fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card ([email protected])
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller ([email protected]), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static int ext4_sync_fs_nojournal(struct super_block *sb, int wait);
  73 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  75 static int ext4_unfreeze(struct super_block *sb);
  76 static int ext4_freeze(struct super_block *sb);
  77 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  78                        const char *dev_name, void *data);
  79 static inline int ext2_feature_set_ok(struct super_block *sb);
  80 static inline int ext3_feature_set_ok(struct super_block *sb);
  81 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
  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 MODULE_ALIAS_FS("ext2");
  96 MODULE_ALIAS("ext2");
  97 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  98 #else
  99 #define IS_EXT2_SB(sb) (0)
 100 #endif
 101
 102
 103 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 104 static struct file_system_type ext3_fs_type = {
 105         .owner          = THIS_MODULE,
 106         .name           = "ext3",
 107         .mount          = ext4_mount,
 108         .kill_sb        = kill_block_super,
 109         .fs_flags       = FS_REQUIRES_DEV,
 110 };
 111 MODULE_ALIAS_FS("ext3");
 112 MODULE_ALIAS("ext3");
 113 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 114 #else
 115 #define IS_EXT3_SB(sb) (0)
 116 #endif
 117
 118 static int ext4_verify_csum_type(struct super_block *sb,
 119                                  struct ext4_super_block *es)
 120 {
 121         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 122                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 123                 return 1;
 124
 125         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 126 }
 127
 128 static __le32 ext4_superblock_csum(struct super_block *sb,
 129                                    struct ext4_super_block *es)
 130 {
 131         struct ext4_sb_info *sbi = EXT4_SB(sb);
 132         int offset = offsetof(struct ext4_super_block, s_checksum);
 133         __u32 csum;
 134
 135         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 136
 137         return cpu_to_le32(csum);
 138 }
 139
 140 int ext4_superblock_csum_verify(struct super_block *sb,
 141                                 struct ext4_super_block *es)
 142 {
 143         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 144                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 145                 return 1;
 146
 147         return es->s_checksum == ext4_superblock_csum(sb, es);
 148 }
 149
 150 void ext4_superblock_csum_set(struct super_block *sb)
 151 {
 152         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 153
 154         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 155                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 156                 return;
 157
 158         es->s_checksum = ext4_superblock_csum(sb, es);
 159 }
 160
 161 void *ext4_kvmalloc(size_t size, gfp_t flags)
 162 {
 163         void *ret;
 164
 165         ret = kmalloc(size, flags);
 166         if (!ret)
 167                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 168         return ret;
 169 }
 170
 171 void *ext4_kvzalloc(size_t size, gfp_t flags)
 172 {
 173         void *ret;
 174
 175         ret = kzalloc(size, flags);
 176         if (!ret)
 177                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 178         return ret;
 179 }
 180
 181 void ext4_kvfree(void *ptr)
 182 {
 183         if (is_vmalloc_addr(ptr))
 184                 vfree(ptr);
 185         else
 186                 kfree(ptr);
 187
 188 }
 189
 190 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 191                                struct ext4_group_desc *bg)
 192 {
 193         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 194                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 195                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 196 }
 197
 198 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 199                                struct ext4_group_desc *bg)
 200 {
 201         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 202                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 203                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 204 }
 205
 206 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 207                               struct ext4_group_desc *bg)
 208 {
 209         return le32_to_cpu(bg->bg_inode_table_lo) |
 210                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 211                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 212 }
 213
 214 __u32 ext4_free_group_clusters(struct super_block *sb,
 215                                struct ext4_group_desc *bg)
 216 {
 217         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 218                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 219                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 220 }
 221
 222 __u32 ext4_free_inodes_count(struct super_block *sb,
 223                               struct ext4_group_desc *bg)
 224 {
 225         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 226                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 227                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 228 }
 229
 230 __u32 ext4_used_dirs_count(struct super_block *sb,
 231                               struct ext4_group_desc *bg)
 232 {
 233         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 234                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 235                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 236 }
 237
 238 __u32 ext4_itable_unused_count(struct super_block *sb,
 239                               struct ext4_group_desc *bg)
 240 {
 241         return le16_to_cpu(bg->bg_itable_unused_lo) |
 242                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 243                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 244 }
 245
 246 void ext4_block_bitmap_set(struct super_block *sb,
 247                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 248 {
 249         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 250         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 251                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 252 }
 253
 254 void ext4_inode_bitmap_set(struct super_block *sb,
 255                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 256 {
 257         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 258         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 259                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 260 }
 261
 262 void ext4_inode_table_set(struct super_block *sb,
 263                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 264 {
 265         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 266         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 267                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 268 }
 269
 270 void ext4_free_group_clusters_set(struct super_block *sb,
 271                                   struct ext4_group_desc *bg, __u32 count)
 272 {
 273         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 274         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 275                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 276 }
 277
 278 void ext4_free_inodes_set(struct super_block *sb,
 279                           struct ext4_group_desc *bg, __u32 count)
 280 {
 281         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 282         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 283                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 284 }
 285
 286 void ext4_used_dirs_set(struct super_block *sb,
 287                           struct ext4_group_desc *bg, __u32 count)
 288 {
 289         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 290         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 291                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 292 }
 293
 294 void ext4_itable_unused_set(struct super_block *sb,
 295                           struct ext4_group_desc *bg, __u32 count)
 296 {
 297         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 298         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 299                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 300 }
 301
 302
 303 static void __save_error_info(struct super_block *sb, const char *func,
 304                             unsigned int line)
 305 {
 306         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 307
 308         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 309         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 310         es->s_last_error_time = cpu_to_le32(get_seconds());
 311         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 312         es->s_last_error_line = cpu_to_le32(line);
 313         if (!es->s_first_error_time) {
 314                 es->s_first_error_time = es->s_last_error_time;
 315                 strncpy(es->s_first_error_func, func,
 316                         sizeof(es->s_first_error_func));
 317                 es->s_first_error_line = cpu_to_le32(line);
 318                 es->s_first_error_ino = es->s_last_error_ino;
 319                 es->s_first_error_block = es->s_last_error_block;
 320         }
 321         /*
 322          * Start the daily error reporting function if it hasn't been
 323          * started already
 324          */
 325         if (!es->s_error_count)
 326                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 327         le32_add_cpu(&es->s_error_count, 1);
 328 }
 329
 330 static void save_error_info(struct super_block *sb, const char *func,
 331                             unsigned int line)
 332 {
 333         __save_error_info(sb, func, line);
 334         ext4_commit_super(sb, 1);
 335 }
 336
 337 /*
 338  * The del_gendisk() function uninitializes the disk-specific data
 339  * structures, including the bdi structure, without telling anyone
 340  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 341  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 342  * This is a kludge to prevent these oops until we can put in a proper
 343  * hook in del_gendisk() to inform the VFS and file system layers.
 344  */
 345 static int block_device_ejected(struct super_block *sb)
 346 {
 347         struct inode *bd_inode = sb->s_bdev->bd_inode;
 348         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 349
 350         return bdi->dev == NULL;
 351 }
 352
 353 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 354 {
 355         struct super_block              *sb = journal->j_private;
 356         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 357         int                             error = is_journal_aborted(journal);
 358         struct ext4_journal_cb_entry    *jce;
 359
 360         BUG_ON(txn->t_state == T_FINISHED);
 361         spin_lock(&sbi->s_md_lock);
 362         while (!list_empty(&txn->t_private_list)) {
 363                 jce = list_entry(txn->t_private_list.next,
 364                                  struct ext4_journal_cb_entry, jce_list);
 365                 list_del_init(&jce->jce_list);
 366                 spin_unlock(&sbi->s_md_lock);
 367                 jce->jce_func(sb, jce, error);
 368                 spin_lock(&sbi->s_md_lock);
 369         }
 370         spin_unlock(&sbi->s_md_lock);
 371 }
 372
 373 /* Deal with the reporting of failure conditions on a filesystem such as
 374  * inconsistencies detected or read IO failures.
 375  *
 376  * On ext2, we can store the error state of the filesystem in the
 377  * superblock.  That is not possible on ext4, because we may have other
 378  * write ordering constraints on the superblock which prevent us from
 379  * writing it out straight away; and given that the journal is about to
 380  * be aborted, we can't rely on the current, or future, transactions to
 381  * write out the superblock safely.
 382  *
 383  * We'll just use the jbd2_journal_abort() error code to record an error in
 384  * the journal instead.  On recovery, the journal will complain about
 385  * that error until we've noted it down and cleared it.
 386  */
 387
 388 static void ext4_handle_error(struct super_block *sb)
 389 {
 390         if (sb->s_flags & MS_RDONLY)
 391                 return;
 392
 393         if (!test_opt(sb, ERRORS_CONT)) {
 394                 journal_t *journal = EXT4_SB(sb)->s_journal;
 395
 396                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 397                 if (journal)
 398                         jbd2_journal_abort(journal, -EIO);
 399         }
 400         if (test_opt(sb, ERRORS_RO)) {
 401                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 402                 /*
 403                  * Make sure updated value of ->s_mount_flags will be visible
 404                  * before ->s_flags update
 405                  */
 406                 smp_wmb();
 407                 sb->s_flags |= MS_RDONLY;
 408         }
 409         if (test_opt(sb, ERRORS_PANIC))
 410                 panic("EXT4-fs (device %s): panic forced after error\n",
 411                         sb->s_id);
 412 }
 413
 414 void __ext4_error(struct super_block *sb, const char *function,
 415                   unsigned int line, const char *fmt, ...)
 416 {
 417         struct va_format vaf;
 418         va_list args;
 419
 420         va_start(args, fmt);
 421         vaf.fmt = fmt;
 422         vaf.va = &args;
 423         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 424                sb->s_id, function, line, current->comm, &vaf);
 425         va_end(args);
 426         save_error_info(sb, function, line);
 427
 428         ext4_handle_error(sb);
 429 }
 430
 431 void __ext4_error_inode(struct inode *inode, const char *function,
 432                         unsigned int line, ext4_fsblk_t block,
 433                         const char *fmt, ...)
 434 {
 435         va_list args;
 436         struct va_format vaf;
 437         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 438
 439         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 440         es->s_last_error_block = cpu_to_le64(block);
 441         save_error_info(inode->i_sb, function, line);
 442         va_start(args, fmt);
 443         vaf.fmt = fmt;
 444         vaf.va = &args;
 445         if (block)
 446                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 447                        "inode #%lu: block %llu: comm %s: %pV\n",
 448                        inode->i_sb->s_id, function, line, inode->i_ino,
 449                        block, current->comm, &vaf);
 450         else
 451                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 452                        "inode #%lu: comm %s: %pV\n",
 453                        inode->i_sb->s_id, function, line, inode->i_ino,
 454                        current->comm, &vaf);
 455         va_end(args);
 456
 457         ext4_handle_error(inode->i_sb);
 458 }
 459
 460 void __ext4_error_file(struct file *file, const char *function,
 461                        unsigned int line, ext4_fsblk_t block,
 462                        const char *fmt, ...)
 463 {
 464         va_list args;
 465         struct va_format vaf;
 466         struct ext4_super_block *es;
 467         struct inode *inode = file_inode(file);
 468         char pathname[80], *path;
 469
 470         es = EXT4_SB(inode->i_sb)->s_es;
 471         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 472         save_error_info(inode->i_sb, function, line);
 473         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 474         if (IS_ERR(path))
 475                 path = "(unknown)";
 476         va_start(args, fmt);
 477         vaf.fmt = fmt;
 478         vaf.va = &args;
 479         if (block)
 480                 printk(KERN_CRIT
 481                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 482                        "block %llu: comm %s: path %s: %pV\n",
 483                        inode->i_sb->s_id, function, line, inode->i_ino,
 484                        block, current->comm, path, &vaf);
 485         else
 486                 printk(KERN_CRIT
 487                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 488                        "comm %s: path %s: %pV\n",
 489                        inode->i_sb->s_id, function, line, inode->i_ino,
 490                        current->comm, path, &vaf);
 491         va_end(args);
 492
 493         ext4_handle_error(inode->i_sb);
 494 }
 495
 496 const char *ext4_decode_error(struct super_block *sb, int errno,
 497                               char nbuf[16])
 498 {
 499         char *errstr = NULL;
 500
 501         switch (errno) {
 502         case -EIO:
 503                 errstr = "IO failure";
 504                 break;
 505         case -ENOMEM:
 506                 errstr = "Out of memory";
 507                 break;
 508         case -EROFS:
 509                 if (!sb || (EXT4_SB(sb)->s_journal &&
 510                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 511                         errstr = "Journal has aborted";
 512                 else
 513                         errstr = "Readonly filesystem";
 514                 break;
 515         default:
 516                 /* If the caller passed in an extra buffer for unknown
 517                  * errors, textualise them now.  Else we just return
 518                  * NULL. */
 519                 if (nbuf) {
 520                         /* Check for truncated error codes... */
 521                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 522                                 errstr = nbuf;
 523                 }
 524                 break;
 525         }
 526
 527         return errstr;
 528 }
 529
 530 /* __ext4_std_error decodes expected errors from journaling functions
 531  * automatically and invokes the appropriate error response.  */
 532
 533 void __ext4_std_error(struct super_block *sb, const char *function,
 534                       unsigned int line, int errno)
 535 {
 536         char nbuf[16];
 537         const char *errstr;
 538
 539         /* Special case: if the error is EROFS, and we're not already
 540          * inside a transaction, then there's really no point in logging
 541          * an error. */
 542         if (errno == -EROFS && journal_current_handle() == NULL &&
 543             (sb->s_flags & MS_RDONLY))
 544                 return;
 545
 546         errstr = ext4_decode_error(sb, errno, nbuf);
 547         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 548                sb->s_id, function, line, errstr);
 549         save_error_info(sb, function, line);
 550
 551         ext4_handle_error(sb);
 552 }
 553
 554 /*
 555  * ext4_abort is a much stronger failure handler than ext4_error.  The
 556  * abort function may be used to deal with unrecoverable failures such
 557  * as journal IO errors or ENOMEM at a critical moment in log management.
 558  *
 559  * We unconditionally force the filesystem into an ABORT|READONLY state,
 560  * unless the error response on the fs has been set to panic in which
 561  * case we take the easy way out and panic immediately.
 562  */
 563
 564 void __ext4_abort(struct super_block *sb, const char *function,
 565                 unsigned int line, const char *fmt, ...)
 566 {
 567         va_list args;
 568
 569         save_error_info(sb, function, line);
 570         va_start(args, fmt);
 571         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 572                function, line);
 573         vprintk(fmt, args);
 574         printk("\n");
 575         va_end(args);
 576
 577         if ((sb->s_flags & MS_RDONLY) == 0) {
 578                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 579                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 580                 /*
 581                  * Make sure updated value of ->s_mount_flags will be visible
 582                  * before ->s_flags update
 583                  */
 584                 smp_wmb();
 585                 sb->s_flags |= MS_RDONLY;
 586                 if (EXT4_SB(sb)->s_journal)
 587                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 588                 save_error_info(sb, function, line);
 589         }
 590         if (test_opt(sb, ERRORS_PANIC))
 591                 panic("EXT4-fs panic from previous error\n");
 592 }
 593
 594 void __ext4_msg(struct super_block *sb,
 595                 const char *prefix, const char *fmt, ...)
 596 {
 597         struct va_format vaf;
 598         va_list args;
 599
 600         va_start(args, fmt);
 601         vaf.fmt = fmt;
 602         vaf.va = &args;
 603         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 604         va_end(args);
 605 }
 606
 607 void __ext4_warning(struct super_block *sb, const char *function,
 608                     unsigned int line, const char *fmt, ...)
 609 {
 610         struct va_format vaf;
 611         va_list args;
 612
 613         va_start(args, fmt);
 614         vaf.fmt = fmt;
 615         vaf.va = &args;
 616         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 617                sb->s_id, function, line, &vaf);
 618         va_end(args);
 619 }
 620
 621 void __ext4_grp_locked_error(const char *function, unsigned int line,
 622                              struct super_block *sb, ext4_group_t grp,
 623                              unsigned long ino, ext4_fsblk_t block,
 624                              const char *fmt, ...)
 625 __releases(bitlock)
 626 __acquires(bitlock)
 627 {
 628         struct va_format vaf;
 629         va_list args;
 630         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 631
 632         es->s_last_error_ino = cpu_to_le32(ino);
 633         es->s_last_error_block = cpu_to_le64(block);
 634         __save_error_info(sb, function, line);
 635
 636         va_start(args, fmt);
 637
 638         vaf.fmt = fmt;
 639         vaf.va = &args;
 640         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 641                sb->s_id, function, line, grp);
 642         if (ino)
 643                 printk(KERN_CONT "inode %lu: ", ino);
 644         if (block)
 645                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 646         printk(KERN_CONT "%pV\n", &vaf);
 647         va_end(args);
 648
 649         if (test_opt(sb, ERRORS_CONT)) {
 650                 ext4_commit_super(sb, 0);
 651                 return;
 652         }
 653
 654         ext4_unlock_group(sb, grp);
 655         ext4_handle_error(sb);
 656         /*
 657          * We only get here in the ERRORS_RO case; relocking the group
 658          * may be dangerous, but nothing bad will happen since the
 659          * filesystem will have already been marked read/only and the
 660          * journal has been aborted.  We return 1 as a hint to callers
 661          * who might what to use the return value from
 662          * ext4_grp_locked_error() to distinguish between the
 663          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 664          * aggressively from the ext4 function in question, with a
 665          * more appropriate error code.
 666          */
 667         ext4_lock_group(sb, grp);
 668         return;
 669 }
 670
 671 void ext4_update_dynamic_rev(struct super_block *sb)
 672 {
 673         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 674
 675         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 676                 return;
 677
 678         ext4_warning(sb,
 679                      "updating to rev %d because of new feature flag, "
 680                      "running e2fsck is recommended",
 681                      EXT4_DYNAMIC_REV);
 682
 683         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 684         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 685         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 686         /* leave es->s_feature_*compat flags alone */
 687         /* es->s_uuid will be set by e2fsck if empty */
 688
 689         /*
 690          * The rest of the superblock fields should be zero, and if not it
 691          * means they are likely already in use, so leave them alone.  We
 692          * can leave it up to e2fsck to clean up any inconsistencies there.
 693          */
 694 }
 695
 696 /*
 697  * Open the external journal device
 698  */
 699 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 700 {
 701         struct block_device *bdev;
 702         char b[BDEVNAME_SIZE];
 703
 704         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 705         if (IS_ERR(bdev))
 706                 goto fail;
 707         return bdev;
 708
 709 fail:
 710         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 711                         __bdevname(dev, b), PTR_ERR(bdev));
 712         return NULL;
 713 }
 714
 715 /*
 716  * Release the journal device
 717  */
 718 static void ext4_blkdev_put(struct block_device *bdev)
 719 {
 720         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 721 }
 722
 723 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 724 {
 725         struct block_device *bdev;
 726         bdev = sbi->journal_bdev;
 727         if (bdev) {
 728                 ext4_blkdev_put(bdev);
 729                 sbi->journal_bdev = NULL;
 730         }
 731 }
 732
 733 static inline struct inode *orphan_list_entry(struct list_head *l)
 734 {
 735         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 736 }
 737
 738 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 739 {
 740         struct list_head *l;
 741
 742         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 743                  le32_to_cpu(sbi->s_es->s_last_orphan));
 744
 745         printk(KERN_ERR "sb_info orphan list:\n");
 746         list_for_each(l, &sbi->s_orphan) {
 747                 struct inode *inode = orphan_list_entry(l);
 748                 printk(KERN_ERR "  "
 749                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 750                        inode->i_sb->s_id, inode->i_ino, inode,
 751                        inode->i_mode, inode->i_nlink,
 752                        NEXT_ORPHAN(inode));
 753         }
 754 }
 755
 756 static void ext4_put_super(struct super_block *sb)
 757 {
 758         struct ext4_sb_info *sbi = EXT4_SB(sb);
 759         struct ext4_super_block *es = sbi->s_es;
 760         int i, err;
 761
 762         ext4_unregister_li_request(sb);
 763         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 764
 765         flush_workqueue(sbi->unrsv_conversion_wq);
 766         flush_workqueue(sbi->rsv_conversion_wq);
 767         destroy_workqueue(sbi->unrsv_conversion_wq);
 768         destroy_workqueue(sbi->rsv_conversion_wq);
 769
 770         if (sbi->s_journal) {
 771                 err = jbd2_journal_destroy(sbi->s_journal);
 772                 sbi->s_journal = NULL;
 773                 if (err < 0)
 774                         ext4_abort(sb, "Couldn't clean up the journal");
 775         }
 776
 777         ext4_es_unregister_shrinker(sbi);
 778         del_timer(&sbi->s_err_report);
 779         ext4_release_system_zone(sb);
 780         ext4_mb_release(sb);
 781         ext4_ext_release(sb);
 782         ext4_xattr_put_super(sb);
 783
 784         if (!(sb->s_flags & MS_RDONLY)) {
 785                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 786                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 787         }
 788         if (!(sb->s_flags & MS_RDONLY))
 789                 ext4_commit_super(sb, 1);
 790
 791         if (sbi->s_proc) {
 792                 remove_proc_entry("options", sbi->s_proc);
 793                 remove_proc_entry(sb->s_id, ext4_proc_root);
 794         }
 795         kobject_del(&sbi->s_kobj);
 796
 797         for (i = 0; i < sbi->s_gdb_count; i++)
 798                 brelse(sbi->s_group_desc[i]);
 799         ext4_kvfree(sbi->s_group_desc);
 800         ext4_kvfree(sbi->s_flex_groups);
 801         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 802         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 803         percpu_counter_destroy(&sbi->s_dirs_counter);
 804         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 805         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
 806         brelse(sbi->s_sbh);
 807 #ifdef CONFIG_QUOTA
 808         for (i = 0; i < MAXQUOTAS; i++)
 809                 kfree(sbi->s_qf_names[i]);
 810 #endif
 811
 812         /* Debugging code just in case the in-memory inode orphan list
 813          * isn't empty.  The on-disk one can be non-empty if we've
 814          * detected an error and taken the fs readonly, but the
 815          * in-memory list had better be clean by this point. */
 816         if (!list_empty(&sbi->s_orphan))
 817                 dump_orphan_list(sb, sbi);
 818         J_ASSERT(list_empty(&sbi->s_orphan));
 819
 820         invalidate_bdev(sb->s_bdev);
 821         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 822                 /*
 823                  * Invalidate the journal device's buffers.  We don't want them
 824                  * floating about in memory - the physical journal device may
 825                  * hotswapped, and it breaks the `ro-after' testing code.
