bfree - ext4_r_blocks_count(es);
4799         if (buf->f_bfree < ext4_r_blocks_count(es))
4800                 buf->f_bavail = 0;
4801         buf->f_files = le32_to_cpu(es->s_inodes_count);
4802         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4803         buf->f_namelen = EXT4_NAME_LEN;
4804         fsid = le64_to_cpup((void *)es->s_uuid) ^
4805                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4806         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4807         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4808
4809         return 0;
4810 }
4811
4812 /* Helper function for writing quotas on sync - we need to start transaction
4813  * before quota file is locked for write. Otherwise the are possible deadlocks:
4814  * Process 1                         Process 2
4815  * ext4_create()                     quota_sync()
4816  *   jbd2_journal_start()                  write_dquot()
4817  *   dquot_initialize()                         down(dqio_mutex)
4818  *     down(dqio_mutex)                    jbd2_journal_start()
4819  *
4820  */
4821
4822 #ifdef CONFIG_QUOTA
4823
4824 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4825 {
4826         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4827 }
4828
4829 static int ext4_write_dquot(struct dquot *dquot)
4830 {
4831         int ret, err;
4832         handle_t *handle;
4833         struct inode *inode;
4834
4835         inode = dquot_to_inode(dquot);
4836         handle = ext4_journal_start(inode,
4837                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4838         if (IS_ERR(handle))
4839                 return PTR_ERR(handle);
4840         ret = dquot_commit(dquot);
4841         err = ext4_journal_stop(handle);
4842         if (!ret)
4843                 ret = err;
4844         return ret;
4845 }
4846
4847 static int ext4_acquire_dquot(struct dquot *dquot)
4848 {
4849         int ret, err;
4850         handle_t *handle;
4851
4852         handle = ext4_journal_start(dquot_to_inode(dquot),
4853                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4854         if (IS_ERR(handle))
4855                 return PTR_ERR(handle);
4856         ret = dquot_acquire(dquot);
4857         err = ext4_journal_stop(handle);
4858         if (!ret)
4859                 ret = err;
4860         return ret;
4861 }
4862
4863 static int ext4_release_dquot(struct dquot *dquot)
4864 {
4865         int ret, err;
4866         handle_t *handle;
4867
4868         handle = ext4_journal_start(dquot_to_inode(dquot),
4869                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4870         if (IS_ERR(handle)) {
4871                 /* Release dquot anyway to avoid endless cycle in dqput() */
4872                 dquot_release(dquot);
4873                 return PTR_ERR(handle);
4874         }
4875         ret = dquot_release(dquot);
4876         err = ext4_journal_stop(handle);
4877         if (!ret)
4878                 ret = err;
4879         return ret;
4880 }
4881
4882 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4883 {
4884         /* Are we journaling quotas? */
4885         if (EXT4_SB(dquot->dq_sb)->s_qf_names[USRQUOTA] ||
4886             EXT4_SB(dquot->dq_sb)->s_qf_names[GRPQUOTA]) {
4887                 dquot_mark_dquot_dirty(dquot);
4888                 return ext4_write_dquot(dquot);
4889         } else {
4890                 return dquot_mark_dquot_dirty(dquot);
4891         }
4892 }
4893
4894 static int ext4_write_info(struct super_block *sb, int type)
4895 {
4896         int ret, err;
4897         handle_t *handle;
4898
4899         /* Data block + inode block */
4900         handle = ext4_journal_start(sb->s_root->d_inode, 2);
4901         if (IS_ERR(handle))
4902                 return PTR_ERR(handle);
4903         ret = dquot_commit_info(sb, type);
4904         err = ext4_journal_stop(handle);
4905         if (!ret)
4906                 ret = err;
4907         return ret;
4908 }
4909
4910 /*
4911  * Turn on quotas during mount time - we need to find
4912  * the quota file and such...
