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 void ext4_blkdev_put(struct block_device *bdev)
 707 {
 708         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 709 }
 710
 711 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 712 {
 713         struct block_device *bdev;
 714         bdev = sbi->journal_bdev;
 715         if (bdev) {
 716                 ext4_blkdev_put(bdev);
 717                 sbi->journal_bdev = NULL;
 718         }
 719 }
 720
 721 static inline struct inode *orphan_list_entry(struct list_head *l)
 722 {
 723         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 724 }
 725
 726 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 727 {
 728         struct list_head *l;
 729
 730         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 731                  le32_to_cpu(sbi->s_es->s_last_orphan));
 732
 733         printk(KERN_ERR "sb_info orphan list:\n");
 734         list_for_each(l, &sbi->s_orphan) {
 735                 struct inode *inode = orphan_list_entry(l);
 736                 printk(KERN_ERR "  "
 737                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 738                        inode->i_sb->s_id, inode->i_ino, inode,
 739                        inode->i_mode, inode->i_nlink,
 740                        NEXT_ORPHAN(inode));
 741         }
 742 }
 743
 744 static void ext4_put_super(struct super_block *sb)
 745 {
 746         struct ext4_sb_info *sbi = EXT4_SB(sb);
 747         struct ext4_super_block *es = sbi->s_es;
 748         int i, err;
 749
 750         ext4_unregister_li_request(sb);
 751         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 752
 753         flush_workqueue(sbi->dio_unwritten_wq);
 754         destroy_workqueue(sbi->dio_unwritten_wq);
 755
 756         if (sbi->s_journal) {
 757                 err = jbd2_journal_destroy(sbi->s_journal);
 758                 sbi->s_journal = NULL;
 759                 if (err < 0)
 760                         ext4_abort(sb, "Couldn't clean up the journal");
 761         }
 762
 763         ext4_es_unregister_shrinker(sb);
 764         del_timer(&sbi->s_err_report);
 765         ext4_release_system_zone(sb);
 766         ext4_mb_release(sb);
 767         ext4_ext_release(sb);
 768         ext4_xattr_put_super(sb);
 769
 770         if (!(sb->s_flags & MS_RDONLY)) {
 771                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 772                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 773         }
 774         if (!(sb->s_flags & MS_RDONLY))
 775                 ext4_commit_super(sb, 1);
 776
 777         if (sbi->s_proc) {
 778                 remove_proc_entry("options", sbi->s_proc);
 779                 remove_proc_entry(sb->s_id, ext4_proc_root);
 780         }
 781         kobject_del(&sbi->s_kobj);
 782
 783         for (i = 0; i < sbi->s_gdb_count; i++)
 784                 brelse(sbi->s_group_desc[i]);
 785         ext4_kvfree(sbi->s_group_desc);
 786         ext4_kvfree(sbi->s_flex_groups);
 787         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 788         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 789         percpu_counter_destroy(&sbi->s_dirs_counter);
 790         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 791         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
 792         brelse(sbi->s_sbh);
 793 #ifdef CONFIG_QUOTA
 794         for (i = 0; i < MAXQUOTAS; i++)
 795                 kfree(sbi->s_qf_names[i]);
 796 #endif
 797
 798         /* Debugging code just in case the in-memory inode orphan list
 799          * isn't empty.  The on-disk one can be non-empty if we've
 800          * detected an error and taken the fs readonly, but the
 801          * in-memory list had better be clean by this point. */
 802         if (!list_empty(&sbi->s_orphan))
 803                 dump_orphan_list(sb, sbi);
 804         J_ASSERT(list_empty(&sbi->s_orphan));
 805
 806         invalidate_bdev(sb->s_bdev);
 807         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 808                 /*
 809                  * Invalidate the journal device's buffers.  We don't want them
 810                  * floating about in memory - the physical journal device may
 811                  * hotswapped, and it breaks the `ro-after' testing code.
 812                  */
 813                 sync_blockdev(sbi->journal_bdev);
 814                 invalidate_bdev(sbi->journal_bdev);
 815                 ext4_blkdev_remove(sbi);
 816         }
 817         if (sbi->s_mmp_tsk)
 818                 kthread_stop(sbi->s_mmp_tsk);
 819         sb->s_fs_info = NULL;
 820         /*
 821          * Now that we are completely done shutting down the
 822          * superblock, we need to actually destroy the kobject.