 826                  */
 827                 sync_blockdev(sbi->journal_bdev);
 828                 invalidate_bdev(sbi->journal_bdev);
 829                 ext4_blkdev_remove(sbi);
 830         }
 831         if (sbi->s_mmp_tsk)
 832                 kthread_stop(sbi->s_mmp_tsk);
 833         sb->s_fs_info = NULL;
 834         /*
 835          * Now that we are completely done shutting down the
 836          * superblock, we need to actually destroy the kobject.
 837          */
 838         kobject_put(&sbi->s_kobj);
 839         wait_for_completion(&sbi->s_kobj_unregister);
 840         if (sbi->s_chksum_driver)
 841                 crypto_free_shash(sbi->s_chksum_driver);
 842         kfree(sbi->s_blockgroup_lock);
 843         kfree(sbi);
 844 }
 845
 846 static struct kmem_cache *ext4_inode_cachep;
 847
 848 /*
 849  * Called inside transaction, so use GFP_NOFS
 850  */
 851 static struct inode *ext4_alloc_inode(struct super_block *sb)
 852 {
 853         struct ext4_inode_info *ei;
 854
 855         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 856         if (!ei)
 857                 return NULL;
 858
 859         ei->vfs_inode.i_version = 1;
 860         INIT_LIST_HEAD(&ei->i_prealloc_list);
 861         spin_lock_init(&ei->i_prealloc_lock);
 862         ext4_es_init_tree(&ei->i_es_tree);
 863         rwlock_init(&ei->i_es_lock);
 864         INIT_LIST_HEAD(&ei->i_es_lru);
 865         ei->i_es_lru_nr = 0;
 866         ei->i_touch_when = 0;
 867         ei->i_reserved_data_blocks = 0;
 868         ei->i_reserved_meta_blocks = 0;
 869         ei->i_allocated_meta_blocks = 0;
 870         ei->i_da_metadata_calc_len = 0;
 871         ei->i_da_metadata_calc_last_lblock = 0;
 872         spin_lock_init(&(ei->i_block_reservation_lock));
 873 #ifdef CONFIG_QUOTA
 874         ei->i_reserved_quota = 0;
 875 #endif
 876         ei->jinode = NULL;
 877         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 878         INIT_LIST_HEAD(&ei->i_unrsv_conversion_list);
 879         spin_lock_init(&ei->i_completed_io_lock);
 880         ei->i_sync_tid = 0;
 881         ei->i_datasync_tid = 0;
 882         atomic_set(&ei->i_ioend_count, 0);
 883         atomic_set(&ei->i_unwritten, 0);
 884         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 885         INIT_WORK(&ei->i_unrsv_conversion_work, ext4_end_io_unrsv_work);
 886
 887         return &ei->vfs_inode;
 888 }
 889
 890 static int ext4_drop_inode(struct inode *inode)
 891 {
 892         int drop = generic_drop_inode(inode);
 893
 894         trace_ext4_drop_inode(inode, drop);
 895         return drop;
 896 }
 897
 898 static void ext4_i_callback(struct rcu_head *head)
 899 {
 900         struct inode *inode = container_of(head, struct inode, i_rcu);
 901         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 902 }
 903
 904 static void ext4_destroy_inode(struct inode *inode)
 905 {
 906         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 907                 ext4_msg(inode->i_sb, KERN_ERR,
 908                          "Inode %lu (%p): orphan list check failed!",
 909                          inode->i_ino, EXT4_I(inode));
 910                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 911                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 912                                 true);
 913                 dump_stack();
 914         }
 915         call_rcu(&inode->i_rcu, ext4_i_callback);
 916 }
 917
 918 static void init_once(void *foo)
 919 {
 920         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 921
 922         INIT_LIST_HEAD(&ei->i_orphan);
 923         init_rwsem(&ei->xattr_sem);
 924         init_rwsem(&ei->i_data_sem);
 925         inode_init_once(&ei->vfs_inode);
 926 }
 927
 928 static int init_inodecache(void)
 929 {
 930         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 931                                              sizeof(struct ext4_inode_info),
 932                                              0, (SLAB_RECLAIM_ACCOUNT|
 933                                                 SLAB_MEM_SPREAD),
 934                                              init_once);
 935         if (ext4_inode_cachep == NULL)
 936                 return -ENOMEM;
 937         return 0;
 938 }
 939
 940 static void destroy_inodecache(void)
 941 {
 942         /*
 943          * Make sure all delayed rcu free inodes are flushed before we
 944          * destroy cache.
 945          */
 946         rcu_barrier();
 947         kmem_cache_destroy(ext4_inode_cachep);
 948 }
 949
 950 void ext4_clear_inode(struct inode *inode)
 951 {
 952         invalidate_inode_buffers(inode);
 953         clear_inode(inode);
 954         dquot_drop(inode);
 955         ext4_discard_preallocations(inode);
 956         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 957         ext4_es_lru_del(inode);
 958         if (EXT4_I(inode)->jinode) {
 959                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 960                                                EXT4_I(inode)->jinode);
 961                 jbd2_free_inode(EXT4_I(inode)->jinode);
 962                 EXT4_I(inode)->jinode = NULL;
 963         }
 964 }
 965
 966 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 967                                         u64 ino, u32 generation)
 968 {
 969         struct inode *inode;
 970
 971         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 972                 return ERR_PTR(-ESTALE);
 973         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 974                 return ERR_PTR(-ESTALE);
 975
 976         /* iget isn't really right if the inode is currently unallocated!!
 977          *
 978          * ext4_read_inode will return a bad_inode if the inode had been
 979          * deleted, so we should be safe.
 980          *
 981          * Currently we don't know the generation for parent directory, so
 982          * a generation of 0 means "accept any"
 983          */
 984         inode = ext4_iget(sb, ino);
 985         if (IS_ERR(inode))
 986                 return ERR_CAST(inode);
 987         if (generation && inode->i_generation != generation) {
 988                 iput(inode);
 989                 return ERR_PTR(-ESTALE);
 990         }
 991
 992         return inode;
 993 }
 994
 995 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 996                                         int fh_len, int fh_type)
 997 {
 998         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 999                                     ext4_nfs_get_inode);
1000 }
1001
1002 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1003                                         int fh_len, int fh_type)
1004 {
1005         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1006                                     ext4_nfs_get_inode);
1007 }
1008
1009 /*
1010  * Try to release metadata pages (indirect blocks, directories) which are
1011  * mapped via the block device.  Since these pages could have journal heads
1012  * which would prevent try_to_free_buffers() from freeing them, we must use
1013  * jbd2 layer's try_to_free_buffers() function to release them.
1014  */
1015 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1016                                  gfp_t wait)
1017 {
1018         journal_t *journal = EXT4_SB(sb)->s_journal;
1019
1020         WARN_ON(PageChecked(page));
1021         if (!page_has_buffers(page))
1022                 return 0;
1023         if (journal)
1024                 return jbd2_journal_try_to_free_buffers(journal, page,
1025                                                         wait & ~__GFP_WAIT);
1026         return try_to_free_buffers(page);
1027 }
1028
1029 #ifdef CONFIG_QUOTA
1030 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1031 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1032
1033 static int ext4_write_dquot(struct dquot *dquot);
1034 static int ext4_acquire_dquot(struct dquot *dquot);
1035 static int ext4_release_dquot(struct dquot *dquot);
1036 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1037 static int ext4_write_info(struct super_block *sb, int type);
1038 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1039                          struct path *path);
1040 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1041                                  int format_id);
1042 static int ext4_quota_off(struct super_block *sb, int type);
1043 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1044 static int ext4_quota_on_mount(struct super_block *sb, int type);
1045 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1046                                size_t len, loff_t off);
1047 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1048                                 const char *data, size_t len, loff_t off);
1049 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1050                              unsigned int flags);
1051 static int ext4_enable_quotas(struct super_block *sb);
1052
1053 static const struct dquot_operations ext4_quota_operations = {
1054         .get_reserved_space = ext4_get_reserved_space,
1055         .write_dquot    = ext4_write_dquot,
1056         .acquire_dquot  = ext4_acquire_dquot,
1057         .release_dquot  = ext4_release_dquot,
1058         .mark_dirty     = ext4_mark_dquot_dirty,
1059         .write_info     = ext4_write_info,
1060         .alloc_dquot    = dquot_alloc,
1061         .destroy_dquot  = dquot_destroy,
1062 };
1063
1064 static const struct quotactl_ops ext4_qctl_operations = {
1065         .quota_on       = ext4_quota_on,
1066         .quota_off      = ext4_quota_off,
1067         .quota_sync     = dquot_quota_sync,
1068         .get_info       = dquot_get_dqinfo,
1069         .set_info       = dquot_set_dqinfo,
1070         .get_dqblk      = dquot_get_dqblk,
1071         .set_dqblk      = dquot_set_dqblk
1072 };
1073
1074 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1075         .quota_on_meta  = ext4_quota_on_sysfile,
1076         .quota_off      = ext4_quota_off_sysfile,
1077         .quota_sync     = dquot_quota_sync,
1078         .get_info       = dquot_get_dqinfo,
1079         .set_info       = dquot_set_dqinfo,
1080         .get_dqblk      = dquot_get_dqblk,
1081         .set_dqblk      = dquot_set_dqblk
1082 };
1083 #endif
1084
1085 static const struct super_operations ext4_sops = {
1086         .alloc_inode    = ext4_alloc_inode,
1087         .destroy_inode  = ext4_destroy_inode,
1088         .write_inode    = ext4_write_inode,
1089         .dirty_inode    = ext4_dirty_inode,
1090         .drop_inode     = ext4_drop_inode,
1091         .evict_inode    = ext4_evict_inode,
1092         .put_super      = ext4_put_super,
1093         .sync_fs        = ext4_sync_fs,
1094         .freeze_fs      = ext4_freeze,
1095         .unfreeze_fs    = ext4_unfreeze,
1096         .statfs         = ext4_statfs,
1097         .remount_fs     = ext4_remount,
1098         .show_options   = ext4_show_options,
1099 #ifdef CONFIG_QUOTA
1100         .quota_read     = ext4_quota_read,
1101         .quota_write    = ext4_quota_write,
1102 #endif
1103         .bdev_try_to_free_page = bdev_try_to_free_page,
1104 };
1105
1106 static const struct super_operations ext4_nojournal_sops = {
1107         .alloc_inode    = ext4_alloc_inode,
1108         .destroy_inode  = ext4_destroy_inode,
1109         .write_inode    = ext4_write_inode,
1110         .dirty_inode    = ext4_dirty_inode,
1111         .drop_inode     = ext4_drop_inode,
1112         .evict_inode    = ext4_evict_inode,
1113         .sync_fs        = ext4_sync_fs_nojournal,
1114         .put_super      = ext4_put_super,
1115         .statfs         = ext4_statfs,
1116         .remount_fs     = ext4_remount,
1117         .show_options   = ext4_show_options,
1118 #ifdef CONFIG_QUOTA
1119         .quota_read     = ext4_quota_read,
1120         .quota_write    = ext4_quota_write,
1121 #endif
1122         .bdev_try_to_free_page = bdev_try_to_free_page,
1123 };
1124
1125 static const struct export_operations ext4_export_ops = {
1126         .fh_to_dentry = ext4_fh_to_dentry,
1127         .fh_to_parent = ext4_fh_to_parent,
1128         .get_parent = ext4_get_parent,
1129 };
1130
1131 enum {
1132         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1133         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1134         Opt_nouid32, Opt_debug, Opt_removed,
1135         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1136         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1137         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1138         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1139         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1140         Opt_data_err_abort, Opt_data_err_ignore,
1141         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1142         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1143         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1144         Opt_usrquota, Opt_grpquota, Opt_i_version,
1145         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1146         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1147         Opt_inode_readahead_blks, Opt_journal_ioprio,
1148         Opt_dioread_nolock, Opt_dioread_lock,
1149         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1150         Opt_max_dir_size_kb,
1151 };
1152
1153 static const match_table_t tokens = {
1154         {Opt_bsd_df, "bsddf"},
1155         {Opt_minix_df, "minixdf"},
1156         {Opt_grpid, "grpid"},
1157         {Opt_grpid, "bsdgroups"},
1158         {Opt_nogrpid, "nogrpid"},
1159         {Opt_nogrpid, "sysvgroups"},
1160         {Opt_resgid, "resgid=%u"},
1161         {Opt_resuid, "resuid=%u"},
1162         {Opt_sb, "sb=%u"},
1163         {Opt_err_cont, "errors=continue"},
1164         {Opt_err_panic, "errors=panic"},
1165         {Opt_err_ro, "errors=remount-ro"},
1166         {Opt_nouid32, "nouid32"},
1167         {Opt_debug, "debug"},
1168         {Opt_removed, "oldalloc"},
1169         {Opt_removed, "orlov"},
1170         {Opt_user_xattr, "user_xattr"},
1171         {Opt_nouser_xattr, "nouser_xattr"},
1172         {Opt_acl, "acl"},
1173         {Opt_noacl, "noacl"},
1174         {Opt_noload, "norecovery"},
1175         {Opt_noload, "noload"},
1176         {Opt_removed, "nobh"},
1177         {Opt_removed, "bh"},
1178         {Opt_commit, "commit=%u"},
1179         {Opt_min_batch_time, "min_batch_time=%u"},
1180         {Opt_max_batch_time, "max_batch_time=%u"},
1181         {Opt_journal_dev, "journal_dev=%u"},
1182         {Opt_journal_checksum, "journal_checksum"},
1183         {Opt_journal_async_commit, "journal_async_commit"},
1184         {Opt_abort, "abort"},
1185         {Opt_data_journal, "data=journal"},
1186         {Opt_data_ordered, "data=ordered"},
1187         {Opt_data_writeback, "data=writeback"},
1188         {Opt_data_err_abort, "data_err=abort"},
1189         {Opt_data_err_ignore, "data_err=ignore"},
1190         {Opt_offusrjquota, "usrjquota="},
1191         {Opt_usrjquota, "usrjquota=%s"},
1192         {Opt_offgrpjquota, "grpjquota="},
1193         {Opt_grpjquota, "grpjquota=%s"},
1194         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1195         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1196         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1197         {Opt_grpquota, "grpquota"},
1198         {Opt_noquota, "noquota"},
1199         {Opt_quota, "quota"},
1200         {Opt_usrquota, "usrquota"},
1201         {Opt_barrier, "barrier=%u"},
1202         {Opt_barrier, "barrier"},
1203         {Opt_nobarrier, "nobarrier"},
1204         {Opt_i_version, "i_version"},
1205         {Opt_stripe, "stripe=%u"},
1206         {Opt_delalloc, "delalloc"},
1207         {Opt_nodelalloc, "nodelalloc"},
1208         {Opt_removed, "mblk_io_submit"},
1209         {Opt_removed, "nomblk_io_submit"},
1210         {Opt_block_validity, "block_validity"},
1211         {Opt_noblock_validity, "noblock_validity"},
1212         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1213         {Opt_journal_ioprio, "journal_ioprio=%u"},
1214         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1215         {Opt_auto_da_alloc, "auto_da_alloc"},
1216         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1217         {Opt_dioread_nolock, "dioread_nolock"},
1218         {Opt_dioread_lock, "dioread_lock"},
1219         {Opt_discard, "discard"},
1220         {Opt_nodiscard, "nodiscard"},
1221         {Opt_init_itable, "init_itable=%u"},
1222         {Opt_init_itable, "init_itable"},
1223         {Opt_noinit_itable, "noinit_itable"},
1224         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1225         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1226         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1227         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1228         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1229         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1230         {Opt_err, NULL},
1231 };
1232
1233 static ext4_fsblk_t get_sb_block(void **data)
1234 {
1235         ext4_fsblk_t    sb_block;
1236         char            *options = (char *) *data;
1237
1238         if (!options || strncmp(options, "sb=", 3) != 0)
1239                 return 1;       /* Default location */
1240
1241         options += 3;
1242         /* TODO: use simple_strtoll with >32bit ext4 */
1243         sb_block = simple_strtoul(options, &options, 0);
1244         if (*options && *options != ',') {
1245                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1246                        (char *) *data);
1247                 return 1;
1248         }
1249         if (*options == ',')
1250                 options++;
1251         *data = (void *) options;
1252
1253         return sb_block;
1254 }
1255
1256 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1257 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1258         "Contact [email protected] if you think we should keep it.\n";
1259
1260 #ifdef CONFIG_QUOTA
1261 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1262 {
1263         struct ext4_sb_info *sbi = EXT4_SB(sb);
1264         char *qname;
1265         int ret = -1;
1266
1267         if (sb_any_quota_loaded(sb) &&
1268                 !sbi->s_qf_names[qtype]) {
1269                 ext4_msg(sb, KERN_ERR,
1270                         "Cannot change journaled "
1271                         "quota options when quota turned on");
1272                 return -1;
1273         }
1274         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1275                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1276                          "when QUOTA feature is enabled");
1277                 return -1;
1278         }
1279         qname = match_strdup(args);
1280         if (!qname) {
1281                 ext4_msg(sb, KERN_ERR,
1282                         "Not enough memory for storing quotafile name");
1283                 return -1;
1284         }
1285         if (sbi->s_qf_names[qtype]) {
1286                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1287                         ret = 1;
1288                 else
1289                         ext4_msg(sb, KERN_ERR,
1290                                  "%s quota file already specified",
1291                                  QTYPE2NAME(qtype));
1292                 goto errout;
1293         }
1294         if (strchr(qname, '/')) {
1295                 ext4_msg(sb, KERN_ERR,
1296                         "quotafile must be on filesystem root");
1297                 goto errout;
1298         }
1299         sbi->s_qf_names[qtype] = qname;
1300         set_opt(sb, QUOTA);
1301         return 1;
1302 errout:
1303         kfree(qname);
1304         return ret;
1305 }
1306
1307 static int clear_qf_name(struct super_block *sb, int qtype)
1308 {
1309
1310         struct ext4_sb_info *sbi = EXT4_SB(sb);
1311
1312         if (sb_any_quota_loaded(sb) &&
1313                 sbi->s_qf_names[qtype]) {
1314                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1315                         " when quota turned on");
1316                 return -1;
1317         }
1318         kfree(sbi->s_qf_names[qtype]);
1319         sbi->s_qf_names[qtype] = NULL;
1320         return 1;
1321 }
1322 #endif
1323
1324 #define MOPT_SET        0x0001
1325 #define MOPT_CLEAR      0x0002
1326 #define MOPT_NOSUPPORT  0x0004
1327 #define MOPT_EXPLICIT   0x0008
1328 #define MOPT_CLEAR_ERR  0x0010
1329 #define MOPT_GTE0       0x0020
1330 #ifdef CONFIG_QUOTA
1331 #define MOPT_Q          0
1332 #define MOPT_QFMT       0x0040
1333 #else
1334 #define MOPT_Q          MOPT_NOSUPPORT
1335 #define MOPT_QFMT       MOPT_NOSUPPORT
1336 #endif
1337 #define MOPT_DATAJ      0x0080
1338 #define MOPT_NO_EXT2    0x0100
1339 #define MOPT_NO_EXT3    0x0200
1340 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1341
1342 static const struct mount_opts {
1343         int     token;
1344         int     mount_opt;
1345         int     flags;
1346 } ext4_mount_opts[] = {
1347         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1348         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1349         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1350         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1351         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1352         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1353         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1354          MOPT_EXT4_ONLY | MOPT_SET},
1355         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1356          MOPT_EXT4_ONLY | MOPT_CLEAR},
1357         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1358         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1359         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1360          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1361         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1362          MOPT_EXT4_ONLY | MOPT_CLEAR},
1363         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1364          MOPT_EXT4_ONLY | MOPT_SET},
1365         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1366                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1367          MOPT_EXT4_ONLY | MOPT_SET},
1368         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1369         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1370         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1371         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1372         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1373          MOPT_NO_EXT2 | MOPT_SET},
1374         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1375          MOPT_NO_EXT2 | MOPT_CLEAR},
1376         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1377         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1378         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1379         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1380         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1381         {Opt_commit, 0, MOPT_GTE0},
1382         {Opt_max_batch_time, 0, MOPT_GTE0},
1383         {Opt_min_batch_time, 0, MOPT_GTE0},
1384         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1385         {Opt_init_itable, 0, MOPT_GTE0},
1386         {Opt_stripe, 0, MOPT_GTE0},
1387         {Opt_resuid, 0, MOPT_GTE0},
1388         {Opt_resgid, 0, MOPT_GTE0},
1389         {Opt_journal_dev, 0, MOPT_GTE0},
1390         {Opt_journal_ioprio, 0, MOPT_GTE0},
1391         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1392         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1393         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1394          MOPT_NO_EXT2 | MOPT_DATAJ},
1395         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1396         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1397 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1398         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1399         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1400 #else
1401         {Opt_acl, 0, MOPT_NOSUPPORT},
1402         {Opt_noacl, 0, MOPT_NOSUPPORT},
1403 #endif
1404         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1405         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1406         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1407         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1408                                                         MOPT_SET | MOPT_Q},
1409         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1410                                                         MOPT_SET | MOPT_Q},
1411         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1412                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1413         {Opt_usrjquota, 0, MOPT_Q},
1414         {Opt_grpjquota, 0, MOPT_Q},
1415         {Opt_offusrjquota, 0, MOPT_Q},
1416         {Opt_offgrpjquota, 0, MOPT_Q},
1417         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1418         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1419         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1420         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1421         {Opt_err, 0, 0}
1422 };
1423
1424 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1425                             substring_t *args, unsigned long *journal_devnum,
1426                             unsigned int *journal_ioprio, int is_remount)
1427 {
1428         struct ext4_sb_info *sbi = EXT4_SB(sb);
1429         const struct mount_opts *m;
1430         kuid_t uid;
1431         kgid_t gid;
1432         int arg = 0;
1433
1434 #ifdef CONFIG_QUOTA
1435         if (token == Opt_usrjquota)
1436                 return set_qf_name(sb, USRQUOTA, &args[0]);
1437         else if (token == Opt_grpjquota)
1438                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1439         else if (token == Opt_offusrjquota)
1440                 return clear_qf_name(sb, USRQUOTA);
1441         else if (token == Opt_offgrpjquota)
1442                 return clear_qf_name(sb, GRPQUOTA);
1443 #endif
1444         switch (token) {
1445         case Opt_noacl:
1446         case Opt_nouser_xattr:
1447                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1448                 break;
1449         case Opt_sb:
1450                 return 1;       /* handled by get_sb_block() */
1451         case Opt_removed:
1452                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1453                 return 1;
1454         case Opt_abort:
1455                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1456                 return 1;
1457         case Opt_i_version:
1458                 sb->s_flags |= MS_I_VERSION;
1459                 return 1;
1460         }
1461
1462         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1463                 if (token == m->token)
1464                         break;
1465
1466         if (m->token == Opt_err) {
1467                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1468                          "or missing value", opt);
1469                 return -1;
1470         }
1471
1472         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1473                 ext4_msg(sb, KERN_ERR,
1474                          "Mount option \"%s\" incompatible with ext2", opt);
1475                 return -1;
1476         }