4913  */
4914 static int ext4_quota_on_mount(struct super_block *sb, int type)
4915 {
4916         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4917                                         EXT4_SB(sb)->s_jquota_fmt, type);
4918 }
4919
4920 /*
4921  * Standard function to be called on quota_on
4922  */
4923 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4924                          struct path *path)
4925 {
4926         int err;
4927
4928         if (!test_opt(sb, QUOTA))
4929                 return -EINVAL;
4930
4931         /* Quotafile not on the same filesystem? */
4932         if (path->dentry->d_sb != sb)
4933                 return -EXDEV;
4934         /* Journaling quota? */
4935         if (EXT4_SB(sb)->s_qf_names[type]) {
4936                 /* Quotafile not in fs root? */
4937                 if (path->dentry->d_parent != sb->s_root)
4938                         ext4_msg(sb, KERN_WARNING,
4939                                 "Quota file not on filesystem root. "
4940                                 "Journaled quota will not work");
4941         }
4942
4943         /*
4944          * When we journal data on quota file, we have to flush journal to see
4945          * all updates to the file when we bypass pagecache...
4946          */
4947         if (EXT4_SB(sb)->s_journal &&
4948             ext4_should_journal_data(path->dentry->d_inode)) {
4949                 /*
4950                  * We don't need to lock updates but journal_flush() could
4951                  * otherwise be livelocked...
4952                  */
4953                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4954                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4955                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4956                 if (err)
4957                         return err;
4958         }
4959
4960         return dquot_quota_on(sb, type, format_id, path);
4961 }
4962
4963 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
4964                              unsigned int flags)
4965 {
4966         int err;
4967         struct inode *qf_inode;
4968         unsigned long qf_inums[MAXQUOTAS] = {
4969                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4970                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4971         };
4972
4973         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
4974
4975         if (!qf_inums[type])
4976                 return -EPERM;
4977
4978         qf_inode = ext4_iget(sb, qf_inums[type]);
4979         if (IS_ERR(qf_inode)) {
4980                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
4981                 return PTR_ERR(qf_inode);
4982         }
4983
4984         err = dquot_enable(qf_inode, type, format_id, flags);
4985         iput(qf_inode);
4986
4987         return err;
4988 }
4989
4990 /* Enable usage tracking for all quota types. */
4991 static int ext4_enable_quotas(struct super_block *sb)
4992 {
4993         int type, err = 0;
4994         unsigned long qf_inums[MAXQUOTAS] = {
4995                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4996                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4997         };
4998
4999         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5000         for (type = 0; type < MAXQUOTAS; type++) {
5001                 if (qf_inums[type]) {
5002                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5003                                                 DQUOT_USAGE_ENABLED);
5004                         if (err) {
5005                                 ext4_warning(sb,
5006                                         "Failed to enable quota (type=%d) "
5007                                         "tracking. Please run e2fsck to fix.",
5008                                         type);
5009                                 return err;
5010                         }
5011                 }
5012         }
5013         return 0;
5014 }
5015
5016 /*
5017  * quota_on function that is used when QUOTA feature is set.
5018  */
5019 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5020                                  int format_id)
5021 {
5022         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5023                 return -EINVAL;
5024
5025         /*
5026          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5027          */
5028         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5029 }
5030
5031 static int ext4_quota_off(struct super_block *sb, int type)
5032 {
5033         struct inode *inode = sb_dqopt(sb)->files[type];
5034         handle_t *handle;
5035
5036         /* Force all delayed allocation blocks to be allocated.
5037          * Caller already holds s_umount sem */
5038         if (test_opt(sb, DELALLOC))
5039                 sync_filesystem(sb);
5040
5041         if (!inode)
5042                 goto out;
5043
5044         /* Update modification times of quota files when userspace can
5045          * start looking at them */
5046         handle = ext4_journal_start(inode, 1);
5047         if (IS_ERR(handle))
5048                 goto out;
5049         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5050         ext4_mark_inode_dirty(handle, inode);
5051         ext4_journal_stop(handle);
5052
5053 out:
5054         return dquot_quota_off(sb, type);
5055 }
5056
5057 /*
5058  * quota_off function that is used when QUOTA feature is set.
