fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  73 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  74 static int ext4_unfreeze(struct super_block *sb);
  75 static int ext4_freeze(struct super_block *sb);
  76 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  77                        const char *dev_name, void *data);
  78 static inline int ext2_feature_set_ok(struct super_block *sb);
  79 static inline int ext3_feature_set_ok(struct super_block *sb);
  80 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  81 static void ext4_destroy_lazyinit_thread(void);
  82 static void ext4_unregister_li_request(struct super_block *sb);
  83 static void ext4_clear_request_list(void);
  84 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
  85
  86 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  87 static struct file_system_type ext2_fs_type = {
  88         .owner          = THIS_MODULE,
  89         .name           = "ext2",
  90         .mount          = ext4_mount,
  91         .kill_sb        = kill_block_super,
  92         .fs_flags       = FS_REQUIRES_DEV,
  93 };
  94 MODULE_ALIAS_FS("ext2");
  95 MODULE_ALIAS("ext2");
  96 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  97 #else
  98 #define IS_EXT2_SB(sb) (0)
  99 #endif
 100
 101
 102 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 103 static struct file_system_type ext3_fs_type = {
 104         .owner          = THIS_MODULE,
 105         .name           = "ext3",
 106         .mount          = ext4_mount,
 107         .kill_sb        = kill_block_super,
 108         .fs_flags       = FS_REQUIRES_DEV,
 109 };
 110 MODULE_ALIAS_FS("ext3");
 111 MODULE_ALIAS("ext3");
 112 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 113 #else
 114 #define IS_EXT3_SB(sb) (0)
 115 #endif
 116
 117 static int ext4_verify_csum_type(struct super_block *sb,
 118                                  struct ext4_super_block *es)
 119 {
 120         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 121                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 122                 return 1;
 123
 124         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 125 }
 126
 127 static __le32 ext4_superblock_csum(struct super_block *sb,
 128                                    struct ext4_super_block *es)
 129 {
 130         struct ext4_sb_info *sbi = EXT4_SB(sb);
 131         int offset = offsetof(struct ext4_super_block, s_checksum);
 132         __u32 csum;
 133
 134         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 135
 136         return cpu_to_le32(csum);
 137 }
 138
 139 int ext4_superblock_csum_verify(struct super_block *sb,
 140                                 struct ext4_super_block *es)
 141 {
 142         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 143                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 144                 return 1;
 145
 146         return es->s_checksum == ext4_superblock_csum(sb, es);
 147 }
 148
 149 void ext4_superblock_csum_set(struct super_block *sb)
 150 {
 151         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 152
 153         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 154                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 155                 return;
 156
 157         es->s_checksum = ext4_superblock_csum(sb, es);
 158 }
 159
 160 void *ext4_kvmalloc(size_t size, gfp_t flags)
 161 {
 162         void *ret;
 163
 164         ret = kmalloc(size, flags);
 165         if (!ret)
 166                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 167         return ret;
 168 }
 169
 170 void *ext4_kvzalloc(size_t size, gfp_t flags)
 171 {
 172         void *ret;
 173
 174         ret = kzalloc(size, flags);
 175         if (!ret)
 176                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 177         return ret;
 178 }
 179
 180 void ext4_kvfree(void *ptr)
 181 {
 182         if (is_vmalloc_addr(ptr))
 183                 vfree(ptr);
 184         else
 185                 kfree(ptr);
 186
 187 }
 188
 189 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 190                                struct ext4_group_desc *bg)
 191 {
 192         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 193                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 194                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 195 }
 196
 197 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 198                                struct ext4_group_desc *bg)
 199 {
 200         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 201                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 202                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 203 }
 204
 205 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 206                               struct ext4_group_desc *bg)
 207 {
 208         return le32_to_cpu(bg->bg_inode_table_lo) |
 209                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 210                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 211 }
 212
 213 __u32 ext4_free_group_clusters(struct super_block *sb,
 214                                struct ext4_group_desc *bg)
 215 {
 216         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 217                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 218                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 219 }
 220
 221 __u32 ext4_free_inodes_count(struct super_block *sb,
 222                               struct ext4_group_desc *bg)
 223 {
 224         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 225                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 226                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 227 }
 228
 229 __u32 ext4_used_dirs_count(struct super_block *sb,
 230                               struct ext4_group_desc *bg)
 231 {
 232         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 233                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 234                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 235 }
 236
 237 __u32 ext4_itable_unused_count(struct super_block *sb,
 238                               struct ext4_group_desc *bg)
 239 {
 240         return le16_to_cpu(bg->bg_itable_unused_lo) |
 241                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 242                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 243 }
 244
 245 void ext4_block_bitmap_set(struct super_block *sb,
 246                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 247 {
 248         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 249         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 250                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 251 }
 252
 253 void ext4_inode_bitmap_set(struct super_block *sb,
 254                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 255 {
 256         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 257         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 258                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 259 }
 260
 261 void ext4_inode_table_set(struct super_block *sb,
 262                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 263 {
 264         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 265         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 266                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 267 }
 268
 269 void ext4_free_group_clusters_set(struct super_block *sb,
 270                                   struct ext4_group_desc *bg, __u32 count)
 271 {
 272         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 273         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 274                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 275 }
 276
 277 void ext4_free_inodes_set(struct super_block *sb,
 278                           struct ext4_group_desc *bg, __u32 count)
 279 {
 280         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 281         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 282                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 283 }
 284
 285 void ext4_used_dirs_set(struct super_block *sb,
 286                           struct ext4_group_desc *bg, __u32 count)
 287 {
 288         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 289         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 290                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 291 }
 292
 293 void ext4_itable_unused_set(struct super_block *sb,
 294                           struct ext4_group_desc *bg, __u32 count)
 295 {
 296         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 297         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 298                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 299 }
 300
 301
 302 static void __save_error_info(struct super_block *sb, const char *func,
 303                             unsigned int line)
 304 {
 305         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 306
 307         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 308         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 309         es->s_last_error_time = cpu_to_le32(get_seconds());
 310         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 311         es->s_last_error_line = cpu_to_le32(line);
 312         if (!es->s_first_error_time) {
 313                 es->s_first_error_time = es->s_last_error_time;
 314                 strncpy(es->s_first_error_func, func,
 315                         sizeof(es->s_first_error_func));
 316                 es->s_first_error_line = cpu_to_le32(line);
 317                 es->s_first_error_ino = es->s_last_error_ino;
 318                 es->s_first_error_block = es->s_last_error_block;
 319         }
 320         /*
 321          * Start the daily error reporting function if it hasn't been
 322          * started already
 323          */
 324         if (!es->s_error_count)
 325                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 326         le32_add_cpu(&es->s_error_count, 1);
 327 }
 328
 329 static void save_error_info(struct super_block *sb, const char *func,
 330                             unsigned int line)
 331 {
 332         __save_error_info(sb, func, line);
 333         ext4_commit_super(sb, 1);
 334 }
 335
 336 /*
 337  * The del_gendisk() function uninitializes the disk-specific data
 338  * structures, including the bdi structure, without telling anyone
 339  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 340  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 341  * This is a kludge to prevent these oops until we can put in a proper
 342  * hook in del_gendisk() to inform the VFS and file system layers.
 343  */
 344 static int block_device_ejected(struct super_block *sb)
 345 {
 346         struct inode *bd_inode = sb->s_bdev->bd_inode;
 347         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 348
 349         return bdi->dev == NULL;
 350 }
 351
 352 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 353 {
 354         struct super_block              *sb = journal->j_private;
 355         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 356         int                             error = is_journal_aborted(journal);
 357         struct ext4_journal_cb_entry    *jce;
 358
 359         BUG_ON(txn->t_state == T_FINISHED);
 360         spin_lock(&sbi->s_md_lock);
 361         while (!list_empty(&txn->t_private_list)) {
 362                 jce = list_entry(txn->t_private_list.next,
 363                                  struct ext4_journal_cb_entry, jce_list);
 364                 list_del_init(&jce->jce_list);
 365                 spin_unlock(&sbi->s_md_lock);
 366                 jce->jce_func(sb, jce, error);
 367                 spin_lock(&sbi->s_md_lock);
 368         }
 369         spin_unlock(&sbi->s_md_lock);
 370 }
 371
 372 /* Deal with the reporting of failure conditions on a filesystem such as
 373  * inconsistencies detected or read IO failures.
 374  *
 375  * On ext2, we can store the error state of the filesystem in the
 376  * superblock.  That is not possible on ext4, because we may have other
 377  * write ordering constraints on the superblock which prevent us from
 378  * writing it out straight away; and given that the journal is about to
 379  * be aborted, we can't rely on the current, or future, transactions to
 380  * write out the superblock safely.
 381  *
 382  * We'll just use the jbd2_journal_abort() error code to record an error in
 383  * the journal instead.  On recovery, the journal will complain about
 384  * that error until we've noted it down and cleared it.
 385  */
 386
 387 static void ext4_handle_error(struct super_block *sb)
 388 {
 389         if (sb->s_flags & MS_RDONLY)
 390                 return;
 391
 392         if (!test_opt(sb, ERRORS_CONT)) {
 393                 journal_t *journal = EXT4_SB(sb)->s_journal;
 394
 395                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 396                 if (journal)
 397                         jbd2_journal_abort(journal, -EIO);
 398         }
 399         if (test_opt(sb, ERRORS_RO)) {
 400                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 401                 sb->s_flags |= MS_RDONLY;
 402         }
 403         if (test_opt(sb, ERRORS_PANIC))
 404                 panic("EXT4-fs (device %s): panic forced after error\n",
 405                         sb->s_id);
 406 }
 407
 408 void __ext4_error(struct super_block *sb, const char *function,
 409                   unsigned int line, const char *fmt, ...)
 410 {
 411         struct va_format vaf;
 412         va_list args;
 413
 414         va_start(args, fmt);
 415         vaf.fmt = fmt;
 416         vaf.va = &args;
 417         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 418                sb->s_id, function, line, current->comm, &vaf);
 419         va_end(args);
 420         save_error_info(sb, function, line);
 421
 422         ext4_handle_error(sb);
 423 }
 424
 425 void ext4_error_inode(struct inode *inode, const char *function,
 426                       unsigned int line, ext4_fsblk_t block,
 427                       const char *fmt, ...)
 428 {
 429         va_list args;
 430         struct va_format vaf;
 431         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 432
 433         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 434         es->s_last_error_block = cpu_to_le64(block);
 435         save_error_info(inode->i_sb, function, line);
 436         va_start(args, fmt);
 437         vaf.fmt = fmt;
 438         vaf.va = &args;
 439         if (block)
 440                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 441                        "inode #%lu: block %llu: comm %s: %pV\n",
 442                        inode->i_sb->s_id, function, line, inode->i_ino,
 443                        block, current->comm, &vaf);
 444         else
 445                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 446                        "inode #%lu: comm %s: %pV\n",
 447                        inode->i_sb->s_id, function, line, inode->i_ino,
 448                        current->comm, &vaf);
 449         va_end(args);
 450
 451         ext4_handle_error(inode->i_sb);
 452 }
 453
 454 void ext4_error_file(struct file *file, const char *function,
 455                      unsigned int line, ext4_fsblk_t block,
 456                      const char *fmt, ...)
 457 {
 458         va_list args;
 459         struct va_format vaf;
 460         struct ext4_super_block *es;
 461         struct inode *inode = file_inode(file);
 462         char pathname[80], *path;
 463
 464         es = EXT4_SB(inode->i_sb)->s_es;
 465         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 466         save_error_info(inode->i_sb, function, line);
 467         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 468         if (IS_ERR(path))
 469                 path = "(unknown)";
 470         va_start(args, fmt);
 471         vaf.fmt = fmt;
 472         vaf.va = &args;
 473         if (block)
 474                 printk(KERN_CRIT
 475                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 476                        "block %llu: comm %s: path %s: %pV\n",
 477                        inode->i_sb->s_id, function, line, inode->i_ino,
 478                        block, current->comm, path, &vaf);
 479         else
 480                 printk(KERN_CRIT
 481                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 482                        "comm %s: path %s: %pV\n",
 483                        inode->i_sb->s_id, function, line, inode->i_ino,
 484                        current->comm, path, &vaf);
 485         va_end(args);
 486
 487         ext4_handle_error(inode->i_sb);
 488 }
 489
 490 const char *ext4_decode_error(struct super_block *sb, int errno,
 491                               char nbuf[16])
 492 {
 493         char *errstr = NULL;
 494
 495         switch (errno) {
 496         case -EIO:
 497                 errstr = "IO failure";
 498                 break;
 499         case -ENOMEM:
 500                 errstr = "Out of memory";
 501                 break;
 502         case -EROFS:
 503                 if (!sb || (EXT4_SB(sb)->s_journal &&
 504                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 505                         errstr = "Journal has aborted";
 506                 else
 507                         errstr = "Readonly filesystem";
 508                 break;
 509         default:
 510                 /* If the caller passed in an extra buffer for unknown
 511                  * errors, textualise them now.  Else we just return
 512                  * NULL. */
 513                 if (nbuf) {
 514                         /* Check for truncated error codes... */
 515                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 516                                 errstr = nbuf;
 517                 }
 518                 break;
 519         }
 520
 521         return errstr;
 522 }
 523
 524 /* __ext4_std_error decodes expected errors from journaling functions
 525  * automatically and invokes the appropriate error response.  */
 526
 527 void __ext4_std_error(struct super_block *sb, const char *function,
 528                       unsigned int line, int errno)
 529 {
 530         char nbuf[16];
 531         const char *errstr;
 532
 533         /* Special case: if the error is EROFS, and we're not already
 534          * inside a transaction, then there's really no point in logging
 535          * an error. */
 536         if (errno == -EROFS && journal_current_handle() == NULL &&
 537             (sb->s_flags & MS_RDONLY))
 538                 return;
 539
 540         errstr = ext4_decode_error(sb, errno, nbuf);
 541         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 542                sb->s_id, function, line, errstr);
 543         save_error_info(sb, function, line);
 544
 545         ext4_handle_error(sb);
 546 }
 547
 548 /*
 549  * ext4_abort is a much stronger failure handler than ext4_error.  The
 550  * abort function may be used to deal with unrecoverable failures such
 551  * as journal IO errors or ENOMEM at a critical moment in log management.
 552  *
 553  * We unconditionally force the filesystem into an ABORT|READONLY state,
 554  * unless the error response on the fs has been set to panic in which
 555  * case we take the easy way out and panic immediately.
 556  */
 557
 558 void __ext4_abort(struct super_block *sb, const char *function,
 559                 unsigned int line, const char *fmt, ...)
 560 {
 561         va_list args;
 562
 563         save_error_info(sb, function, line);
 564         va_start(args, fmt);
 565         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 566                function, line);
 567         vprintk(fmt, args);
 568         printk("\n");
 569         va_end(args);
 570
 571         if ((sb->s_flags & MS_RDONLY) == 0) {
 572                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 573                 sb->s_flags |= MS_RDONLY;
 574                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 575                 if (EXT4_SB(sb)->s_journal)
 576                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 577                 save_error_info(sb, function, line);
 578         }
 579         if (test_opt(sb, ERRORS_PANIC))
 580                 panic("EXT4-fs panic from previous error\n");
 581 }
 582
 583 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 584 {
 585         struct va_format vaf;
 586         va_list args;
 587
 588         va_start(args, fmt);
 589         vaf.fmt = fmt;
 590         vaf.va = &args;
 591         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 592         va_end(args);
 593 }
 594
 595 void __ext4_warning(struct super_block *sb, const char *function,
 596                     unsigned int line, const char *fmt, ...)
 597 {
 598         struct va_format vaf;
 599         va_list args;
 600
 601         va_start(args, fmt);
 602         vaf.fmt = fmt;
 603         vaf.va = &args;
 604         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 605                sb->s_id, function, line, &vaf);
 606         va_end(args);
 607 }
 608
 609 void __ext4_grp_locked_error(const char *function, unsigned int line,
 610                              struct super_block *sb, ext4_group_t grp,
 611                              unsigned long ino, ext4_fsblk_t block,
 612                              const char *fmt, ...)
 613 __releases(bitlock)
 614 __acquires(bitlock)
 615 {
 616         struct va_format vaf;
 617         va_list args;
 618         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 619
 620         es->s_last_error_ino = cpu_to_le32(ino);
 621         es->s_last_error_block = cpu_to_le64(block);
 622         __save_error_info(sb, function, line);
 623
 624         va_start(args, fmt);
 625
 626         vaf.fmt = fmt;
 627         vaf.va = &args;
 628         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 629                sb->s_id, function, line, grp);
 630         if (ino)
 631                 printk(KERN_CONT "inode %lu: ", ino);
 632         if (block)
 633                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 634         printk(KERN_CONT "%pV\n", &vaf);
 635         va_end(args);
 636
 637         if (test_opt(sb, ERRORS_CONT)) {
 638                 ext4_commit_super(sb, 0);
 639                 return;
 640         }
 641
 642         ext4_unlock_group(sb, grp);
 643         ext4_handle_error(sb);
 644         /*
 645          * We only get here in the ERRORS_RO case; relocking the group
 646          * may be dangerous, but nothing bad will happen since the
 647          * filesystem will have already been marked read/only and the
 648          * journal has been aborted.  We return 1 as a hint to callers
 649          * who might what to use the return value from
 650          * ext4_grp_locked_error() to distinguish between the
 651          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 652          * aggressively from the ext4 function in question, with a
 653          * more appropriate error code.
 654          */
 655         ext4_lock_group(sb, grp);
 656         return;
 657 }
 658
 659 void ext4_update_dynamic_rev(struct super_block *sb)
 660 {
 661         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 662
 663         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 664                 return;
 665
 666         ext4_warning(sb,
 667                      "updating to rev %d because of new feature flag, "
 668                      "running e2fsck is recommended",
 669                      EXT4_DYNAMIC_REV);
 670
 671         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 672         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 673         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 674         /* leave es->s_feature_*compat flags alone */
 675         /* es->s_uuid will be set by e2fsck if empty */
 676
 677         /*
 678          * The rest of the superblock fields should be zero, and if not it
 679          * means they are likely already in use, so leave them alone.  We
 680          * can leave it up to e2fsck to clean up any inconsistencies there.
 681          */
 682 }
 683
 684 /*
 685  * Open the external journal device
 686  */
 687 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 688 {
 689         struct block_device *bdev;
 690         char b[BDEVNAME_SIZE];
 691
 692         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 693         if (IS_ERR(bdev))
 694                 goto fail;
 695         return bdev;
 696
 697 fail:
 698         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 699                         __bdevname(dev, b), PTR_ERR(bdev));
 700         return NULL;
 701 }
 702
 703 /*
 704  * Release the journal device
 705  */
 706 static int ext4_blkdev_put(struct block_device *bdev)
 707 {
 708         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 709 }
 710
 711 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 712 {
 713         struct block_device *bdev;
 714         int ret = -ENODEV;
 715
 716         bdev = sbi->journal_bdev;
 717         if (bdev) {
 718                 ret = ext4_blkdev_put(bdev);
 719                 sbi->journal_bdev = NULL;
 720         }
 721         return ret;
 722 }
 723
 724 static inline struct inode *orphan_list_entry(struct list_head *l)
 725 {
 726         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 727 }
 728
 729 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 730 {
 731         struct list_head *l;
 732
 733         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 734                  le32_to_cpu(sbi->s_es->s_last_orphan));
 735
 736         printk(KERN_ERR "sb_info orphan list:\n");
 737         list_for_each(l, &sbi->s_orphan) {
 738                 struct inode *inode = orphan_list_entry(l);
 739                 printk(KERN_ERR "  "
 740                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 741                        inode->i_sb->s_id, inode->i_ino, inode,
 742                        inode->i_mode, inode->i_nlink,
 743                        NEXT_ORPHAN(inode));
 744         }
 745 }
 746
 747 static void ext4_put_super(struct super_block *sb)
 748 {
 749         struct ext4_sb_info *sbi = EXT4_SB(sb);
 750         struct ext4_super_block *es = sbi->s_es;
 751         int i, err;
 752
 753         ext4_unregister_li_request(sb);
 754         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 755
 756         flush_workqueue(sbi->dio_unwritten_wq);
 757         destroy_workqueue(sbi->dio_unwritten_wq);
 758
 759         if (sbi->s_journal) {
 760                 err = jbd2_journal_destroy(sbi->s_journal);
 761                 sbi->s_journal = NULL;
 762                 if (err < 0)
 763                         ext4_abort(sb, "Couldn't clean up the journal");
 764         }
 765
 766         ext4_es_unregister_shrinker(sb);
 767         del_timer(&sbi->s_err_report);
 768         ext4_release_system_zone(sb);
 769         ext4_mb_release(sb);
 770         ext4_ext_release(sb);
 771         ext4_xattr_put_super(sb);
 772
 773         if (!(sb->s_flags & MS_RDONLY)) {
 774                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 775                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 776         }
 777         if (!(sb->s_flags & MS_RDONLY))
 778                 ext4_commit_super(sb, 1);
 779
 780         if (sbi->s_proc) {
 781                 remove_proc_entry("options", sbi->s_proc);
 782                 remove_proc_entry(sb->s_id, ext4_proc_root);
 783         }
 784         kobject_del(&sbi->s_kobj);
 785
 786         for (i = 0; i < sbi->s_gdb_count; i++)
 787                 brelse(sbi->s_group_desc[i]);
 788         ext4_kvfree(sbi->s_group_desc);
 789         ext4_kvfree(sbi->s_flex_groups);
 790         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 791         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 792         percpu_counter_destroy(&sbi->s_dirs_counter);
 793         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 794         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
 795         brelse(sbi->s_sbh);
 796 #ifdef CONFIG_QUOTA
 797         for (i = 0; i < MAXQUOTAS; i++)
 798                 kfree(sbi->s_qf_names[i]);
 799 #endif
 800
 801         /* Debugging code just in case the in-memory inode orphan list
 802          * isn't empty.  The on-disk one can be non-empty if we've
 803          * detected an error and taken the fs readonly, but the
 804          * in-memory list had better be clean by this point. */
 805         if (!list_empty(&sbi->s_orphan))
 806                 dump_orphan_list(sb, sbi);
 807         J_ASSERT(list_empty(&sbi->s_orphan));
 808
 809         invalidate_bdev(sb->s_bdev);
 810         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 811                 /*
 812                  * Invalidate the journal device's buffers.  We don't want them
 813                  * floating about in memory - the physical journal device may
 814                  * hotswapped, and it breaks the `ro-after' testing code.