1477         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1478                 ext4_msg(sb, KERN_ERR,
1479                          "Mount option \"%s\" incompatible with ext3", opt);
1480                 return -1;
1481         }
1482
1483         if (args->from && match_int(args, &arg))
1484                 return -1;
1485         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1486                 return -1;
1487         if (m->flags & MOPT_EXPLICIT)
1488                 set_opt2(sb, EXPLICIT_DELALLOC);
1489         if (m->flags & MOPT_CLEAR_ERR)
1490                 clear_opt(sb, ERRORS_MASK);
1491         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1492                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1493                          "options when quota turned on");
1494                 return -1;
1495         }
1496
1497         if (m->flags & MOPT_NOSUPPORT) {
1498                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1499         } else if (token == Opt_commit) {
1500                 if (arg == 0)
1501                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1502                 sbi->s_commit_interval = HZ * arg;
1503         } else if (token == Opt_max_batch_time) {
1504                 if (arg == 0)
1505                         arg = EXT4_DEF_MAX_BATCH_TIME;
1506                 sbi->s_max_batch_time = arg;
1507         } else if (token == Opt_min_batch_time) {
1508                 sbi->s_min_batch_time = arg;
1509         } else if (token == Opt_inode_readahead_blks) {
1510                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1511                         ext4_msg(sb, KERN_ERR,
1512                                  "EXT4-fs: inode_readahead_blks must be "
1513                                  "0 or a power of 2 smaller than 2^31");
1514                         return -1;
1515                 }
1516                 sbi->s_inode_readahead_blks = arg;
1517         } else if (token == Opt_init_itable) {
1518                 set_opt(sb, INIT_INODE_TABLE);
1519                 if (!args->from)
1520                         arg = EXT4_DEF_LI_WAIT_MULT;
1521                 sbi->s_li_wait_mult = arg;
1522         } else if (token == Opt_max_dir_size_kb) {
1523                 sbi->s_max_dir_size_kb = arg;
1524         } else if (token == Opt_stripe) {
1525                 sbi->s_stripe = arg;
1526         } else if (token == Opt_resuid) {
1527                 uid = make_kuid(current_user_ns(), arg);
1528                 if (!uid_valid(uid)) {
1529                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1530                         return -1;
1531                 }
1532                 sbi->s_resuid = uid;
1533         } else if (token == Opt_resgid) {
1534                 gid = make_kgid(current_user_ns(), arg);
1535                 if (!gid_valid(gid)) {
1536                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1537                         return -1;
1538                 }
1539                 sbi->s_resgid = gid;
1540         } else if (token == Opt_journal_dev) {
1541                 if (is_remount) {
1542                         ext4_msg(sb, KERN_ERR,
1543                                  "Cannot specify journal on remount");
1544                         return -1;
1545                 }
1546                 *journal_devnum = arg;
1547         } else if (token == Opt_journal_ioprio) {
1548                 if (arg > 7) {
1549                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1550                                  " (must be 0-7)");
1551                         return -1;
1552                 }
1553                 *journal_ioprio =
1554                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1555         } else if (m->flags & MOPT_DATAJ) {
1556                 if (is_remount) {
1557                         if (!sbi->s_journal)
1558                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1559                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1560                                 ext4_msg(sb, KERN_ERR,
1561                                          "Cannot change data mode on remount");
1562                                 return -1;
1563                         }
1564                 } else {
1565                         clear_opt(sb, DATA_FLAGS);
1566                         sbi->s_mount_opt |= m->mount_opt;
1567                 }
1568 #ifdef CONFIG_QUOTA
1569         } else if (m->flags & MOPT_QFMT) {
1570                 if (sb_any_quota_loaded(sb) &&
1571                     sbi->s_jquota_fmt != m->mount_opt) {
1572                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1573                                  "quota options when quota turned on");
1574                         return -1;
1575                 }
1576                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1577                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1578                         ext4_msg(sb, KERN_ERR,
1579                                  "Cannot set journaled quota options "
1580                                  "when QUOTA feature is enabled");
1581                         return -1;
1582                 }
1583                 sbi->s_jquota_fmt = m->mount_opt;
1584 #endif
1585         } else {
1586                 if (!args->from)
1587                         arg = 1;
1588                 if (m->flags & MOPT_CLEAR)
1589                         arg = !arg;
1590                 else if (unlikely(!(m->flags & MOPT_SET))) {
1591                         ext4_msg(sb, KERN_WARNING,
1592                                  "buggy handling of option %s", opt);
1593                         WARN_ON(1);
1594                         return -1;
1595                 }
1596                 if (arg != 0)
1597                         sbi->s_mount_opt |= m->mount_opt;
1598                 else
1599                         sbi->s_mount_opt &= ~m->mount_opt;
1600         }
1601         return 1;
1602 }
1603
1604 static int parse_options(char *options, struct super_block *sb,
1605                          unsigned long *journal_devnum,
1606                          unsigned int *journal_ioprio,
1607                          int is_remount)
1608 {
1609         struct ext4_sb_info *sbi = EXT4_SB(sb);
1610         char *p;
1611         substring_t args[MAX_OPT_ARGS];
1612         int token;
1613
1614         if (!options)
1615                 return 1;
1616
1617         while ((p = strsep(&options, ",")) != NULL) {
1618                 if (!*p)
1619                         continue;
1620                 /*
1621                  * Initialize args struct so we know whether arg was
1622                  * found; some options take optional arguments.
1623                  */
1624                 args[0].to = args[0].from = NULL;
1625                 token = match_token(p, tokens, args);
1626                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1627                                      journal_ioprio, is_remount) < 0)
1628                         return 0;
1629         }
1630 #ifdef CONFIG_QUOTA
1631         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1632             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1633                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1634                          "feature is enabled");
1635                 return 0;
1636         }
1637         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1638                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1639                         clear_opt(sb, USRQUOTA);
1640
1641                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1642                         clear_opt(sb, GRPQUOTA);
1643
1644                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1645                         ext4_msg(sb, KERN_ERR, "old and new quota "
1646                                         "format mixing");
1647                         return 0;
1648                 }
1649
1650                 if (!sbi->s_jquota_fmt) {
1651                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1652                                         "not specified");
1653                         return 0;
1654                 }
1655         } else {
1656                 if (sbi->s_jquota_fmt) {
1657                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1658                                         "specified with no journaling "
1659                                         "enabled");
1660                         return 0;
1661                 }
1662         }
1663 #endif
1664         if (test_opt(sb, DIOREAD_NOLOCK)) {
1665                 int blocksize =
1666                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1667
1668                 if (blocksize < PAGE_CACHE_SIZE) {
1669                         ext4_msg(sb, KERN_ERR, "can't mount with "
1670                                  "dioread_nolock if block size != PAGE_SIZE");
1671                         return 0;
1672                 }
1673         }
1674         return 1;
1675 }
1676
1677 static inline void ext4_show_quota_options(struct seq_file *seq,
1678                                            struct super_block *sb)
1679 {
1680 #if defined(CONFIG_QUOTA)
1681         struct ext4_sb_info *sbi = EXT4_SB(sb);
1682
1683         if (sbi->s_jquota_fmt) {
1684                 char *fmtname = "";
1685
1686                 switch (sbi->s_jquota_fmt) {
1687                 case QFMT_VFS_OLD:
1688                         fmtname = "vfsold";
1689                         break;
1690                 case QFMT_VFS_V0:
1691                         fmtname = "vfsv0";
1692                         break;
1693                 case QFMT_VFS_V1:
1694                         fmtname = "vfsv1";
1695                         break;
1696                 }
1697                 seq_printf(seq, ",jqfmt=%s", fmtname);
1698         }
1699
1700         if (sbi->s_qf_names[USRQUOTA])
1701                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1702
1703         if (sbi->s_qf_names[GRPQUOTA])
1704                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1705 #endif
1706 }
1707
1708 static const char *token2str(int token)
1709 {
1710         const struct match_token *t;
1711
1712         for (t = tokens; t->token != Opt_err; t++)
1713                 if (t->token == token && !strchr(t->pattern, '='))
1714                         break;
1715         return t->pattern;
1716 }
1717
1718 /*
1719  * Show an option if
1720  *  - it's set to a non-default value OR
1721  *  - if the per-sb default is different from the global default
1722  */
1723 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1724                               int nodefs)
1725 {
1726         struct ext4_sb_info *sbi = EXT4_SB(sb);
1727         struct ext4_super_block *es = sbi->s_es;
1728         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1729         const struct mount_opts *m;
1730         char sep = nodefs ? '\n' : ',';
1731
1732 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1733 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1734
1735         if (sbi->s_sb_block != 1)
1736                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1737
1738         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1739                 int want_set = m->flags & MOPT_SET;
1740                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1741                     (m->flags & MOPT_CLEAR_ERR))
1742                         continue;
1743                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1744                         continue; /* skip if same as the default */
1745                 if ((want_set &&
1746                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1747                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1748                         continue; /* select Opt_noFoo vs Opt_Foo */
1749                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1750         }
1751
1752         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1753             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1754                 SEQ_OPTS_PRINT("resuid=%u",
1755                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1756         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1757             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1758                 SEQ_OPTS_PRINT("resgid=%u",
1759                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1760         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1761         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1762                 SEQ_OPTS_PUTS("errors=remount-ro");
1763         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1764                 SEQ_OPTS_PUTS("errors=continue");
1765         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1766                 SEQ_OPTS_PUTS("errors=panic");
1767         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1768                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1769         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1770                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1771         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1772                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1773         if (sb->s_flags & MS_I_VERSION)
1774                 SEQ_OPTS_PUTS("i_version");
1775         if (nodefs || sbi->s_stripe)
1776                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1777         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1778                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1779                         SEQ_OPTS_PUTS("data=journal");
1780                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1781                         SEQ_OPTS_PUTS("data=ordered");
1782                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1783                         SEQ_OPTS_PUTS("data=writeback");
1784         }
1785         if (nodefs ||
1786             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1787                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1788                                sbi->s_inode_readahead_blks);
1789
1790         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1791                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1792                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1793         if (nodefs || sbi->s_max_dir_size_kb)
1794                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1795
1796         ext4_show_quota_options(seq, sb);
1797         return 0;
1798 }
1799
1800 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1801 {
1802         return _ext4_show_options(seq, root->d_sb, 0);
1803 }
1804
1805 static int options_seq_show(struct seq_file *seq, void *offset)
1806 {
1807         struct super_block *sb = seq->private;
1808         int rc;
1809
1810         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1811         rc = _ext4_show_options(seq, sb, 1);
1812         seq_puts(seq, "\n");
1813         return rc;
1814 }
1815
1816 static int options_open_fs(struct inode *inode, struct file *file)
1817 {
1818         return single_open(file, options_seq_show, PDE_DATA(inode));
1819 }
1820
1821 static const struct file_operations ext4_seq_options_fops = {
1822         .owner = THIS_MODULE,
1823         .open = options_open_fs,
1824         .read = seq_read,
1825         .llseek = seq_lseek,
1826         .release = single_release,
1827 };
1828
1829 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1830                             int read_only)
1831 {
1832         struct ext4_sb_info *sbi = EXT4_SB(sb);
1833         int res = 0;
1834
1835         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1836                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1837                          "forcing read-only mode");
1838                 res = MS_RDONLY;
1839         }
1840         if (read_only)
1841                 goto done;
1842         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1843                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1844                          "running e2fsck is recommended");
1845         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1846                 ext4_msg(sb, KERN_WARNING,
1847                          "warning: mounting fs with errors, "
1848                          "running e2fsck is recommended");
1849         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1850                  le16_to_cpu(es->s_mnt_count) >=
1851                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1852                 ext4_msg(sb, KERN_WARNING,
1853                          "warning: maximal mount count reached, "
1854                          "running e2fsck is recommended");
1855         else if (le32_to_cpu(es->s_checkinterval) &&
1856                 (le32_to_cpu(es->s_lastcheck) +
1857                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1858                 ext4_msg(sb, KERN_WARNING,
1859                          "warning: checktime reached, "
1860                          "running e2fsck is recommended");
1861         if (!sbi->s_journal)
1862                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1863         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1864                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1865         le16_add_cpu(&es->s_mnt_count, 1);
1866         es->s_mtime = cpu_to_le32(get_seconds());
1867         ext4_update_dynamic_rev(sb);
1868         if (sbi->s_journal)
1869                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1870
1871         ext4_commit_super(sb, 1);
1872 done:
1873         if (test_opt(sb, DEBUG))
1874                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1875                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1876                         sb->s_blocksize,
1877                         sbi->s_groups_count,
1878                         EXT4_BLOCKS_PER_GROUP(sb),
1879                         EXT4_INODES_PER_GROUP(sb),
1880                         sbi->s_mount_opt, sbi->s_mount_opt2);
1881
1882         cleancache_init_fs(sb);
1883         return res;
1884 }
1885
1886 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1887 {
1888         struct ext4_sb_info *sbi = EXT4_SB(sb);
1889         struct flex_groups *new_groups;
1890         int size;
1891
1892         if (!sbi->s_log_groups_per_flex)
1893                 return 0;
1894
1895         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1896         if (size <= sbi->s_flex_groups_allocated)
1897                 return 0;
1898
1899         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1900         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1901         if (!new_groups) {
1902                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1903                          size / (int) sizeof(struct flex_groups));
1904                 return -ENOMEM;
1905         }
1906
1907         if (sbi->s_flex_groups) {
1908                 memcpy(new_groups, sbi->s_flex_groups,
1909                        (sbi->s_flex_groups_allocated *
1910                         sizeof(struct flex_groups)));
1911                 ext4_kvfree(sbi->s_flex_groups);
1912         }
1913         sbi->s_flex_groups = new_groups;
1914         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1915         return 0;
1916 }
1917
1918 static int ext4_fill_flex_info(struct super_block *sb)
1919 {
1920         struct ext4_sb_info *sbi = EXT4_SB(sb);
1921         struct ext4_group_desc *gdp = NULL;
1922         ext4_group_t flex_group;
1923         int i, err;
1924
1925         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1926         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1927                 sbi->s_log_groups_per_flex = 0;
1928                 return 1;
1929         }
1930
1931         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1932         if (err)
1933                 goto failed;
1934
1935         for (i = 0; i < sbi->s_groups_count; i++) {
1936                 gdp = ext4_get_group_desc(sb, i, NULL);
1937
1938                 flex_group = ext4_flex_group(sbi, i);
1939                 atomic_add(ext4_free_inodes_count(sb, gdp),
1940                            &sbi->s_flex_groups[flex_group].free_inodes);
1941                 atomic64_add(ext4_free_group_clusters(sb, gdp),
1942                              &sbi->s_flex_groups[flex_group].free_clusters);
1943                 atomic_add(ext4_used_dirs_count(sb, gdp),
1944                            &sbi->s_flex_groups[flex_group].used_dirs);
1945         }
1946
1947         return 1;
1948 failed:
1949         return 0;
1950 }
1951
1952 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1953                                    struct ext4_group_desc *gdp)
1954 {
1955         int offset;
1956         __u16 crc = 0;
1957         __le32 le_group = cpu_to_le32(block_group);
1958
1959         if ((sbi->s_es->s_feature_ro_compat &
1960              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1961                 /* Use new metadata_csum algorithm */
1962                 __le16 save_csum;
1963                 __u32 csum32;
1964
1965                 save_csum = gdp->bg_checksum;
1966                 gdp->bg_checksum = 0;
1967                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1968                                      sizeof(le_group));
1969                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1970                                      sbi->s_desc_size);
1971                 gdp->bg_checksum = save_csum;
1972
1973                 crc = csum32 & 0xFFFF;
1974                 goto out;
1975         }
1976
1977         /* old crc16 code */
1978         offset = offsetof(struct ext4_group_desc, bg_checksum);
1979
1980         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1981         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1982         crc = crc16(crc, (__u8 *)gdp, offset);
1983         offset += sizeof(gdp->bg_checksum); /* skip checksum */
1984         /* for checksum of struct ext4_group_desc do the rest...*/
1985         if ((sbi->s_es->s_feature_incompat &
1986              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1987             offset < le16_to_cpu(sbi->s_es->s_desc_size))
1988                 crc = crc16(crc, (__u8 *)gdp + offset,
1989                             le16_to_cpu(sbi->s_es->s_desc_size) -
1990                                 offset);
1991
1992 out:
1993         return cpu_to_le16(crc);
1994 }
1995
1996 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
1997                                 struct ext4_group_desc *gdp)
1998 {
1999         if (ext4_has_group_desc_csum(sb) &&
2000             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2001                                                       block_group, gdp)))
2002                 return 0;
2003
2004         return 1;
2005 }
2006
2007 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2008                               struct ext4_group_desc *gdp)
2009 {
2010         if (!ext4_has_group_desc_csum(sb))
2011                 return;
2012         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2013 }
2014
2015 /* Called at mount-time, super-block is locked */
2016 static int ext4_check_descriptors(struct super_block *sb,
2017                                   ext4_group_t *first_not_zeroed)
2018 {
2019         struct ext4_sb_info *sbi = EXT4_SB(sb);
2020         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2021         ext4_fsblk_t last_block;
2022         ext4_fsblk_t block_bitmap;
2023         ext4_fsblk_t inode_bitmap;
2024         ext4_fsblk_t inode_table;
2025         int flexbg_flag = 0;
2026         ext4_group_t i, grp = sbi->s_groups_count;
2027
2028         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2029                 flexbg_flag = 1;
2030
2031         ext4_debug("Checking group descriptors");
2032
2033         for (i = 0; i < sbi->s_groups_count; i++) {
2034                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2035
2036                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2037                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2038                 else
2039                         last_block = first_block +
2040                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2041
2042                 if ((grp == sbi->s_groups_count) &&
2043                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2044                         grp = i;
2045
2046                 block_bitmap = ext4_block_bitmap(sb, gdp);
2047                 if (block_bitmap < first_block || block_bitmap > last_block) {
2048                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2049                                "Block bitmap for group %u not in group "
2050                                "(block %llu)!", i, block_bitmap);
2051                         return 0;
2052                 }
2053                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2054                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2055                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2056                                "Inode bitmap for group %u not in group "
2057                                "(block %llu)!", i, inode_bitmap);
2058                         return 0;
2059                 }
2060                 inode_table = ext4_inode_table(sb, gdp);
2061                 if (inode_table < first_block ||
2062                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2063                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2064                                "Inode table for group %u not in group "
2065                                "(block %llu)!", i, inode_table);
2066                         return 0;
2067                 }
2068                 ext4_lock_group(sb, i);
2069                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2070                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2071                                  "Checksum for group %u failed (%u!=%u)",
2072                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2073                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2074                         if (!(sb->s_flags & MS_RDONLY)) {
2075                                 ext4_unlock_group(sb, i);
2076                                 return 0;
2077                         }
2078                 }
2079                 ext4_unlock_group(sb, i);
2080                 if (!flexbg_flag)
2081                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2082         }
2083         if (NULL != first_not_zeroed)
2084                 *first_not_zeroed = grp;
2085
2086         ext4_free_blocks_count_set(sbi->s_es,
2087                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2088         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2089         return 1;
2090 }
2091
2092 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2093  * the superblock) which were deleted from all directories, but held open by
2094  * a process at the time of a crash.  We walk the list and try to delete these
2095  * inodes at recovery time (only with a read-write filesystem).