 823          */
 824         kobject_put(&sbi->s_kobj);
 825         wait_for_completion(&sbi->s_kobj_unregister);
 826         if (sbi->s_chksum_driver)
 827                 crypto_free_shash(sbi->s_chksum_driver);
 828         kfree(sbi->s_blockgroup_lock);
 829         kfree(sbi);
 830 }
 831
 832 static struct kmem_cache *ext4_inode_cachep;
 833
 834 /*
 835  * Called inside transaction, so use GFP_NOFS
 836  */
 837 static struct inode *ext4_alloc_inode(struct super_block *sb)
 838 {
 839         struct ext4_inode_info *ei;
 840
 841         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 842         if (!ei)
 843                 return NULL;
 844
 845         ei->vfs_inode.i_version = 1;
 846         INIT_LIST_HEAD(&ei->i_prealloc_list);
 847         spin_lock_init(&ei->i_prealloc_lock);
 848         ext4_es_init_tree(&ei->i_es_tree);
 849         rwlock_init(&ei->i_es_lock);
 850         INIT_LIST_HEAD(&ei->i_es_lru);
 851         ei->i_es_lru_nr = 0;
 852         ei->i_reserved_data_blocks = 0;
 853         ei->i_reserved_meta_blocks = 0;
 854         ei->i_allocated_meta_blocks = 0;
 855         ei->i_da_metadata_calc_len = 0;
 856         ei->i_da_metadata_calc_last_lblock = 0;
 857         spin_lock_init(&(ei->i_block_reservation_lock));
 858 #ifdef CONFIG_QUOTA
 859         ei->i_reserved_quota = 0;
 860 #endif
 861         ei->jinode = NULL;
 862         INIT_LIST_HEAD(&ei->i_completed_io_list);
 863         spin_lock_init(&ei->i_completed_io_lock);
 864         ei->i_sync_tid = 0;
 865         ei->i_datasync_tid = 0;
 866         atomic_set(&ei->i_ioend_count, 0);
 867         atomic_set(&ei->i_unwritten, 0);
 868         INIT_WORK(&ei->i_unwritten_work, ext4_end_io_work);
 869
 870         return &ei->vfs_inode;
 871 }
 872
 873 static int ext4_drop_inode(struct inode *inode)
 874 {
 875         int drop = generic_drop_inode(inode);
 876
 877         trace_ext4_drop_inode(inode, drop);
 878         return drop;
 879 }
 880
 881 static void ext4_i_callback(struct rcu_head *head)
 882 {
 883         struct inode *inode = container_of(head, struct inode, i_rcu);
 884         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 885 }
 886
 887 static void ext4_destroy_inode(struct inode *inode)
 888 {
 889         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 890                 ext4_msg(inode->i_sb, KERN_ERR,
 891                          "Inode %lu (%p): orphan list check failed!",
 892                          inode->i_ino, EXT4_I(inode));
 893                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 894                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 895                                 true);
 896                 dump_stack();
 897         }
 898         call_rcu(&inode->i_rcu, ext4_i_callback);
 899 }
 900
 901 static void init_once(void *foo)
 902 {
 903         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 904
 905         INIT_LIST_HEAD(&ei->i_orphan);
 906         init_rwsem(&ei->xattr_sem);
 907         init_rwsem(&ei->i_data_sem);
 908         inode_init_once(&ei->vfs_inode);
 909 }
 910
 911 static int init_inodecache(void)
 912 {
 913         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 914                                              sizeof(struct ext4_inode_info),
 915                                              0, (SLAB_RECLAIM_ACCOUNT|
 916                                                 SLAB_MEM_SPREAD),
 917                                              init_once);
 918         if (ext4_inode_cachep == NULL)
 919                 return -ENOMEM;
 920         return 0;
 921 }
 922
 923 static void destroy_inodecache(void)
 924 {
 925         /*
 926          * Make sure all delayed rcu free inodes are flushed before we
 927          * destroy cache.
 928          */
 929         rcu_barrier();
 930         kmem_cache_destroy(ext4_inode_cachep);
 931 }
 932
 933 void ext4_clear_inode(struct inode *inode)
 934 {
 935         invalidate_inode_buffers(inode);
 936         clear_inode(inode);
 937         dquot_drop(inode);
 938         ext4_discard_preallocations(inode);
 939         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 940         ext4_es_lru_del(inode);
 941         if (EXT4_I(inode)->jinode) {
 942                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 943                                                EXT4_I(inode)->jinode);
 944                 jbd2_free_inode(EXT4_I(inode)->jinode);
 945                 EXT4_I(inode)->jinode = NULL;
 946         }
 947 }
 948
 949 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 950                                         u64 ino, u32 generation)
 951 {
 952         struct inode *inode;
 953
 954         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 955                 return ERR_PTR(-ESTALE);
 956         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 957                 return ERR_PTR(-ESTALE);
 958
 959         /* iget isn't really right if the inode is currently unallocated!!