 815                  */
 816                 sync_blockdev(sbi->journal_bdev);
 817                 invalidate_bdev(sbi->journal_bdev);
 818                 ext4_blkdev_remove(sbi);
 819         }
 820         if (sbi->s_mmp_tsk)
 821                 kthread_stop(sbi->s_mmp_tsk);
 822         sb->s_fs_info = NULL;
 823         /*
 824          * Now that we are completely done shutting down the
 825          * superblock, we need to actually destroy the kobject.
 826          */
 827         kobject_put(&sbi->s_kobj);
 828         wait_for_completion(&sbi->s_kobj_unregister);
 829         if (sbi->s_chksum_driver)
 830                 crypto_free_shash(sbi->s_chksum_driver);
 831         kfree(sbi->s_blockgroup_lock);
 832         kfree(sbi);
 833 }
 834
 835 static struct kmem_cache *ext4_inode_cachep;
 836
 837 /*
 838  * Called inside transaction, so use GFP_NOFS
 839  */
 840 static struct inode *ext4_alloc_inode(struct super_block *sb)
 841 {
 842         struct ext4_inode_info *ei;
 843
 844         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 845         if (!ei)
 846                 return NULL;
 847
 848         ei->vfs_inode.i_version = 1;
 849         INIT_LIST_HEAD(&ei->i_prealloc_list);
 850         spin_lock_init(&ei->i_prealloc_lock);
 851         ext4_es_init_tree(&ei->i_es_tree);
 852         rwlock_init(&ei->i_es_lock);
 853         INIT_LIST_HEAD(&ei->i_es_lru);
 854         ei->i_es_lru_nr = 0;
 855         ei->i_reserved_data_blocks = 0;
 856         ei->i_reserved_meta_blocks = 0;
 857         ei->i_allocated_meta_blocks = 0;
 858         ei->i_da_metadata_calc_len = 0;
 859         ei->i_da_metadata_calc_last_lblock = 0;
 860         spin_lock_init(&(ei->i_block_reservation_lock));
 861 #ifdef CONFIG_QUOTA
 862         ei->i_reserved_quota = 0;
 863 #endif
 864         ei->jinode = NULL;
 865         INIT_LIST_HEAD(&ei->i_completed_io_list);
 866         spin_lock_init(&ei->i_completed_io_lock);
 867         ei->i_sync_tid = 0;
 868         ei->i_datasync_tid = 0;
 869         atomic_set(&ei->i_ioend_count, 0);
 870         atomic_set(&ei->i_unwritten, 0);
 871         INIT_WORK(&ei->i_unwritten_work, ext4_end_io_work);
 872
 873         return &ei->vfs_inode;
 874 }
 875
 876 static int ext4_drop_inode(struct inode *inode)
 877 {
 878         int drop = generic_drop_inode(inode);
 879
 880         trace_ext4_drop_inode(inode, drop);
 881         return drop;
 882 }
 883
 884 static void ext4_i_callback(struct rcu_head *head)
 885 {
 886         struct inode *inode = container_of(head, struct inode, i_rcu);
 887         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 888 }
 889
 890 static void ext4_destroy_inode(struct inode *inode)
 891 {
 892         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 893                 ext4_msg(inode->i_sb, KERN_ERR,
 894                          "Inode %lu (%p): orphan list check failed!",
 895                          inode->i_ino, EXT4_I(inode));
 896                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 897                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 898                                 true);
 899                 dump_stack();
 900         }
 901         call_rcu(&inode->i_rcu, ext4_i_callback);
 902 }
 903
 904 static void init_once(void *foo)
 905 {
 906         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 907
 908         INIT_LIST_HEAD(&ei->i_orphan);
 909         init_rwsem(&ei->xattr_sem);
 910         init_rwsem(&ei->i_data_sem);
 911         inode_init_once(&ei->vfs_inode);
 912 }
 913
 914 static int init_inodecache(void)
 915 {
 916         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 917                                              sizeof(struct ext4_inode_info),
 918                                              0, (SLAB_RECLAIM_ACCOUNT|
 919                                                 SLAB_MEM_SPREAD),
 920                                              init_once);
 921         if (ext4_inode_cachep == NULL)
 922                 return -ENOMEM;
 923         return 0;
 924 }
 925
 926 static void destroy_inodecache(void)
 927 {
 928         /*
 929          * Make sure all delayed rcu free inodes are flushed before we
 930          * destroy cache.
 931          */
 932         rcu_barrier();
 933         kmem_cache_destroy(ext4_inode_cachep);
 934 }
 935
 936 void ext4_clear_inode(struct inode *inode)
 937 {
 938         invalidate_inode_buffers(inode);
 939         clear_inode(inode);
 940         dquot_drop(inode);
 941         ext4_discard_preallocations(inode);
 942         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 943         ext4_es_lru_del(inode);
 944         if (EXT4_I(inode)->jinode) {
 945                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 946                                                EXT4_I(inode)->jinode);
 947                 jbd2_free_inode(EXT4_I(inode)->jinode);
 948                 EXT4_I(inode)->jinode = NULL;
 949         }
 950 }
 951
 952 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 953                                         u64 ino, u32 generation)
 954 {
 955         struct inode *inode;
 956
 957         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 958                 return ERR_PTR(-ESTALE);
 959         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 960                 return ERR_PTR(-ESTALE);
 961
 962         /* iget isn't really right if the inode is currently unallocated!!
 963          *
 964          * ext4_read_inode will return a bad_inode if the inode had been
 965          * deleted, so we should be safe.
 966          *
 967          * Currently we don't know the generation for parent directory, so
 968          * a generation of 0 means "accept any"
 969          */
 970         inode = ext4_iget(sb, ino);
 971         if (IS_ERR(inode))
 972                 return ERR_CAST(inode);
 973         if (generation && inode->i_generation != generation) {
 974                 iput(inode);
 975                 return ERR_PTR(-ESTALE);
 976         }
 977
 978         return inode;
 979 }
 980
 981 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 982                                         int fh_len, int fh_type)
 983 {
 984         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 985                                     ext4_nfs_get_inode);
 986 }
 987
 988 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 989                                         int fh_len, int fh_type)
 990 {
 991         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 992                                     ext4_nfs_get_inode);
 993 }
 994
 995 /*
 996  * Try to release metadata pages (indirect blocks, directories) which are
 997  * mapped via the block device.  Since these pages could have journal heads
 998  * which would prevent try_to_free_buffers() from freeing them, we must use
 999  * jbd2 layer's try_to_free_buffers() function to release them.
1000  */
1001 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1002                                  gfp_t wait)
1003 {
1004         journal_t *journal = EXT4_SB(sb)->s_journal;
1005
1006         WARN_ON(PageChecked(page));
1007         if (!page_has_buffers(page))
1008                 return 0;
1009         if (journal)
1010                 return jbd2_journal_try_to_free_buffers(journal, page,
1011                                                         wait & ~__GFP_WAIT);
1012         return try_to_free_buffers(page);
1013 }
1014
1015 #ifdef CONFIG_QUOTA
1016 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1017 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1018
1019 static int ext4_write_dquot(struct dquot *dquot);
1020 static int ext4_acquire_dquot(struct dquot *dquot);
1021 static int ext4_release_dquot(struct dquot *dquot);
1022 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1023 static int ext4_write_info(struct super_block *sb, int type);
1024 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1025                          struct path *path);
1026 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1027                                  int format_id);
1028 static int ext4_quota_off(struct super_block *sb, int type);
1029 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1030 static int ext4_quota_on_mount(struct super_block *sb, int type);
1031 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1032                                size_t len, loff_t off);
1033 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1034                                 const char *data, size_t len, loff_t off);
1035 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1036                              unsigned int flags);
1037 static int ext4_enable_quotas(struct super_block *sb);
1038
1039 static const struct dquot_operations ext4_quota_operations = {
1040         .get_reserved_space = ext4_get_reserved_space,
1041         .write_dquot    = ext4_write_dquot,
1042         .acquire_dquot  = ext4_acquire_dquot,
1043         .release_dquot  = ext4_release_dquot,
1044         .mark_dirty     = ext4_mark_dquot_dirty,
1045         .write_info     = ext4_write_info,
1046         .alloc_dquot    = dquot_alloc,
1047         .destroy_dquot  = dquot_destroy,
1048 };
1049
1050 static const struct quotactl_ops ext4_qctl_operations = {
1051         .quota_on       = ext4_quota_on,
1052         .quota_off      = ext4_quota_off,
1053         .quota_sync     = dquot_quota_sync,
1054         .get_info       = dquot_get_dqinfo,
1055         .set_info       = dquot_set_dqinfo,
1056         .get_dqblk      = dquot_get_dqblk,
1057         .set_dqblk      = dquot_set_dqblk
1058 };
1059
1060 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1061         .quota_on_meta  = ext4_quota_on_sysfile,
1062         .quota_off      = ext4_quota_off_sysfile,
1063         .quota_sync     = dquot_quota_sync,
1064         .get_info       = dquot_get_dqinfo,
1065         .set_info       = dquot_set_dqinfo,
1066         .get_dqblk      = dquot_get_dqblk,
1067         .set_dqblk      = dquot_set_dqblk
1068 };
1069 #endif
1070
1071 static const struct super_operations ext4_sops = {
1072         .alloc_inode    = ext4_alloc_inode,
1073         .destroy_inode  = ext4_destroy_inode,
1074         .write_inode    = ext4_write_inode,
1075         .dirty_inode    = ext4_dirty_inode,
1076         .drop_inode     = ext4_drop_inode,
1077         .evict_inode    = ext4_evict_inode,
1078         .put_super      = ext4_put_super,
1079         .sync_fs        = ext4_sync_fs,
1080         .freeze_fs      = ext4_freeze,
1081         .unfreeze_fs    = ext4_unfreeze,
1082         .statfs         = ext4_statfs,
1083         .remount_fs     = ext4_remount,
1084         .show_options   = ext4_show_options,
1085 #ifdef CONFIG_QUOTA
1086         .quota_read     = ext4_quota_read,
1087         .quota_write    = ext4_quota_write,
1088 #endif
1089         .bdev_try_to_free_page = bdev_try_to_free_page,
1090 };
1091
1092 static const struct super_operations ext4_nojournal_sops = {
1093         .alloc_inode    = ext4_alloc_inode,
1094         .destroy_inode  = ext4_destroy_inode,
1095         .write_inode    = ext4_write_inode,
1096         .dirty_inode    = ext4_dirty_inode,
1097         .drop_inode     = ext4_drop_inode,
1098         .evict_inode    = ext4_evict_inode,
1099         .put_super      = ext4_put_super,
1100         .statfs         = ext4_statfs,
1101         .remount_fs     = ext4_remount,
1102         .show_options   = ext4_show_options,
1103 #ifdef CONFIG_QUOTA
1104         .quota_read     = ext4_quota_read,
1105         .quota_write    = ext4_quota_write,
1106 #endif
1107         .bdev_try_to_free_page = bdev_try_to_free_page,
1108 };
1109
1110 static const struct export_operations ext4_export_ops = {
1111         .fh_to_dentry = ext4_fh_to_dentry,
1112         .fh_to_parent = ext4_fh_to_parent,
1113         .get_parent = ext4_get_parent,
1114 };
1115
1116 enum {
1117         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1118         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1119         Opt_nouid32, Opt_debug, Opt_removed,
1120         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1121         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1122         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1123         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1124         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1125         Opt_data_err_abort, Opt_data_err_ignore,
1126         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1127         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1128         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1129         Opt_usrquota, Opt_grpquota, Opt_i_version,
1130         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1131         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1132         Opt_inode_readahead_blks, Opt_journal_ioprio,
1133         Opt_dioread_nolock, Opt_dioread_lock,
1134         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1135         Opt_max_dir_size_kb,
1136 };
1137
1138 static const match_table_t tokens = {
1139         {Opt_bsd_df, "bsddf"},
1140         {Opt_minix_df, "minixdf"},
1141         {Opt_grpid, "grpid"},
1142         {Opt_grpid, "bsdgroups"},
1143         {Opt_nogrpid, "nogrpid"},
1144         {Opt_nogrpid, "sysvgroups"},
1145         {Opt_resgid, "resgid=%u"},
1146         {Opt_resuid, "resuid=%u"},
1147         {Opt_sb, "sb=%u"},
1148         {Opt_err_cont, "errors=continue"},
1149         {Opt_err_panic, "errors=panic"},
1150         {Opt_err_ro, "errors=remount-ro"},
1151         {Opt_nouid32, "nouid32"},
1152         {Opt_debug, "debug"},
1153         {Opt_removed, "oldalloc"},
1154         {Opt_removed, "orlov"},
1155         {Opt_user_xattr, "user_xattr"},
1156         {Opt_nouser_xattr, "nouser_xattr"},
1157         {Opt_acl, "acl"},
1158         {Opt_noacl, "noacl"},
1159         {Opt_noload, "norecovery"},
1160         {Opt_noload, "noload"},
1161         {Opt_removed, "nobh"},
1162         {Opt_removed, "bh"},
1163         {Opt_commit, "commit=%u"},
1164         {Opt_min_batch_time, "min_batch_time=%u"},
1165         {Opt_max_batch_time, "max_batch_time=%u"},
1166         {Opt_journal_dev, "journal_dev=%u"},
1167         {Opt_journal_checksum, "journal_checksum"},
1168         {Opt_journal_async_commit, "journal_async_commit"},
1169         {Opt_abort, "abort"},
1170         {Opt_data_journal, "data=journal"},
1171         {Opt_data_ordered, "data=ordered"},
1172         {Opt_data_writeback, "data=writeback"},
1173         {Opt_data_err_abort, "data_err=abort"},
1174         {Opt_data_err_ignore, "data_err=ignore"},
1175         {Opt_offusrjquota, "usrjquota="},
1176         {Opt_usrjquota, "usrjquota=%s"},
1177         {Opt_offgrpjquota, "grpjquota="},
1178         {Opt_grpjquota, "grpjquota=%s"},
1179         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1180         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1181         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1182         {Opt_grpquota, "grpquota"},
1183         {Opt_noquota, "noquota"},
1184         {Opt_quota, "quota"},
1185         {Opt_usrquota, "usrquota"},
1186         {Opt_barrier, "barrier=%u"},
1187         {Opt_barrier, "barrier"},
1188         {Opt_nobarrier, "nobarrier"},
1189         {Opt_i_version, "i_version"},
1190         {Opt_stripe, "stripe=%u"},
1191         {Opt_delalloc, "delalloc"},
1192         {Opt_nodelalloc, "nodelalloc"},
1193         {Opt_removed, "mblk_io_submit"},
1194         {Opt_removed, "nomblk_io_submit"},
1195         {Opt_block_validity, "block_validity"},
1196         {Opt_noblock_validity, "noblock_validity"},
1197         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1198         {Opt_journal_ioprio, "journal_ioprio=%u"},
1199         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1200         {Opt_auto_da_alloc, "auto_da_alloc"},
1201         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1202         {Opt_dioread_nolock, "dioread_nolock"},
1203         {Opt_dioread_lock, "dioread_lock"},
1204         {Opt_discard, "discard"},
1205         {Opt_nodiscard, "nodiscard"},
1206         {Opt_init_itable, "init_itable=%u"},
1207         {Opt_init_itable, "init_itable"},
1208         {Opt_noinit_itable, "noinit_itable"},
1209         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1210         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1211         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1212         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1213         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1214         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1215         {Opt_err, NULL},
1216 };
1217
1218 static ext4_fsblk_t get_sb_block(void **data)
1219 {
1220         ext4_fsblk_t    sb_block;
1221         char            *options = (char *) *data;
1222
1223         if (!options || strncmp(options, "sb=", 3) != 0)
1224                 return 1;       /* Default location */
1225
1226         options += 3;
1227         /* TODO: use simple_strtoll with >32bit ext4 */
1228         sb_block = simple_strtoul(options, &options, 0);
1229         if (*options && *options != ',') {
1230                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1231                        (char *) *data);
1232                 return 1;
1233         }
1234         if (*options == ',')
1235                 options++;
1236         *data = (void *) options;
1237
1238         return sb_block;
1239 }
1240
1241 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1242 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1243         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1244
1245 #ifdef CONFIG_QUOTA
1246 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1247 {
1248         struct ext4_sb_info *sbi = EXT4_SB(sb);
1249         char *qname;
1250         int ret = -1;
1251
1252         if (sb_any_quota_loaded(sb) &&
1253                 !sbi->s_qf_names[qtype]) {
1254                 ext4_msg(sb, KERN_ERR,
1255                         "Cannot change journaled "
1256                         "quota options when quota turned on");
1257                 return -1;
1258         }
1259         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1260                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1261                          "when QUOTA feature is enabled");
1262                 return -1;
1263         }
1264         qname = match_strdup(args);
1265         if (!qname) {
1266                 ext4_msg(sb, KERN_ERR,
1267                         "Not enough memory for storing quotafile name");
1268                 return -1;
1269         }
1270         if (sbi->s_qf_names[qtype]) {
1271                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1272                         ret = 1;
1273                 else
1274                         ext4_msg(sb, KERN_ERR,
1275                                  "%s quota file already specified",
1276                                  QTYPE2NAME(qtype));
1277                 goto errout;
1278         }
1279         if (strchr(qname, '/')) {
1280                 ext4_msg(sb, KERN_ERR,
1281                         "quotafile must be on filesystem root");
1282                 goto errout;
1283         }
1284         sbi->s_qf_names[qtype] = qname;
1285         set_opt(sb, QUOTA);
1286         return 1;
1287 errout:
1288         kfree(qname);
1289         return ret;
1290 }
1291
1292 static int clear_qf_name(struct super_block *sb, int qtype)
1293 {
1294
1295         struct ext4_sb_info *sbi = EXT4_SB(sb);
1296
1297         if (sb_any_quota_loaded(sb) &&
1298                 sbi->s_qf_names[qtype]) {
1299                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1300                         " when quota turned on");
1301                 return -1;
1302         }
1303         kfree(sbi->s_qf_names[qtype]);
1304         sbi->s_qf_names[qtype] = NULL;
1305         return 1;
1306 }
1307 #endif
1308
1309 #define MOPT_SET        0x0001
1310 #define MOPT_CLEAR      0x0002
1311 #define MOPT_NOSUPPORT  0x0004
1312 #define MOPT_EXPLICIT   0x0008
1313 #define MOPT_CLEAR_ERR  0x0010
1314 #define MOPT_GTE0       0x0020
1315 #ifdef CONFIG_QUOTA
1316 #define MOPT_Q          0
1317 #define MOPT_QFMT       0x0040
1318 #else
1319 #define MOPT_Q          MOPT_NOSUPPORT
1320 #define MOPT_QFMT       MOPT_NOSUPPORT
1321 #endif
1322 #define MOPT_DATAJ      0x0080
1323 #define MOPT_NO_EXT2    0x0100
1324 #define MOPT_NO_EXT3    0x0200
1325 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1326
1327 static const struct mount_opts {
1328         int     token;
1329         int     mount_opt;
1330         int     flags;
1331 } ext4_mount_opts[] = {
1332         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1333         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1334         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1335         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1336         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1337         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1338         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1339          MOPT_EXT4_ONLY | MOPT_SET},
1340         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1341          MOPT_EXT4_ONLY | MOPT_CLEAR},
1342         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1343         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1344         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1345          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1346         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1347          MOPT_EXT4_ONLY | MOPT_CLEAR | MOPT_EXPLICIT},
1348         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1349          MOPT_EXT4_ONLY | MOPT_SET},
1350         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1351                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1352          MOPT_EXT4_ONLY | MOPT_SET},
1353         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1354         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1355         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1356         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1357         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1358          MOPT_NO_EXT2 | MOPT_SET},
1359         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1360          MOPT_NO_EXT2 | MOPT_CLEAR},
1361         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1362         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1363         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1364         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1365         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1366         {Opt_commit, 0, MOPT_GTE0},
1367         {Opt_max_batch_time, 0, MOPT_GTE0},
1368         {Opt_min_batch_time, 0, MOPT_GTE0},
1369         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1370         {Opt_init_itable, 0, MOPT_GTE0},
1371         {Opt_stripe, 0, MOPT_GTE0},
1372         {Opt_resuid, 0, MOPT_GTE0},
1373         {Opt_resgid, 0, MOPT_GTE0},
1374         {Opt_journal_dev, 0, MOPT_GTE0},
1375         {Opt_journal_ioprio, 0, MOPT_GTE0},
1376         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1377         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1378         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1379          MOPT_NO_EXT2 | MOPT_DATAJ},
1380         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1381         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1382 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1383         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1384         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1385 #else
1386         {Opt_acl, 0, MOPT_NOSUPPORT},
1387         {Opt_noacl, 0, MOPT_NOSUPPORT},
1388 #endif
1389         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1390         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1391         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1392         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1393                                                         MOPT_SET | MOPT_Q},
1394         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1395                                                         MOPT_SET | MOPT_Q},
1396         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1397                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1398         {Opt_usrjquota, 0, MOPT_Q},
1399         {Opt_grpjquota, 0, MOPT_Q},
1400         {Opt_offusrjquota, 0, MOPT_Q},
1401         {Opt_offgrpjquota, 0, MOPT_Q},
1402         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1403         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1404         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1405         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1406         {Opt_err, 0, 0}
1407 };
1408
1409 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1410                             substring_t *args, unsigned long *journal_devnum,
1411                             unsigned int *journal_ioprio, int is_remount)
1412 {
1413         struct ext4_sb_info *sbi = EXT4_SB(sb);
1414         const struct mount_opts *m;
1415         kuid_t uid;
1416         kgid_t gid;
1417         int arg = 0;
1418
1419 #ifdef CONFIG_QUOTA
1420         if (token == Opt_usrjquota)
1421                 return set_qf_name(sb, USRQUOTA, &args[0]);
1422         else if (token == Opt_grpjquota)
1423                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1424         else if (token == Opt_offusrjquota)
1425                 return clear_qf_name(sb, USRQUOTA);
1426         else if (token == Opt_offgrpjquota)
1427                 return clear_qf_name(sb, GRPQUOTA);
1428 #endif
1429         switch (token) {
1430         case Opt_noacl:
1431         case Opt_nouser_xattr:
1432                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1433                 break;
1434         case Opt_sb:
1435                 return 1;       /* handled by get_sb_block() */
1436         case Opt_removed:
1437                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1438                 return 1;
1439         case Opt_abort:
1440                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1441                 return 1;
1442         case Opt_i_version:
1443                 sb->s_flags |= MS_I_VERSION;
1444                 return 1;
1445         }
1446
1447         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1448                 if (token == m->token)
1449                         break;
1450
1451         if (m->token == Opt_err) {
1452                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1453                          "or missing value", opt);
1454                 return -1;
1455         }
1456
1457         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1458                 ext4_msg(sb, KERN_ERR,
1459                          "Mount option \"%s\" incompatible with ext2", opt);
1460                 return -1;
1461         }
1462         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1463                 ext4_msg(sb, KERN_ERR,
1464                          "Mount option \"%s\" incompatible with ext3", opt);
1465                 return -1;
1466         }
1467
1468         if (args->from && match_int(args, &arg))