2096  *
2097  * In order to keep the orphan inode chain consistent during traversal (in
2098  * case of crash during recovery), we link each inode into the superblock
2099  * orphan list_head and handle it the same way as an inode deletion during
2100  * normal operation (which journals the operations for us).
2101  *
2102  * We only do an iget() and an iput() on each inode, which is very safe if we
2103  * accidentally point at an in-use or already deleted inode.  The worst that
2104  * can happen in this case is that we get a "bit already cleared" message from
2105  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2106  * e2fsck was run on this filesystem, and it must have already done the orphan
2107  * inode cleanup for us, so we can safely abort without any further action.
2108  */
2109 static void ext4_orphan_cleanup(struct super_block *sb,
2110                                 struct ext4_super_block *es)
2111 {
2112         unsigned int s_flags = sb->s_flags;
2113         int nr_orphans = 0, nr_truncates = 0;
2114 #ifdef CONFIG_QUOTA
2115         int i;
2116 #endif
2117         if (!es->s_last_orphan) {
2118                 jbd_debug(4, "no orphan inodes to clean up\n");
2119                 return;
2120         }
2121
2122         if (bdev_read_only(sb->s_bdev)) {
2123                 ext4_msg(sb, KERN_ERR, "write access "
2124                         "unavailable, skipping orphan cleanup");
2125                 return;
2126         }
2127
2128         /* Check if feature set would not allow a r/w mount */
2129         if (!ext4_feature_set_ok(sb, 0)) {
2130                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2131                          "unknown ROCOMPAT features");
2132                 return;
2133         }
2134
2135         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2136                 /* don't clear list on RO mount w/ errors */
2137                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2138                         jbd_debug(1, "Errors on filesystem, "
2139                                   "clearing orphan list.\n");
2140                         es->s_last_orphan = 0;
2141                 }
2142                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2143                 return;
2144         }
2145
2146         if (s_flags & MS_RDONLY) {
2147                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2148                 sb->s_flags &= ~MS_RDONLY;
2149         }
2150 #ifdef CONFIG_QUOTA
2151         /* Needed for iput() to work correctly and not trash data */
2152         sb->s_flags |= MS_ACTIVE;
2153         /* Turn on quotas so that they are updated correctly */
2154         for (i = 0; i < MAXQUOTAS; i++) {
2155                 if (EXT4_SB(sb)->s_qf_names[i]) {
2156                         int ret = ext4_quota_on_mount(sb, i);
2157                         if (ret < 0)
2158                                 ext4_msg(sb, KERN_ERR,
2159                                         "Cannot turn on journaled "
2160                                         "quota: error %d", ret);
2161                 }
2162         }
2163 #endif
2164
2165         while (es->s_last_orphan) {
2166                 struct inode *inode;
2167
2168                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2169                 if (IS_ERR(inode)) {
2170                         es->s_last_orphan = 0;
2171                         break;
2172                 }
2173
2174                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2175                 dquot_initialize(inode);
2176                 if (inode->i_nlink) {
2177                         if (test_opt(sb, DEBUG))
2178                                 ext4_msg(sb, KERN_DEBUG,
2179                                         "%s: truncating inode %lu to %lld bytes",
2180                                         __func__, inode->i_ino, inode->i_size);
2181                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2182                                   inode->i_ino, inode->i_size);
2183                         mutex_lock(&inode->i_mutex);
2184                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2185                         ext4_truncate(inode);
2186                         mutex_unlock(&inode->i_mutex);
2187                         nr_truncates++;
2188                 } else {
2189                         if (test_opt(sb, DEBUG))
2190                                 ext4_msg(sb, KERN_DEBUG,
2191                                         "%s: deleting unreferenced inode %lu",
2192                                         __func__, inode->i_ino);
2193                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2194                                   inode->i_ino);
2195                         nr_orphans++;
2196                 }
2197                 iput(inode);  /* The delete magic happens here! */
2198         }
2199
2200 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2201
2202         if (nr_orphans)
2203                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2204                        PLURAL(nr_orphans));
2205         if (nr_truncates)
2206                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2207                        PLURAL(nr_truncates));
2208 #ifdef CONFIG_QUOTA
2209         /* Turn quotas off */
2210         for (i = 0; i < MAXQUOTAS; i++) {
2211                 if (sb_dqopt(sb)->files[i])
2212                         dquot_quota_off(sb, i);
2213         }
2214 #endif
2215         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2216 }
2217
2218 /*
2219  * Maximal extent format file size.
2220  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2221  * extent format containers, within a sector_t, and within i_blocks
2222  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2223  * so that won't be a limiting factor.
2224  *
2225  * However there is other limiting factor. We do store extents in the form
2226  * of starting block and length, hence the resulting length of the extent
2227  * covering maximum file size must fit into on-disk format containers as
2228  * well. Given that length is always by 1 unit bigger than max unit (because
2229  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2230  *
2231  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2232  */
2233 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2234 {
2235         loff_t res;
2236         loff_t upper_limit = MAX_LFS_FILESIZE;
2237
2238         /* small i_blocks in vfs inode? */
2239         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2240                 /*
2241                  * CONFIG_LBDAF is not enabled implies the inode
2242                  * i_block represent total blocks in 512 bytes
2243                  * 32 == size of vfs inode i_blocks * 8
2244                  */
2245                 upper_limit = (1LL << 32) - 1;
2246
2247                 /* total blocks in file system block size */
2248                 upper_limit >>= (blkbits - 9);
2249                 upper_limit <<= blkbits;
2250         }
2251
2252         /*
2253          * 32-bit extent-start container, ee_block. We lower the maxbytes
2254          * by one fs block, so ee_len can cover the extent of maximum file
2255          * size
2256          */
2257         res = (1LL << 32) - 1;
2258         res <<= blkbits;
2259
2260         /* Sanity check against vm- & vfs- imposed limits */
2261         if (res > upper_limit)
2262                 res = upper_limit;
2263
2264         return res;
2265 }
2266
2267 /*
2268  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2269  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2270  * We need to be 1 filesystem block less than the 2^48 sector limit.
2271  */
2272 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2273 {
2274         loff_t res = EXT4_NDIR_BLOCKS;
2275         int meta_blocks;
2276         loff_t upper_limit;
2277         /* This is calculated to be the largest file size for a dense, block
2278          * mapped file such that the file's total number of 512-byte sectors,
2279          * including data and all indirect blocks, does not exceed (2^48 - 1).
2280          *
2281          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2282          * number of 512-byte sectors of the file.
2283          */
2284
2285         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2286                 /*
2287                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2288                  * the inode i_block field represents total file blocks in
2289                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2290                  */
2291                 upper_limit = (1LL << 32) - 1;
2292
2293                 /* total blocks in file system block size */
2294                 upper_limit >>= (bits - 9);
2295
2296         } else {
2297                 /*
2298                  * We use 48 bit ext4_inode i_blocks
2299                  * With EXT4_HUGE_FILE_FL set the i_blocks
2300                  * represent total number of blocks in
2301                  * file system block size
2302                  */
2303                 upper_limit = (1LL << 48) - 1;
2304
2305         }
2306
2307         /* indirect blocks */
2308         meta_blocks = 1;
2309         /* double indirect blocks */
2310         meta_blocks += 1 + (1LL << (bits-2));
2311         /* tripple indirect blocks */
2312         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2313
2314         upper_limit -= meta_blocks;
2315         upper_limit <<= bits;
2316
2317         res += 1LL << (bits-2);
2318         res += 1LL << (2*(bits-2));
2319         res += 1LL << (3*(bits-2));
2320         res <<= bits;
2321         if (res > upper_limit)
2322                 res = upper_limit;
2323
2324         if (res > MAX_LFS_FILESIZE)
2325                 res = MAX_LFS_FILESIZE;
2326
2327         return res;
2328 }
2329
2330 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2331                                    ext4_fsblk_t logical_sb_block, int nr)
2332 {
2333         struct ext4_sb_info *sbi = EXT4_SB(sb);
2334         ext4_group_t bg, first_meta_bg;
2335         int has_super = 0;
2336
2337         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2338
2339         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2340             nr < first_meta_bg)
2341                 return logical_sb_block + nr + 1;
2342         bg = sbi->s_desc_per_block * nr;
2343         if (ext4_bg_has_super(sb, bg))
2344                 has_super = 1;
2345
2346         return (has_super + ext4_group_first_block_no(sb, bg));
2347 }
2348
2349 /**
2350  * ext4_get_stripe_size: Get the stripe size.
2351  * @sbi: In memory super block info
2352  *
2353  * If we have specified it via mount option, then
2354  * use the mount option value. If the value specified at mount time is
2355  * greater than the blocks per group use the super block value.
2356  * If the super block value is greater than blocks per group return 0.
2357  * Allocator needs it be less than blocks per group.
2358  *
2359  */
2360 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2361 {
2362         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2363         unsigned long stripe_width =
2364                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2365         int ret;
2366
2367         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2368                 ret = sbi->s_stripe;
2369         else if (stripe_width <= sbi->s_blocks_per_group)
2370                 ret = stripe_width;
2371         else if (stride <= sbi->s_blocks_per_group)
2372                 ret = stride;
2373         else
2374                 ret = 0;
2375
2376         /*
2377          * If the stripe width is 1, this makes no sense and
2378          * we set it to 0 to turn off stripe handling code.
2379          */
2380         if (ret <= 1)
2381                 ret = 0;
2382
2383         return ret;
2384 }
2385
2386 /* sysfs supprt */
2387
2388 struct ext4_attr {
2389         struct attribute attr;
2390         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2391         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2392                          const char *, size_t);
2393         union {
2394                 int offset;
2395                 int deprecated_val;
2396         } u;
2397 };
2398
2399 static int parse_strtoull(const char *buf,
2400                 unsigned long long max, unsigned long long *value)
2401 {
2402         int ret;
2403
2404         ret = kstrtoull(skip_spaces(buf), 0, value);
2405         if (!ret && *value > max)
2406                 ret = -EINVAL;
2407         return ret;
2408 }
2409
2410 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2411                                               struct ext4_sb_info *sbi,
2412                                               char *buf)
2413 {
2414         return snprintf(buf, PAGE_SIZE, "%llu\n",
2415                 (s64) EXT4_C2B(sbi,
2416                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2417 }
2418
2419 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2420                                          struct ext4_sb_info *sbi, char *buf)
2421 {
2422         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2423
2424         if (!sb->s_bdev->bd_part)
2425                 return snprintf(buf, PAGE_SIZE, "0\n");
2426         return snprintf(buf, PAGE_SIZE, "%lu\n",
2427                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2428                          sbi->s_sectors_written_start) >> 1);
2429 }
2430
2431 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2432                                           struct ext4_sb_info *sbi, char *buf)
2433 {
2434         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2435
2436         if (!sb->s_bdev->bd_part)
2437                 return snprintf(buf, PAGE_SIZE, "0\n");
2438         return snprintf(buf, PAGE_SIZE, "%llu\n",
2439                         (unsigned long long)(sbi->s_kbytes_written +
2440                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2441                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2442 }
2443
2444 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2445                                           struct ext4_sb_info *sbi,
2446                                           const char *buf, size_t count)
2447 {
2448         unsigned long t;
2449         int ret;
2450
2451         ret = kstrtoul(skip_spaces(buf), 0, &t);
2452         if (ret)
2453                 return ret;
2454
2455         if (t && (!is_power_of_2(t) || t > 0x40000000))
2456                 return -EINVAL;
2457
2458         sbi->s_inode_readahead_blks = t;
2459         return count;
2460 }
2461
2462 static ssize_t sbi_ui_show(struct ext4_attr *a,
2463                            struct ext4_sb_info *sbi, char *buf)
2464 {
2465         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2466
2467         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2468 }
2469
2470 static ssize_t sbi_ui_store(struct ext4_attr *a,
2471                             struct ext4_sb_info *sbi,
2472                             const char *buf, size_t count)
2473 {
2474         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2475         unsigned long t;
2476         int ret;
2477
2478         ret = kstrtoul(skip_spaces(buf), 0, &t);
2479         if (ret)
2480                 return ret;
2481         *ui = t;
2482         return count;
2483 }
2484
2485 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2486                                   struct ext4_sb_info *sbi, char *buf)
2487 {
2488         return snprintf(buf, PAGE_SIZE, "%llu\n",
2489                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2490 }
2491
2492 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2493                                    struct ext4_sb_info *sbi,
2494                                    const char *buf, size_t count)
2495 {
2496         unsigned long long val;
2497         int ret;
2498
2499         if (parse_strtoull(buf, -1ULL, &val))
2500                 return -EINVAL;
2501         ret = ext4_reserve_clusters(sbi, val);
2502
2503         return ret ? ret : count;
2504 }
2505
2506 static ssize_t trigger_test_error(struct ext4_attr *a,
2507                                   struct ext4_sb_info *sbi,
2508                                   const char *buf, size_t count)
2509 {
2510         int len = count;
2511
2512         if (!capable(CAP_SYS_ADMIN))
2513                 return -EPERM;
2514
2515         if (len && buf[len-1] == '\n')
2516                 len--;
2517
2518         if (len)
2519                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2520         return count;
2521 }
2522
2523 static ssize_t sbi_deprecated_show(struct ext4_attr *a,
2524                                    struct ext4_sb_info *sbi, char *buf)
2525 {
2526         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
2527 }
2528
2529 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2530 static struct ext4_attr ext4_attr_##_name = {                   \
2531         .attr = {.name = __stringify(_name), .mode = _mode },   \
2532         .show   = _show,                                        \
2533         .store  = _store,                                       \
2534         .u = {                                                  \
2535                 .offset = offsetof(struct ext4_sb_info, _elname),\
2536         },                                                      \
2537 }
2538 #define EXT4_ATTR(name, mode, show, store) \
2539 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2540
2541 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2542 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2543 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2544 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2545         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2546 #define ATTR_LIST(name) &ext4_attr_##name.attr
2547 #define EXT4_DEPRECATED_ATTR(_name, _val)       \
2548 static struct ext4_attr ext4_attr_##_name = {                   \
2549         .attr = {.name = __stringify(_name), .mode = 0444 },    \
2550         .show   = sbi_deprecated_show,                          \
2551         .u = {                                                  \
2552                 .deprecated_val = _val,                         \
2553         },                                                      \
2554 }
2555
2556 EXT4_RO_ATTR(delayed_allocation_blocks);
2557 EXT4_RO_ATTR(session_write_kbytes);
2558 EXT4_RO_ATTR(lifetime_write_kbytes);
2559 EXT4_RW_ATTR(reserved_clusters);
2560 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2561                  inode_readahead_blks_store, s_inode_readahead_blks);
2562 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2563 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2564 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2565 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2566 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2567 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2568 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2569 EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
2570 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2571 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2572
2573 static struct attribute *ext4_attrs[] = {
2574         ATTR_LIST(delayed_allocation_blocks),
2575         ATTR_LIST(session_write_kbytes),
2576         ATTR_LIST(lifetime_write_kbytes),
2577         ATTR_LIST(reserved_clusters),
2578         ATTR_LIST(inode_readahead_blks),
2579         ATTR_LIST(inode_goal),
2580         ATTR_LIST(mb_stats),
2581         ATTR_LIST(mb_max_to_scan),
2582         ATTR_LIST(mb_min_to_scan),
2583         ATTR_LIST(mb_order2_req),
2584         ATTR_LIST(mb_stream_req),
2585         ATTR_LIST(mb_group_prealloc),
2586         ATTR_LIST(max_writeback_mb_bump),
2587         ATTR_LIST(extent_max_zeroout_kb),
2588         ATTR_LIST(trigger_fs_error),
2589         NULL,
2590 };
2591
2592 /* Features this copy of ext4 supports */
2593 EXT4_INFO_ATTR(lazy_itable_init);
2594 EXT4_INFO_ATTR(batched_discard);
2595 EXT4_INFO_ATTR(meta_bg_resize);
2596
2597 static struct attribute *ext4_feat_attrs[] = {
2598         ATTR_LIST(lazy_itable_init),
2599         ATTR_LIST(batched_discard),
2600         ATTR_LIST(meta_bg_resize),
2601         NULL,
2602 };
2603
2604 static ssize_t ext4_attr_show(struct kobject *kobj,
2605                               struct attribute *attr, char *buf)
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->show ? a->show(a, sbi, buf) : 0;
2612 }
2613
2614 static ssize_t ext4_attr_store(struct kobject *kobj,
2615                                struct attribute *attr,
2616                                const char *buf, size_t len)
2617 {
2618         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2619                                                 s_kobj);
2620         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2621
2622         return a->store ? a->store(a, sbi, buf, len) : 0;
2623 }
2624
2625 static void ext4_sb_release(struct kobject *kobj)
2626 {
2627         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2628                                                 s_kobj);
2629         complete(&sbi->s_kobj_unregister);
2630 }
2631
2632 static const struct sysfs_ops ext4_attr_ops = {
2633         .show   = ext4_attr_show,
2634         .store  = ext4_attr_store,
2635 };
2636
2637 static struct kobj_type ext4_ktype = {
2638         .default_attrs  = ext4_attrs,
2639         .sysfs_ops      = &ext4_attr_ops,
2640         .release        = ext4_sb_release,
2641 };
2642
2643 static void ext4_feat_release(struct kobject *kobj)
2644 {
2645         complete(&ext4_feat->f_kobj_unregister);
2646 }
2647
2648 static struct kobj_type ext4_feat_ktype = {
2649         .default_attrs  = ext4_feat_attrs,
2650         .sysfs_ops      = &ext4_attr_ops,
2651         .release        = ext4_feat_release,
2652 };
2653
2654 /*
2655  * Check whether this filesystem can be mounted based on
2656  * the features present and the RDONLY/RDWR mount requested.
2657  * Returns 1 if this filesystem can be mounted as requested,
2658  * 0 if it cannot be.