5059  */
5060 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5061 {
5062         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5063                 return -EINVAL;
5064
5065         /* Disable only the limits. */
5066         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5067 }
5068
5069 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5070  * acquiring the locks... As quota files are never truncated and quota code
5071  * itself serializes the operations (and no one else should touch the files)
5072  * we don't have to be afraid of races */
5073 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5074                                size_t len, loff_t off)
5075 {
5076         struct inode *inode = sb_dqopt(sb)->files[type];
5077         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5078         int err = 0;
5079         int offset = off & (sb->s_blocksize - 1);
5080         int tocopy;
5081         size_t toread;
5082         struct buffer_head *bh;
5083         loff_t i_size = i_size_read(inode);
5084
5085         if (off > i_size)
5086                 return 0;
5087         if (off+len > i_size)
5088                 len = i_size-off;
5089         toread = len;
5090         while (toread > 0) {
5091                 tocopy = sb->s_blocksize - offset < toread ?
5092                                 sb->s_blocksize - offset : toread;
5093                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5094                 if (err)
5095                         return err;
5096                 if (!bh)        /* A hole? */
5097                         memset(data, 0, tocopy);
5098                 else
5099                         memcpy(data, bh->b_data+offset, tocopy);
5100                 brelse(bh);
5101                 offset = 0;
5102                 toread -= tocopy;
5103                 data += tocopy;
5104                 blk++;
5105         }
5106         return len;
5107 }
5108
5109 /* Write to quotafile (we know the transaction is already started and has
5110  * enough credits) */
5111 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5112                                 const char *data, size_t len, loff_t off)
5113 {
5114         struct inode *inode = sb_dqopt(sb)->files[type];
5115         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5116         int err = 0;
5117         int offset = off & (sb->s_blocksize - 1);
5118         struct buffer_head *bh;
5119         handle_t *handle = journal_current_handle();
5120
5121         if (EXT4_SB(sb)->s_journal && !handle) {
5122                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5123                         " cancelled because transaction is not started",
5124                         (unsigned long long)off, (unsigned long long)len);
5125                 return -EIO;
5126         }
5127         /*
5128          * Since we account only one data block in transaction credits,
5129          * then it is impossible to cross a block boundary.
5130          */
5131         if (sb->s_blocksize - offset < len) {
5132                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5133                         " cancelled because not block aligned",
5134                         (unsigned long long)off, (unsigned long long)len);
5135                 return -EIO;
5136         }
5137
5138         bh = ext4_bread(handle, inode, blk, 1, &err);
5139         if (!bh)
5140                 goto out;
5141         err = ext4_journal_get_write_access(handle, bh);
5142         if (err) {
5143                 brelse(bh);
5144                 goto out;
5145         }
5146         lock_buffer(bh);
5147         memcpy(bh->b_data+offset, data, len);
5148         flush_dcache_page(bh->b_page);
5149         unlock_buffer(bh);
5150         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5151         brelse(bh);
5152 out:
5153         if (err)
5154                 return err;
5155         if (inode->i_size < off + len) {
5156                 i_size_write(inode, off + len);
5157                 EXT4_I(inode)->i_disksize = inode->i_size;
5158                 ext4_mark_inode_dirty(handle, inode);
5159         }
5160         return len;
5161 }
5162
5163 #endif
5164
5165 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5166                        const char *dev_name, void *data)
5167 {
5168         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5169 }
5170
5171 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5172 static inline void register_as_ext2(void)
5173 {
5174         int err = register_filesystem(&ext2_fs_type);
5175         if (err)
5176                 printk(KERN_WARNING
5177                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5178 }
5179
5180 static inline void unregister_as_ext2(void)
5181 {
5182         unregister_filesystem(&ext2_fs_type);
5183 }
5184
5185 static inline int ext2_feature_set_ok(struct super_block *sb)
5186 {
5187         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5188                 return 0;
5189         if (sb->s_flags & MS_RDONLY)
5190                 return 1;
5191         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5192                 return 0;
5193         return 1;
5194 }
5195 MODULE_ALIAS("ext2");
5196 #else
5197 static inline void register_as_ext2(void) { }