1469                 return -1;
1470         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1471                 return -1;
1472         if (m->flags & MOPT_EXPLICIT)
1473                 set_opt2(sb, EXPLICIT_DELALLOC);
1474         if (m->flags & MOPT_CLEAR_ERR)
1475                 clear_opt(sb, ERRORS_MASK);
1476         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1477                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1478                          "options when quota turned on");
1479                 return -1;
1480         }
1481
1482         if (m->flags & MOPT_NOSUPPORT) {
1483                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1484         } else if (token == Opt_commit) {
1485                 if (arg == 0)
1486                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1487                 sbi->s_commit_interval = HZ * arg;
1488         } else if (token == Opt_max_batch_time) {
1489                 if (arg == 0)
1490                         arg = EXT4_DEF_MAX_BATCH_TIME;
1491                 sbi->s_max_batch_time = arg;
1492         } else if (token == Opt_min_batch_time) {
1493                 sbi->s_min_batch_time = arg;
1494         } else if (token == Opt_inode_readahead_blks) {
1495                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1496                         ext4_msg(sb, KERN_ERR,
1497                                  "EXT4-fs: inode_readahead_blks must be "
1498                                  "0 or a power of 2 smaller than 2^31");
1499                         return -1;
1500                 }
1501                 sbi->s_inode_readahead_blks = arg;
1502         } else if (token == Opt_init_itable) {
1503                 set_opt(sb, INIT_INODE_TABLE);
1504                 if (!args->from)
1505                         arg = EXT4_DEF_LI_WAIT_MULT;
1506                 sbi->s_li_wait_mult = arg;
1507         } else if (token == Opt_max_dir_size_kb) {
1508                 sbi->s_max_dir_size_kb = arg;
1509         } else if (token == Opt_stripe) {
1510                 sbi->s_stripe = arg;
1511         } else if (token == Opt_resuid) {
1512                 uid = make_kuid(current_user_ns(), arg);
1513                 if (!uid_valid(uid)) {
1514                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1515                         return -1;
1516                 }
1517                 sbi->s_resuid = uid;
1518         } else if (token == Opt_resgid) {
1519                 gid = make_kgid(current_user_ns(), arg);
1520                 if (!gid_valid(gid)) {
1521                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1522                         return -1;
1523                 }
1524                 sbi->s_resgid = gid;
1525         } else if (token == Opt_journal_dev) {
1526                 if (is_remount) {
1527                         ext4_msg(sb, KERN_ERR,
1528                                  "Cannot specify journal on remount");
1529                         return -1;
1530                 }
1531                 *journal_devnum = arg;
1532         } else if (token == Opt_journal_ioprio) {
1533                 if (arg > 7) {
1534                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1535                                  " (must be 0-7)");
1536                         return -1;
1537                 }
1538                 *journal_ioprio =
1539                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1540         } else if (m->flags & MOPT_DATAJ) {
1541                 if (is_remount) {
1542                         if (!sbi->s_journal)
1543                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1544                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1545                                 ext4_msg(sb, KERN_ERR,
1546                                          "Cannot change data mode on remount");
1547                                 return -1;
1548                         }
1549                 } else {
1550                         clear_opt(sb, DATA_FLAGS);
1551                         sbi->s_mount_opt |= m->mount_opt;
1552                 }
1553 #ifdef CONFIG_QUOTA
1554         } else if (m->flags & MOPT_QFMT) {
1555                 if (sb_any_quota_loaded(sb) &&
1556                     sbi->s_jquota_fmt != m->mount_opt) {
1557                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1558                                  "quota options when quota turned on");
1559                         return -1;
1560                 }
1561                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1562                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1563                         ext4_msg(sb, KERN_ERR,
1564                                  "Cannot set journaled quota options "
1565                                  "when QUOTA feature is enabled");
1566                         return -1;
1567                 }
1568                 sbi->s_jquota_fmt = m->mount_opt;
1569 #endif
1570         } else {
1571                 if (!args->from)
1572                         arg = 1;
1573                 if (m->flags & MOPT_CLEAR)
1574                         arg = !arg;
1575                 else if (unlikely(!(m->flags & MOPT_SET))) {
1576                         ext4_msg(sb, KERN_WARNING,
1577                                  "buggy handling of option %s", opt);
1578                         WARN_ON(1);
1579                         return -1;
1580                 }
1581                 if (arg != 0)
1582                         sbi->s_mount_opt |= m->mount_opt;
1583                 else
1584                         sbi->s_mount_opt &= ~m->mount_opt;
1585         }
1586         return 1;
1587 }
1588
1589 static int parse_options(char *options, struct super_block *sb,
1590                          unsigned long *journal_devnum,
1591                          unsigned int *journal_ioprio,
1592                          int is_remount)
1593 {
1594         struct ext4_sb_info *sbi = EXT4_SB(sb);
1595         char *p;
1596         substring_t args[MAX_OPT_ARGS];
1597         int token;
1598
1599         if (!options)
1600                 return 1;
1601
1602         while ((p = strsep(&options, ",")) != NULL) {
1603                 if (!*p)
1604                         continue;
1605                 /*
1606                  * Initialize args struct so we know whether arg was
1607                  * found; some options take optional arguments.
1608                  */
1609                 args[0].to = args[0].from = NULL;
1610                 token = match_token(p, tokens, args);
1611                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1612                                      journal_ioprio, is_remount) < 0)
1613                         return 0;
1614         }
1615 #ifdef CONFIG_QUOTA
1616         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1617             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1618                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1619                          "feature is enabled");
1620                 return 0;
1621         }
1622         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1623                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1624                         clear_opt(sb, USRQUOTA);
1625
1626                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1627                         clear_opt(sb, GRPQUOTA);
1628
1629                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1630                         ext4_msg(sb, KERN_ERR, "old and new quota "
1631                                         "format mixing");
1632                         return 0;
1633                 }
1634
1635                 if (!sbi->s_jquota_fmt) {
1636                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1637                                         "not specified");
1638                         return 0;
1639                 }
1640         } else {
1641                 if (sbi->s_jquota_fmt) {
1642                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1643                                         "specified with no journaling "
1644                                         "enabled");
1645                         return 0;
1646                 }
1647         }
1648 #endif
1649         if (test_opt(sb, DIOREAD_NOLOCK)) {
1650                 int blocksize =
1651                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1652
1653                 if (blocksize < PAGE_CACHE_SIZE) {
1654                         ext4_msg(sb, KERN_ERR, "can't mount with "
1655                                  "dioread_nolock if block size != PAGE_SIZE");
1656                         return 0;
1657                 }
1658         }
1659         return 1;
1660 }
1661
1662 static inline void ext4_show_quota_options(struct seq_file *seq,
1663                                            struct super_block *sb)
1664 {
1665 #if defined(CONFIG_QUOTA)
1666         struct ext4_sb_info *sbi = EXT4_SB(sb);
1667
1668         if (sbi->s_jquota_fmt) {
1669                 char *fmtname = "";
1670
1671                 switch (sbi->s_jquota_fmt) {
1672                 case QFMT_VFS_OLD:
1673                         fmtname = "vfsold";
1674                         break;
1675                 case QFMT_VFS_V0:
1676                         fmtname = "vfsv0";
1677                         break;
1678                 case QFMT_VFS_V1:
1679                         fmtname = "vfsv1";
1680                         break;
1681                 }
1682                 seq_printf(seq, ",jqfmt=%s", fmtname);
1683         }
1684
1685         if (sbi->s_qf_names[USRQUOTA])
1686                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1687
1688         if (sbi->s_qf_names[GRPQUOTA])
1689                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1690
1691         if (test_opt(sb, USRQUOTA))
1692                 seq_puts(seq, ",usrquota");
1693
1694         if (test_opt(sb, GRPQUOTA))
1695                 seq_puts(seq, ",grpquota");
1696 #endif
1697 }
1698
1699 static const char *token2str(int token)
1700 {
1701         const struct match_token *t;
1702
1703         for (t = tokens; t->token != Opt_err; t++)
1704                 if (t->token == token && !strchr(t->pattern, '='))
1705                         break;
1706         return t->pattern;
1707 }
1708
1709 /*
1710  * Show an option if
1711  *  - it's set to a non-default value OR
1712  *  - if the per-sb default is different from the global default
1713  */
1714 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1715                               int nodefs)
1716 {
1717         struct ext4_sb_info *sbi = EXT4_SB(sb);
1718         struct ext4_super_block *es = sbi->s_es;
1719         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1720         const struct mount_opts *m;
1721         char sep = nodefs ? '\n' : ',';
1722
1723 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1724 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1725
1726         if (sbi->s_sb_block != 1)
1727                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1728
1729         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1730                 int want_set = m->flags & MOPT_SET;
1731                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1732                     (m->flags & MOPT_CLEAR_ERR))
1733                         continue;
1734                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1735                         continue; /* skip if same as the default */
1736                 if ((want_set &&
1737                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1738                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1739                         continue; /* select Opt_noFoo vs Opt_Foo */
1740                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1741         }
1742
1743         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1744             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1745                 SEQ_OPTS_PRINT("resuid=%u",
1746                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1747         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1748             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1749                 SEQ_OPTS_PRINT("resgid=%u",
1750                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1751         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1752         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1753                 SEQ_OPTS_PUTS("errors=remount-ro");
1754         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1755                 SEQ_OPTS_PUTS("errors=continue");
1756         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1757                 SEQ_OPTS_PUTS("errors=panic");
1758         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1759                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1760         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1761                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1762         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1763                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1764         if (sb->s_flags & MS_I_VERSION)
1765                 SEQ_OPTS_PUTS("i_version");
1766         if (nodefs || sbi->s_stripe)
1767                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1768         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1769                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1770                         SEQ_OPTS_PUTS("data=journal");
1771                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1772                         SEQ_OPTS_PUTS("data=ordered");
1773                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1774                         SEQ_OPTS_PUTS("data=writeback");
1775         }
1776         if (nodefs ||
1777             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1778                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1779                                sbi->s_inode_readahead_blks);
1780
1781         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1782                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1783                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1784         if (nodefs || sbi->s_max_dir_size_kb)
1785                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1786
1787         ext4_show_quota_options(seq, sb);
1788         return 0;
1789 }
1790
1791 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1792 {
1793         return _ext4_show_options(seq, root->d_sb, 0);
1794 }
1795
1796 static int options_seq_show(struct seq_file *seq, void *offset)
1797 {
1798         struct super_block *sb = seq->private;
1799         int rc;
1800
1801         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1802         rc = _ext4_show_options(seq, sb, 1);
1803         seq_puts(seq, "\n");
1804         return rc;
1805 }
1806
1807 static int options_open_fs(struct inode *inode, struct file *file)
1808 {
1809         return single_open(file, options_seq_show, PDE(inode)->data);
1810 }
1811
1812 static const struct file_operations ext4_seq_options_fops = {
1813         .owner = THIS_MODULE,
1814         .open = options_open_fs,
1815         .read = seq_read,
1816         .llseek = seq_lseek,
1817         .release = single_release,
1818 };
1819
1820 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1821                             int read_only)
1822 {
1823         struct ext4_sb_info *sbi = EXT4_SB(sb);
1824         int res = 0;
1825
1826         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1827                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1828                          "forcing read-only mode");
1829                 res = MS_RDONLY;
1830         }
1831         if (read_only)
1832                 goto done;
1833         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1834                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1835                          "running e2fsck is recommended");
1836         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1837                 ext4_msg(sb, KERN_WARNING,
1838                          "warning: mounting fs with errors, "
1839                          "running e2fsck is recommended");
1840         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1841                  le16_to_cpu(es->s_mnt_count) >=
1842                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1843                 ext4_msg(sb, KERN_WARNING,
1844                          "warning: maximal mount count reached, "
1845                          "running e2fsck is recommended");
1846         else if (le32_to_cpu(es->s_checkinterval) &&
1847                 (le32_to_cpu(es->s_lastcheck) +
1848                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1849                 ext4_msg(sb, KERN_WARNING,
1850                          "warning: checktime reached, "
1851                          "running e2fsck is recommended");
1852         if (!sbi->s_journal)
1853                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1854         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1855                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1856         le16_add_cpu(&es->s_mnt_count, 1);
1857         es->s_mtime = cpu_to_le32(get_seconds());
1858         ext4_update_dynamic_rev(sb);
1859         if (sbi->s_journal)
1860                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1861
1862         ext4_commit_super(sb, 1);
1863 done:
1864         if (test_opt(sb, DEBUG))
1865                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1866                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1867                         sb->s_blocksize,
1868                         sbi->s_groups_count,
1869                         EXT4_BLOCKS_PER_GROUP(sb),
1870                         EXT4_INODES_PER_GROUP(sb),
1871                         sbi->s_mount_opt, sbi->s_mount_opt2);
1872
1873         cleancache_init_fs(sb);
1874         return res;
1875 }
1876
1877 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1878 {
1879         struct ext4_sb_info *sbi = EXT4_SB(sb);
1880         struct flex_groups *new_groups;
1881         int size;
1882
1883         if (!sbi->s_log_groups_per_flex)
1884                 return 0;
1885
1886         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1887         if (size <= sbi->s_flex_groups_allocated)
1888                 return 0;
1889
1890         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1891         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1892         if (!new_groups) {
1893                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1894                          size / (int) sizeof(struct flex_groups));
1895                 return -ENOMEM;
1896         }
1897
1898         if (sbi->s_flex_groups) {
1899                 memcpy(new_groups, sbi->s_flex_groups,
1900                        (sbi->s_flex_groups_allocated *
1901                         sizeof(struct flex_groups)));
1902                 ext4_kvfree(sbi->s_flex_groups);
1903         }
1904         sbi->s_flex_groups = new_groups;
1905         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1906         return 0;
1907 }
1908
1909 static int ext4_fill_flex_info(struct super_block *sb)
1910 {
1911         struct ext4_sb_info *sbi = EXT4_SB(sb);
1912         struct ext4_group_desc *gdp = NULL;
1913         ext4_group_t flex_group;
1914         unsigned int groups_per_flex = 0;
1915         int i, err;
1916
1917         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1918         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1919                 sbi->s_log_groups_per_flex = 0;
1920                 return 1;
1921         }
1922         groups_per_flex = 1U << sbi->s_log_groups_per_flex;
1923
1924         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1925         if (err)
1926                 goto failed;
1927
1928         for (i = 0; i < sbi->s_groups_count; i++) {
1929                 gdp = ext4_get_group_desc(sb, i, NULL);
1930
1931                 flex_group = ext4_flex_group(sbi, i);
1932                 atomic_add(ext4_free_inodes_count(sb, gdp),
1933                            &sbi->s_flex_groups[flex_group].free_inodes);
1934                 atomic64_add(ext4_free_group_clusters(sb, gdp),
1935                              &sbi->s_flex_groups[flex_group].free_clusters);
1936                 atomic_add(ext4_used_dirs_count(sb, gdp),
1937                            &sbi->s_flex_groups[flex_group].used_dirs);
1938         }
1939
1940         return 1;
1941 failed:
1942         return 0;
1943 }
1944
1945 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1946                                    struct ext4_group_desc *gdp)
1947 {
1948         int offset;
1949         __u16 crc = 0;
1950         __le32 le_group = cpu_to_le32(block_group);
1951
1952         if ((sbi->s_es->s_feature_ro_compat &
1953              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1954                 /* Use new metadata_csum algorithm */
1955                 __le16 save_csum;
1956                 __u32 csum32;
1957
1958                 save_csum = gdp->bg_checksum;
1959                 gdp->bg_checksum = 0;
1960                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1961                                      sizeof(le_group));
1962                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1963                                      sbi->s_desc_size);
1964                 gdp->bg_checksum = save_csum;
1965
1966                 crc = csum32 & 0xFFFF;
1967                 goto out;
1968         }
1969
1970         /* old crc16 code */
1971         offset = offsetof(struct ext4_group_desc, bg_checksum);
1972
1973         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1974         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1975         crc = crc16(crc, (__u8 *)gdp, offset);
1976         offset += sizeof(gdp->bg_checksum); /* skip checksum */
1977         /* for checksum of struct ext4_group_desc do the rest...*/
1978         if ((sbi->s_es->s_feature_incompat &
1979              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1980             offset < le16_to_cpu(sbi->s_es->s_desc_size))
1981                 crc = crc16(crc, (__u8 *)gdp + offset,
1982                             le16_to_cpu(sbi->s_es->s_desc_size) -
1983                                 offset);
1984
1985 out:
1986         return cpu_to_le16(crc);
1987 }
1988
1989 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
1990                                 struct ext4_group_desc *gdp)
1991 {
1992         if (ext4_has_group_desc_csum(sb) &&
1993             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
1994                                                       block_group, gdp)))
1995                 return 0;
1996
1997         return 1;
1998 }
1999
2000 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2001                               struct ext4_group_desc *gdp)
2002 {
2003         if (!ext4_has_group_desc_csum(sb))
2004                 return;
2005         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2006 }
2007
2008 /* Called at mount-time, super-block is locked */
2009 static int ext4_check_descriptors(struct super_block *sb,
2010                                   ext4_group_t *first_not_zeroed)
2011 {
2012         struct ext4_sb_info *sbi = EXT4_SB(sb);
2013         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2014         ext4_fsblk_t last_block;
2015         ext4_fsblk_t block_bitmap;
2016         ext4_fsblk_t inode_bitmap;
2017         ext4_fsblk_t inode_table;
2018         int flexbg_flag = 0;
2019         ext4_group_t i, grp = sbi->s_groups_count;
2020
2021         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2022                 flexbg_flag = 1;
2023
2024         ext4_debug("Checking group descriptors");
2025
2026         for (i = 0; i < sbi->s_groups_count; i++) {
2027                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2028
2029                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2030                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2031                 else
2032                         last_block = first_block +
2033                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2034
2035                 if ((grp == sbi->s_groups_count) &&
2036                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2037                         grp = i;
2038
2039                 block_bitmap = ext4_block_bitmap(sb, gdp);
2040                 if (block_bitmap < first_block || block_bitmap > last_block) {
2041                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2042                                "Block bitmap for group %u not in group "
2043                                "(block %llu)!", i, block_bitmap);
2044                         return 0;
2045                 }
2046                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2047                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2048                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2049                                "Inode bitmap for group %u not in group "
2050                                "(block %llu)!", i, inode_bitmap);
2051                         return 0;
2052                 }
2053                 inode_table = ext4_inode_table(sb, gdp);
2054                 if (inode_table < first_block ||
2055                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2056                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2057                                "Inode table for group %u not in group "
2058                                "(block %llu)!", i, inode_table);
2059                         return 0;
2060                 }
2061                 ext4_lock_group(sb, i);
2062                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2063                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2064                                  "Checksum for group %u failed (%u!=%u)",
2065                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2066                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2067                         if (!(sb->s_flags & MS_RDONLY)) {
2068                                 ext4_unlock_group(sb, i);
2069                                 return 0;
2070                         }
2071                 }
2072                 ext4_unlock_group(sb, i);
2073                 if (!flexbg_flag)
2074                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2075         }
2076         if (NULL != first_not_zeroed)
2077                 *first_not_zeroed = grp;
2078
2079         ext4_free_blocks_count_set(sbi->s_es,
2080                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2081         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2082         return 1;
2083 }
2084
2085 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2086  * the superblock) which were deleted from all directories, but held open by
2087  * a process at the time of a crash.  We walk the list and try to delete these
2088  * inodes at recovery time (only with a read-write filesystem).