2659  */
2660 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2661 {
2662         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2663                 ext4_msg(sb, KERN_ERR,
2664                         "Couldn't mount because of "
2665                         "unsupported optional features (%x)",
2666                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2667                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2668                 return 0;
2669         }
2670
2671         if (readonly)
2672                 return 1;
2673
2674         /* Check that feature set is OK for a read-write mount */
2675         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2676                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2677                          "unsupported optional features (%x)",
2678                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2679                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2680                 return 0;
2681         }
2682         /*
2683          * Large file size enabled file system can only be mounted
2684          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2685          */
2686         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2687                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2688                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2689                                  "cannot be mounted RDWR without "
2690                                  "CONFIG_LBDAF");
2691                         return 0;
2692                 }
2693         }
2694         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2695             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2696                 ext4_msg(sb, KERN_ERR,
2697                          "Can't support bigalloc feature without "
2698                          "extents feature\n");
2699                 return 0;
2700         }
2701
2702 #ifndef CONFIG_QUOTA
2703         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2704             !readonly) {
2705                 ext4_msg(sb, KERN_ERR,
2706                          "Filesystem with quota feature cannot be mounted RDWR "
2707                          "without CONFIG_QUOTA");
2708                 return 0;
2709         }
2710 #endif  /* CONFIG_QUOTA */
2711         return 1;
2712 }
2713
2714 /*
2715  * This function is called once a day if we have errors logged
2716  * on the file system
2717  */
2718 static void print_daily_error_info(unsigned long arg)
2719 {
2720         struct super_block *sb = (struct super_block *) arg;
2721         struct ext4_sb_info *sbi;
2722         struct ext4_super_block *es;
2723
2724         sbi = EXT4_SB(sb);
2725         es = sbi->s_es;
2726
2727         if (es->s_error_count)
2728                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2729                          le32_to_cpu(es->s_error_count));
2730         if (es->s_first_error_time) {
2731                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2732                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2733                        (int) sizeof(es->s_first_error_func),
2734                        es->s_first_error_func,
2735                        le32_to_cpu(es->s_first_error_line));
2736                 if (es->s_first_error_ino)
2737                         printk(": inode %u",
2738                                le32_to_cpu(es->s_first_error_ino));
2739                 if (es->s_first_error_block)
2740                         printk(": block %llu", (unsigned long long)
2741                                le64_to_cpu(es->s_first_error_block));
2742                 printk("\n");
2743         }
2744         if (es->s_last_error_time) {
2745                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2746                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2747                        (int) sizeof(es->s_last_error_func),
2748                        es->s_last_error_func,
2749                        le32_to_cpu(es->s_last_error_line));
2750                 if (es->s_last_error_ino)
2751                         printk(": inode %u",
2752                                le32_to_cpu(es->s_last_error_ino));
2753                 if (es->s_last_error_block)
2754                         printk(": block %llu", (unsigned long long)
2755                                le64_to_cpu(es->s_last_error_block));
2756                 printk("\n");
2757         }
2758         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2759 }
2760
2761 /* Find next suitable group and run ext4_init_inode_table */
2762 static int ext4_run_li_request(struct ext4_li_request *elr)
2763 {
2764         struct ext4_group_desc *gdp = NULL;
2765         ext4_group_t group, ngroups;
2766         struct super_block *sb;
2767         unsigned long timeout = 0;
2768         int ret = 0;
2769
2770         sb = elr->lr_super;
2771         ngroups = EXT4_SB(sb)->s_groups_count;
2772
2773         sb_start_write(sb);
2774         for (group = elr->lr_next_group; group < ngroups; group++) {
2775                 gdp = ext4_get_group_desc(sb, group, NULL);
2776                 if (!gdp) {
2777                         ret = 1;
2778                         break;
2779                 }
2780
2781                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2782                         break;
2783         }
2784
2785         if (group >= ngroups)
2786                 ret = 1;
2787
2788         if (!ret) {
2789                 timeout = jiffies;
2790                 ret = ext4_init_inode_table(sb, group,
2791                                             elr->lr_timeout ? 0 : 1);
2792                 if (elr->lr_timeout == 0) {
2793                         timeout = (jiffies - timeout) *
2794                                   elr->lr_sbi->s_li_wait_mult;
2795                         elr->lr_timeout = timeout;
2796                 }
2797                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2798                 elr->lr_next_group = group + 1;
2799         }
2800         sb_end_write(sb);
2801
2802         return ret;
2803 }
2804
2805 /*
2806  * Remove lr_request from the list_request and free the
2807  * request structure. Should be called with li_list_mtx held
2808  */
2809 static void ext4_remove_li_request(struct ext4_li_request *elr)
2810 {
2811         struct ext4_sb_info *sbi;
2812
2813         if (!elr)
2814                 return;
2815
2816         sbi = elr->lr_sbi;
2817
2818         list_del(&elr->lr_request);
2819         sbi->s_li_request = NULL;
2820         kfree(elr);
2821 }
2822
2823 static void ext4_unregister_li_request(struct super_block *sb)
2824 {
2825         mutex_lock(&ext4_li_mtx);
2826         if (!ext4_li_info) {
2827                 mutex_unlock(&ext4_li_mtx);
2828                 return;
2829         }
2830
2831         mutex_lock(&ext4_li_info->li_list_mtx);
2832         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2833         mutex_unlock(&ext4_li_info->li_list_mtx);
2834         mutex_unlock(&ext4_li_mtx);
2835 }
2836
2837 static struct task_struct *ext4_lazyinit_task;
2838
2839 /*
2840  * This is the function where ext4lazyinit thread lives. It walks
2841  * through the request list searching for next scheduled filesystem.
2842  * When such a fs is found, run the lazy initialization request
2843  * (ext4_rn_li_request) and keep track of the time spend in this
2844  * function. Based on that time we compute next schedule time of
2845  * the request. When walking through the list is complete, compute
2846  * next waking time and put itself into sleep.
2847  */
2848 static int ext4_lazyinit_thread(void *arg)
2849 {
2850         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2851         struct list_head *pos, *n;
2852         struct ext4_li_request *elr;
2853         unsigned long next_wakeup, cur;
2854
2855         BUG_ON(NULL == eli);
2856
2857 cont_thread:
2858         while (true) {
2859                 next_wakeup = MAX_JIFFY_OFFSET;
2860
2861                 mutex_lock(&eli->li_list_mtx);
2862                 if (list_empty(&eli->li_request_list)) {
2863                         mutex_unlock(&eli->li_list_mtx);
2864                         goto exit_thread;
2865                 }
2866
2867                 list_for_each_safe(pos, n, &eli->li_request_list) {
2868                         elr = list_entry(pos, struct ext4_li_request,
2869                                          lr_request);
2870
2871                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2872                                 if (ext4_run_li_request(elr) != 0) {
2873                                         /* error, remove the lazy_init job */
2874                                         ext4_remove_li_request(elr);
2875                                         continue;
2876                                 }
2877                         }
2878
2879                         if (time_before(elr->lr_next_sched, next_wakeup))
2880                                 next_wakeup = elr->lr_next_sched;
2881                 }
2882                 mutex_unlock(&eli->li_list_mtx);
2883
2884                 try_to_freeze();
2885
2886                 cur = jiffies;
2887                 if ((time_after_eq(cur, next_wakeup)) ||
2888                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2889                         cond_resched();
2890                         continue;
2891                 }
2892
2893                 schedule_timeout_interruptible(next_wakeup - cur);
2894
2895                 if (kthread_should_stop()) {
2896                         ext4_clear_request_list();
2897                         goto exit_thread;
2898                 }
2899         }
2900
2901 exit_thread:
2902         /*
2903          * It looks like the request list is empty, but we need
2904          * to check it under the li_list_mtx lock, to prevent any
2905          * additions into it, and of course we should lock ext4_li_mtx
2906          * to atomically free the list and ext4_li_info, because at
2907          * this point another ext4 filesystem could be registering
2908          * new one.
2909          */
2910         mutex_lock(&ext4_li_mtx);
2911         mutex_lock(&eli->li_list_mtx);
2912         if (!list_empty(&eli->li_request_list)) {
2913                 mutex_unlock(&eli->li_list_mtx);
2914                 mutex_unlock(&ext4_li_mtx);
2915                 goto cont_thread;
2916         }
2917         mutex_unlock(&eli->li_list_mtx);
2918         kfree(ext4_li_info);
2919         ext4_li_info = NULL;
2920         mutex_unlock(&ext4_li_mtx);
2921
2922         return 0;
2923 }
2924
2925 static void ext4_clear_request_list(void)
2926 {
2927         struct list_head *pos, *n;
2928         struct ext4_li_request *elr;
2929
2930         mutex_lock(&ext4_li_info->li_list_mtx);
2931         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2932                 elr = list_entry(pos, struct ext4_li_request,
2933                                  lr_request);
2934                 ext4_remove_li_request(elr);
2935         }
2936         mutex_unlock(&ext4_li_info->li_list_mtx);
2937 }
2938
2939 static int ext4_run_lazyinit_thread(void)
2940 {
2941         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2942                                          ext4_li_info, "ext4lazyinit");
2943         if (IS_ERR(ext4_lazyinit_task)) {
2944                 int err = PTR_ERR(ext4_lazyinit_task);
2945                 ext4_clear_request_list();
2946                 kfree(ext4_li_info);
2947                 ext4_li_info = NULL;
2948                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2949                                  "initialization thread\n",
2950                                  err);
2951                 return err;
2952         }
2953         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2954         return 0;
2955 }
2956
2957 /*
2958  * Check whether it make sense to run itable init. thread or not.
2959  * If there is at least one uninitialized inode table, return
2960  * corresponding group number, else the loop goes through all
2961  * groups and return total number of groups.
2962  */
2963 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2964 {
2965         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2966         struct ext4_group_desc *gdp = NULL;
2967
2968         for (group = 0; group < ngroups; group++) {
2969                 gdp = ext4_get_group_desc(sb, group, NULL);
2970                 if (!gdp)
2971                         continue;
2972
2973                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2974                         break;
2975         }
2976
2977         return group;
2978 }
2979
2980 static int ext4_li_info_new(void)
2981 {
2982         struct ext4_lazy_init *eli = NULL;
2983
2984         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2985         if (!eli)
2986                 return -ENOMEM;
2987
2988         INIT_LIST_HEAD(&eli->li_request_list);
2989         mutex_init(&eli->li_list_mtx);
2990
2991         eli->li_state |= EXT4_LAZYINIT_QUIT;
2992
2993         ext4_li_info = eli;
2994
2995         return 0;
2996 }
2997
2998 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2999                                             ext4_group_t start)
3000 {
3001         struct ext4_sb_info *sbi = EXT4_SB(sb);
3002         struct ext4_li_request *elr;
3003         unsigned long rnd;
3004
3005         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3006         if (!elr)
3007                 return NULL;
3008
3009         elr->lr_super = sb;
3010         elr->lr_sbi = sbi;
3011         elr->lr_next_group = start;
3012
3013         /*
3014          * Randomize first schedule time of the request to
3015          * spread the inode table initialization requests
3016          * better.
3017          */
3018         get_random_bytes(&rnd, sizeof(rnd));
3019         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3020                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3021
3022         return elr;
3023 }
3024
3025 int ext4_register_li_request(struct super_block *sb,
3026                              ext4_group_t first_not_zeroed)
3027 {
3028         struct ext4_sb_info *sbi = EXT4_SB(sb);
3029         struct ext4_li_request *elr = NULL;
3030         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3031         int ret = 0;
3032
3033         mutex_lock(&ext4_li_mtx);
3034         if (sbi->s_li_request != NULL) {
3035                 /*
3036                  * Reset timeout so it can be computed again, because
3037                  * s_li_wait_mult might have changed.
3038                  */
3039                 sbi->s_li_request->lr_timeout = 0;
3040                 goto out;
3041         }
3042
3043         if (first_not_zeroed == ngroups ||
3044             (sb->s_flags & MS_RDONLY) ||
3045             !test_opt(sb, INIT_INODE_TABLE))
3046                 goto out;
3047
3048         elr = ext4_li_request_new(sb, first_not_zeroed);
3049         if (!elr) {
3050                 ret = -ENOMEM;
3051                 goto out;
3052         }
3053
3054         if (NULL == ext4_li_info) {
3055                 ret = ext4_li_info_new();
3056                 if (ret)
3057                         goto out;
3058         }
3059
3060         mutex_lock(&ext4_li_info->li_list_mtx);
3061         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3062         mutex_unlock(&ext4_li_info->li_list_mtx);
3063
3064         sbi->s_li_request = elr;
3065         /*
3066          * set elr to NULL here since it has been inserted to
3067          * the request_list and the removal and free of it is
3068          * handled by ext4_clear_request_list from now on.
3069          */
3070         elr = NULL;
3071
3072         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3073                 ret = ext4_run_lazyinit_thread();
3074                 if (ret)
3075                         goto out;
3076         }
3077 out:
3078         mutex_unlock(&ext4_li_mtx);
3079         if (ret)
3080                 kfree(elr);
3081         return ret;
3082 }
3083
3084 /*
3085  * We do not need to lock anything since this is called on
3086  * module unload.
3087  */
3088 static void ext4_destroy_lazyinit_thread(void)
3089 {
3090         /*
3091          * If thread exited earlier
3092          * there's nothing to be done.
3093          */
3094         if (!ext4_li_info || !ext4_lazyinit_task)
3095                 return;
3096
3097         kthread_stop(ext4_lazyinit_task);
3098 }
3099
3100 static int set_journal_csum_feature_set(struct super_block *sb)
3101 {
3102         int ret = 1;
3103         int compat, incompat;
3104         struct ext4_sb_info *sbi = EXT4_SB(sb);
3105
3106         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3107                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3108                 /* journal checksum v2 */
3109                 compat = 0;
3110                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3111         } else {
3112                 /* journal checksum v1 */
3113                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3114                 incompat = 0;
3115         }
3116
3117         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3118                 ret = jbd2_journal_set_features(sbi->s_journal,
3119                                 compat, 0,
3120                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3121                                 incompat);
3122         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3123                 ret = jbd2_journal_set_features(sbi->s_journal,
3124                                 compat, 0,
3125                                 incompat);
3126                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3127                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3128         } else {
3129                 jbd2_journal_clear_features(sbi->s_journal,
3130                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3131                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3132                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3133         }
3134
3135         return ret;
3136 }
3137
3138 /*
3139  * Note: calculating the overhead so we can be compatible with
3140  * historical BSD practice is quite difficult in the face of
3141  * clusters/bigalloc.  This is because multiple metadata blocks from
3142  * different block group can end up in the same allocation cluster.
3143  * Calculating the exact overhead in the face of clustered allocation
3144  * requires either O(all block bitmaps) in memory or O(number of block
3145  * groups**2) in time.  We will still calculate the superblock for
3146  * older file systems --- and if we come across with a bigalloc file
3147  * system with zero in s_overhead_clusters the estimate will be close to
3148  * correct especially for very large cluster sizes --- but for newer
3149  * file systems, it's better to calculate this figure once at mkfs
3150  * time, and store it in the superblock.  If the superblock value is
3151  * present (even for non-bigalloc file systems), we will use it.
3152  */
3153 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3154                           char *buf)
3155 {
3156         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3157         struct ext4_group_desc  *gdp;
3158         ext4_fsblk_t            first_block, last_block, b;
3159         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3160         int                     s, j, count = 0;
3161
3162         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3163                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3164                         sbi->s_itb_per_group + 2);
3165
3166         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3167                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3168         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3169         for (i = 0; i < ngroups; i++) {
3170                 gdp = ext4_get_group_desc(sb, i, NULL);
3171                 b = ext4_block_bitmap(sb, gdp);
3172                 if (b >= first_block && b <= last_block) {
3173                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3174                         count++;
3175                 }
3176                 b = ext4_inode_bitmap(sb, gdp);
3177                 if (b >= first_block && b <= last_block) {
3178                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3179                         count++;
3180                 }
3181                 b = ext4_inode_table(sb, gdp);
3182                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3183                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3184                                 int c = EXT4_B2C(sbi, b - first_block);
3185                                 ext4_set_bit(c, buf);
3186                                 count++;
3187                         }
3188                 if (i != grp)
3189                         continue;
3190                 s = 0;
3191                 if (ext4_bg_has_super(sb, grp)) {
3192                         ext4_set_bit(s++, buf);
3193                         count++;
3194                 }
3195                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3196                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3197                         count++;
3198                 }
3199         }
3200         if (!count)
3201                 return 0;
3202         return EXT4_CLUSTERS_PER_GROUP(sb) -
3203                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3204 }
3205
3206 /*
3207  * Compute the overhead and stash it in sbi->s_overhead
3208  */
3209 int ext4_calculate_overhead(struct super_block *sb)
3210 {
3211         struct ext4_sb_info *sbi = EXT4_SB(sb);
3212         struct ext4_super_block *es = sbi->s_es;
3213         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3214         ext4_fsblk_t overhead = 0;
3215         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3216
3217         if (!buf)
3218                 return -ENOMEM;
3219
3220         /*
3221          * Compute the overhead (FS structures).  This is constant
3222          * for a given filesystem unless the number of block groups
3223          * changes so we cache the previous value until it does.
3224          */
3225
3226         /*
3227          * All of the blocks before first_data_block are overhead
3228          */
3229         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3230
3231         /*
3232          * Add the overhead found in each block group
3233          */
3234         for (i = 0; i < ngroups; i++) {
3235                 int blks;
3236
3237                 blks = count_overhead(sb, i, buf);
3238                 overhead += blks;
3239                 if (blks)
3240                         memset(buf, 0, PAGE_SIZE);
3241                 cond_resched();
3242         }
3243         /* Add the journal blocks as well */
3244         if (sbi->s_journal)
3245                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3246
3247         sbi->s_overhead = overhead;
3248         smp_wmb();
3249         free_page((unsigned long) buf);
3250         return 0;
3251 }
3252
3253
3254 static ext4_fsblk_t ext4_calculate_resv_clusters(struct ext4_sb_info *sbi)
3255 {
3256         ext4_fsblk_t resv_clusters;
3257
3258         /*
3259          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3260          * This should cover the situations where we can not afford to run
3261          * out of space like for example punch hole, or converting
3262          * uninitialized extents in delalloc path. In most cases such
3263          * allocation would require 1, or 2 blocks, higher numbers are
3264          * very rare.
3265          */
3266         resv_clusters = ext4_blocks_count(sbi->s_es) >> sbi->s_cluster_bits;
3267
3268         do_div(resv_clusters, 50);
3269         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3270
3271         return resv_clusters;
3272 }
3273
3274
3275 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3276 {
3277         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3278                                 sbi->s_cluster_bits;
3279
3280         if (count >= clusters)
3281                 return -EINVAL;
3282
3283         atomic64_set(&sbi->s_resv_clusters, count);
3284         return 0;
3285 }
3286
3287 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3288 {
3289         char *orig_data = kstrdup(data, GFP_KERNEL);
3290         struct buffer_head *bh;
3291         struct ext4_super_block *es = NULL;
3292         struct ext4_sb_info *sbi;
3293         ext4_fsblk_t block;
3294         ext4_fsblk_t sb_block = get_sb_block(&data);
3295         ext4_fsblk_t logical_sb_block;
3296         unsigned long offset = 0;
3297         unsigned long journal_devnum = 0;
3298         unsigned long def_mount_opts;
3299         struct inode *root;
3300         char *cp;
3301         const char *descr;
3302         int ret = -ENOMEM;
3303         int blocksize, clustersize;
3304         unsigned int db_count;
3305         unsigned int i;
3306         int needs_recovery, has_huge_files, has_bigalloc;
3307         __u64 blocks_count;
3308         int err = 0;
3309         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3310         ext4_group_t first_not_zeroed;
3311
3312         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3313         if (!sbi)
3314                 goto out_free_orig;
3315
3316         sbi->s_blockgroup_lock =
3317                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3318         if (!sbi->s_blockgroup_lock) {
3319                 kfree(sbi);
3320                 goto out_free_orig;
3321         }
3322         sb->s_fs_info = sbi;
3323         sbi->s_sb = sb;
3324         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3325         sbi->s_sb_block = sb_block;
3326         if (sb->s_bdev->bd_part)
3327                 sbi->s_sectors_written_start =
3328                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3329
3330         /* Cleanup superblock name */
3331         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3332                 *cp = '!';
3333
3334         /* -EINVAL is default */
3335         ret = -EINVAL;
3336         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3337         if (!blocksize) {
3338                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3339                 goto out_fail;
3340         }
3341
3342         /*
3343          * The ext4 superblock will not be buffer aligned for other than 1kB
3344          * block sizes.  We need to calculate the offset from buffer start.
3345          */
3346         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3347                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3348                 offset = do_div(logical_sb_block, blocksize);
3349         } else {
3350                 logical_sb_block = sb_block;
3351         }
3352
3353         if (!(bh = sb_bread(sb, logical_sb_block))) {
3354                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3355                 goto out_fail;
3356         }
3357         /*
3358          * Note: s_es must be initialized as soon as possible because
3359          *       some ext4 macro-instructions depend on its value
3360          */
3361         es = (struct ext4_super_block *) (bh->b_data + offset);
3362         sbi->s_es = es;
3363         sb->s_magic = le16_to_cpu(es->s_magic);
3364         if (sb->s_magic != EXT4_SUPER_MAGIC)
3365                 goto cantfind_ext4;
3366         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3367
3368         /* Warn if metadata_csum and gdt_csum are both set. */
3369         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3370                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3371             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3372                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3373                              "redundant flags; please run fsck.");
3374
3375         /* Check for a known checksum algorithm */
3376         if (!ext4_verify_csum_type(sb, es)) {
3377                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3378                          "unknown checksum algorithm.");
3379                 silent = 1;
3380                 goto cantfind_ext4;
3381         }
3382
3383         /* Load the checksum driver */
3384         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3385                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3386                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3387                 if (IS_ERR(sbi->s_chksum_driver)) {
3388                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3389                         ret = PTR_ERR(sbi->s_chksum_driver);
3390                         sbi->s_chksum_driver = NULL;
3391                         goto failed_mount;
3392                 }
3393         }
3394
3395         /* Check superblock checksum */
3396         if (!ext4_superblock_csum_verify(sb, es)) {
3397                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3398                          "invalid superblock checksum.  Run e2fsck?");
3399                 silent = 1;
3400                 goto cantfind_ext4;
3401         }
3402
3403         /* Precompute checksum seed for all metadata */
3404         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3405                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3406                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3407                                                sizeof(es->s_uuid));
3408
3409         /* Set defaults before we parse the mount options */
3410         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3411         set_opt(sb, INIT_INODE_TABLE);
3412         if (def_mount_opts & EXT4_DEFM_DEBUG)
3413                 set_opt(sb, DEBUG);
3414         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3415                 set_opt(sb, GRPID);
3416         if (def_mount_opts & EXT4_DEFM_UID16)
3417                 set_opt(sb, NO_UID32);
3418         /* xattr user namespace & acls are now defaulted on */
3419         set_opt(sb, XATTR_USER);
3420 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3421         set_opt(sb, POSIX_ACL);
3422 #endif
3423         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3424                 set_opt(sb, JOURNAL_DATA);
3425         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3426                 set_opt(sb, ORDERED_DATA);
3427         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3428                 set_opt(sb, WRITEBACK_DATA);
3429
3430         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3431                 set_opt(sb, ERRORS_PANIC);
3432         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3433                 set_opt(sb, ERRORS_CONT);
3434         else
3435                 set_opt(sb, ERRORS_RO);
3436         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3437                 set_opt(sb, BLOCK_VALIDITY);
3438         if (def_mount_opts & EXT4_DEFM_DISCARD)
3439                 set_opt(sb, DISCARD);
3440
3441         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3442         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3443         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3444         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3445         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3446
3447         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3448                 set_opt(sb, BARRIER);
3449
3450         /*
3451          * enable delayed allocation by default
3452          * Use -o nodelalloc to turn it off
3453          */
3454         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3455             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3456                 set_opt(sb, DELALLOC);
3457
3458         /*
3459          * set default s_li_wait_mult for lazyinit, for the case there is
3460          * no mount option specified.