5198 static inline void unregister_as_ext2(void) { }
5199 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5200 #endif
5201
5202 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5203 static inline void register_as_ext3(void)
5204 {
5205         int err = register_filesystem(&ext3_fs_type);
5206         if (err)
5207                 printk(KERN_WARNING
5208                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5209 }
5210
5211 static inline void unregister_as_ext3(void)
5212 {
5213         unregister_filesystem(&ext3_fs_type);
5214 }
5215
5216 static inline int ext3_feature_set_ok(struct super_block *sb)
5217 {
5218         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5219                 return 0;
5220         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5221                 return 0;
5222         if (sb->s_flags & MS_RDONLY)
5223                 return 1;
5224         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5225                 return 0;
5226         return 1;
5227 }
5228 MODULE_ALIAS("ext3");
5229 #else
5230 static inline void register_as_ext3(void) { }
5231 static inline void unregister_as_ext3(void) { }
5232 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5233 #endif
5234
5235 static struct file_system_type ext4_fs_type = {
5236         .owner          = THIS_MODULE,
5237         .name           = "ext4",
5238         .mount          = ext4_mount,
5239         .kill_sb        = kill_block_super,
5240         .fs_flags       = FS_REQUIRES_DEV,
5241 };
5242
5243 static int __init ext4_init_feat_adverts(void)
5244 {
5245         struct ext4_features *ef;
5246         int ret = -ENOMEM;
5247
5248         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5249         if (!ef)
5250                 goto out;
5251
5252         ef->f_kobj.kset = ext4_kset;
5253         init_completion(&ef->f_kobj_unregister);
5254         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5255                                    "features");
5256         if (ret) {
5257                 kfree(ef);
5258                 goto out;
5259         }
5260
5261         ext4_feat = ef;
5262         ret = 0;
5263 out:
5264         return ret;
5265 }
5266
5267 static void ext4_exit_feat_adverts(void)
5268 {
5269         kobject_put(&ext4_feat->f_kobj);
5270         wait_for_completion(&ext4_feat->f_kobj_unregister);
5271         kfree(ext4_feat);
5272 }
5273
5274 /* Shared across all ext4 file systems */
5275 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5276 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5277
5278 static int __init ext4_init_fs(void)
5279 {
5280         int i, err;
5281
5282         ext4_li_info = NULL;
5283         mutex_init(&ext4_li_mtx);
5284
5285         ext4_check_flag_values();
5286
5287         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5288                 mutex_init(&ext4__aio_mutex[i]);
5289                 init_waitqueue_head(&ext4__ioend_wq[i]);
5290         }
5291
5292         err = ext4_init_pageio();
5293         if (err)
5294                 return err;
5295         err = ext4_init_system_zone();
5296         if (err)
5297                 goto out6;
5298         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5299         if (!ext4_kset) {
5300                 err = -ENOMEM;
5301                 goto out5;
5302         }
5303         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5304
5305         err = ext4_init_feat_adverts();
5306         if (err)
5307                 goto out4;
5308
5309         err = ext4_init_mballoc();
5310         if (err)
5311                 goto out3;
5312
5313         err = ext4_init_xattr();
5314         if (err)
5315                 goto out2;
5316         err = init_inodecache();
5317         if (err)
5318                 goto out1;
5319         register_as_ext3();
5320         register_as_ext2();
5321         err = register_filesystem(&ext4_fs_type);
5322         if (err)
5323                 goto out;
5324
5325         return 0;
5326 out:
5327         unregister_as_ext2();
5328         unregister_as_ext3();
5329         destroy_inodecache();
5330 out1:
5331         ext4_exit_xattr();
5332 out2:
5333         ext4_exit_mballoc();
5334 out3:
5335         ext4_exit_feat_adverts();
5336 out4:
5337         if (ext4_proc_root)
5338                 remove_proc_entry("fs/ext4", NULL);
5339         kset_unregister(ext4_kset);
5340 out5:
5341         ext4_exit_system_zone();
5342 out6:
5343         ext4_exit_pageio();
5344         return err;
5345 }
5346
5347 static void __exit ext4_exit_fs(void)
5348 {
5349         ext4_destroy_lazyinit_thread();
5350         unregister_as_ext2();
5351         unregister_as_ext3();
5352         unregister_filesystem(&ext4_fs_type);
5353         destroy_inodecache();
5354         ext4_exit_xattr();
5355         ext4_exit_mballoc();
5356         ext4_exit_feat_adverts();
5357         remove_proc_entry("fs/ext4", NULL);
5358         kset_unregister(ext4_kset);
5359         ext4_exit_system_zone();
5360         ext4_exit_pageio();
5361 }
5362
5363 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5364 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5365 MODULE_LICENSE("GPL");
5366 module_init(ext4_init_fs)
5367 module_exit(ext4_exit_fs)