2089  *
2090  * In order to keep the orphan inode chain consistent during traversal (in
2091  * case of crash during recovery), we link each inode into the superblock
2092  * orphan list_head and handle it the same way as an inode deletion during
2093  * normal operation (which journals the operations for us).
2094  *
2095  * We only do an iget() and an iput() on each inode, which is very safe if we
2096  * accidentally point at an in-use or already deleted inode.  The worst that
2097  * can happen in this case is that we get a "bit already cleared" message from
2098  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2099  * e2fsck was run on this filesystem, and it must have already done the orphan
2100  * inode cleanup for us, so we can safely abort without any further action.
2101  */
2102 static void ext4_orphan_cleanup(struct super_block *sb,
2103                                 struct ext4_super_block *es)
2104 {
2105         unsigned int s_flags = sb->s_flags;
2106         int nr_orphans = 0, nr_truncates = 0;
2107 #ifdef CONFIG_QUOTA
2108         int i;
2109 #endif
2110         if (!es->s_last_orphan) {
2111                 jbd_debug(4, "no orphan inodes to clean up\n");
2112                 return;
2113         }
2114
2115         if (bdev_read_only(sb->s_bdev)) {
2116                 ext4_msg(sb, KERN_ERR, "write access "
2117                         "unavailable, skipping orphan cleanup");
2118                 return;
2119         }
2120
2121         /* Check if feature set would not allow a r/w mount */
2122         if (!ext4_feature_set_ok(sb, 0)) {
2123                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2124                          "unknown ROCOMPAT features");
2125                 return;
2126         }
2127
2128         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2129                 /* don't clear list on RO mount w/ errors */
2130                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2131                         jbd_debug(1, "Errors on filesystem, "
2132                                   "clearing orphan list.\n");
2133                         es->s_last_orphan = 0;
2134                 }
2135                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2136                 return;
2137         }
2138
2139         if (s_flags & MS_RDONLY) {
2140                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2141                 sb->s_flags &= ~MS_RDONLY;
2142         }
2143 #ifdef CONFIG_QUOTA
2144         /* Needed for iput() to work correctly and not trash data */
2145         sb->s_flags |= MS_ACTIVE;
2146         /* Turn on quotas so that they are updated correctly */
2147         for (i = 0; i < MAXQUOTAS; i++) {
2148                 if (EXT4_SB(sb)->s_qf_names[i]) {
2149                         int ret = ext4_quota_on_mount(sb, i);
2150                         if (ret < 0)
2151                                 ext4_msg(sb, KERN_ERR,
2152                                         "Cannot turn on journaled "
2153                                         "quota: error %d", ret);
2154                 }
2155         }
2156 #endif
2157
2158         while (es->s_last_orphan) {
2159                 struct inode *inode;
2160
2161                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2162                 if (IS_ERR(inode)) {
2163                         es->s_last_orphan = 0;
2164                         break;
2165                 }
2166
2167                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2168                 dquot_initialize(inode);
2169                 if (inode->i_nlink) {
2170                         ext4_msg(sb, KERN_DEBUG,
2171                                 "%s: truncating inode %lu to %lld bytes",
2172                                 __func__, inode->i_ino, inode->i_size);
2173                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2174                                   inode->i_ino, inode->i_size);
2175                         mutex_lock(&inode->i_mutex);
2176                         ext4_truncate(inode);
2177                         mutex_unlock(&inode->i_mutex);
2178                         nr_truncates++;
2179                 } else {
2180                         ext4_msg(sb, KERN_DEBUG,
2181                                 "%s: deleting unreferenced inode %lu",
2182                                 __func__, inode->i_ino);
2183                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2184                                   inode->i_ino);
2185                         nr_orphans++;
2186                 }
2187                 iput(inode);  /* The delete magic happens here! */
2188         }
2189
2190 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2191
2192         if (nr_orphans)
2193                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2194                        PLURAL(nr_orphans));
2195         if (nr_truncates)
2196                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2197                        PLURAL(nr_truncates));
2198 #ifdef CONFIG_QUOTA
2199         /* Turn quotas off */
2200         for (i = 0; i < MAXQUOTAS; i++) {
2201                 if (sb_dqopt(sb)->files[i])
2202                         dquot_quota_off(sb, i);
2203         }
2204 #endif
2205         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2206 }
2207
2208 /*
2209  * Maximal extent format file size.
2210  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2211  * extent format containers, within a sector_t, and within i_blocks
2212  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2213  * so that won't be a limiting factor.
2214  *
2215  * However there is other limiting factor. We do store extents in the form
2216  * of starting block and length, hence the resulting length of the extent
2217  * covering maximum file size must fit into on-disk format containers as
2218  * well. Given that length is always by 1 unit bigger than max unit (because
2219  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2220  *
2221  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2222  */
2223 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2224 {
2225         loff_t res;
2226         loff_t upper_limit = MAX_LFS_FILESIZE;
2227
2228         /* small i_blocks in vfs inode? */
2229         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2230                 /*
2231                  * CONFIG_LBDAF is not enabled implies the inode
2232                  * i_block represent total blocks in 512 bytes
2233                  * 32 == size of vfs inode i_blocks * 8
2234                  */
2235                 upper_limit = (1LL << 32) - 1;
2236
2237                 /* total blocks in file system block size */
2238                 upper_limit >>= (blkbits - 9);
2239                 upper_limit <<= blkbits;
2240         }
2241
2242         /*
2243          * 32-bit extent-start container, ee_block. We lower the maxbytes
2244          * by one fs block, so ee_len can cover the extent of maximum file
2245          * size
2246          */
2247         res = (1LL << 32) - 1;
2248         res <<= blkbits;
2249
2250         /* Sanity check against vm- & vfs- imposed limits */
2251         if (res > upper_limit)
2252                 res = upper_limit;
2253
2254         return res;
2255 }
2256
2257 /*
2258  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2259  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2260  * We need to be 1 filesystem block less than the 2^48 sector limit.
2261  */
2262 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2263 {
2264         loff_t res = EXT4_NDIR_BLOCKS;
2265         int meta_blocks;
2266         loff_t upper_limit;
2267         /* This is calculated to be the largest file size for a dense, block
2268          * mapped file such that the file's total number of 512-byte sectors,
2269          * including data and all indirect blocks, does not exceed (2^48 - 1).
2270          *
2271          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2272          * number of 512-byte sectors of the file.
2273          */
2274
2275         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2276                 /*
2277                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2278                  * the inode i_block field represents total file blocks in
2279                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2280                  */
2281                 upper_limit = (1LL << 32) - 1;
2282
2283                 /* total blocks in file system block size */
2284                 upper_limit >>= (bits - 9);
2285
2286         } else {
2287                 /*
2288                  * We use 48 bit ext4_inode i_blocks
2289                  * With EXT4_HUGE_FILE_FL set the i_blocks
2290                  * represent total number of blocks in
2291                  * file system block size
2292                  */
2293                 upper_limit = (1LL << 48) - 1;
2294
2295         }
2296
2297         /* indirect blocks */
2298         meta_blocks = 1;
2299         /* double indirect blocks */
2300         meta_blocks += 1 + (1LL << (bits-2));
2301         /* tripple indirect blocks */
2302         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2303
2304         upper_limit -= meta_blocks;
2305         upper_limit <<= bits;
2306
2307         res += 1LL << (bits-2);
2308         res += 1LL << (2*(bits-2));
2309         res += 1LL << (3*(bits-2));
2310         res <<= bits;
2311         if (res > upper_limit)
2312                 res = upper_limit;
2313
2314         if (res > MAX_LFS_FILESIZE)
2315                 res = MAX_LFS_FILESIZE;
2316
2317         return res;
2318 }
2319
2320 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2321                                    ext4_fsblk_t logical_sb_block, int nr)
2322 {
2323         struct ext4_sb_info *sbi = EXT4_SB(sb);
2324         ext4_group_t bg, first_meta_bg;
2325         int has_super = 0;
2326
2327         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2328
2329         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2330             nr < first_meta_bg)
2331                 return logical_sb_block + nr + 1;
2332         bg = sbi->s_desc_per_block * nr;
2333         if (ext4_bg_has_super(sb, bg))
2334                 has_super = 1;
2335
2336         return (has_super + ext4_group_first_block_no(sb, bg));
2337 }
2338
2339 /**
2340  * ext4_get_stripe_size: Get the stripe size.
2341  * @sbi: In memory super block info
2342  *
2343  * If we have specified it via mount option, then
2344  * use the mount option value. If the value specified at mount time is
2345  * greater than the blocks per group use the super block value.
2346  * If the super block value is greater than blocks per group return 0.
2347  * Allocator needs it be less than blocks per group.
2348  *
2349  */
2350 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2351 {
2352         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2353         unsigned long stripe_width =
2354                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2355         int ret;
2356
2357         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2358                 ret = sbi->s_stripe;
2359         else if (stripe_width <= sbi->s_blocks_per_group)
2360                 ret = stripe_width;
2361         else if (stride <= sbi->s_blocks_per_group)
2362                 ret = stride;
2363         else
2364                 ret = 0;
2365
2366         /*
2367          * If the stripe width is 1, this makes no sense and
2368          * we set it to 0 to turn off stripe handling code.
2369          */
2370         if (ret <= 1)
2371                 ret = 0;
2372
2373         return ret;
2374 }
2375
2376 /* sysfs supprt */
2377
2378 struct ext4_attr {
2379         struct attribute attr;
2380         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2381         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2382                          const char *, size_t);
2383         int offset;
2384 };
2385
2386 static int parse_strtoull(const char *buf,
2387                 unsigned long long max, unsigned long long *value)
2388 {
2389         int ret;
2390
2391         ret = kstrtoull(skip_spaces(buf), 0, value);
2392         if (!ret && *value > max)
2393                 ret = -EINVAL;
2394         return ret;
2395 }
2396
2397 static int parse_strtoul(const char *buf,
2398                 unsigned long max, unsigned long *value)
2399 {
2400         char *endp;
2401
2402         *value = simple_strtoul(skip_spaces(buf), &endp, 0);
2403         endp = skip_spaces(endp);
2404         if (*endp || *value > max)
2405                 return -EINVAL;
2406
2407         return 0;
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
2450         if (parse_strtoul(buf, 0x40000000, &t))
2451                 return -EINVAL;
2452
2453         if (t && !is_power_of_2(t))
2454                 return -EINVAL;
2455
2456         sbi->s_inode_readahead_blks = t;
2457         return count;
2458 }
2459
2460 static ssize_t sbi_ui_show(struct ext4_attr *a,
2461                            struct ext4_sb_info *sbi, char *buf)
2462 {
2463         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2464
2465         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2466 }
2467
2468 static ssize_t sbi_ui_store(struct ext4_attr *a,
2469                             struct ext4_sb_info *sbi,
2470                             const char *buf, size_t count)
2471 {
2472         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2473         unsigned long t;
2474
2475         if (parse_strtoul(buf, 0xffffffff, &t))
2476                 return -EINVAL;
2477         *ui = t;
2478         return count;
2479 }
2480
2481 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2482                                   struct ext4_sb_info *sbi, char *buf)
2483 {
2484         return snprintf(buf, PAGE_SIZE, "%llu\n",
2485                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2486 }
2487
2488 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2489                                    struct ext4_sb_info *sbi,
2490                                    const char *buf, size_t count)
2491 {
2492         unsigned long long val;
2493         int ret;
2494
2495         if (parse_strtoull(buf, -1ULL, &val))
2496                 return -EINVAL;
2497         ret = ext4_reserve_clusters(sbi, val);
2498
2499         return ret ? ret : count;
2500 }
2501
2502 static ssize_t trigger_test_error(struct ext4_attr *a,
2503                                   struct ext4_sb_info *sbi,
2504                                   const char *buf, size_t count)
2505 {
2506         int len = count;
2507
2508         if (!capable(CAP_SYS_ADMIN))
2509                 return -EPERM;
2510
2511         if (len && buf[len-1] == '\n')
2512                 len--;
2513
2514         if (len)
2515                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2516         return count;
2517 }
2518
2519 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2520 static struct ext4_attr ext4_attr_##_name = {                   \
2521         .attr = {.name = __stringify(_name), .mode = _mode },   \
2522         .show   = _show,                                        \
2523         .store  = _store,                                       \
2524         .offset = offsetof(struct ext4_sb_info, _elname),       \
2525 }
2526 #define EXT4_ATTR(name, mode, show, store) \
2527 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2528
2529 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2530 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2531 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2532 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2533         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2534 #define ATTR_LIST(name) &ext4_attr_##name.attr
2535
2536 EXT4_RO_ATTR(delayed_allocation_blocks);
2537 EXT4_RO_ATTR(session_write_kbytes);
2538 EXT4_RO_ATTR(lifetime_write_kbytes);
2539 EXT4_RW_ATTR(reserved_clusters);
2540 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2541                  inode_readahead_blks_store, s_inode_readahead_blks);
2542 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2543 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2544 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2545 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2546 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2547 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2548 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2549 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2550 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2551 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2552
2553 static struct attribute *ext4_attrs[] = {
2554         ATTR_LIST(delayed_allocation_blocks),
2555         ATTR_LIST(session_write_kbytes),
2556         ATTR_LIST(lifetime_write_kbytes),
2557         ATTR_LIST(reserved_clusters),
2558         ATTR_LIST(inode_readahead_blks),
2559         ATTR_LIST(inode_goal),
2560         ATTR_LIST(mb_stats),
2561         ATTR_LIST(mb_max_to_scan),
2562         ATTR_LIST(mb_min_to_scan),
2563         ATTR_LIST(mb_order2_req),
2564         ATTR_LIST(mb_stream_req),
2565         ATTR_LIST(mb_group_prealloc),
2566         ATTR_LIST(max_writeback_mb_bump),
2567         ATTR_LIST(extent_max_zeroout_kb),
2568         ATTR_LIST(trigger_fs_error),
2569         NULL,
2570 };
2571
2572 /* Features this copy of ext4 supports */
2573 EXT4_INFO_ATTR(lazy_itable_init);
2574 EXT4_INFO_ATTR(batched_discard);
2575 EXT4_INFO_ATTR(meta_bg_resize);
2576
2577 static struct attribute *ext4_feat_attrs[] = {
2578         ATTR_LIST(lazy_itable_init),
2579         ATTR_LIST(batched_discard),
2580         ATTR_LIST(meta_bg_resize),
2581         NULL,
2582 };
2583
2584 static ssize_t ext4_attr_show(struct kobject *kobj,
2585                               struct attribute *attr, char *buf)
2586 {
2587         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2588                                                 s_kobj);
2589         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2590
2591         return a->show ? a->show(a, sbi, buf) : 0;
2592 }
2593
2594 static ssize_t ext4_attr_store(struct kobject *kobj,
2595                                struct attribute *attr,
2596                                const char *buf, size_t len)
2597 {
2598         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2599                                                 s_kobj);
2600         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2601
2602         return a->store ? a->store(a, sbi, buf, len) : 0;
2603 }
2604
2605 static void ext4_sb_release(struct kobject *kobj)
2606 {
2607         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2608                                                 s_kobj);
2609         complete(&sbi->s_kobj_unregister);
2610 }
2611
2612 static const struct sysfs_ops ext4_attr_ops = {
2613         .show   = ext4_attr_show,
2614         .store  = ext4_attr_store,
2615 };
2616
2617 static struct kobj_type ext4_ktype = {
2618         .default_attrs  = ext4_attrs,
2619         .sysfs_ops      = &ext4_attr_ops,
2620         .release        = ext4_sb_release,
2621 };
2622
2623 static void ext4_feat_release(struct kobject *kobj)
2624 {
2625         complete(&ext4_feat->f_kobj_unregister);
2626 }
2627
2628 static struct kobj_type ext4_feat_ktype = {
2629         .default_attrs  = ext4_feat_attrs,
2630         .sysfs_ops      = &ext4_attr_ops,
2631         .release        = ext4_feat_release,
2632 };
2633
2634 /*
2635  * Check whether this filesystem can be mounted based on
2636  * the features present and the RDONLY/RDWR mount requested.
2637  * Returns 1 if this filesystem can be mounted as requested,
2638  * 0 if it cannot be.