3461          */
3462         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3463
3464         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3465                            &journal_devnum, &journal_ioprio, 0)) {
3466                 ext4_msg(sb, KERN_WARNING,
3467                          "failed to parse options in superblock: %s",
3468                          sbi->s_es->s_mount_opts);
3469         }
3470         sbi->s_def_mount_opt = sbi->s_mount_opt;
3471         if (!parse_options((char *) data, sb, &journal_devnum,
3472                            &journal_ioprio, 0))
3473                 goto failed_mount;
3474
3475         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3476                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3477                             "with data=journal disables delayed "
3478                             "allocation and O_DIRECT support!\n");
3479                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3480                         ext4_msg(sb, KERN_ERR, "can't mount with "
3481                                  "both data=journal and delalloc");
3482                         goto failed_mount;
3483                 }
3484                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3485                         ext4_msg(sb, KERN_ERR, "can't mount with "
3486                                  "both data=journal and dioread_nolock");
3487                         goto failed_mount;
3488                 }
3489                 if (test_opt(sb, DELALLOC))
3490                         clear_opt(sb, DELALLOC);
3491         }
3492
3493         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3494                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3495
3496         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3497             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3498              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3499              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3500                 ext4_msg(sb, KERN_WARNING,
3501                        "feature flags set on rev 0 fs, "
3502                        "running e2fsck is recommended");
3503
3504         if (IS_EXT2_SB(sb)) {
3505                 if (ext2_feature_set_ok(sb))
3506                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3507                                  "using the ext4 subsystem");
3508                 else {
3509                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3510                                  "to feature incompatibilities");
3511                         goto failed_mount;
3512                 }
3513         }
3514
3515         if (IS_EXT3_SB(sb)) {
3516                 if (ext3_feature_set_ok(sb))
3517                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3518                                  "using the ext4 subsystem");
3519                 else {
3520                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3521                                  "to feature incompatibilities");
3522                         goto failed_mount;
3523                 }
3524         }
3525
3526         /*
3527          * Check feature flags regardless of the revision level, since we
3528          * previously didn't change the revision level when setting the flags,
3529          * so there is a chance incompat flags are set on a rev 0 filesystem.
3530          */
3531         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3532                 goto failed_mount;
3533
3534         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3535         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3536             blocksize > EXT4_MAX_BLOCK_SIZE) {
3537                 ext4_msg(sb, KERN_ERR,
3538                        "Unsupported filesystem blocksize %d", blocksize);
3539                 goto failed_mount;
3540         }
3541
3542         if (sb->s_blocksize != blocksize) {
3543                 /* Validate the filesystem blocksize */
3544                 if (!sb_set_blocksize(sb, blocksize)) {
3545                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3546                                         blocksize);
3547                         goto failed_mount;
3548                 }
3549
3550                 brelse(bh);
3551                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3552                 offset = do_div(logical_sb_block, blocksize);
3553                 bh = sb_bread(sb, logical_sb_block);
3554                 if (!bh) {
3555                         ext4_msg(sb, KERN_ERR,
3556                                "Can't read superblock on 2nd try");
3557                         goto failed_mount;
3558                 }
3559                 es = (struct ext4_super_block *)(bh->b_data + offset);
3560                 sbi->s_es = es;
3561                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3562                         ext4_msg(sb, KERN_ERR,
3563                                "Magic mismatch, very weird!");
3564                         goto failed_mount;
3565                 }
3566         }
3567
3568         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3569                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3570         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3571                                                       has_huge_files);
3572         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3573
3574         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3575                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3576                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3577         } else {
3578                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3579                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3580                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3581                     (!is_power_of_2(sbi->s_inode_size)) ||
3582                     (sbi->s_inode_size > blocksize)) {
3583                         ext4_msg(sb, KERN_ERR,
3584                                "unsupported inode size: %d",
3585                                sbi->s_inode_size);
3586                         goto failed_mount;
3587                 }
3588                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3589                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3590         }
3591
3592         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3593         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3594                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3595                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3596                     !is_power_of_2(sbi->s_desc_size)) {
3597                         ext4_msg(sb, KERN_ERR,
3598                                "unsupported descriptor size %lu",
3599                                sbi->s_desc_size);
3600                         goto failed_mount;
3601                 }
3602         } else
3603                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3604
3605         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3606         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3607         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3608                 goto cantfind_ext4;
3609
3610         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3611         if (sbi->s_inodes_per_block == 0)
3612                 goto cantfind_ext4;
3613         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3614                                         sbi->s_inodes_per_block;
3615         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3616         sbi->s_sbh = bh;
3617         sbi->s_mount_state = le16_to_cpu(es->s_state);
3618         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3619         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3620
3621         for (i = 0; i < 4; i++)
3622                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3623         sbi->s_def_hash_version = es->s_def_hash_version;
3624         i = le32_to_cpu(es->s_flags);
3625         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3626                 sbi->s_hash_unsigned = 3;
3627         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3628 #ifdef __CHAR_UNSIGNED__
3629                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3630                 sbi->s_hash_unsigned = 3;
3631 #else
3632                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3633 #endif
3634         }
3635
3636         /* Handle clustersize */
3637         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3638         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3639                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3640         if (has_bigalloc) {
3641                 if (clustersize < blocksize) {
3642                         ext4_msg(sb, KERN_ERR,
3643                                  "cluster size (%d) smaller than "
3644                                  "block size (%d)", clustersize, blocksize);
3645                         goto failed_mount;
3646                 }
3647                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3648                         le32_to_cpu(es->s_log_block_size);
3649                 sbi->s_clusters_per_group =
3650                         le32_to_cpu(es->s_clusters_per_group);
3651                 if (sbi->s_clusters_per_group > blocksize * 8) {
3652                         ext4_msg(sb, KERN_ERR,
3653                                  "#clusters per group too big: %lu",
3654                                  sbi->s_clusters_per_group);
3655                         goto failed_mount;
3656                 }
3657                 if (sbi->s_blocks_per_group !=
3658                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3659                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3660                                  "clusters per group (%lu) inconsistent",
3661                                  sbi->s_blocks_per_group,
3662                                  sbi->s_clusters_per_group);
3663                         goto failed_mount;
3664                 }
3665         } else {
3666                 if (clustersize != blocksize) {
3667                         ext4_warning(sb, "fragment/cluster size (%d) != "
3668                                      "block size (%d)", clustersize,
3669                                      blocksize);
3670                         clustersize = blocksize;
3671                 }
3672                 if (sbi->s_blocks_per_group > blocksize * 8) {
3673                         ext4_msg(sb, KERN_ERR,
3674                                  "#blocks per group too big: %lu",
3675                                  sbi->s_blocks_per_group);
3676                         goto failed_mount;
3677                 }
3678                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3679                 sbi->s_cluster_bits = 0;
3680         }
3681         sbi->s_cluster_ratio = clustersize / blocksize;
3682
3683         if (sbi->s_inodes_per_group > blocksize * 8) {
3684                 ext4_msg(sb, KERN_ERR,
3685                        "#inodes per group too big: %lu",
3686                        sbi->s_inodes_per_group);
3687                 goto failed_mount;
3688         }
3689
3690         /* Do we have standard group size of clustersize * 8 blocks ? */
3691         if (sbi->s_blocks_per_group == clustersize << 3)
3692                 set_opt2(sb, STD_GROUP_SIZE);
3693
3694         /*
3695          * Test whether we have more sectors than will fit in sector_t,
3696          * and whether the max offset is addressable by the page cache.
3697          */
3698         err = generic_check_addressable(sb->s_blocksize_bits,
3699                                         ext4_blocks_count(es));
3700         if (err) {
3701                 ext4_msg(sb, KERN_ERR, "filesystem"
3702                          " too large to mount safely on this system");
3703                 if (sizeof(sector_t) < 8)
3704                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3705                 goto failed_mount;
3706         }
3707
3708         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3709                 goto cantfind_ext4;
3710
3711         /* check blocks count against device size */
3712         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3713         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3714                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3715                        "exceeds size of device (%llu blocks)",
3716                        ext4_blocks_count(es), blocks_count);
3717                 goto failed_mount;
3718         }
3719
3720         /*
3721          * It makes no sense for the first data block to be beyond the end
3722          * of the filesystem.
3723          */
3724         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3725                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3726                          "block %u is beyond end of filesystem (%llu)",
3727                          le32_to_cpu(es->s_first_data_block),
3728                          ext4_blocks_count(es));
3729                 goto failed_mount;
3730         }
3731         blocks_count = (ext4_blocks_count(es) -
3732                         le32_to_cpu(es->s_first_data_block) +
3733                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3734         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3735         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3736                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3737                        "(block count %llu, first data block %u, "
3738                        "blocks per group %lu)", sbi->s_groups_count,
3739                        ext4_blocks_count(es),
3740                        le32_to_cpu(es->s_first_data_block),
3741                        EXT4_BLOCKS_PER_GROUP(sb));
3742                 goto failed_mount;
3743         }
3744         sbi->s_groups_count = blocks_count;
3745         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3746                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3747         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3748                    EXT4_DESC_PER_BLOCK(sb);
3749         sbi->s_group_desc = ext4_kvmalloc(db_count *
3750                                           sizeof(struct buffer_head *),
3751                                           GFP_KERNEL);
3752         if (sbi->s_group_desc == NULL) {
3753                 ext4_msg(sb, KERN_ERR, "not enough memory");
3754                 ret = -ENOMEM;
3755                 goto failed_mount;
3756         }
3757
3758         if (ext4_proc_root)
3759                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3760
3761         if (sbi->s_proc)
3762                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3763                                  &ext4_seq_options_fops, sb);
3764
3765         bgl_lock_init(sbi->s_blockgroup_lock);
3766
3767         for (i = 0; i < db_count; i++) {
3768                 block = descriptor_loc(sb, logical_sb_block, i);
3769                 sbi->s_group_desc[i] = sb_bread(sb, block);
3770                 if (!sbi->s_group_desc[i]) {
3771                         ext4_msg(sb, KERN_ERR,
3772                                "can't read group descriptor %d", i);
3773                         db_count = i;
3774                         goto failed_mount2;
3775                 }
3776         }
3777         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3778                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3779                 goto failed_mount2;
3780         }
3781         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3782                 if (!ext4_fill_flex_info(sb)) {
3783                         ext4_msg(sb, KERN_ERR,
3784                                "unable to initialize "
3785                                "flex_bg meta info!");
3786                         goto failed_mount2;
3787                 }
3788
3789         sbi->s_gdb_count = db_count;
3790         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3791         spin_lock_init(&sbi->s_next_gen_lock);
3792
3793         init_timer(&sbi->s_err_report);
3794         sbi->s_err_report.function = print_daily_error_info;
3795         sbi->s_err_report.data = (unsigned long) sb;
3796
3797         /* Register extent status tree shrinker */
3798         ext4_es_register_shrinker(sbi);
3799
3800         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3801                         ext4_count_free_clusters(sb));
3802         if (!err) {
3803                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3804                                 ext4_count_free_inodes(sb));
3805         }
3806         if (!err) {
3807                 err = percpu_counter_init(&sbi->s_dirs_counter,
3808                                 ext4_count_dirs(sb));
3809         }
3810         if (!err) {
3811                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3812         }
3813         if (!err) {
3814                 err = percpu_counter_init(&sbi->s_extent_cache_cnt, 0);
3815         }
3816         if (err) {
3817                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3818                 goto failed_mount3;
3819         }
3820
3821         sbi->s_stripe = ext4_get_stripe_size(sbi);
3822         sbi->s_extent_max_zeroout_kb = 32;
3823
3824         /*
3825          * set up enough so that it can read an inode
3826          */
3827         if (!test_opt(sb, NOLOAD) &&
3828             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3829                 sb->s_op = &ext4_sops;
3830         else
3831                 sb->s_op = &ext4_nojournal_sops;
3832         sb->s_export_op = &ext4_export_ops;
3833         sb->s_xattr = ext4_xattr_handlers;
3834 #ifdef CONFIG_QUOTA
3835         sb->dq_op = &ext4_quota_operations;
3836         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3837                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3838         else
3839                 sb->s_qcop = &ext4_qctl_operations;
3840 #endif
3841         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3842
3843         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3844         mutex_init(&sbi->s_orphan_lock);
3845
3846         sb->s_root = NULL;
3847
3848         needs_recovery = (es->s_last_orphan != 0 ||
3849                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3850                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3851
3852         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3853             !(sb->s_flags & MS_RDONLY))
3854                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3855                         goto failed_mount3;
3856
3857         /*
3858          * The first inode we look at is the journal inode.  Don't try
3859          * root first: it may be modified in the journal!
3860          */
3861         if (!test_opt(sb, NOLOAD) &&
3862             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3863                 if (ext4_load_journal(sb, es, journal_devnum))
3864                         goto failed_mount3;
3865         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3866               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3867                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3868                        "suppressed and not mounted read-only");
3869                 goto failed_mount_wq;
3870         } else {
3871                 clear_opt(sb, DATA_FLAGS);
3872                 sbi->s_journal = NULL;
3873                 needs_recovery = 0;
3874                 goto no_journal;
3875         }
3876
3877         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3878             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3879                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3880                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3881                 goto failed_mount_wq;
3882         }
3883
3884         if (!set_journal_csum_feature_set(sb)) {
3885                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3886                          "feature set");
3887                 goto failed_mount_wq;
3888         }
3889
3890         /* We have now updated the journal if required, so we can
3891          * validate the data journaling mode. */
3892         switch (test_opt(sb, DATA_FLAGS)) {
3893         case 0:
3894                 /* No mode set, assume a default based on the journal
3895                  * capabilities: ORDERED_DATA if the journal can
3896                  * cope, else JOURNAL_DATA
3897                  */
3898                 if (jbd2_journal_check_available_features
3899                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3900                         set_opt(sb, ORDERED_DATA);
3901                 else
3902                         set_opt(sb, JOURNAL_DATA);
3903                 break;
3904
3905         case EXT4_MOUNT_ORDERED_DATA:
3906         case EXT4_MOUNT_WRITEBACK_DATA:
3907                 if (!jbd2_journal_check_available_features
3908                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3909                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3910                                "requested data journaling mode");
3911                         goto failed_mount_wq;
3912                 }
3913         default:
3914                 break;
3915         }
3916         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3917
3918         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3919
3920         /*
3921          * The journal may have updated the bg summary counts, so we
3922          * need to update the global counters.
3923          */
3924         percpu_counter_set(&sbi->s_freeclusters_counter,
3925                            ext4_count_free_clusters(sb));
3926         percpu_counter_set(&sbi->s_freeinodes_counter,
3927                            ext4_count_free_inodes(sb));
3928         percpu_counter_set(&sbi->s_dirs_counter,
3929                            ext4_count_dirs(sb));
3930         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3931
3932 no_journal:
3933         /*
3934          * Get the # of file system overhead blocks from the
3935          * superblock if present.
3936          */
3937         if (es->s_overhead_clusters)
3938                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3939         else {
3940                 err = ext4_calculate_overhead(sb);
3941                 if (err)
3942                         goto failed_mount_wq;
3943         }
3944
3945         /*
3946          * The maximum number of concurrent works can be high and
3947          * concurrency isn't really necessary.  Limit it to 1.
3948          */
3949         EXT4_SB(sb)->rsv_conversion_wq =
3950                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3951         if (!EXT4_SB(sb)->rsv_conversion_wq) {
3952                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3953                 ret = -ENOMEM;
3954                 goto failed_mount4;
3955         }
3956
3957         EXT4_SB(sb)->unrsv_conversion_wq =
3958                 alloc_workqueue("ext4-unrsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3959         if (!EXT4_SB(sb)->unrsv_conversion_wq) {
3960                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3961                 ret = -ENOMEM;
3962                 goto failed_mount4;
3963         }
3964
3965         /*
3966          * The jbd2_journal_load will have done any necessary log recovery,
3967          * so we can safely mount the rest of the filesystem now.
3968          */
3969
3970         root = ext4_iget(sb, EXT4_ROOT_INO);
3971         if (IS_ERR(root)) {
3972                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3973                 ret = PTR_ERR(root);
3974                 root = NULL;
3975                 goto failed_mount4;
3976         }
3977         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3978                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3979                 iput(root);
3980                 goto failed_mount4;
3981         }
3982         sb->s_root = d_make_root(root);
3983         if (!sb->s_root) {
3984                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3985                 ret = -ENOMEM;
3986                 goto failed_mount4;
3987         }
3988
3989         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3990                 sb->s_flags |= MS_RDONLY;
3991
3992         /* determine the minimum size of new large inodes, if present */
3993         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3994                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3995                                                      EXT4_GOOD_OLD_INODE_SIZE;
3996                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3997                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3998                         if (sbi->s_want_extra_isize <
3999                             le16_to_cpu(es->s_want_extra_isize))
4000                                 sbi->s_want_extra_isize =
4001                                         le16_to_cpu(es->s_want_extra_isize);
4002                         if (sbi->s_want_extra_isize <
4003                             le16_to_cpu(es->s_min_extra_isize))
4004                                 sbi->s_want_extra_isize =
4005                                         le16_to_cpu(es->s_min_extra_isize);
4006                 }
4007         }
4008         /* Check if enough inode space is available */
4009         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4010                                                         sbi->s_inode_size) {
4011                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4012                                                        EXT4_GOOD_OLD_INODE_SIZE;
4013                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4014                          "available");
4015         }
4016
4017         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sbi));
4018         if (err) {
4019                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
4020                          "reserved pool", ext4_calculate_resv_clusters(sbi));
4021                 goto failed_mount4a;
4022         }
4023
4024         err = ext4_setup_system_zone(sb);
4025         if (err) {
4026                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4027                          "zone (%d)", err);
4028                 goto failed_mount4a;
4029         }
4030
4031         ext4_ext_init(sb);
4032         err = ext4_mb_init(sb);
4033         if (err) {
4034                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4035                          err);
4036                 goto failed_mount5;
4037         }
4038
4039         err = ext4_register_li_request(sb, first_not_zeroed);
4040         if (err)
4041                 goto failed_mount6;
4042
4043         sbi->s_kobj.kset = ext4_kset;
4044         init_completion(&sbi->s_kobj_unregister);
4045         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4046                                    "%s", sb->s_id);
4047         if (err)
4048                 goto failed_mount7;
4049
4050 #ifdef CONFIG_QUOTA
4051         /* Enable quota usage during mount. */
4052         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4053             !(sb->s_flags & MS_RDONLY)) {
4054                 err = ext4_enable_quotas(sb);
4055                 if (err)
4056                         goto failed_mount8;
4057         }
4058 #endif  /* CONFIG_QUOTA */
4059
4060         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4061         ext4_orphan_cleanup(sb, es);
4062         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4063         if (needs_recovery) {
4064                 ext4_msg(sb, KERN_INFO, "recovery complete");
4065                 ext4_mark_recovery_complete(sb, es);
4066         }
4067         if (EXT4_SB(sb)->s_journal) {
4068                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4069                         descr = " journalled data mode";
4070                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4071                         descr = " ordered data mode";
4072                 else
4073                         descr = " writeback data mode";
4074         } else
4075                 descr = "out journal";
4076
4077         if (test_opt(sb, DISCARD)) {
4078                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4079                 if (!blk_queue_discard(q))
4080                         ext4_msg(sb, KERN_WARNING,
4081                                  "mounting with \"discard\" option, but "
4082                                  "the device does not support discard");
4083         }
4084
4085         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4086                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4087                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4088
4089         if (es->s_error_count)
4090                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4091
4092         kfree(orig_data);
4093         return 0;
4094
4095 cantfind_ext4:
4096         if (!silent)
4097                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4098         goto failed_mount;
4099
4100 #ifdef CONFIG_QUOTA
4101 failed_mount8:
4102         kobject_del(&sbi->s_kobj);
4103 #endif
4104 failed_mount7:
4105         ext4_unregister_li_request(sb);
4106 failed_mount6:
4107         ext4_mb_release(sb);
4108 failed_mount5:
4109         ext4_ext_release(sb);
4110         ext4_release_system_zone(sb);
4111 failed_mount4a:
4112         dput(sb->s_root);
4113         sb->s_root = NULL;
4114 failed_mount4:
4115         ext4_msg(sb, KERN_ERR, "mount failed");
4116         if (EXT4_SB(sb)->rsv_conversion_wq)
4117                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4118         if (EXT4_SB(sb)->unrsv_conversion_wq)
4119                 destroy_workqueue(EXT4_SB(sb)->unrsv_conversion_wq);
4120 failed_mount_wq:
4121         if (sbi->s_journal) {
4122                 jbd2_journal_destroy(sbi->s_journal);
4123                 sbi->s_journal = NULL;
4124         }
4125 failed_mount3:
4126         ext4_es_unregister_shrinker(sbi);
4127         del_timer(&sbi->s_err_report);
4128         if (sbi->s_flex_groups)
4129                 ext4_kvfree(sbi->s_flex_groups);
4130         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4131         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4132         percpu_counter_destroy(&sbi->s_dirs_counter);
4133         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4134         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
4135         if (sbi->s_mmp_tsk)
4136                 kthread_stop(sbi->s_mmp_tsk);
4137 failed_mount2:
4138         for (i = 0; i < db_count; i++)
4139                 brelse(sbi->s_group_desc[i]);
4140         ext4_kvfree(sbi->s_group_desc);
4141 failed_mount:
4142         if (sbi->s_chksum_driver)
4143                 crypto_free_shash(sbi->s_chksum_driver);
4144         if (sbi->s_proc) {
4145                 remove_proc_entry("options", sbi->s_proc);
4146                 remove_proc_entry(sb->s_id, ext4_proc_root);
4147         }
4148 #ifdef CONFIG_QUOTA
4149         for (i = 0; i < MAXQUOTAS; i++)
4150                 kfree(sbi->s_qf_names[i]);
4151 #endif
4152         ext4_blkdev_remove(sbi);
4153         brelse(bh);
4154 out_fail:
4155         sb->s_fs_info = NULL;
4156         kfree(sbi->s_blockgroup_lock);
4157         kfree(sbi);
4158 out_free_orig:
4159         kfree(orig_data);
4160         return err ? err : ret;
4161 }
4162
4163 /*
4164  * Setup any per-fs journal parameters now.  We'll do this both on
4165  * initial mount, once the journal has been initialised but before we've
4166  * done any recovery; and again on any subsequent remount.