 960          *
 961          * ext4_read_inode will return a bad_inode if the inode had been
 962          * deleted, so we should be safe.
 963          *
 964          * Currently we don't know the generation for parent directory, so
 965          * a generation of 0 means "accept any"
 966          */
 967         inode = ext4_iget(sb, ino);
 968         if (IS_ERR(inode))
 969                 return ERR_CAST(inode);
 970         if (generation && inode->i_generation != generation) {
 971                 iput(inode);
 972                 return ERR_PTR(-ESTALE);
 973         }
 974
 975         return inode;
 976 }
 977
 978 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 979                                         int fh_len, int fh_type)
 980 {
 981         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 982                                     ext4_nfs_get_inode);
 983 }
 984
 985 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 986                                         int fh_len, int fh_type)
 987 {
 988         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 989                                     ext4_nfs_get_inode);
 990 }
 991
 992 /*
 993  * Try to release metadata pages (indirect blocks, directories) which are
 994  * mapped via the block device.  Since these pages could have journal heads
 995  * which would prevent try_to_free_buffers() from freeing them, we must use
 996  * jbd2 layer's try_to_free_buffers() function to release them.
 997  */
 998 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
 999                                  gfp_t wait)
1000 {
1001         journal_t *journal = EXT4_SB(sb)->s_journal;
1002
1003         WARN_ON(PageChecked(page));
1004         if (!page_has_buffers(page))
1005                 return 0;
1006         if (journal)
1007                 return jbd2_journal_try_to_free_buffers(journal, page,
1008                                                         wait & ~__GFP_WAIT);
1009         return try_to_free_buffers(page);
1010 }
1011
1012 #ifdef CONFIG_QUOTA
1013 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1014 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1015
1016 static int ext4_write_dquot(struct dquot *dquot);
1017 static int ext4_acquire_dquot(struct dquot *dquot);
1018 static int ext4_release_dquot(struct dquot *dquot);
1019 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1020 static int ext4_write_info(struct super_block *sb, int type);
1021 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1022                          struct path *path);
1023 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1024                                  int format_id);
1025 static int ext4_quota_off(struct super_block *sb, int type);
1026 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1027 static int ext4_quota_on_mount(struct super_block *sb, int type);
1028 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1029                                size_t len, loff_t off);
1030 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1031                                 const char *data, size_t len, loff_t off);
1032 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1033                              unsigned int flags);
1034 static int ext4_enable_quotas(struct super_block *sb);
1035
1036 static const struct dquot_operations ext4_quota_operations = {
1037         .get_reserved_space = ext4_get_reserved_space,
1038         .write_dquot    = ext4_write_dquot,
1039         .acquire_dquot  = ext4_acquire_dquot,
1040         .release_dquot  = ext4_release_dquot,
1041         .mark_dirty     = ext4_mark_dquot_dirty,
1042         .write_info     = ext4_write_info,
1043         .alloc_dquot    = dquot_alloc,
1044         .destroy_dquot  = dquot_destroy,
1045 };
1046
1047 static const struct quotactl_ops ext4_qctl_operations = {
1048         .quota_on       = ext4_quota_on,
1049         .quota_off      = ext4_quota_off,
1050         .quota_sync     = dquot_quota_sync,
1051         .get_info       = dquot_get_dqinfo,
1052         .set_info       = dquot_set_dqinfo,
1053         .get_dqblk      = dquot_get_dqblk,
1054         .set_dqblk      = dquot_set_dqblk
1055 };
1056
1057 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1058         .quota_on_meta  = ext4_quota_on_sysfile,
1059         .quota_off      = ext4_quota_off_sysfile,
1060         .quota_sync     = dquot_quota_sync,
1061         .get_info       = dquot_get_dqinfo,
1062         .set_info       = dquot_set_dqinfo,