2639  */
2640 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2641 {
2642         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2643                 ext4_msg(sb, KERN_ERR,
2644                         "Couldn't mount because of "
2645                         "unsupported optional features (%x)",
2646                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2647                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2648                 return 0;
2649         }
2650
2651         if (readonly)
2652                 return 1;
2653
2654         /* Check that feature set is OK for a read-write mount */
2655         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2656                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2657                          "unsupported optional features (%x)",
2658                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2659                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2660                 return 0;
2661         }
2662         /*
2663          * Large file size enabled file system can only be mounted
2664          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2665          */
2666         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2667                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2668                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2669                                  "cannot be mounted RDWR without "
2670                                  "CONFIG_LBDAF");
2671                         return 0;
2672                 }
2673         }
2674         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2675             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2676                 ext4_msg(sb, KERN_ERR,
2677                          "Can't support bigalloc feature without "
2678                          "extents feature\n");
2679                 return 0;
2680         }
2681
2682 #ifndef CONFIG_QUOTA
2683         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2684             !readonly) {
2685                 ext4_msg(sb, KERN_ERR,
2686                          "Filesystem with quota feature cannot be mounted RDWR "
2687                          "without CONFIG_QUOTA");
2688                 return 0;
2689         }
2690 #endif  /* CONFIG_QUOTA */
2691         return 1;
2692 }
2693
2694 /*
2695  * This function is called once a day if we have errors logged
2696  * on the file system
2697  */
2698 static void print_daily_error_info(unsigned long arg)
2699 {
2700         struct super_block *sb = (struct super_block *) arg;
2701         struct ext4_sb_info *sbi;
2702         struct ext4_super_block *es;
2703
2704         sbi = EXT4_SB(sb);
2705         es = sbi->s_es;
2706
2707         if (es->s_error_count)
2708                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2709                          le32_to_cpu(es->s_error_count));
2710         if (es->s_first_error_time) {
2711                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2712                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2713                        (int) sizeof(es->s_first_error_func),
2714                        es->s_first_error_func,
2715                        le32_to_cpu(es->s_first_error_line));
2716                 if (es->s_first_error_ino)
2717                         printk(": inode %u",
2718                                le32_to_cpu(es->s_first_error_ino));
2719                 if (es->s_first_error_block)
2720                         printk(": block %llu", (unsigned long long)
2721                                le64_to_cpu(es->s_first_error_block));
2722                 printk("\n");
2723         }
2724         if (es->s_last_error_time) {
2725                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2726                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2727                        (int) sizeof(es->s_last_error_func),
2728                        es->s_last_error_func,
2729                        le32_to_cpu(es->s_last_error_line));
2730                 if (es->s_last_error_ino)
2731                         printk(": inode %u",
2732                                le32_to_cpu(es->s_last_error_ino));
2733                 if (es->s_last_error_block)
2734                         printk(": block %llu", (unsigned long long)
2735                                le64_to_cpu(es->s_last_error_block));
2736                 printk("\n");
2737         }
2738         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2739 }
2740
2741 /* Find next suitable group and run ext4_init_inode_table */
2742 static int ext4_run_li_request(struct ext4_li_request *elr)
2743 {
2744         struct ext4_group_desc *gdp = NULL;
2745         ext4_group_t group, ngroups;
2746         struct super_block *sb;
2747         unsigned long timeout = 0;
2748         int ret = 0;
2749
2750         sb = elr->lr_super;
2751         ngroups = EXT4_SB(sb)->s_groups_count;
2752
2753         sb_start_write(sb);
2754         for (group = elr->lr_next_group; group < ngroups; group++) {
2755                 gdp = ext4_get_group_desc(sb, group, NULL);
2756                 if (!gdp) {
2757                         ret = 1;
2758                         break;
2759                 }
2760
2761                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2762                         break;
2763         }
2764
2765         if (group >= ngroups)
2766                 ret = 1;
2767
2768         if (!ret) {
2769                 timeout = jiffies;
2770                 ret = ext4_init_inode_table(sb, group,
2771                                             elr->lr_timeout ? 0 : 1);
2772                 if (elr->lr_timeout == 0) {
2773                         timeout = (jiffies - timeout) *
2774                                   elr->lr_sbi->s_li_wait_mult;
2775                         elr->lr_timeout = timeout;
2776                 }
2777                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2778                 elr->lr_next_group = group + 1;
2779         }
2780         sb_end_write(sb);
2781
2782         return ret;
2783 }
2784
2785 /*
2786  * Remove lr_request from the list_request and free the
2787  * request structure. Should be called with li_list_mtx held
2788  */
2789 static void ext4_remove_li_request(struct ext4_li_request *elr)
2790 {
2791         struct ext4_sb_info *sbi;
2792
2793         if (!elr)
2794                 return;
2795
2796         sbi = elr->lr_sbi;
2797
2798         list_del(&elr->lr_request);
2799         sbi->s_li_request = NULL;
2800         kfree(elr);
2801 }
2802
2803 static void ext4_unregister_li_request(struct super_block *sb)
2804 {
2805         mutex_lock(&ext4_li_mtx);
2806         if (!ext4_li_info) {
2807                 mutex_unlock(&ext4_li_mtx);
2808                 return;
2809         }
2810
2811         mutex_lock(&ext4_li_info->li_list_mtx);
2812         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2813         mutex_unlock(&ext4_li_info->li_list_mtx);
2814         mutex_unlock(&ext4_li_mtx);
2815 }
2816
2817 static struct task_struct *ext4_lazyinit_task;
2818
2819 /*
2820  * This is the function where ext4lazyinit thread lives. It walks
2821  * through the request list searching for next scheduled filesystem.
2822  * When such a fs is found, run the lazy initialization request
2823  * (ext4_rn_li_request) and keep track of the time spend in this
2824  * function. Based on that time we compute next schedule time of
2825  * the request. When walking through the list is complete, compute
2826  * next waking time and put itself into sleep.
2827  */
2828 static int ext4_lazyinit_thread(void *arg)
2829 {
2830         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2831         struct list_head *pos, *n;
2832         struct ext4_li_request *elr;
2833         unsigned long next_wakeup, cur;
2834
2835         BUG_ON(NULL == eli);
2836
2837 cont_thread:
2838         while (true) {
2839                 next_wakeup = MAX_JIFFY_OFFSET;
2840
2841                 mutex_lock(&eli->li_list_mtx);
2842                 if (list_empty(&eli->li_request_list)) {
2843                         mutex_unlock(&eli->li_list_mtx);
2844                         goto exit_thread;
2845                 }
2846
2847                 list_for_each_safe(pos, n, &eli->li_request_list) {
2848                         elr = list_entry(pos, struct ext4_li_request,
2849                                          lr_request);
2850
2851                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2852                                 if (ext4_run_li_request(elr) != 0) {
2853                                         /* error, remove the lazy_init job */
2854                                         ext4_remove_li_request(elr);
2855                                         continue;
2856                                 }
2857                         }
2858
2859                         if (time_before(elr->lr_next_sched, next_wakeup))
2860                                 next_wakeup = elr->lr_next_sched;
2861                 }
2862                 mutex_unlock(&eli->li_list_mtx);
2863
2864                 try_to_freeze();
2865
2866                 cur = jiffies;
2867                 if ((time_after_eq(cur, next_wakeup)) ||
2868                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2869                         cond_resched();
2870                         continue;
2871                 }
2872
2873                 schedule_timeout_interruptible(next_wakeup - cur);
2874
2875                 if (kthread_should_stop()) {
2876                         ext4_clear_request_list();
2877                         goto exit_thread;
2878                 }
2879         }
2880
2881 exit_thread:
2882         /*
2883          * It looks like the request list is empty, but we need
2884          * to check it under the li_list_mtx lock, to prevent any
2885          * additions into it, and of course we should lock ext4_li_mtx
2886          * to atomically free the list and ext4_li_info, because at
2887          * this point another ext4 filesystem could be registering
2888          * new one.
2889          */
2890         mutex_lock(&ext4_li_mtx);
2891         mutex_lock(&eli->li_list_mtx);
2892         if (!list_empty(&eli->li_request_list)) {
2893                 mutex_unlock(&eli->li_list_mtx);
2894                 mutex_unlock(&ext4_li_mtx);
2895                 goto cont_thread;
2896         }
2897         mutex_unlock(&eli->li_list_mtx);
2898         kfree(ext4_li_info);
2899         ext4_li_info = NULL;
2900         mutex_unlock(&ext4_li_mtx);
2901
2902         return 0;
2903 }
2904
2905 static void ext4_clear_request_list(void)
2906 {
2907         struct list_head *pos, *n;
2908         struct ext4_li_request *elr;
2909
2910         mutex_lock(&ext4_li_info->li_list_mtx);
2911         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2912                 elr = list_entry(pos, struct ext4_li_request,
2913                                  lr_request);
2914                 ext4_remove_li_request(elr);
2915         }
2916         mutex_unlock(&ext4_li_info->li_list_mtx);
2917 }
2918
2919 static int ext4_run_lazyinit_thread(void)
2920 {
2921         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2922                                          ext4_li_info, "ext4lazyinit");
2923         if (IS_ERR(ext4_lazyinit_task)) {
2924                 int err = PTR_ERR(ext4_lazyinit_task);
2925                 ext4_clear_request_list();
2926                 kfree(ext4_li_info);
2927                 ext4_li_info = NULL;
2928                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2929                                  "initialization thread\n",
2930                                  err);
2931                 return err;
2932         }
2933         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2934         return 0;
2935 }
2936
2937 /*
2938  * Check whether it make sense to run itable init. thread or not.
2939  * If there is at least one uninitialized inode table, return
2940  * corresponding group number, else the loop goes through all
2941  * groups and return total number of groups.
2942  */
2943 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2944 {
2945         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2946         struct ext4_group_desc *gdp = NULL;
2947
2948         for (group = 0; group < ngroups; group++) {
2949                 gdp = ext4_get_group_desc(sb, group, NULL);
2950                 if (!gdp)
2951                         continue;
2952
2953                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2954                         break;
2955         }
2956
2957         return group;
2958 }
2959
2960 static int ext4_li_info_new(void)
2961 {
2962         struct ext4_lazy_init *eli = NULL;
2963
2964         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2965         if (!eli)
2966                 return -ENOMEM;
2967
2968         INIT_LIST_HEAD(&eli->li_request_list);
2969         mutex_init(&eli->li_list_mtx);
2970
2971         eli->li_state |= EXT4_LAZYINIT_QUIT;
2972
2973         ext4_li_info = eli;
2974
2975         return 0;
2976 }
2977
2978 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2979                                             ext4_group_t start)
2980 {
2981         struct ext4_sb_info *sbi = EXT4_SB(sb);
2982         struct ext4_li_request *elr;
2983         unsigned long rnd;
2984
2985         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2986         if (!elr)
2987                 return NULL;
2988
2989         elr->lr_super = sb;
2990         elr->lr_sbi = sbi;
2991         elr->lr_next_group = start;
2992
2993         /*
2994          * Randomize first schedule time of the request to
2995          * spread the inode table initialization requests
2996          * better.
2997          */
2998         get_random_bytes(&rnd, sizeof(rnd));
2999         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3000                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3001
3002         return elr;
3003 }
3004
3005 int ext4_register_li_request(struct super_block *sb,
3006                              ext4_group_t first_not_zeroed)
3007 {
3008         struct ext4_sb_info *sbi = EXT4_SB(sb);
3009         struct ext4_li_request *elr = NULL;
3010         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3011         int ret = 0;
3012
3013         mutex_lock(&ext4_li_mtx);
3014         if (sbi->s_li_request != NULL) {
3015                 /*
3016                  * Reset timeout so it can be computed again, because
3017                  * s_li_wait_mult might have changed.
3018                  */
3019                 sbi->s_li_request->lr_timeout = 0;
3020                 goto out;
3021         }
3022
3023         if (first_not_zeroed == ngroups ||
3024             (sb->s_flags & MS_RDONLY) ||
3025             !test_opt(sb, INIT_INODE_TABLE))
3026                 goto out;
3027
3028         elr = ext4_li_request_new(sb, first_not_zeroed);
3029         if (!elr) {
3030                 ret = -ENOMEM;
3031                 goto out;
3032         }
3033
3034         if (NULL == ext4_li_info) {
3035                 ret = ext4_li_info_new();
3036                 if (ret)
3037                         goto out;
3038         }
3039
3040         mutex_lock(&ext4_li_info->li_list_mtx);
3041         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3042         mutex_unlock(&ext4_li_info->li_list_mtx);
3043
3044         sbi->s_li_request = elr;
3045         /*
3046          * set elr to NULL here since it has been inserted to
3047          * the request_list and the removal and free of it is
3048          * handled by ext4_clear_request_list from now on.
3049          */
3050         elr = NULL;
3051
3052         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3053                 ret = ext4_run_lazyinit_thread();
3054                 if (ret)
3055                         goto out;
3056         }
3057 out:
3058         mutex_unlock(&ext4_li_mtx);
3059         if (ret)
3060                 kfree(elr);
3061         return ret;
3062 }
3063
3064 /*
3065  * We do not need to lock anything since this is called on
3066  * module unload.
3067  */
3068 static void ext4_destroy_lazyinit_thread(void)
3069 {
3070         /*
3071          * If thread exited earlier
3072          * there's nothing to be done.
3073          */
3074         if (!ext4_li_info || !ext4_lazyinit_task)
3075                 return;
3076
3077         kthread_stop(ext4_lazyinit_task);
3078 }
3079
3080 static int set_journal_csum_feature_set(struct super_block *sb)
3081 {
3082         int ret = 1;
3083         int compat, incompat;
3084         struct ext4_sb_info *sbi = EXT4_SB(sb);
3085
3086         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3087                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3088                 /* journal checksum v2 */
3089                 compat = 0;
3090                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3091         } else {
3092                 /* journal checksum v1 */
3093                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3094                 incompat = 0;
3095         }
3096
3097         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3098                 ret = jbd2_journal_set_features(sbi->s_journal,
3099                                 compat, 0,
3100                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3101                                 incompat);
3102         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3103                 ret = jbd2_journal_set_features(sbi->s_journal,
3104                                 compat, 0,
3105                                 incompat);
3106                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3107                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3108         } else {
3109                 jbd2_journal_clear_features(sbi->s_journal,
3110                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3111                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3112                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3113         }
3114
3115         return ret;
3116 }
3117
3118 /*
3119  * Note: calculating the overhead so we can be compatible with
3120  * historical BSD practice is quite difficult in the face of
3121  * clusters/bigalloc.  This is because multiple metadata blocks from
3122  * different block group can end up in the same allocation cluster.
3123  * Calculating the exact overhead in the face of clustered allocation
3124  * requires either O(all block bitmaps) in memory or O(number of block
3125  * groups**2) in time.  We will still calculate the superblock for
3126  * older file systems --- and if we come across with a bigalloc file
3127  * system with zero in s_overhead_clusters the estimate will be close to
3128  * correct especially for very large cluster sizes --- but for newer
3129  * file systems, it's better to calculate this figure once at mkfs
3130  * time, and store it in the superblock.  If the superblock value is
3131  * present (even for non-bigalloc file systems), we will use it.
3132  */
3133 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3134                           char *buf)
3135 {
3136         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3137         struct ext4_group_desc  *gdp;
3138         ext4_fsblk_t            first_block, last_block, b;
3139         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3140         int                     s, j, count = 0;
3141
3142         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3143                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3144                         sbi->s_itb_per_group + 2);
3145
3146         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3147                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3148         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3149         for (i = 0; i < ngroups; i++) {
3150                 gdp = ext4_get_group_desc(sb, i, NULL);
3151                 b = ext4_block_bitmap(sb, gdp);
3152                 if (b >= first_block && b <= last_block) {
3153                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3154                         count++;
3155                 }
3156                 b = ext4_inode_bitmap(sb, gdp);
3157                 if (b >= first_block && b <= last_block) {
3158                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3159                         count++;
3160                 }
3161                 b = ext4_inode_table(sb, gdp);
3162                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3163                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3164                                 int c = EXT4_B2C(sbi, b - first_block);
3165                                 ext4_set_bit(c, buf);
3166                                 count++;
3167                         }
3168                 if (i != grp)
3169                         continue;
3170                 s = 0;
3171                 if (ext4_bg_has_super(sb, grp)) {
3172                         ext4_set_bit(s++, buf);
3173                         count++;
3174                 }
3175                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3176                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3177                         count++;
3178                 }
3179         }
3180         if (!count)
3181                 return 0;
3182         return EXT4_CLUSTERS_PER_GROUP(sb) -
3183                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3184 }
3185
3186 /*
3187  * Compute the overhead and stash it in sbi->s_overhead
3188  */
3189 int ext4_calculate_overhead(struct super_block *sb)
3190 {
3191         struct ext4_sb_info *sbi = EXT4_SB(sb);
3192         struct ext4_super_block *es = sbi->s_es;
3193         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3194         ext4_fsblk_t overhead = 0;
3195         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3196
3197         if (!buf)
3198                 return -ENOMEM;
3199
3200         /*
3201          * Compute the overhead (FS structures).  This is constant
3202          * for a given filesystem unless the number of block groups
3203          * changes so we cache the previous value until it does.
3204          */
3205
3206         /*
3207          * All of the blocks before first_data_block are overhead
3208          */
3209         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3210
3211         /*
3212          * Add the overhead found in each block group
3213          */
3214         for (i = 0; i < ngroups; i++) {
3215                 int blks;
3216
3217                 blks = count_overhead(sb, i, buf);
3218                 overhead += blks;
3219                 if (blks)
3220                         memset(buf, 0, PAGE_SIZE);
3221                 cond_resched();
3222         }
3223         /* Add the journal blocks as well */
3224         if (sbi->s_journal)
3225                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3226
3227         sbi->s_overhead = overhead;
3228         smp_wmb();
3229         free_page((unsigned long) buf);
3230         return 0;
3231 }
3232
3233
3234 static ext4_fsblk_t ext4_calculate_resv_clusters(struct ext4_sb_info *sbi)
3235 {
3236         ext4_fsblk_t resv_clusters;
3237
3238         /*
3239          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3240          * This should cover the situations where we can not afford to run
3241          * out of space like for example punch hole, or converting
3242          * uninitialized extents in delalloc path. In most cases such
3243          * allocation would require 1, or 2 blocks, higher numbers are
3244          * very rare.
3245          */
3246         resv_clusters = ext4_blocks_count(sbi->s_es) >> sbi->s_cluster_bits;
3247
3248         do_div(resv_clusters, 50);
3249         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3250
3251         return resv_clusters;
3252 }
3253
3254
3255 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3256 {
3257         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3258                                 sbi->s_cluster_bits;
3259
3260         if (count >= clusters)
3261                 return -EINVAL;
3262
3263         atomic64_set(&sbi->s_resv_clusters, count);
3264         return 0;
3265 }
3266
3267 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3268 {
3269         char *orig_data = kstrdup(data, GFP_KERNEL);
3270         struct buffer_head *bh;
3271         struct ext4_super_block *es = NULL;
3272         struct ext4_sb_info *sbi;
3273         ext4_fsblk_t block;
3274         ext4_fsblk_t sb_block = get_sb_block(&data);
3275         ext4_fsblk_t logical_sb_block;
3276         unsigned long offset = 0;
3277         unsigned long journal_devnum = 0;
3278         unsigned long def_mount_opts;
3279         struct inode *root;
3280         char *cp;
3281         const char *descr;
3282         int ret = -ENOMEM;
3283         int blocksize, clustersize;
3284         unsigned int db_count;
3285         unsigned int i;
3286         int needs_recovery, has_huge_files, has_bigalloc;
3287         __u64 blocks_count;
3288         int err = 0;
3289         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3290         ext4_group_t first_not_zeroed;
3291
3292         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3293         if (!sbi)
3294                 goto out_free_orig;
3295
3296         sbi->s_blockgroup_lock =
3297                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3298         if (!sbi->s_blockgroup_lock) {
3299                 kfree(sbi);
3300                 goto out_free_orig;
3301         }
3302         sb->s_fs_info = sbi;
3303         sbi->s_sb = sb;
3304         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3305         sbi->s_sb_block = sb_block;
3306         if (sb->s_bdev->bd_part)
3307                 sbi->s_sectors_written_start =
3308                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3309
3310         /* Cleanup superblock name */
3311         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3312                 *cp = '!';
3313
3314         /* -EINVAL is default */
3315         ret = -EINVAL;
3316         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3317         if (!blocksize) {
3318                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3319                 goto out_fail;
3320         }
3321
3322         /*
3323          * The ext4 superblock will not be buffer aligned for other than 1kB
3324          * block sizes.  We need to calculate the offset from buffer start.