4167  */
4168 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4169 {
4170         struct ext4_sb_info *sbi = EXT4_SB(sb);
4171
4172         journal->j_commit_interval = sbi->s_commit_interval;
4173         journal->j_min_batch_time = sbi->s_min_batch_time;
4174         journal->j_max_batch_time = sbi->s_max_batch_time;
4175
4176         write_lock(&journal->j_state_lock);
4177         if (test_opt(sb, BARRIER))
4178                 journal->j_flags |= JBD2_BARRIER;
4179         else
4180                 journal->j_flags &= ~JBD2_BARRIER;
4181         if (test_opt(sb, DATA_ERR_ABORT))
4182                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4183         else
4184                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4185         write_unlock(&journal->j_state_lock);
4186 }
4187
4188 static journal_t *ext4_get_journal(struct super_block *sb,
4189                                    unsigned int journal_inum)
4190 {
4191         struct inode *journal_inode;
4192         journal_t *journal;
4193
4194         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4195
4196         /* First, test for the existence of a valid inode on disk.  Bad
4197          * things happen if we iget() an unused inode, as the subsequent
4198          * iput() will try to delete it. */
4199
4200         journal_inode = ext4_iget(sb, journal_inum);
4201         if (IS_ERR(journal_inode)) {
4202                 ext4_msg(sb, KERN_ERR, "no journal found");
4203                 return NULL;
4204         }
4205         if (!journal_inode->i_nlink) {
4206                 make_bad_inode(journal_inode);
4207                 iput(journal_inode);
4208                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4209                 return NULL;
4210         }
4211
4212         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4213                   journal_inode, journal_inode->i_size);
4214         if (!S_ISREG(journal_inode->i_mode)) {
4215                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4216                 iput(journal_inode);
4217                 return NULL;
4218         }
4219
4220         journal = jbd2_journal_init_inode(journal_inode);
4221         if (!journal) {
4222                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4223                 iput(journal_inode);
4224                 return NULL;
4225         }
4226         journal->j_private = sb;
4227         ext4_init_journal_params(sb, journal);
4228         return journal;
4229 }
4230
4231 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4232                                        dev_t j_dev)
4233 {
4234         struct buffer_head *bh;
4235         journal_t *journal;
4236         ext4_fsblk_t start;
4237         ext4_fsblk_t len;
4238         int hblock, blocksize;
4239         ext4_fsblk_t sb_block;
4240         unsigned long offset;
4241         struct ext4_super_block *es;
4242         struct block_device *bdev;
4243
4244         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4245
4246         bdev = ext4_blkdev_get(j_dev, sb);
4247         if (bdev == NULL)
4248                 return NULL;
4249
4250         blocksize = sb->s_blocksize;
4251         hblock = bdev_logical_block_size(bdev);
4252         if (blocksize < hblock) {
4253                 ext4_msg(sb, KERN_ERR,
4254                         "blocksize too small for journal device");
4255                 goto out_bdev;
4256         }
4257
4258         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4259         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4260         set_blocksize(bdev, blocksize);
4261         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4262                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4263                        "external journal");
4264                 goto out_bdev;
4265         }
4266
4267         es = (struct ext4_super_block *) (bh->b_data + offset);
4268         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4269             !(le32_to_cpu(es->s_feature_incompat) &
4270               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4271                 ext4_msg(sb, KERN_ERR, "external journal has "
4272                                         "bad superblock");
4273                 brelse(bh);
4274                 goto out_bdev;
4275         }
4276
4277         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4278                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4279                 brelse(bh);
4280                 goto out_bdev;
4281         }
4282
4283         len = ext4_blocks_count(es);
4284         start = sb_block + 1;
4285         brelse(bh);     /* we're done with the superblock */
4286
4287         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4288                                         start, len, blocksize);
4289         if (!journal) {
4290                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4291                 goto out_bdev;
4292         }
4293         journal->j_private = sb;
4294         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4295         wait_on_buffer(journal->j_sb_buffer);
4296         if (!buffer_uptodate(journal->j_sb_buffer)) {
4297                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4298                 goto out_journal;
4299         }
4300         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4301                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4302                                         "user (unsupported) - %d",
4303                         be32_to_cpu(journal->j_superblock->s_nr_users));
4304                 goto out_journal;
4305         }
4306         EXT4_SB(sb)->journal_bdev = bdev;
4307         ext4_init_journal_params(sb, journal);
4308         return journal;
4309
4310 out_journal:
4311         jbd2_journal_destroy(journal);
4312 out_bdev:
4313         ext4_blkdev_put(bdev);
4314         return NULL;
4315 }
4316
4317 static int ext4_load_journal(struct super_block *sb,
4318                              struct ext4_super_block *es,
4319                              unsigned long journal_devnum)
4320 {
4321         journal_t *journal;
4322         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4323         dev_t journal_dev;
4324         int err = 0;
4325         int really_read_only;
4326
4327         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4328
4329         if (journal_devnum &&
4330             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4331                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4332                         "numbers have changed");
4333                 journal_dev = new_decode_dev(journal_devnum);
4334         } else
4335                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4336
4337         really_read_only = bdev_read_only(sb->s_bdev);
4338
4339         /*
4340          * Are we loading a blank journal or performing recovery after a
4341          * crash?  For recovery, we need to check in advance whether we
4342          * can get read-write access to the device.
4343          */
4344         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4345                 if (sb->s_flags & MS_RDONLY) {
4346                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4347                                         "required on readonly filesystem");
4348                         if (really_read_only) {
4349                                 ext4_msg(sb, KERN_ERR, "write access "
4350                                         "unavailable, cannot proceed");
4351                                 return -EROFS;
4352                         }
4353                         ext4_msg(sb, KERN_INFO, "write access will "
4354                                "be enabled during recovery");
4355                 }
4356         }
4357
4358         if (journal_inum && journal_dev) {
4359                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4360                        "and inode journals!");
4361                 return -EINVAL;
4362         }
4363
4364         if (journal_inum) {
4365                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4366                         return -EINVAL;
4367         } else {
4368                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4369                         return -EINVAL;
4370         }
4371
4372         if (!(journal->j_flags & JBD2_BARRIER))
4373                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4374
4375         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4376                 err = jbd2_journal_wipe(journal, !really_read_only);
4377         if (!err) {
4378                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4379                 if (save)
4380                         memcpy(save, ((char *) es) +
4381                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4382                 err = jbd2_journal_load(journal);
4383                 if (save)
4384                         memcpy(((char *) es) + EXT4_S_ERR_START,
4385                                save, EXT4_S_ERR_LEN);
4386                 kfree(save);
4387         }
4388
4389         if (err) {
4390                 ext4_msg(sb, KERN_ERR, "error loading journal");
4391                 jbd2_journal_destroy(journal);
4392                 return err;
4393         }
4394
4395         EXT4_SB(sb)->s_journal = journal;
4396         ext4_clear_journal_err(sb, es);
4397
4398         if (!really_read_only && journal_devnum &&
4399             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4400                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4401
4402                 /* Make sure we flush the recovery flag to disk. */
4403                 ext4_commit_super(sb, 1);
4404         }
4405
4406         return 0;
4407 }
4408
4409 static int ext4_commit_super(struct super_block *sb, int sync)
4410 {
4411         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4412         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4413         int error = 0;
4414
4415         if (!sbh || block_device_ejected(sb))
4416                 return error;
4417         if (buffer_write_io_error(sbh)) {
4418                 /*
4419                  * Oh, dear.  A previous attempt to write the
4420                  * superblock failed.  This could happen because the
4421                  * USB device was yanked out.  Or it could happen to
4422                  * be a transient write error and maybe the block will
4423                  * be remapped.  Nothing we can do but to retry the
4424                  * write and hope for the best.
4425                  */
4426                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4427                        "superblock detected");
4428                 clear_buffer_write_io_error(sbh);
4429                 set_buffer_uptodate(sbh);
4430         }
4431         /*
4432          * If the file system is mounted read-only, don't update the
4433          * superblock write time.  This avoids updating the superblock
4434          * write time when we are mounting the root file system
4435          * read/only but we need to replay the journal; at that point,
4436          * for people who are east of GMT and who make their clock
4437          * tick in localtime for Windows bug-for-bug compatibility,
4438          * the clock is set in the future, and this will cause e2fsck
4439          * to complain and force a full file system check.
4440          */
4441         if (!(sb->s_flags & MS_RDONLY))
4442                 es->s_wtime = cpu_to_le32(get_seconds());
4443         if (sb->s_bdev->bd_part)
4444                 es->s_kbytes_written =
4445                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4446                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4447                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4448         else
4449                 es->s_kbytes_written =
4450                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4451         ext4_free_blocks_count_set(es,
4452                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4453                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4454         es->s_free_inodes_count =
4455                 cpu_to_le32(percpu_counter_sum_positive(
4456                                 &EXT4_SB(sb)->s_freeinodes_counter));
4457         BUFFER_TRACE(sbh, "marking dirty");
4458         ext4_superblock_csum_set(sb);
4459         mark_buffer_dirty(sbh);
4460         if (sync) {
4461                 error = sync_dirty_buffer(sbh);
4462                 if (error)
4463                         return error;
4464
4465                 error = buffer_write_io_error(sbh);
4466                 if (error) {
4467                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4468                                "superblock");
4469                         clear_buffer_write_io_error(sbh);
4470                         set_buffer_uptodate(sbh);
4471                 }
4472         }
4473         return error;
4474 }
4475
4476 /*
4477  * Have we just finished recovery?  If so, and if we are mounting (or
4478  * remounting) the filesystem readonly, then we will end up with a
4479  * consistent fs on disk.  Record that fact.
4480  */
4481 static void ext4_mark_recovery_complete(struct super_block *sb,
4482                                         struct ext4_super_block *es)
4483 {
4484         journal_t *journal = EXT4_SB(sb)->s_journal;
4485
4486         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4487                 BUG_ON(journal != NULL);
4488                 return;
4489         }
4490         jbd2_journal_lock_updates(journal);
4491         if (jbd2_journal_flush(journal) < 0)
4492                 goto out;
4493
4494         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4495             sb->s_flags & MS_RDONLY) {
4496                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4497                 ext4_commit_super(sb, 1);
4498         }
4499
4500 out:
4501         jbd2_journal_unlock_updates(journal);
4502 }
4503
4504 /*
4505  * If we are mounting (or read-write remounting) a filesystem whose journal
4506  * has recorded an error from a previous lifetime, move that error to the
4507  * main filesystem now.
4508  */
4509 static void ext4_clear_journal_err(struct super_block *sb,
4510                                    struct ext4_super_block *es)
4511 {
4512         journal_t *journal;
4513         int j_errno;
4514         const char *errstr;
4515
4516         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4517
4518         journal = EXT4_SB(sb)->s_journal;
4519
4520         /*
4521          * Now check for any error status which may have been recorded in the
4522          * journal by a prior ext4_error() or ext4_abort()
4523          */
4524
4525         j_errno = jbd2_journal_errno(journal);
4526         if (j_errno) {
4527                 char nbuf[16];
4528
4529                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4530                 ext4_warning(sb, "Filesystem error recorded "
4531                              "from previous mount: %s", errstr);
4532                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4533
4534                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4535                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4536                 ext4_commit_super(sb, 1);
4537
4538                 jbd2_journal_clear_err(journal);
4539                 jbd2_journal_update_sb_errno(journal);
4540         }
4541 }
4542
4543 /*
4544  * Force the running and committing transactions to commit,
4545  * and wait on the commit.
4546  */
4547 int ext4_force_commit(struct super_block *sb)
4548 {
4549         journal_t *journal;
4550
4551         if (sb->s_flags & MS_RDONLY)
4552                 return 0;
4553
4554         journal = EXT4_SB(sb)->s_journal;
4555         return ext4_journal_force_commit(journal);
4556 }
4557
4558 static int ext4_sync_fs(struct super_block *sb, int wait)
4559 {
4560         int ret = 0;
4561         tid_t target;
4562         bool needs_barrier = false;
4563         struct ext4_sb_info *sbi = EXT4_SB(sb);
4564
4565         trace_ext4_sync_fs(sb, wait);
4566         flush_workqueue(sbi->rsv_conversion_wq);
4567         flush_workqueue(sbi->unrsv_conversion_wq);
4568         /*
4569          * Writeback quota in non-journalled quota case - journalled quota has
4570          * no dirty dquots
4571          */
4572         dquot_writeback_dquots(sb, -1);
4573         /*
4574          * Data writeback is possible w/o journal transaction, so barrier must
4575          * being sent at the end of the function. But we can skip it if
4576          * transaction_commit will do it for us.
4577          */
4578         target = jbd2_get_latest_transaction(sbi->s_journal);
4579         if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4580             !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4581                 needs_barrier = true;
4582
4583         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4584                 if (wait)
4585                         ret = jbd2_log_wait_commit(sbi->s_journal, target);
4586         }
4587         if (needs_barrier) {
4588                 int err;
4589                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4590                 if (!ret)
4591                         ret = err;
4592         }
4593
4594         return ret;
4595 }
4596
4597 static int ext4_sync_fs_nojournal(struct super_block *sb, int wait)
4598 {
4599         int ret = 0;
4600
4601         trace_ext4_sync_fs(sb, wait);
4602         flush_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4603         flush_workqueue(EXT4_SB(sb)->unrsv_conversion_wq);
4604         dquot_writeback_dquots(sb, -1);
4605         if (wait && test_opt(sb, BARRIER))
4606                 ret = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4607
4608         return ret;
4609 }
4610
4611 /*
4612  * LVM calls this function before a (read-only) snapshot is created.  This
4613  * gives us a chance to flush the journal completely and mark the fs clean.
4614  *
4615  * Note that only this function cannot bring a filesystem to be in a clean
4616  * state independently. It relies on upper layer to stop all data & metadata
4617  * modifications.
4618  */
4619 static int ext4_freeze(struct super_block *sb)
4620 {
4621         int error = 0;
4622         journal_t *journal;
4623
4624         if (sb->s_flags & MS_RDONLY)
4625                 return 0;
4626
4627         journal = EXT4_SB(sb)->s_journal;
4628
4629         /* Now we set up the journal barrier. */
4630         jbd2_journal_lock_updates(journal);
4631
4632         /*
4633          * Don't clear the needs_recovery flag if we failed to flush
4634          * the journal.
4635          */
4636         error = jbd2_journal_flush(journal);
4637         if (error < 0)
4638                 goto out;
4639
4640         /* Journal blocked and flushed, clear needs_recovery flag. */
4641         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4642         error = ext4_commit_super(sb, 1);
4643 out:
4644         /* we rely on upper layer to stop further updates */
4645         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4646         return error;
4647 }
4648
4649 /*
4650  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4651  * flag here, even though the filesystem is not technically dirty yet.
4652  */
4653 static int ext4_unfreeze(struct super_block *sb)
4654 {
4655         if (sb->s_flags & MS_RDONLY)
4656                 return 0;
4657
4658         /* Reset the needs_recovery flag before the fs is unlocked. */
4659         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4660         ext4_commit_super(sb, 1);
4661         return 0;
4662 }
4663
4664 /*
4665  * Structure to save mount options for ext4_remount's benefit
4666  */
4667 struct ext4_mount_options {
4668         unsigned long s_mount_opt;
4669         unsigned long s_mount_opt2;
4670         kuid_t s_resuid;
4671         kgid_t s_resgid;
4672         unsigned long s_commit_interval;
4673         u32 s_min_batch_time, s_max_batch_time;
4674 #ifdef CONFIG_QUOTA
4675         int s_jquota_fmt;
4676         char *s_qf_names[MAXQUOTAS];
4677 #endif
4678 };
4679
4680 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4681 {
4682         struct ext4_super_block *es;
4683         struct ext4_sb_info *sbi = EXT4_SB(sb);
4684         unsigned long old_sb_flags;
4685         struct ext4_mount_options old_opts;
4686         int enable_quota = 0;
4687         ext4_group_t g;
4688         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4689         int err = 0;
4690 #ifdef CONFIG_QUOTA
4691         int i, j;
4692 #endif
4693         char *orig_data = kstrdup(data, GFP_KERNEL);
4694
4695         /* Store the original options */
4696         old_sb_flags = sb->s_flags;
4697         old_opts.s_mount_opt = sbi->s_mount_opt;
4698         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4699         old_opts.s_resuid = sbi->s_resuid;
4700         old_opts.s_resgid = sbi->s_resgid;
4701         old_opts.s_commit_interval = sbi->s_commit_interval;
4702         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4703         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4704 #ifdef CONFIG_QUOTA
4705         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4706         for (i = 0; i < MAXQUOTAS; i++)
4707                 if (sbi->s_qf_names[i]) {
4708                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4709                                                          GFP_KERNEL);
4710                         if (!old_opts.s_qf_names[i]) {
4711                                 for (j = 0; j < i; j++)
4712                                         kfree(old_opts.s_qf_names[j]);
4713                                 kfree(orig_data);
4714                                 return -ENOMEM;
4715                         }
4716                 } else
4717                         old_opts.s_qf_names[i] = NULL;
4718 #endif
4719         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4720                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4721
4722         /*
4723          * Allow the "check" option to be passed as a remount option.
4724          */
4725         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4726                 err = -EINVAL;
4727                 goto restore_opts;
4728         }
4729
4730         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4731                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4732                         ext4_msg(sb, KERN_ERR, "can't mount with "
4733                                  "both data=journal and delalloc");
4734                         err = -EINVAL;
4735                         goto restore_opts;
4736                 }
4737                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4738                         ext4_msg(sb, KERN_ERR, "can't mount with "
4739                                  "both data=journal and dioread_nolock");
4740                         err = -EINVAL;
4741                         goto restore_opts;
4742                 }
4743         }
4744
4745         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4746                 ext4_abort(sb, "Abort forced by user");
4747
4748         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4749                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4750
4751         es = sbi->s_es;
4752
4753         if (sbi->s_journal) {
4754                 ext4_init_journal_params(sb, sbi->s_journal);
4755                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4756         }
4757
4758         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4759                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4760                         err = -EROFS;
4761                         goto restore_opts;
4762                 }
4763
4764                 if (*flags & MS_RDONLY) {
4765                         err = dquot_suspend(sb, -1);
4766                         if (err < 0)
4767                                 goto restore_opts;
4768
4769                         /*
4770                          * First of all, the unconditional stuff we have to do
4771                          * to disable replay of the journal when we next remount
4772                          */
4773                         sb->s_flags |= MS_RDONLY;
4774
4775                         /*
4776                          * OK, test if we are remounting a valid rw partition
4777                          * readonly, and if so set the rdonly flag and then
4778                          * mark the partition as valid again.