3325          */
3326         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3327                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3328                 offset = do_div(logical_sb_block, blocksize);
3329         } else {
3330                 logical_sb_block = sb_block;
3331         }
3332
3333         if (!(bh = sb_bread(sb, logical_sb_block))) {
3334                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3335                 goto out_fail;
3336         }
3337         /*
3338          * Note: s_es must be initialized as soon as possible because
3339          *       some ext4 macro-instructions depend on its value
3340          */
3341         es = (struct ext4_super_block *) (bh->b_data + offset);
3342         sbi->s_es = es;
3343         sb->s_magic = le16_to_cpu(es->s_magic);
3344         if (sb->s_magic != EXT4_SUPER_MAGIC)
3345                 goto cantfind_ext4;
3346         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3347
3348         /* Warn if metadata_csum and gdt_csum are both set. */
3349         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3350                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3351             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3352                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3353                              "redundant flags; please run fsck.");
3354
3355         /* Check for a known checksum algorithm */
3356         if (!ext4_verify_csum_type(sb, es)) {
3357                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3358                          "unknown checksum algorithm.");
3359                 silent = 1;
3360                 goto cantfind_ext4;
3361         }
3362
3363         /* Load the checksum driver */
3364         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3365                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3366                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3367                 if (IS_ERR(sbi->s_chksum_driver)) {
3368                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3369                         ret = PTR_ERR(sbi->s_chksum_driver);
3370                         sbi->s_chksum_driver = NULL;
3371                         goto failed_mount;
3372                 }
3373         }
3374
3375         /* Check superblock checksum */
3376         if (!ext4_superblock_csum_verify(sb, es)) {
3377                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3378                          "invalid superblock checksum.  Run e2fsck?");
3379                 silent = 1;
3380                 goto cantfind_ext4;
3381         }
3382
3383         /* Precompute checksum seed for all metadata */
3384         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3385                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3386                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3387                                                sizeof(es->s_uuid));
3388
3389         /* Set defaults before we parse the mount options */
3390         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3391         set_opt(sb, INIT_INODE_TABLE);
3392         if (def_mount_opts & EXT4_DEFM_DEBUG)
3393                 set_opt(sb, DEBUG);
3394         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3395                 set_opt(sb, GRPID);
3396         if (def_mount_opts & EXT4_DEFM_UID16)
3397                 set_opt(sb, NO_UID32);
3398         /* xattr user namespace & acls are now defaulted on */
3399         set_opt(sb, XATTR_USER);
3400 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3401         set_opt(sb, POSIX_ACL);
3402 #endif
3403         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3404                 set_opt(sb, JOURNAL_DATA);
3405         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3406                 set_opt(sb, ORDERED_DATA);
3407         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3408                 set_opt(sb, WRITEBACK_DATA);
3409
3410         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3411                 set_opt(sb, ERRORS_PANIC);
3412         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3413                 set_opt(sb, ERRORS_CONT);
3414         else
3415                 set_opt(sb, ERRORS_RO);
3416         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3417                 set_opt(sb, BLOCK_VALIDITY);
3418         if (def_mount_opts & EXT4_DEFM_DISCARD)
3419                 set_opt(sb, DISCARD);
3420
3421         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3422         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3423         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3424         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3425         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3426
3427         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3428                 set_opt(sb, BARRIER);
3429
3430         /*
3431          * enable delayed allocation by default
3432          * Use -o nodelalloc to turn it off
3433          */
3434         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3435             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3436                 set_opt(sb, DELALLOC);
3437
3438         /*
3439          * set default s_li_wait_mult for lazyinit, for the case there is
3440          * no mount option specified.
3441          */
3442         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3443
3444         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3445                            &journal_devnum, &journal_ioprio, 0)) {
3446                 ext4_msg(sb, KERN_WARNING,
3447                          "failed to parse options in superblock: %s",
3448                          sbi->s_es->s_mount_opts);
3449         }
3450         sbi->s_def_mount_opt = sbi->s_mount_opt;
3451         if (!parse_options((char *) data, sb, &journal_devnum,
3452                            &journal_ioprio, 0))
3453                 goto failed_mount;
3454
3455         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3456                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3457                             "with data=journal disables delayed "
3458                             "allocation and O_DIRECT support!\n");
3459                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3460                         ext4_msg(sb, KERN_ERR, "can't mount with "
3461                                  "both data=journal and delalloc");
3462                         goto failed_mount;
3463                 }
3464                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3465                         ext4_msg(sb, KERN_ERR, "can't mount with "
3466                                  "both data=journal and delalloc");
3467                         goto failed_mount;
3468                 }
3469                 if (test_opt(sb, DELALLOC))
3470                         clear_opt(sb, DELALLOC);
3471         }
3472
3473         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3474                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3475
3476         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3477             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3478              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3479              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3480                 ext4_msg(sb, KERN_WARNING,
3481                        "feature flags set on rev 0 fs, "
3482                        "running e2fsck is recommended");
3483
3484         if (IS_EXT2_SB(sb)) {
3485                 if (ext2_feature_set_ok(sb))
3486                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3487                                  "using the ext4 subsystem");
3488                 else {
3489                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3490                                  "to feature incompatibilities");
3491                         goto failed_mount;
3492                 }
3493         }
3494
3495         if (IS_EXT3_SB(sb)) {
3496                 if (ext3_feature_set_ok(sb))
3497                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3498                                  "using the ext4 subsystem");
3499                 else {
3500                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3501                                  "to feature incompatibilities");
3502                         goto failed_mount;
3503                 }
3504         }
3505
3506         /*
3507          * Check feature flags regardless of the revision level, since we
3508          * previously didn't change the revision level when setting the flags,
3509          * so there is a chance incompat flags are set on a rev 0 filesystem.
3510          */
3511         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3512                 goto failed_mount;
3513
3514         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3515         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3516             blocksize > EXT4_MAX_BLOCK_SIZE) {
3517                 ext4_msg(sb, KERN_ERR,
3518                        "Unsupported filesystem blocksize %d", blocksize);
3519                 goto failed_mount;
3520         }
3521
3522         if (sb->s_blocksize != blocksize) {
3523                 /* Validate the filesystem blocksize */
3524                 if (!sb_set_blocksize(sb, blocksize)) {
3525                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3526                                         blocksize);
3527                         goto failed_mount;
3528                 }
3529
3530                 brelse(bh);
3531                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3532                 offset = do_div(logical_sb_block, blocksize);
3533                 bh = sb_bread(sb, logical_sb_block);
3534                 if (!bh) {
3535                         ext4_msg(sb, KERN_ERR,
3536                                "Can't read superblock on 2nd try");
3537                         goto failed_mount;
3538                 }
3539                 es = (struct ext4_super_block *)(bh->b_data + offset);
3540                 sbi->s_es = es;
3541                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3542                         ext4_msg(sb, KERN_ERR,
3543                                "Magic mismatch, very weird!");
3544                         goto failed_mount;
3545                 }
3546         }
3547
3548         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3549                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3550         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3551                                                       has_huge_files);
3552         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3553
3554         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3555                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3556                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3557         } else {
3558                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3559                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3560                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3561                     (!is_power_of_2(sbi->s_inode_size)) ||
3562                     (sbi->s_inode_size > blocksize)) {
3563                         ext4_msg(sb, KERN_ERR,
3564                                "unsupported inode size: %d",
3565                                sbi->s_inode_size);
3566                         goto failed_mount;
3567                 }
3568                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3569                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3570         }
3571
3572         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3573         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3574                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3575                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3576                     !is_power_of_2(sbi->s_desc_size)) {
3577                         ext4_msg(sb, KERN_ERR,
3578                                "unsupported descriptor size %lu",
3579                                sbi->s_desc_size);
3580                         goto failed_mount;
3581                 }
3582         } else
3583                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3584
3585         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3586         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3587         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3588                 goto cantfind_ext4;
3589
3590         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3591         if (sbi->s_inodes_per_block == 0)
3592                 goto cantfind_ext4;
3593         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3594                                         sbi->s_inodes_per_block;
3595         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3596         sbi->s_sbh = bh;
3597         sbi->s_mount_state = le16_to_cpu(es->s_state);
3598         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3599         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3600
3601         /* Do we have standard group size of blocksize * 8 blocks ? */
3602         if (sbi->s_blocks_per_group == blocksize << 3)
3603                 set_opt2(sb, STD_GROUP_SIZE);
3604
3605         for (i = 0; i < 4; i++)
3606                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3607         sbi->s_def_hash_version = es->s_def_hash_version;
3608         i = le32_to_cpu(es->s_flags);
3609         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3610                 sbi->s_hash_unsigned = 3;
3611         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3612 #ifdef __CHAR_UNSIGNED__
3613                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3614                 sbi->s_hash_unsigned = 3;
3615 #else
3616                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3617 #endif
3618         }
3619
3620         /* Handle clustersize */
3621         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3622         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3623                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3624         if (has_bigalloc) {
3625                 if (clustersize < blocksize) {
3626                         ext4_msg(sb, KERN_ERR,
3627                                  "cluster size (%d) smaller than "
3628                                  "block size (%d)", clustersize, blocksize);
3629                         goto failed_mount;
3630                 }
3631                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3632                         le32_to_cpu(es->s_log_block_size);
3633                 sbi->s_clusters_per_group =
3634                         le32_to_cpu(es->s_clusters_per_group);
3635                 if (sbi->s_clusters_per_group > blocksize * 8) {
3636                         ext4_msg(sb, KERN_ERR,
3637                                  "#clusters per group too big: %lu",
3638                                  sbi->s_clusters_per_group);
3639                         goto failed_mount;
3640                 }
3641                 if (sbi->s_blocks_per_group !=
3642                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3643                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3644                                  "clusters per group (%lu) inconsistent",
3645                                  sbi->s_blocks_per_group,
3646                                  sbi->s_clusters_per_group);
3647                         goto failed_mount;
3648                 }
3649         } else {
3650                 if (clustersize != blocksize) {
3651                         ext4_warning(sb, "fragment/cluster size (%d) != "
3652                                      "block size (%d)", clustersize,
3653                                      blocksize);
3654                         clustersize = blocksize;
3655                 }
3656                 if (sbi->s_blocks_per_group > blocksize * 8) {
3657                         ext4_msg(sb, KERN_ERR,
3658                                  "#blocks per group too big: %lu",
3659                                  sbi->s_blocks_per_group);
3660                         goto failed_mount;
3661                 }
3662                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3663                 sbi->s_cluster_bits = 0;
3664         }
3665         sbi->s_cluster_ratio = clustersize / blocksize;
3666
3667         if (sbi->s_inodes_per_group > blocksize * 8) {
3668                 ext4_msg(sb, KERN_ERR,
3669                        "#inodes per group too big: %lu",
3670                        sbi->s_inodes_per_group);
3671                 goto failed_mount;
3672         }
3673
3674         /*
3675          * Test whether we have more sectors than will fit in sector_t,
3676          * and whether the max offset is addressable by the page cache.
3677          */
3678         err = generic_check_addressable(sb->s_blocksize_bits,
3679                                         ext4_blocks_count(es));
3680         if (err) {
3681                 ext4_msg(sb, KERN_ERR, "filesystem"
3682                          " too large to mount safely on this system");
3683                 if (sizeof(sector_t) < 8)
3684                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3685                 goto failed_mount;
3686         }
3687
3688         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3689                 goto cantfind_ext4;
3690
3691         /* check blocks count against device size */
3692         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3693         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3694                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3695                        "exceeds size of device (%llu blocks)",
3696                        ext4_blocks_count(es), blocks_count);
3697                 goto failed_mount;
3698         }
3699
3700         /*
3701          * It makes no sense for the first data block to be beyond the end
3702          * of the filesystem.
3703          */
3704         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3705                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3706                          "block %u is beyond end of filesystem (%llu)",
3707                          le32_to_cpu(es->s_first_data_block),
3708                          ext4_blocks_count(es));
3709                 goto failed_mount;
3710         }
3711         blocks_count = (ext4_blocks_count(es) -
3712                         le32_to_cpu(es->s_first_data_block) +
3713                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3714         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3715         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3716                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3717                        "(block count %llu, first data block %u, "
3718                        "blocks per group %lu)", sbi->s_groups_count,
3719                        ext4_blocks_count(es),
3720                        le32_to_cpu(es->s_first_data_block),
3721                        EXT4_BLOCKS_PER_GROUP(sb));
3722                 goto failed_mount;
3723         }
3724         sbi->s_groups_count = blocks_count;
3725         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3726                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3727         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3728                    EXT4_DESC_PER_BLOCK(sb);
3729         sbi->s_group_desc = ext4_kvmalloc(db_count *
3730                                           sizeof(struct buffer_head *),
3731                                           GFP_KERNEL);
3732         if (sbi->s_group_desc == NULL) {
3733                 ext4_msg(sb, KERN_ERR, "not enough memory");
3734                 ret = -ENOMEM;
3735                 goto failed_mount;
3736         }
3737
3738         if (ext4_proc_root)
3739                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3740
3741         if (sbi->s_proc)
3742                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3743                                  &ext4_seq_options_fops, sb);
3744
3745         bgl_lock_init(sbi->s_blockgroup_lock);
3746
3747         for (i = 0; i < db_count; i++) {
3748                 block = descriptor_loc(sb, logical_sb_block, i);
3749                 sbi->s_group_desc[i] = sb_bread(sb, block);
3750                 if (!sbi->s_group_desc[i]) {
3751                         ext4_msg(sb, KERN_ERR,
3752                                "can't read group descriptor %d", i);
3753                         db_count = i;
3754                         goto failed_mount2;
3755                 }
3756         }
3757         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3758                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3759                 goto failed_mount2;
3760         }
3761         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3762                 if (!ext4_fill_flex_info(sb)) {
3763                         ext4_msg(sb, KERN_ERR,
3764                                "unable to initialize "
3765                                "flex_bg meta info!");
3766                         goto failed_mount2;
3767                 }
3768
3769         sbi->s_gdb_count = db_count;
3770         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3771         spin_lock_init(&sbi->s_next_gen_lock);
3772
3773         init_timer(&sbi->s_err_report);
3774         sbi->s_err_report.function = print_daily_error_info;
3775         sbi->s_err_report.data = (unsigned long) sb;
3776
3777         /* Register extent status tree shrinker */
3778         ext4_es_register_shrinker(sb);
3779
3780         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3781                         ext4_count_free_clusters(sb));
3782         if (!err) {
3783                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3784                                 ext4_count_free_inodes(sb));
3785         }
3786         if (!err) {
3787                 err = percpu_counter_init(&sbi->s_dirs_counter,
3788                                 ext4_count_dirs(sb));
3789         }
3790         if (!err) {
3791                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3792         }
3793         if (!err) {
3794                 err = percpu_counter_init(&sbi->s_extent_cache_cnt, 0);
3795         }
3796         if (err) {
3797                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3798                 goto failed_mount3;
3799         }
3800
3801         sbi->s_stripe = ext4_get_stripe_size(sbi);
3802         sbi->s_max_writeback_mb_bump = 128;
3803         sbi->s_extent_max_zeroout_kb = 32;
3804
3805         /*
3806          * set up enough so that it can read an inode
3807          */
3808         if (!test_opt(sb, NOLOAD) &&
3809             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3810                 sb->s_op = &ext4_sops;
3811         else
3812                 sb->s_op = &ext4_nojournal_sops;
3813         sb->s_export_op = &ext4_export_ops;
3814         sb->s_xattr = ext4_xattr_handlers;
3815 #ifdef CONFIG_QUOTA
3816         sb->dq_op = &ext4_quota_operations;
3817         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3818                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3819         else
3820                 sb->s_qcop = &ext4_qctl_operations;
3821 #endif
3822         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3823
3824         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3825         mutex_init(&sbi->s_orphan_lock);
3826
3827         sb->s_root = NULL;
3828
3829         needs_recovery = (es->s_last_orphan != 0 ||
3830                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3831                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3832
3833         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3834             !(sb->s_flags & MS_RDONLY))
3835                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3836                         goto failed_mount3;
3837
3838         /*
3839          * The first inode we look at is the journal inode.  Don't try
3840          * root first: it may be modified in the journal!
3841          */
3842         if (!test_opt(sb, NOLOAD) &&
3843             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3844                 if (ext4_load_journal(sb, es, journal_devnum))
3845                         goto failed_mount3;
3846         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3847               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3848                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3849                        "suppressed and not mounted read-only");
3850                 goto failed_mount_wq;
3851         } else {
3852                 clear_opt(sb, DATA_FLAGS);
3853                 sbi->s_journal = NULL;
3854                 needs_recovery = 0;
3855                 goto no_journal;
3856         }
3857
3858         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3859             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3860                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3861                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3862                 goto failed_mount_wq;
3863         }
3864
3865         if (!set_journal_csum_feature_set(sb)) {
3866                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3867                          "feature set");
3868                 goto failed_mount_wq;
3869         }
3870
3871         /* We have now updated the journal if required, so we can
3872          * validate the data journaling mode. */
3873         switch (test_opt(sb, DATA_FLAGS)) {
3874         case 0:
3875                 /* No mode set, assume a default based on the journal
3876                  * capabilities: ORDERED_DATA if the journal can
3877                  * cope, else JOURNAL_DATA
3878                  */
3879                 if (jbd2_journal_check_available_features
3880                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3881                         set_opt(sb, ORDERED_DATA);
3882                 else
3883                         set_opt(sb, JOURNAL_DATA);
3884                 break;
3885
3886         case EXT4_MOUNT_ORDERED_DATA:
3887         case EXT4_MOUNT_WRITEBACK_DATA:
3888                 if (!jbd2_journal_check_available_features
3889                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3890                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3891                                "requested data journaling mode");
3892                         goto failed_mount_wq;
3893                 }
3894         default:
3895                 break;
3896         }
3897         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3898
3899         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3900
3901         /*
3902          * The journal may have updated the bg summary counts, so we
3903          * need to update the global counters.
3904          */
3905         percpu_counter_set(&sbi->s_freeclusters_counter,
3906                            ext4_count_free_clusters(sb));
3907         percpu_counter_set(&sbi->s_freeinodes_counter,
3908                            ext4_count_free_inodes(sb));
3909         percpu_counter_set(&sbi->s_dirs_counter,
3910                            ext4_count_dirs(sb));
3911         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3912
3913 no_journal:
3914         /*
3915          * Get the # of file system overhead blocks from the
3916          * superblock if present.
3917          */
3918         if (es->s_overhead_clusters)
3919                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3920         else {
3921                 err = ext4_calculate_overhead(sb);
3922                 if (err)
3923                         goto failed_mount_wq;
3924         }
3925
3926         /*
3927          * The maximum number of concurrent works can be high and
3928          * concurrency isn't really necessary.  Limit it to 1.
3929          */
3930         EXT4_SB(sb)->dio_unwritten_wq =
3931                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3932         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3933                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3934                 ret = -ENOMEM;
3935                 goto failed_mount_wq;
3936         }
3937
3938         /*
3939          * The jbd2_journal_load will have done any necessary log recovery,
3940          * so we can safely mount the rest of the filesystem now.
3941          */
3942
3943         root = ext4_iget(sb, EXT4_ROOT_INO);
3944         if (IS_ERR(root)) {
3945                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3946                 ret = PTR_ERR(root);
3947                 root = NULL;
3948                 goto failed_mount4;
3949         }
3950         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3951                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3952                 iput(root);
3953                 goto failed_mount4;
3954         }
3955         sb->s_root = d_make_root(root);
3956         if (!sb->s_root) {
3957                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3958                 ret = -ENOMEM;
3959                 goto failed_mount4;
3960         }
3961
3962         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3963                 sb->s_flags |= MS_RDONLY;
3964
3965         /* determine the minimum size of new large inodes, if present */
3966         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3967                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3968                                                      EXT4_GOOD_OLD_INODE_SIZE;
3969                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3970                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3971                         if (sbi->s_want_extra_isize <
3972                             le16_to_cpu(es->s_want_extra_isize))
3973                                 sbi->s_want_extra_isize =
3974                                         le16_to_cpu(es->s_want_extra_isize);
3975                         if (sbi->s_want_extra_isize <
3976                             le16_to_cpu(es->s_min_extra_isize))
3977                                 sbi->s_want_extra_isize =
3978                                         le16_to_cpu(es->s_min_extra_isize);
3979                 }
3980         }
3981         /* Check if enough inode space is available */
3982         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3983                                                         sbi->s_inode_size) {
3984                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3985                                                        EXT4_GOOD_OLD_INODE_SIZE;
3986                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3987                          "available");
3988         }
3989
3990         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sbi));
3991         if (err) {
3992                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
3993                          "reserved pool", ext4_calculate_resv_clusters(sbi));
3994                 goto failed_mount4a;
3995         }
3996
3997         err = ext4_setup_system_zone(sb);
3998         if (err) {
3999                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4000                          "zone (%d)", err);
4001                 goto failed_mount4a;
4002         }
4003
4004         ext4_ext_init(sb);
4005         err = ext4_mb_init(sb);
4006         if (err) {
4007                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4008                          err);
4009                 goto failed_mount5;
4010         }
4011
4012         err = ext4_register_li_request(sb, first_not_zeroed);
4013         if (err)
4014                 goto failed_mount6;
4015
4016         sbi->s_kobj.kset = ext4_kset;
4017         init_completion(&sbi->s_kobj_unregister);
4018         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4019                                    "%s", sb->s_id);
4020         if (err)
4021                 goto failed_mount7;
4022
4023 #ifdef CONFIG_QUOTA
4024         /* Enable quota usage during mount. */
4025         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4026             !(sb->s_flags & MS_RDONLY)) {
4027                 err = ext4_enable_quotas(sb);
4028                 if (err)
4029                         goto failed_mount8;
4030         }
4031 #endif  /* CONFIG_QUOTA */
4032
4033         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4034         ext4_orphan_cleanup(sb, es);
4035         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4036         if (needs_recovery) {
4037                 ext4_msg(sb, KERN_INFO, "recovery complete");
4038                 ext4_mark_recovery_complete(sb, es);
4039         }
4040         if (EXT4_SB(sb)->s_journal) {
4041                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4042                         descr = " journalled data mode";
4043                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4044                         descr = " ordered data mode";
4045                 else
4046                         descr = " writeback data mode";
4047         } else
4048                 descr = "out journal";
4049
4050         if (test_opt(sb, DISCARD)) {
4051                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4052                 if (!blk_queue_discard(q))
4053                         ext4_msg(sb, KERN_WARNING,
4054                                  "mounting with \"discard\" option, but "
4055                                  "the device does not support discard");
4056         }
4057
4058         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4059                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4060                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4061
4062         if (es->s_error_count)
4063                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4064
4065         kfree(orig_data);
4066         return 0;
4067
4068 cantfind_ext4:
4069         if (!silent)
4070                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4071         goto failed_mount;
4072
4073 #ifdef CONFIG_QUOTA
4074 failed_mount8:
4075         kobject_del(&sbi->s_kobj);
4076 #endif
4077 failed_mount7:
4078         ext4_unregister_li_request(sb);
4079 failed_mount6:
4080         ext4_mb_release(sb);
4081 failed_mount5:
4082         ext4_ext_release(sb);
4083         ext4_release_system_zone(sb);
4084 failed_mount4a:
4085         dput(sb->s_root);
4086         sb->s_root = NULL;
4087 failed_mount4:
4088         ext4_msg(sb, KERN_ERR, "mount failed");
4089         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4090 failed_mount_wq:
4091         if (sbi->s_journal) {
4092                 jbd2_journal_destroy(sbi->s_journal);
4093                 sbi->s_journal = NULL;
4094         }
4095 failed_mount3:
4096         ext4_es_unregister_shrinker(sb);
4097         del_timer(&sbi->s_err_report);
4098         if (sbi->s_flex_groups)
4099                 ext4_kvfree(sbi->s_flex_groups);
4100         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4101         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4102         percpu_counter_destroy(&sbi->s_dirs_counter);
4103         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4104         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
4105         if (sbi->s_mmp_tsk)
4106                 kthread_stop(sbi->s_mmp_tsk);
4107 failed_mount2:
4108         for (i = 0; i < db_count; i++)
4109                 brelse(sbi->s_group_desc[i]);
4110         ext4_kvfree(sbi->s_group_desc);
4111 failed_mount:
4112         if (sbi->s_chksum_driver)
4113                 crypto_free_shash(sbi->s_chksum_driver);
4114         if (sbi->s_proc) {
4115                 remove_proc_entry("options", sbi->s_proc);
4116                 remove_proc_entry(sb->s_id, ext4_proc_root);
4117         }
4118 #ifdef CONFIG_QUOTA
4119         for (i = 0; i < MAXQUOTAS; i++)
4120                 kfree(sbi->s_qf_names[i]);
4121 #endif
4122         ext4_blkdev_remove(sbi);
4123         brelse(bh);
4124 out_fail:
4125         sb->s_fs_info = NULL;
4126         kfree(sbi->s_blockgroup_lock);
4127         kfree(sbi);
4128 out_free_orig:
4129         kfree(orig_data);
4130         return err ? err : ret;
4131 }
4132
4133 /*
4134  * Setup any per-fs journal parameters now.  We'll do this both on
4135  * initial mount, once the journal has been initialised but before we've
4136  * done any recovery; and again on any subsequent remount.
4137  */
4138 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4139 {
4140         struct ext4_sb_info *sbi = EXT4_SB(sb);
4141
4142         journal->j_commit_interval = sbi->s_commit_interval;
4143         journal->j_min_batch_time = sbi->s_min_batch_time;
4144         journal->j_max_batch_time = sbi->s_max_batch_time;
4145
4146         write_lock(&journal->j_state_lock);
4147         if (test_opt(sb, BARRIER))
4148                 journal->j_flags |= JBD2_BARRIER;
4149         else
4150                 journal->j_flags &= ~JBD2_BARRIER;
4151         if (test_opt(sb, DATA_ERR_ABORT))
4152                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4153         else
4154                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4155         write_unlock(&journal->j_state_lock);
4156 }
4157
4158 static journal_t *ext4_get_journal(struct super_block *sb,
4159                                    unsigned int journal_inum)
4160 {
4161         struct inode *journal_inode;
4162         journal_t *journal;
4163
4164         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4165
4166         /* First, test for the existence of a valid inode on disk.  Bad
4167          * things happen if we iget() an unused inode, as the subsequent
4168          * iput() will try to delete it. */
4169
4170         journal_inode = ext4_iget(sb, journal_inum);
4171         if (IS_ERR(journal_inode)) {
4172                 ext4_msg(sb, KERN_ERR, "no journal found");
4173                 return NULL;
4174         }
4175         if (!journal_inode->i_nlink) {
4176                 make_bad_inode(journal_inode);
4177                 iput(journal_inode);
4178                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4179                 return NULL;
4180         }
4181
4182         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4183                   journal_inode, journal_inode->i_size);
4184         if (!S_ISREG(journal_inode->i_mode)) {
4185                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4186                 iput(journal_inode);
4187                 return NULL;
4188         }
4189
4190         journal = jbd2_journal_init_inode(journal_inode);
4191         if (!journal) {
4192                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4193                 iput(journal_inode);
4194                 return NULL;
4195         }
4196         journal->j_private = sb;
4197         ext4_init_journal_params(sb, journal);
4198         return journal;
4199 }
4200
4201 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4202                                        dev_t j_dev)
4203 {
4204         struct buffer_head *bh;
4205         journal_t *journal;
4206         ext4_fsblk_t start;
4207         ext4_fsblk_t len;
4208         int hblock, blocksize;
4209         ext4_fsblk_t sb_block;
4210         unsigned long offset;
4211         struct ext4_super_block *es;
4212         struct block_device *bdev;
4213
4214         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4215
4216         bdev = ext4_blkdev_get(j_dev, sb);
4217         if (bdev == NULL)
4218                 return NULL;
4219
4220         blocksize = sb->s_blocksize;
4221         hblock = bdev_logical_block_size(bdev);
4222         if (blocksize < hblock) {
4223                 ext4_msg(sb, KERN_ERR,
4224                         "blocksize too small for journal device");
4225                 goto out_bdev;
4226         }
4227
4228         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4229         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4230         set_blocksize(bdev, blocksize);
4231         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4232                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4233                        "external journal");
4234                 goto out_bdev;
4235         }
4236
4237         es = (struct ext4_super_block *) (bh->b_data + offset);
4238         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4239             !(le32_to_cpu(es->s_feature_incompat) &
4240               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4241                 ext4_msg(sb, KERN_ERR, "external journal has "
4242                                         "bad superblock");
4243                 brelse(bh);
4244                 goto out_bdev;
4245         }
4246
4247         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4248                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4249                 brelse(bh);
4250                 goto out_bdev;
4251         }
4252
4253         len = ext4_blocks_count(es);
4254         start = sb_block + 1;
4255         brelse(bh);     /* we're done with the superblock */
4256
4257         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4258                                         start, len, blocksize);
4259         if (!journal) {
4260                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4261                 goto out_bdev;
4262         }
4263         journal->j_private = sb;
4264         ll_rw_block(READ, 1, &journal->j_sb_buffer);
4265         wait_on_buffer(journal->j_sb_buffer);
4266         if (!buffer_uptodate(journal->j_sb_buffer)) {
4267                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4268                 goto out_journal;
4269         }
4270         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4271                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4272                                         "user (unsupported) - %d",
4273                         be32_to_cpu(journal->j_superblock->s_nr_users));
4274                 goto out_journal;
4275         }
4276         EXT4_SB(sb)->journal_bdev = bdev;
4277         ext4_init_journal_params(sb, journal);
4278         return journal;
4279
4280 out_journal:
4281         jbd2_journal_destroy(journal);
4282 out_bdev:
4283         ext4_blkdev_put(bdev);
4284         return NULL;
4285 }
4286
4287 static int ext4_load_journal(struct super_block *sb,
4288                              struct ext4_super_block *es,
4289                              unsigned long journal_devnum)
4290 {
4291         journal_t *journal;
4292         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4293         dev_t journal_dev;
4294         int err = 0;
4295         int really_read_only;
4296
4297         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4298
4299         if (journal_devnum &&
4300             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4301                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4302                         "numbers have changed");
4303                 journal_dev = new_decode_dev(journal_devnum);
4304         } else
4305                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4306
4307         really_read_only = bdev_read_only(sb->s_bdev);
4308
4309         /*
4310          * Are we loading a blank journal or performing recovery after a
4311          * crash?  For recovery, we need to check in advance whether we
4312          * can get read-write access to the device.
4313          */
4314         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4315                 if (sb->s_flags & MS_RDONLY) {
4316                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4317                                         "required on readonly filesystem");
4318                         if (really_read_only) {
4319                                 ext4_msg(sb, KERN_ERR, "write access "
4320                                         "unavailable, cannot proceed");
4321                                 return -EROFS;
4322                         }
4323                         ext4_msg(sb, KERN_INFO, "write access will "
4324                                "be enabled during recovery");
4325                 }
4326         }
4327
4328         if (journal_inum && journal_dev) {
4329                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4330                        "and inode journals!");
4331                 return -EINVAL;
4332         }
4333
4334         if (journal_inum) {
4335                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4336                         return -EINVAL;
4337         } else {
4338                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4339                         return -EINVAL;
4340         }
4341
4342         if (!(journal->j_flags & JBD2_BARRIER))
4343                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4344
4345         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4346                 err = jbd2_journal_wipe(journal, !really_read_only);
4347         if (!err) {
4348                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4349                 if (save)
4350                         memcpy(save, ((char *) es) +
4351                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4352                 err = jbd2_journal_load(journal);
4353                 if (save)
4354                         memcpy(((char *) es) + EXT4_S_ERR_START,
4355                                save, EXT4_S_ERR_LEN);
4356                 kfree(save);
4357         }
4358
4359         if (err) {
4360                 ext4_msg(sb, KERN_ERR, "error loading journal");
4361                 jbd2_journal_destroy(journal);
4362                 return err;
4363         }
4364
4365         EXT4_SB(sb)->s_journal = journal;
4366         ext4_clear_journal_err(sb, es);
4367
4368         if (!really_read_only && journal_devnum &&
4369             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4370                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4371
4372                 /* Make sure we flush the recovery flag to disk. */
4373                 ext4_commit_super(sb, 1);
4374         }
4375
4376         return 0;
4377 }
4378
4379 static int ext4_commit_super(struct super_block *sb, int sync)
4380 {
4381         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4382         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4383         int error = 0;
4384
4385         if (!sbh || block_device_ejected(sb))
4386                 return error;
4387         if (buffer_write_io_error(sbh)) {
4388                 /*
4389                  * Oh, dear.  A previous attempt to write the
4390                  * superblock failed.  This could happen because the
4391                  * USB device was yanked out.  Or it could happen to
4392                  * be a transient write error and maybe the block will
4393                  * be remapped.  Nothing we can do but to retry the
4394                  * write and hope for the best.
4395                  */
4396                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4397                        "superblock detected");
4398                 clear_buffer_write_io_error(sbh);
4399                 set_buffer_uptodate(sbh);
4400         }
4401         /*
4402          * If the file system is mounted read-only, don't update the
4403          * superblock write time.  This avoids updating the superblock
4404          * write time when we are mounting the root file system
4405          * read/only but we need to replay the journal; at that point,
4406          * for people who are east of GMT and who make their clock
4407          * tick in localtime for Windows bug-for-bug compatibility,
4408          * the clock is set in the future, and this will cause e2fsck
4409          * to complain and force a full file system check.
4410          */
4411         if (!(sb->s_flags & MS_RDONLY))
4412                 es->s_wtime = cpu_to_le32(get_seconds());
4413         if (sb->s_bdev->bd_part)
4414                 es->s_kbytes_written =
4415                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4416                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4417                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4418         else
4419                 es->s_kbytes_written =
4420                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4421         ext4_free_blocks_count_set(es,
4422                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4423                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4424         es->s_free_inodes_count =
4425                 cpu_to_le32(percpu_counter_sum_positive(
4426                                 &EXT4_SB(sb)->s_freeinodes_counter));
4427         BUFFER_TRACE(sbh, "marking dirty");
4428         ext4_superblock_csum_set(sb);
4429         mark_buffer_dirty(sbh);
4430         if (sync) {
4431                 error = sync_dirty_buffer(sbh);
4432                 if (error)
4433                         return error;
4434
4435                 error = buffer_write_io_error(sbh);
4436                 if (error) {
4437                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4438                                "superblock");
4439                         clear_buffer_write_io_error(sbh);
4440                         set_buffer_uptodate(sbh);
4441                 }
4442         }
4443         return error;
4444 }
4445
4446 /*
4447  * Have we just finished recovery?  If so, and if we are mounting (or
4448  * remounting) the filesystem readonly, then we will end up with a
4449  * consistent fs on disk.  Record that fact.
4450  */
4451 static void ext4_mark_recovery_complete(struct super_block *sb,
4452                                         struct ext4_super_block *es)
4453 {
4454         journal_t *journal = EXT4_SB(sb)->s_journal;
4455
4456         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4457                 BUG_ON(journal != NULL);
4458                 return;
4459         }
4460         jbd2_journal_lock_updates(journal);
4461         if (jbd2_journal_flush(journal) < 0)
4462                 goto out;
4463
4464         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4465             sb->s_flags & MS_RDONLY) {
4466                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4467                 ext4_commit_super(sb, 1);
4468         }
4469
4470 out:
4471         jbd2_journal_unlock_updates(journal);
4472 }
4473
4474 /*
4475  * If we are mounting (or read-write remounting) a filesystem whose journal
4476  * has recorded an error from a previous lifetime, move that error to the
4477  * main filesystem now.
4478  */
4479 static void ext4_clear_journal_err(struct super_block *sb,
4480                                    struct ext4_super_block *es)
4481 {
4482         journal_t *journal;
4483         int j_errno;
4484         const char *errstr;
4485
4486         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4487
4488         journal = EXT4_SB(sb)->s_journal;
4489
4490         /*
4491          * Now check for any error status which may have been recorded in the
4492          * journal by a prior ext4_error() or ext4_abort()
4493          */
4494
4495         j_errno = jbd2_journal_errno(journal);
4496         if (j_errno) {
4497                 char nbuf[16];
4498
4499                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4500                 ext4_warning(sb, "Filesystem error recorded "
4501                              "from previous mount: %s", errstr);
4502                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4503
4504                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4505                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4506                 ext4_commit_super(sb, 1);
4507
4508                 jbd2_journal_clear_err(journal);
4509                 jbd2_journal_update_sb_errno(journal);
4510         }
4511 }
4512
4513 /*
4514  * Force the running and committing transactions to commit,
4515  * and wait on the commit.
4516  */
4517 int ext4_force_commit(struct super_block *sb)
4518 {
4519         journal_t *journal;
4520
4521         if (sb->s_flags & MS_RDONLY)
4522                 return 0;
4523
4524         journal = EXT4_SB(sb)->s_journal;
4525         return ext4_journal_force_commit(journal);
4526 }
4527
4528 static int ext4_sync_fs(struct super_block *sb, int wait)
4529 {
4530         int ret = 0;
4531         tid_t target;
4532         struct ext4_sb_info *sbi = EXT4_SB(sb);
4533
4534         trace_ext4_sync_fs(sb, wait);
4535         flush_workqueue(sbi->dio_unwritten_wq);
4536         /*
4537          * Writeback quota in non-journalled quota case - journalled quota has
4538          * no dirty dquots
4539          */
4540         dquot_writeback_dquots(sb, -1);
4541         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4542                 if (wait)
4543                         jbd2_log_wait_commit(sbi->s_journal, target);
4544         }
4545         return ret;
4546 }
4547
4548 /*
4549  * LVM calls this function before a (read-only) snapshot is created.  This
4550  * gives us a chance to flush the journal completely and mark the fs clean.
4551  *
4552  * Note that only this function cannot bring a filesystem to be in a clean
4553  * state independently. It relies on upper layer to stop all data & metadata
4554  * modifications.
4555  */
4556 static int ext4_freeze(struct super_block *sb)
4557 {
4558         int error = 0;
4559         journal_t *journal;
4560
4561         if (sb->s_flags & MS_RDONLY)
4562                 return 0;
4563
4564         journal = EXT4_SB(sb)->s_journal;
4565
4566         /* Now we set up the journal barrier. */
4567         jbd2_journal_lock_updates(journal);
4568
4569         /*
4570          * Don't clear the needs_recovery flag if we failed to flush
4571          * the journal.
4572          */
4573         error = jbd2_journal_flush(journal);
4574         if (error < 0)
4575                 goto out;
4576
4577         /* Journal blocked and flushed, clear needs_recovery flag. */
4578         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4579         error = ext4_commit_super(sb, 1);
4580 out:
4581         /* we rely on upper layer to stop further updates */
4582         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4583         return error;
4584 }
4585
4586 /*
4587  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4588  * flag here, even though the filesystem is not technically dirty yet.
4589  */
4590 static int ext4_unfreeze(struct super_block *sb)
4591 {
4592         if (sb->s_flags & MS_RDONLY)
4593                 return 0;
4594
4595         /* Reset the needs_recovery flag before the fs is unlocked. */
4596         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4597         ext4_commit_super(sb, 1);
4598         return 0;
4599 }
4600
4601 /*
4602  * Structure to save mount options for ext4_remount's benefit
4603  */
4604 struct ext4_mount_options {
4605         unsigned long s_mount_opt;
4606         unsigned long s_mount_opt2;
4607         kuid_t s_resuid;
4608         kgid_t s_resgid;
4609         unsigned long s_commit_interval;
4610         u32 s_min_batch_time, s_max_batch_time;
4611 #ifdef CONFIG_QUOTA
4612         int s_jquota_fmt;
4613         char *s_qf_names[MAXQUOTAS];
4614 #endif
4615 };
4616
4617 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4618 {
4619         struct ext4_super_block *es;
4620         struct ext4_sb_info *sbi = EXT4_SB(sb);
4621         unsigned long old_sb_flags;
4622         struct ext4_mount_options old_opts;
4623         int enable_quota = 0;
4624         ext4_group_t g;
4625         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4626         int err = 0;
4627 #ifdef CONFIG_QUOTA
4628         int i, j;
4629 #endif
4630         char *orig_data = kstrdup(data, GFP_KERNEL);
4631
4632         /* Store the original options */
4633         old_sb_flags = sb->s_flags;
4634         old_opts.s_mount_opt = sbi->s_mount_opt;
4635         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4636         old_opts.s_resuid = sbi->s_resuid;
4637         old_opts.s_resgid = sbi->s_resgid;
4638         old_opts.s_commit_interval = sbi->s_commit_interval;
4639         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4640         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4641 #ifdef CONFIG_QUOTA
4642         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4643         for (i = 0; i < MAXQUOTAS; i++)
4644                 if (sbi->s_qf_names[i]) {
4645                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4646                                                          GFP_KERNEL);
4647                         if (!old_opts.s_qf_names[i]) {
4648                                 for (j = 0; j < i; j++)
4649                                         kfree(old_opts.s_qf_names[j]);
4650                                 kfree(orig_data);
4651                                 return -ENOMEM;
4652                         }
4653                 } else
4654                         old_opts.s_qf_names[i] = NULL;
4655 #endif
4656         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4657                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4658
4659         /*
4660          * Allow the "check" option to be passed as a remount option.
4661          */
4662         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4663                 err = -EINVAL;
4664                 goto restore_opts;
4665         }
4666
4667         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4668                 ext4_abort(sb, "Abort forced by user");
4669
4670         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4671                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4672
4673         es = sbi->s_es;
4674
4675         if (sbi->s_journal) {
4676                 ext4_init_journal_params(sb, sbi->s_journal);
4677                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4678         }
4679
4680         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4681                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4682                         err = -EROFS;
4683                         goto restore_opts;
4684                 }
4685
4686                 if (*flags & MS_RDONLY) {
4687                         err = dquot_suspend(sb, -1);
4688                         if (err < 0)
4689                                 goto restore_opts;
4690
4691                         /*
4692                          * First of all, the unconditional stuff we have to do
4693                          * to disable replay of the journal when we next remount
4694                          */
4695                         sb->s_flags |= MS_RDONLY;
4696
4697                         /*
4698                          * OK, test if we are remounting a valid rw partition
4699                          * readonly, and if so set the rdonly flag and then
4700                          * mark the partition as valid again.
4701                          */
4702                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4703                             (sbi->s_mount_state & EXT4_VALID_FS))
4704                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4705
4706                         if (sbi->s_journal)
4707                                 ext4_mark_recovery_complete(sb, es);
4708                 } else {
4709                         /* Make sure we can mount this feature set readwrite */
4710                         if (!ext4_feature_set_ok(sb, 0)) {
4711                                 err = -EROFS;
4712                                 goto restore_opts;
4713                         }
4714                         /*
4715                          * Make sure the group descriptor checksums
4716                          * are sane.  If they aren't, refuse to remount r/w.