4779                          */
4780                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4781                             (sbi->s_mount_state & EXT4_VALID_FS))
4782                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4783
4784                         if (sbi->s_journal)
4785                                 ext4_mark_recovery_complete(sb, es);
4786                 } else {
4787                         /* Make sure we can mount this feature set readwrite */
4788                         if (!ext4_feature_set_ok(sb, 0)) {
4789                                 err = -EROFS;
4790                                 goto restore_opts;
4791                         }
4792                         /*
4793                          * Make sure the group descriptor checksums
4794                          * are sane.  If they aren't, refuse to remount r/w.
4795                          */
4796                         for (g = 0; g < sbi->s_groups_count; g++) {
4797                                 struct ext4_group_desc *gdp =
4798                                         ext4_get_group_desc(sb, g, NULL);
4799
4800                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4801                                         ext4_msg(sb, KERN_ERR,
4802                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4803                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4804                                                le16_to_cpu(gdp->bg_checksum));
4805                                         err = -EINVAL;
4806                                         goto restore_opts;
4807                                 }
4808                         }
4809
4810                         /*
4811                          * If we have an unprocessed orphan list hanging
4812                          * around from a previously readonly bdev mount,
4813                          * require a full umount/remount for now.
4814                          */
4815                         if (es->s_last_orphan) {
4816                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4817                                        "remount RDWR because of unprocessed "
4818                                        "orphan inode list.  Please "
4819                                        "umount/remount instead");
4820                                 err = -EINVAL;
4821                                 goto restore_opts;
4822                         }
4823
4824                         /*
4825                          * Mounting a RDONLY partition read-write, so reread
4826                          * and store the current valid flag.  (It may have
4827                          * been changed by e2fsck since we originally mounted
4828                          * the partition.)
4829                          */
4830                         if (sbi->s_journal)
4831                                 ext4_clear_journal_err(sb, es);
4832                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4833                         if (!ext4_setup_super(sb, es, 0))
4834                                 sb->s_flags &= ~MS_RDONLY;
4835                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4836                                                      EXT4_FEATURE_INCOMPAT_MMP))
4837                                 if (ext4_multi_mount_protect(sb,
4838                                                 le64_to_cpu(es->s_mmp_block))) {
4839                                         err = -EROFS;
4840                                         goto restore_opts;
4841                                 }
4842                         enable_quota = 1;
4843                 }
4844         }
4845
4846         /*
4847          * Reinitialize lazy itable initialization thread based on
4848          * current settings
4849          */
4850         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4851                 ext4_unregister_li_request(sb);
4852         else {
4853                 ext4_group_t first_not_zeroed;
4854                 first_not_zeroed = ext4_has_uninit_itable(sb);
4855                 ext4_register_li_request(sb, first_not_zeroed);
4856         }
4857
4858         ext4_setup_system_zone(sb);
4859         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4860                 ext4_commit_super(sb, 1);
4861
4862 #ifdef CONFIG_QUOTA
4863         /* Release old quota file names */
4864         for (i = 0; i < MAXQUOTAS; i++)
4865                 kfree(old_opts.s_qf_names[i]);
4866         if (enable_quota) {
4867                 if (sb_any_quota_suspended(sb))
4868                         dquot_resume(sb, -1);
4869                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4870                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4871                         err = ext4_enable_quotas(sb);
4872                         if (err)
4873                                 goto restore_opts;
4874                 }
4875         }
4876 #endif
4877
4878         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4879         kfree(orig_data);
4880         return 0;
4881
4882 restore_opts:
4883         sb->s_flags = old_sb_flags;
4884         sbi->s_mount_opt = old_opts.s_mount_opt;
4885         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4886         sbi->s_resuid = old_opts.s_resuid;
4887         sbi->s_resgid = old_opts.s_resgid;
4888         sbi->s_commit_interval = old_opts.s_commit_interval;
4889         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4890         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4891 #ifdef CONFIG_QUOTA
4892         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4893         for (i = 0; i < MAXQUOTAS; i++) {
4894                 kfree(sbi->s_qf_names[i]);
4895                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4896         }
4897 #endif
4898         kfree(orig_data);
4899         return err;
4900 }
4901
4902 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4903 {
4904         struct super_block *sb = dentry->d_sb;
4905         struct ext4_sb_info *sbi = EXT4_SB(sb);
4906         struct ext4_super_block *es = sbi->s_es;
4907         ext4_fsblk_t overhead = 0, resv_blocks;
4908         u64 fsid;
4909         s64 bfree;
4910         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4911
4912         if (!test_opt(sb, MINIX_DF))
4913                 overhead = sbi->s_overhead;
4914
4915         buf->f_type = EXT4_SUPER_MAGIC;
4916         buf->f_bsize = sb->s_blocksize;
4917         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4918         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4919                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4920         /* prevent underflow in case that few free space is available */
4921         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4922         buf->f_bavail = buf->f_bfree -
4923                         (ext4_r_blocks_count(es) + resv_blocks);
4924         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4925                 buf->f_bavail = 0;
4926         buf->f_files = le32_to_cpu(es->s_inodes_count);
4927         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4928         buf->f_namelen = EXT4_NAME_LEN;
4929         fsid = le64_to_cpup((void *)es->s_uuid) ^
4930                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4931         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4932         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4933
4934         return 0;
4935 }
4936
4937 /* Helper function for writing quotas on sync - we need to start transaction
4938  * before quota file is locked for write. Otherwise the are possible deadlocks:
4939  * Process 1                         Process 2
4940  * ext4_create()                     quota_sync()
4941  *   jbd2_journal_start()                  write_dquot()
4942  *   dquot_initialize()                         down(dqio_mutex)
4943  *     down(dqio_mutex)                    jbd2_journal_start()
4944  *
4945  */
4946
4947 #ifdef CONFIG_QUOTA
4948
4949 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4950 {
4951         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4952 }
4953
4954 static int ext4_write_dquot(struct dquot *dquot)
4955 {
4956         int ret, err;
4957         handle_t *handle;
4958         struct inode *inode;
4959
4960         inode = dquot_to_inode(dquot);
4961         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4962                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4963         if (IS_ERR(handle))
4964                 return PTR_ERR(handle);
4965         ret = dquot_commit(dquot);
4966         err = ext4_journal_stop(handle);
4967         if (!ret)
4968                 ret = err;
4969         return ret;
4970 }
4971
4972 static int ext4_acquire_dquot(struct dquot *dquot)
4973 {
4974         int ret, err;
4975         handle_t *handle;
4976
4977         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4978                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4979         if (IS_ERR(handle))
4980                 return PTR_ERR(handle);
4981         ret = dquot_acquire(dquot);
4982         err = ext4_journal_stop(handle);
4983         if (!ret)
4984                 ret = err;
4985         return ret;
4986 }
4987
4988 static int ext4_release_dquot(struct dquot *dquot)
4989 {
4990         int ret, err;
4991         handle_t *handle;
4992
4993         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4994                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4995         if (IS_ERR(handle)) {
4996                 /* Release dquot anyway to avoid endless cycle in dqput() */
4997                 dquot_release(dquot);
4998                 return PTR_ERR(handle);
4999         }
5000         ret = dquot_release(dquot);
5001         err = ext4_journal_stop(handle);
5002         if (!ret)
5003                 ret = err;
5004         return ret;
5005 }
5006
5007 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5008 {
5009         struct super_block *sb = dquot->dq_sb;
5010         struct ext4_sb_info *sbi = EXT4_SB(sb);
5011
5012         /* Are we journaling quotas? */
5013         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
5014             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5015                 dquot_mark_dquot_dirty(dquot);
5016                 return ext4_write_dquot(dquot);
5017         } else {
5018                 return dquot_mark_dquot_dirty(dquot);
5019         }
5020 }
5021
5022 static int ext4_write_info(struct super_block *sb, int type)
5023 {
5024         int ret, err;
5025         handle_t *handle;
5026
5027         /* Data block + inode block */
5028         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
5029         if (IS_ERR(handle))
5030                 return PTR_ERR(handle);
5031         ret = dquot_commit_info(sb, type);
5032         err = ext4_journal_stop(handle);
5033         if (!ret)
5034                 ret = err;
5035         return ret;
5036 }
5037
5038 /*
5039  * Turn on quotas during mount time - we need to find
5040  * the quota file and such...
5041  */
5042 static int ext4_quota_on_mount(struct super_block *sb, int type)
5043 {
5044         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5045                                         EXT4_SB(sb)->s_jquota_fmt, type);
5046 }
5047
5048 /*
5049  * Standard function to be called on quota_on
5050  */
5051 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5052                          struct path *path)
5053 {
5054         int err;
5055
5056         if (!test_opt(sb, QUOTA))
5057                 return -EINVAL;
5058
5059         /* Quotafile not on the same filesystem? */
5060         if (path->dentry->d_sb != sb)
5061                 return -EXDEV;
5062         /* Journaling quota? */
5063         if (EXT4_SB(sb)->s_qf_names[type]) {
5064                 /* Quotafile not in fs root? */
5065                 if (path->dentry->d_parent != sb->s_root)
5066                         ext4_msg(sb, KERN_WARNING,
5067                                 "Quota file not on filesystem root. "
5068                                 "Journaled quota will not work");
5069         }
5070
5071         /*
5072          * When we journal data on quota file, we have to flush journal to see
5073          * all updates to the file when we bypass pagecache...
5074          */
5075         if (EXT4_SB(sb)->s_journal &&
5076             ext4_should_journal_data(path->dentry->d_inode)) {
5077                 /*
5078                  * We don't need to lock updates but journal_flush() could
5079                  * otherwise be livelocked...
5080                  */
5081                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5082                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5083                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5084                 if (err)
5085                         return err;
5086         }
5087
5088         return dquot_quota_on(sb, type, format_id, path);
5089 }
5090
5091 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5092                              unsigned int flags)
5093 {
5094         int err;
5095         struct inode *qf_inode;
5096         unsigned long qf_inums[MAXQUOTAS] = {
5097                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5098                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5099         };
5100
5101         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5102
5103         if (!qf_inums[type])
5104                 return -EPERM;
5105
5106         qf_inode = ext4_iget(sb, qf_inums[type]);
5107         if (IS_ERR(qf_inode)) {
5108                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5109                 return PTR_ERR(qf_inode);
5110         }
5111
5112         /* Don't account quota for quota files to avoid recursion */
5113         qf_inode->i_flags |= S_NOQUOTA;
5114         err = dquot_enable(qf_inode, type, format_id, flags);
5115         iput(qf_inode);
5116
5117         return err;
5118 }
5119
5120 /* Enable usage tracking for all quota types. */
5121 static int ext4_enable_quotas(struct super_block *sb)
5122 {
5123         int type, err = 0;
5124         unsigned long qf_inums[MAXQUOTAS] = {
5125                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5126                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5127         };
5128
5129         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5130         for (type = 0; type < MAXQUOTAS; type++) {
5131                 if (qf_inums[type]) {
5132                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5133                                                 DQUOT_USAGE_ENABLED);
5134                         if (err) {
5135                                 ext4_warning(sb,
5136                                         "Failed to enable quota tracking "
5137                                         "(type=%d, err=%d). Please run "
5138                                         "e2fsck to fix.", type, err);
5139                                 return err;
5140                         }
5141                 }
5142         }
5143         return 0;
5144 }
5145
5146 /*
5147  * quota_on function that is used when QUOTA feature is set.
5148  */
5149 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5150                                  int format_id)
5151 {
5152         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5153                 return -EINVAL;
5154
5155         /*
5156          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5157          */
5158         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5159 }
5160
5161 static int ext4_quota_off(struct super_block *sb, int type)
5162 {
5163         struct inode *inode = sb_dqopt(sb)->files[type];
5164         handle_t *handle;
5165
5166         /* Force all delayed allocation blocks to be allocated.
5167          * Caller already holds s_umount sem */
5168         if (test_opt(sb, DELALLOC))
5169                 sync_filesystem(sb);
5170
5171         if (!inode)
5172                 goto out;
5173
5174         /* Update modification times of quota files when userspace can
5175          * start looking at them */
5176         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5177         if (IS_ERR(handle))
5178                 goto out;
5179         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5180         ext4_mark_inode_dirty(handle, inode);
5181         ext4_journal_stop(handle);
5182
5183 out:
5184         return dquot_quota_off(sb, type);
5185 }
5186
5187 /*
5188  * quota_off function that is used when QUOTA feature is set.
5189  */
5190 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5191 {
5192         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5193                 return -EINVAL;
5194
5195         /* Disable only the limits. */
5196         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5197 }
5198
5199 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5200  * acquiring the locks... As quota files are never truncated and quota code
5201  * itself serializes the operations (and no one else should touch the files)
5202  * we don't have to be afraid of races */
5203 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5204                                size_t len, loff_t off)
5205 {
5206         struct inode *inode = sb_dqopt(sb)->files[type];
5207         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5208         int err = 0;
5209         int offset = off & (sb->s_blocksize - 1);
5210         int tocopy;
5211         size_t toread;
5212         struct buffer_head *bh;
5213         loff_t i_size = i_size_read(inode);
5214
5215         if (off > i_size)
5216                 return 0;
5217         if (off+len > i_size)
5218                 len = i_size-off;
5219         toread = len;
5220         while (toread > 0) {
5221                 tocopy = sb->s_blocksize - offset < toread ?
5222                                 sb->s_blocksize - offset : toread;
5223                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5224                 if (err)
5225                         return err;
5226                 if (!bh)        /* A hole? */
5227                         memset(data, 0, tocopy);
5228                 else
5229                         memcpy(data, bh->b_data+offset, tocopy);
5230                 brelse(bh);
5231                 offset = 0;
5232                 toread -= tocopy;
5233                 data += tocopy;
5234                 blk++;
5235         }
5236         return len;
5237 }
5238
5239 /* Write to quotafile (we know the transaction is already started and has
5240  * enough credits) */
5241 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5242                                 const char *data, size_t len, loff_t off)
5243 {
5244         struct inode *inode = sb_dqopt(sb)->files[type];
5245         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5246         int err = 0;
5247         int offset = off & (sb->s_blocksize - 1);
5248         struct buffer_head *bh;
5249         handle_t *handle = journal_current_handle();
5250
5251         if (EXT4_SB(sb)->s_journal && !handle) {
5252                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5253                         " cancelled because transaction is not started",
5254                         (unsigned long long)off, (unsigned long long)len);
5255                 return -EIO;
5256         }
5257         /*
5258          * Since we account only one data block in transaction credits,
5259          * then it is impossible to cross a block boundary.
5260          */
5261         if (sb->s_blocksize - offset < len) {
5262                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5263                         " cancelled because not block aligned",
5264                         (unsigned long long)off, (unsigned long long)len);
5265                 return -EIO;
5266         }
5267
5268         bh = ext4_bread(handle, inode, blk, 1, &err);
5269         if (!bh)
5270                 goto out;
5271         err = ext4_journal_get_write_access(handle, bh);
5272         if (err) {
5273                 brelse(bh);
5274                 goto out;
5275         }
5276         lock_buffer(bh);
5277         memcpy(bh->b_data+offset, data, len);
5278         flush_dcache_page(bh->b_page);
5279         unlock_buffer(bh);
5280         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5281         brelse(bh);
5282 out:
5283         if (err)
5284                 return err;
5285         if (inode->i_size < off + len) {
5286                 i_size_write(inode, off + len);
5287                 EXT4_I(inode)->i_disksize = inode->i_size;
5288                 ext4_mark_inode_dirty(handle, inode);
5289         }
5290         return len;
5291 }
5292
5293 #endif
5294
5295 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5296                        const char *dev_name, void *data)
5297 {
5298         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5299 }
5300
5301 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5302 static inline void register_as_ext2(void)
5303 {
5304         int err = register_filesystem(&ext2_fs_type);
5305         if (err)
5306                 printk(KERN_WARNING
5307                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5308 }
5309
5310 static inline void unregister_as_ext2(void)
5311 {
5312         unregister_filesystem(&ext2_fs_type);
5313 }
5314
5315 static inline int ext2_feature_set_ok(struct super_block *sb)
5316 {
5317         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5318                 return 0;
5319         if (sb->s_flags & MS_RDONLY)
5320                 return 1;
5321         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5322                 return 0;
5323         return 1;
5324 }
5325 #else
5326 static inline void register_as_ext2(void) { }
5327 static inline void unregister_as_ext2(void) { }
5328 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5329 #endif
5330
5331 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5332 static inline void register_as_ext3(void)
5333 {
5334         int err = register_filesystem(&ext3_fs_type);
5335         if (err)
5336                 printk(KERN_WARNING
5337                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5338 }
5339
5340 static inline void unregister_as_ext3(void)
5341 {
5342         unregister_filesystem(&ext3_fs_type);
5343 }
5344
5345 static inline int ext3_feature_set_ok(struct super_block *sb)
5346 {
5347         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5348                 return 0;
5349         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5350                 return 0;
5351         if (sb->s_flags & MS_RDONLY)
5352                 return 1;
5353         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5354                 return 0;
5355         return 1;
5356 }
5357 #else
5358 static inline void register_as_ext3(void) { }
5359 static inline void unregister_as_ext3(void) { }
5360 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5361 #endif
5362
5363 static struct file_system_type ext4_fs_type = {
5364         .owner          = THIS_MODULE,
5365         .name           = "ext4",
5366         .mount          = ext4_mount,
5367         .kill_sb        = kill_block_super,
5368         .fs_flags       = FS_REQUIRES_DEV,
5369 };
5370 MODULE_ALIAS_FS("ext4");
5371
5372 static int __init ext4_init_feat_adverts(void)
5373 {
5374         struct ext4_features *ef;
5375         int ret = -ENOMEM;
5376
5377         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5378         if (!ef)
5379                 goto out;
5380
5381         ef->f_kobj.kset = ext4_kset;
5382         init_completion(&ef->f_kobj_unregister);
5383         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5384                                    "features");
5385         if (ret) {
5386                 kfree(ef);
5387                 goto out;
5388         }
5389
5390         ext4_feat = ef;
5391         ret = 0;
5392 out:
5393         return ret;
5394 }
5395
5396 static void ext4_exit_feat_adverts(void)
5397 {
5398         kobject_put(&ext4_feat->f_kobj);
5399         wait_for_completion(&ext4_feat->f_kobj_unregister);
5400         kfree(ext4_feat);
5401 }
5402
5403 /* Shared across all ext4 file systems */
5404 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5405 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5406
5407 static int __init ext4_init_fs(void)
5408 {
5409         int i, err;
5410
5411         ext4_li_info = NULL;
5412         mutex_init(&ext4_li_mtx);
5413
5414         /* Build-time check for flags consistency */
5415         ext4_check_flag_values();
5416
5417         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5418                 mutex_init(&ext4__aio_mutex[i]);
5419                 init_waitqueue_head(&ext4__ioend_wq[i]);
5420         }
5421
5422         err = ext4_init_es();
5423         if (err)
5424                 return err;
5425
5426         err = ext4_init_pageio();
5427         if (err)
5428                 goto out7;
5429
5430         err = ext4_init_system_zone();
5431         if (err)
5432                 goto out6;
5433         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5434         if (!ext4_kset) {
5435                 err = -ENOMEM;
5436                 goto out5;
5437         }
5438         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5439
5440         err = ext4_init_feat_adverts();
5441         if (err)
5442                 goto out4;
5443
5444         err = ext4_init_mballoc();
5445         if (err)
5446                 goto out3;
5447
5448         err = ext4_init_xattr();
5449         if (err)
5450                 goto out2;
5451         err = init_inodecache();
5452         if (err)
5453                 goto out1;
5454         register_as_ext3();
5455         register_as_ext2();
5456         err = register_filesystem(&ext4_fs_type);
5457         if (err)
5458                 goto out;
5459
5460         return 0;
5461 out:
5462         unregister_as_ext2();
5463         unregister_as_ext3();
5464         destroy_inodecache();
5465 out1:
5466         ext4_exit_xattr();
5467 out2:
5468         ext4_exit_mballoc();
5469 out3:
5470         ext4_exit_feat_adverts();
5471 out4:
5472         if (ext4_proc_root)
5473                 remove_proc_entry("fs/ext4", NULL);
5474         kset_unregister(ext4_kset);
5475 out5:
5476         ext4_exit_system_zone();
5477 out6:
5478         ext4_exit_pageio();
5479 out7:
5480         ext4_exit_es();
5481
5482         return err;
5483 }
5484
5485 static void __exit ext4_exit_fs(void)
5486 {
5487         ext4_destroy_lazyinit_thread();
5488         unregister_as_ext2();
5489         unregister_as_ext3();
5490         unregister_filesystem(&ext4_fs_type);
5491         destroy_inodecache();
5492         ext4_exit_xattr();
5493         ext4_exit_mballoc();
5494         ext4_exit_feat_adverts();
5495         remove_proc_entry("fs/ext4", NULL);
5496         kset_unregister(ext4_kset);
5497         ext4_exit_system_zone();
5498         ext4_exit_pageio();
5499         ext4_exit_es();
5500 }
5501
5502 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5503 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5504 MODULE_LICENSE("GPL");
5505 module_init(ext4_init_fs)
5506 module_exit(ext4_exit_fs)