4717                          */
4718                         for (g = 0; g < sbi->s_groups_count; g++) {
4719                                 struct ext4_group_desc *gdp =
4720                                         ext4_get_group_desc(sb, g, NULL);
4721
4722                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4723                                         ext4_msg(sb, KERN_ERR,
4724                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4725                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4726                                                le16_to_cpu(gdp->bg_checksum));
4727                                         err = -EINVAL;
4728                                         goto restore_opts;
4729                                 }
4730                         }
4731
4732                         /*
4733                          * If we have an unprocessed orphan list hanging
4734                          * around from a previously readonly bdev mount,
4735                          * require a full umount/remount for now.
4736                          */
4737                         if (es->s_last_orphan) {
4738                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4739                                        "remount RDWR because of unprocessed "
4740                                        "orphan inode list.  Please "
4741                                        "umount/remount instead");
4742                                 err = -EINVAL;
4743                                 goto restore_opts;
4744                         }
4745
4746                         /*
4747                          * Mounting a RDONLY partition read-write, so reread
4748                          * and store the current valid flag.  (It may have
4749                          * been changed by e2fsck since we originally mounted
4750                          * the partition.)
4751                          */
4752                         if (sbi->s_journal)
4753                                 ext4_clear_journal_err(sb, es);
4754                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4755                         if (!ext4_setup_super(sb, es, 0))
4756                                 sb->s_flags &= ~MS_RDONLY;
4757                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4758                                                      EXT4_FEATURE_INCOMPAT_MMP))
4759                                 if (ext4_multi_mount_protect(sb,
4760                                                 le64_to_cpu(es->s_mmp_block))) {
4761                                         err = -EROFS;
4762                                         goto restore_opts;
4763                                 }
4764                         enable_quota = 1;
4765                 }
4766         }
4767
4768         /*
4769          * Reinitialize lazy itable initialization thread based on
4770          * current settings
4771          */
4772         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4773                 ext4_unregister_li_request(sb);
4774         else {
4775                 ext4_group_t first_not_zeroed;
4776                 first_not_zeroed = ext4_has_uninit_itable(sb);
4777                 ext4_register_li_request(sb, first_not_zeroed);
4778         }
4779
4780         ext4_setup_system_zone(sb);
4781         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4782                 ext4_commit_super(sb, 1);
4783
4784 #ifdef CONFIG_QUOTA
4785         /* Release old quota file names */
4786         for (i = 0; i < MAXQUOTAS; i++)
4787                 kfree(old_opts.s_qf_names[i]);
4788         if (enable_quota) {
4789                 if (sb_any_quota_suspended(sb))
4790                         dquot_resume(sb, -1);
4791                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4792                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4793                         err = ext4_enable_quotas(sb);
4794                         if (err)
4795                                 goto restore_opts;
4796                 }
4797         }
4798 #endif
4799
4800         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4801         kfree(orig_data);
4802         return 0;
4803
4804 restore_opts:
4805         sb->s_flags = old_sb_flags;
4806         sbi->s_mount_opt = old_opts.s_mount_opt;
4807         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4808         sbi->s_resuid = old_opts.s_resuid;
4809         sbi->s_resgid = old_opts.s_resgid;
4810         sbi->s_commit_interval = old_opts.s_commit_interval;
4811         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4812         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4813 #ifdef CONFIG_QUOTA
4814         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4815         for (i = 0; i < MAXQUOTAS; i++) {
4816                 kfree(sbi->s_qf_names[i]);
4817                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4818         }
4819 #endif
4820         kfree(orig_data);
4821         return err;
4822 }
4823
4824 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4825 {
4826         struct super_block *sb = dentry->d_sb;
4827         struct ext4_sb_info *sbi = EXT4_SB(sb);
4828         struct ext4_super_block *es = sbi->s_es;
4829         ext4_fsblk_t overhead = 0, resv_blocks;
4830         u64 fsid;
4831         s64 bfree;
4832         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4833
4834         if (!test_opt(sb, MINIX_DF))
4835                 overhead = sbi->s_overhead;
4836
4837         buf->f_type = EXT4_SUPER_MAGIC;
4838         buf->f_bsize = sb->s_blocksize;
4839         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4840         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4841                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4842         /* prevent underflow in case that few free space is available */
4843         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4844         buf->f_bavail = buf->f_bfree -
4845                         (ext4_r_blocks_count(es) + resv_blocks);
4846         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4847                 buf->f_bavail = 0;
4848         buf->f_files = le32_to_cpu(es->s_inodes_count);
4849         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4850         buf->f_namelen = EXT4_NAME_LEN;
4851         fsid = le64_to_cpup((void *)es->s_uuid) ^
4852                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4853         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4854         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4855
4856         return 0;
4857 }
4858
4859 /* Helper function for writing quotas on sync - we need to start transaction
4860  * before quota file is locked for write. Otherwise the are possible deadlocks:
4861  * Process 1                         Process 2
4862  * ext4_create()                     quota_sync()
4863  *   jbd2_journal_start()                  write_dquot()
4864  *   dquot_initialize()                         down(dqio_mutex)
4865  *     down(dqio_mutex)                    jbd2_journal_start()
4866  *
4867  */
4868
4869 #ifdef CONFIG_QUOTA
4870
4871 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4872 {
4873         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4874 }
4875
4876 static int ext4_write_dquot(struct dquot *dquot)
4877 {
4878         int ret, err;
4879         handle_t *handle;
4880         struct inode *inode;
4881
4882         inode = dquot_to_inode(dquot);
4883         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4884                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4885         if (IS_ERR(handle))
4886                 return PTR_ERR(handle);
4887         ret = dquot_commit(dquot);
4888         err = ext4_journal_stop(handle);
4889         if (!ret)
4890                 ret = err;
4891         return ret;
4892 }
4893
4894 static int ext4_acquire_dquot(struct dquot *dquot)
4895 {
4896         int ret, err;
4897         handle_t *handle;
4898
4899         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4900                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4901         if (IS_ERR(handle))
4902                 return PTR_ERR(handle);
4903         ret = dquot_acquire(dquot);
4904         err = ext4_journal_stop(handle);
4905         if (!ret)
4906                 ret = err;
4907         return ret;
4908 }
4909
4910 static int ext4_release_dquot(struct dquot *dquot)
4911 {
4912         int ret, err;
4913         handle_t *handle;
4914
4915         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4916                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4917         if (IS_ERR(handle)) {
4918                 /* Release dquot anyway to avoid endless cycle in dqput() */
4919                 dquot_release(dquot);
4920                 return PTR_ERR(handle);
4921         }
4922         ret = dquot_release(dquot);
4923         err = ext4_journal_stop(handle);
4924         if (!ret)
4925                 ret = err;
4926         return ret;
4927 }
4928
4929 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4930 {
4931         struct super_block *sb = dquot->dq_sb;
4932         struct ext4_sb_info *sbi = EXT4_SB(sb);
4933
4934         /* Are we journaling quotas? */
4935         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
4936             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4937                 dquot_mark_dquot_dirty(dquot);
4938                 return ext4_write_dquot(dquot);
4939         } else {
4940                 return dquot_mark_dquot_dirty(dquot);
4941         }
4942 }
4943
4944 static int ext4_write_info(struct super_block *sb, int type)
4945 {
4946         int ret, err;
4947         handle_t *handle;
4948
4949         /* Data block + inode block */
4950         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
4951         if (IS_ERR(handle))
4952                 return PTR_ERR(handle);
4953         ret = dquot_commit_info(sb, type);
4954         err = ext4_journal_stop(handle);
4955         if (!ret)
4956                 ret = err;
4957         return ret;
4958 }
4959
4960 /*
4961  * Turn on quotas during mount time - we need to find
4962  * the quota file and such...
4963  */
4964 static int ext4_quota_on_mount(struct super_block *sb, int type)
4965 {
4966         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4967                                         EXT4_SB(sb)->s_jquota_fmt, type);
4968 }
4969
4970 /*
4971  * Standard function to be called on quota_on
4972  */
4973 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4974                          struct path *path)
4975 {
4976         int err;
4977
4978         if (!test_opt(sb, QUOTA))
4979                 return -EINVAL;
4980
4981         /* Quotafile not on the same filesystem? */
4982         if (path->dentry->d_sb != sb)
4983                 return -EXDEV;
4984         /* Journaling quota? */
4985         if (EXT4_SB(sb)->s_qf_names[type]) {
4986                 /* Quotafile not in fs root? */
4987                 if (path->dentry->d_parent != sb->s_root)
4988                         ext4_msg(sb, KERN_WARNING,
4989                                 "Quota file not on filesystem root. "
4990                                 "Journaled quota will not work");
4991         }
4992
4993         /*
4994          * When we journal data on quota file, we have to flush journal to see
4995          * all updates to the file when we bypass pagecache...
4996          */
4997         if (EXT4_SB(sb)->s_journal &&
4998             ext4_should_journal_data(path->dentry->d_inode)) {
4999                 /*
5000                  * We don't need to lock updates but journal_flush() could
5001                  * otherwise be livelocked...
5002                  */
5003                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5004                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5005                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5006                 if (err)
5007                         return err;
5008         }
5009
5010         return dquot_quota_on(sb, type, format_id, path);
5011 }
5012
5013 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5014                              unsigned int flags)
5015 {
5016         int err;
5017         struct inode *qf_inode;
5018         unsigned long qf_inums[MAXQUOTAS] = {
5019                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5020                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5021         };
5022
5023         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5024
5025         if (!qf_inums[type])
5026                 return -EPERM;
5027
5028         qf_inode = ext4_iget(sb, qf_inums[type]);
5029         if (IS_ERR(qf_inode)) {
5030                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5031                 return PTR_ERR(qf_inode);
5032         }
5033
5034         /* Don't account quota for quota files to avoid recursion */
5035         qf_inode->i_flags |= S_NOQUOTA;
5036         err = dquot_enable(qf_inode, type, format_id, flags);
5037         iput(qf_inode);
5038
5039         return err;
5040 }
5041
5042 /* Enable usage tracking for all quota types. */
5043 static int ext4_enable_quotas(struct super_block *sb)
5044 {
5045         int type, err = 0;
5046         unsigned long qf_inums[MAXQUOTAS] = {
5047                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5048                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5049         };
5050
5051         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5052         for (type = 0; type < MAXQUOTAS; type++) {
5053                 if (qf_inums[type]) {
5054                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5055                                                 DQUOT_USAGE_ENABLED);
5056                         if (err) {
5057                                 ext4_warning(sb,
5058                                         "Failed to enable quota tracking "
5059                                         "(type=%d, err=%d). Please run "
5060                                         "e2fsck to fix.", type, err);
5061                                 return err;
5062                         }
5063                 }
5064         }
5065         return 0;
5066 }
5067
5068 /*
5069  * quota_on function that is used when QUOTA feature is set.
5070  */
5071 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5072                                  int format_id)
5073 {
5074         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5075                 return -EINVAL;
5076
5077         /*
5078          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5079          */
5080         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5081 }
5082
5083 static int ext4_quota_off(struct super_block *sb, int type)
5084 {
5085         struct inode *inode = sb_dqopt(sb)->files[type];
5086         handle_t *handle;
5087
5088         /* Force all delayed allocation blocks to be allocated.
5089          * Caller already holds s_umount sem */
5090         if (test_opt(sb, DELALLOC))
5091                 sync_filesystem(sb);
5092
5093         if (!inode)
5094                 goto out;
5095
5096         /* Update modification times of quota files when userspace can
5097          * start looking at them */
5098         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5099         if (IS_ERR(handle))
5100                 goto out;
5101         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5102         ext4_mark_inode_dirty(handle, inode);
5103         ext4_journal_stop(handle);
5104
5105 out:
5106         return dquot_quota_off(sb, type);
5107 }
5108
5109 /*
5110  * quota_off function that is used when QUOTA feature is set.
5111  */
5112 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5113 {
5114         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5115                 return -EINVAL;
5116
5117         /* Disable only the limits. */
5118         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5119 }
5120
5121 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5122  * acquiring the locks... As quota files are never truncated and quota code
5123  * itself serializes the operations (and no one else should touch the files)
5124  * we don't have to be afraid of races */
5125 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5126                                size_t len, loff_t off)
5127 {
5128         struct inode *inode = sb_dqopt(sb)->files[type];
5129         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5130         int err = 0;
5131         int offset = off & (sb->s_blocksize - 1);
5132         int tocopy;
5133         size_t toread;
5134         struct buffer_head *bh;
5135         loff_t i_size = i_size_read(inode);
5136
5137         if (off > i_size)
5138                 return 0;
5139         if (off+len > i_size)
5140                 len = i_size-off;
5141         toread = len;
5142         while (toread > 0) {
5143                 tocopy = sb->s_blocksize - offset < toread ?
5144                                 sb->s_blocksize - offset : toread;
5145                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5146                 if (err)
5147                         return err;
5148                 if (!bh)        /* A hole? */
5149                         memset(data, 0, tocopy);
5150                 else
5151                         memcpy(data, bh->b_data+offset, tocopy);
5152                 brelse(bh);
5153                 offset = 0;
5154                 toread -= tocopy;
5155                 data += tocopy;
5156                 blk++;
5157         }
5158         return len;
5159 }
5160
5161 /* Write to quotafile (we know the transaction is already started and has
5162  * enough credits) */
5163 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5164                                 const char *data, size_t len, loff_t off)
5165 {
5166         struct inode *inode = sb_dqopt(sb)->files[type];
5167         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5168         int err = 0;
5169         int offset = off & (sb->s_blocksize - 1);
5170         struct buffer_head *bh;
5171         handle_t *handle = journal_current_handle();
5172
5173         if (EXT4_SB(sb)->s_journal && !handle) {
5174                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5175                         " cancelled because transaction is not started",
5176                         (unsigned long long)off, (unsigned long long)len);
5177                 return -EIO;
5178         }
5179         /*
5180          * Since we account only one data block in transaction credits,
5181          * then it is impossible to cross a block boundary.
5182          */
5183         if (sb->s_blocksize - offset < len) {
5184                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5185                         " cancelled because not block aligned",
5186                         (unsigned long long)off, (unsigned long long)len);
5187                 return -EIO;
5188         }
5189
5190         bh = ext4_bread(handle, inode, blk, 1, &err);
5191         if (!bh)
5192                 goto out;
5193         err = ext4_journal_get_write_access(handle, bh);
5194         if (err) {
5195                 brelse(bh);
5196                 goto out;
5197         }
5198         lock_buffer(bh);
5199         memcpy(bh->b_data+offset, data, len);
5200         flush_dcache_page(bh->b_page);
5201         unlock_buffer(bh);
5202         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5203         brelse(bh);
5204 out:
5205         if (err)
5206                 return err;
5207         if (inode->i_size < off + len) {
5208                 i_size_write(inode, off + len);
5209                 EXT4_I(inode)->i_disksize = inode->i_size;
5210                 ext4_mark_inode_dirty(handle, inode);
5211         }
5212         return len;
5213 }
5214
5215 #endif
5216
5217 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5218                        const char *dev_name, void *data)
5219 {
5220         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5221 }
5222
5223 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5224 static inline void register_as_ext2(void)
5225 {
5226         int err = register_filesystem(&ext2_fs_type);
5227         if (err)
5228                 printk(KERN_WARNING
5229                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5230 }
5231
5232 static inline void unregister_as_ext2(void)
5233 {
5234         unregister_filesystem(&ext2_fs_type);
5235 }
5236
5237 static inline int ext2_feature_set_ok(struct super_block *sb)
5238 {
5239         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5240                 return 0;
5241         if (sb->s_flags & MS_RDONLY)
5242                 return 1;
5243         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5244                 return 0;
5245         return 1;
5246 }
5247 #else
5248 static inline void register_as_ext2(void) { }
5249 static inline void unregister_as_ext2(void) { }
5250 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5251 #endif
5252
5253 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5254 static inline void register_as_ext3(void)
5255 {
5256         int err = register_filesystem(&ext3_fs_type);
5257         if (err)
5258                 printk(KERN_WARNING
5259                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5260 }
5261
5262 static inline void unregister_as_ext3(void)
5263 {
5264         unregister_filesystem(&ext3_fs_type);
5265 }
5266
5267 static inline int ext3_feature_set_ok(struct super_block *sb)
5268 {
5269         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5270                 return 0;
5271         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5272                 return 0;
5273         if (sb->s_flags & MS_RDONLY)
5274                 return 1;
5275         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5276                 return 0;
5277         return 1;
5278 }
5279 #else
5280 static inline void register_as_ext3(void) { }
5281 static inline void unregister_as_ext3(void) { }
5282 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5283 #endif
5284
5285 static struct file_system_type ext4_fs_type = {
5286         .owner          = THIS_MODULE,
5287         .name           = "ext4",
5288         .mount          = ext4_mount,
5289         .kill_sb        = kill_block_super,
5290         .fs_flags       = FS_REQUIRES_DEV,
5291 };
5292 MODULE_ALIAS_FS("ext4");
5293
5294 static int __init ext4_init_feat_adverts(void)
5295 {
5296         struct ext4_features *ef;
5297         int ret = -ENOMEM;
5298
5299         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5300         if (!ef)
5301                 goto out;
5302
5303         ef->f_kobj.kset = ext4_kset;
5304         init_completion(&ef->f_kobj_unregister);
5305         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5306                                    "features");
5307         if (ret) {
5308                 kfree(ef);
5309                 goto out;
5310         }
5311
5312         ext4_feat = ef;
5313         ret = 0;
5314 out:
5315         return ret;
5316 }
5317
5318 static void ext4_exit_feat_adverts(void)
5319 {
5320         kobject_put(&ext4_feat->f_kobj);
5321         wait_for_completion(&ext4_feat->f_kobj_unregister);
5322         kfree(ext4_feat);
5323 }
5324
5325 /* Shared across all ext4 file systems */
5326 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5327 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5328
5329 static int __init ext4_init_fs(void)
5330 {
5331         int i, err;
5332
5333         ext4_li_info = NULL;
5334         mutex_init(&ext4_li_mtx);
5335
5336         /* Build-time check for flags consistency */
5337         ext4_check_flag_values();
5338
5339         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5340                 mutex_init(&ext4__aio_mutex[i]);
5341                 init_waitqueue_head(&ext4__ioend_wq[i]);
5342         }
5343
5344         err = ext4_init_es();
5345         if (err)
5346                 return err;
5347
5348         err = ext4_init_pageio();
5349         if (err)
5350                 goto out7;
5351
5352         err = ext4_init_system_zone();
5353         if (err)
5354                 goto out6;
5355         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5356         if (!ext4_kset) {
5357                 err = -ENOMEM;
5358                 goto out5;
5359         }
5360         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5361
5362         err = ext4_init_feat_adverts();
5363         if (err)
5364                 goto out4;
5365
5366         err = ext4_init_mballoc();
5367         if (err)
5368                 goto out3;
5369
5370         err = ext4_init_xattr();
5371         if (err)
5372                 goto out2;
5373         err = init_inodecache();
5374         if (err)
5375                 goto out1;
5376         register_as_ext3();
5377         register_as_ext2();
5378         err = register_filesystem(&ext4_fs_type);
5379         if (err)
5380                 goto out;
5381
5382         return 0;
5383 out:
5384         unregister_as_ext2();
5385         unregister_as_ext3();
5386         destroy_inodecache();
5387 out1:
5388         ext4_exit_xattr();
5389 out2:
5390         ext4_exit_mballoc();
5391 out3:
5392         ext4_exit_feat_adverts();
5393 out4:
5394         if (ext4_proc_root)
5395                 remove_proc_entry("fs/ext4", NULL);
5396         kset_unregister(ext4_kset);
5397 out5:
5398         ext4_exit_system_zone();
5399 out6:
5400         ext4_exit_pageio();
5401 out7:
5402         ext4_exit_es();
5403
5404         return err;
5405 }
5406
5407 static void __exit ext4_exit_fs(void)
5408 {
5409         ext4_destroy_lazyinit_thread();
5410         unregister_as_ext2();
5411         unregister_as_ext3();
5412         unregister_filesystem(&ext4_fs_type);
5413         destroy_inodecache();
5414         ext4_exit_xattr();
5415         ext4_exit_mballoc();
5416         ext4_exit_feat_adverts();
5417         remove_proc_entry("fs/ext4", NULL);
5418         kset_unregister(ext4_kset);
5419         ext4_exit_system_zone();
5420         ext4_exit_pageio();
5421 }
5422
5423 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5424 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5425 MODULE_LICENSE("GPL");
5426 module_init(ext4_init_fs)
5427 module_exit(ext4_exit_fs)