1063         .get_dqblk      = dquot_get_dqblk,
1064         .set_dqblk      = dquot_set_dqblk
1065 };
1066 #endif
1067
1068 static const struct super_operations ext4_sops = {
1069         .alloc_inode    = ext4_alloc_inode,
1070         .destroy_inode  = ext4_destroy_inode,
1071         .write_inode    = ext4_write_inode,
1072         .dirty_inode    = ext4_dirty_inode,
1073         .drop_inode     = ext4_drop_inode,
1074         .evict_inode    = ext4_evict_inode,
1075         .put_super      = ext4_put_super,
1076         .sync_fs        = ext4_sync_fs,
1077         .freeze_fs      = ext4_freeze,
1078         .unfreeze_fs    = ext4_unfreeze,
1079         .statfs         = ext4_statfs,
1080         .remount_fs     = ext4_remount,
1081         .show_options   = ext4_show_options,
1082 #ifdef CONFIG_QUOTA
1083         .quota_read     = ext4_quota_read,
1084         .quota_write    = ext4_quota_write,
1085 #endif
1086         .bdev_try_to_free_page = bdev_try_to_free_page,
1087 };
1088
1089 static const struct super_operations ext4_nojournal_sops = {
1090         .alloc_inode    = ext4_alloc_inode,
1091         .destroy_inode  = ext4_destroy_inode,
1092         .write_inode    = ext4_write_inode,
1093         .dirty_inode    = ext4_dirty_inode,
1094         .drop_inode     = ext4_drop_inode,
1095         .evict_inode    = ext4_evict_inode,
1096         .put_super      = ext4_put_super,
1097         .statfs         = ext4_statfs,
1098         .remount_fs     = ext4_remount,
1099         .show_options   = ext4_show_options,
1100 #ifdef CONFIG_QUOTA
1101         .quota_read     = ext4_quota_read,
1102         .quota_write    = ext4_quota_write,
1103 #endif
1104         .bdev_try_to_free_page = bdev_try_to_free_page,
1105 };
1106
1107 static const struct export_operations ext4_export_ops = {
1108         .fh_to_dentry = ext4_fh_to_dentry,
1109         .fh_to_parent = ext4_fh_to_parent,
1110         .get_parent = ext4_get_parent,
1111 };
1112
1113 enum {
1114         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1115         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1116         Opt_nouid32, Opt_debug, Opt_removed,
1117         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1118         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1119         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1120         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1121         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1122         Opt_data_err_abort, Opt_data_err_ignore,
1123         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1124         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1125         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1126         Opt_usrquota, Opt_grpquota, Opt_i_version,
1127         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1128         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1129         Opt_inode_readahead_blks, Opt_journal_ioprio,
1130         Opt_dioread_nolock, Opt_dioread_lock,
1131         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1132         Opt_max_dir_size_kb,
1133 };
1134
1135 static const match_table_t tokens = {
1136         {Opt_bsd_df, "bsddf"},
1137         {Opt_minix_df, "minixdf"},
1138         {Opt_grpid, "grpid"},
1139         {Opt_grpid, "bsdgroups"},
1140         {Opt_nogrpid, "nogrpid"},
1141         {Opt_nogrpid, "sysvgroups"},
1142         {Opt_resgid, "resgid=%u"},
1143         {Opt_resuid, "resuid=%u"},
1144         {Opt_sb, "sb=%u"},
1145         {Opt_err_cont, "errors=continue"},
1146         {Opt_err_panic, "errors=panic"},
1147         {Opt_err_ro, "errors=remount-ro"},
1148         {Opt_nouid32, "nouid32"},
1149         {Opt_debug, "debug"},
1150         {Opt_removed, "oldalloc"},
1151         {Opt_removed, "orlov"},
1152         {Opt_user_xattr, "user_xattr"},
1153         {Opt_nouser_xattr, "nouser_xattr"},
1154         {Opt_acl, "acl"},
1155         {Opt_noacl, "noacl"},
1156         {Opt_noload, "norecovery"},
1157         {Opt_noload, "noload"},
1158         {Opt_removed, "nobh"},
1159         {Opt_removed, "bh"},
1160         {Opt_commit, "commit=%u"},
1161         {Opt_min_batch_time, "min_batch_time=%u"},
1162         {Opt_max_batch_time, "max_batch_time=%u"},
1163         {Opt_journal_dev, "journal_dev=%u"},
1164         {Opt_journal_checksum, "journal_checksum"},
1165         {Opt_journal_async_commit, "journal_async_commit"},
1166         {Opt_abort, "abort"},
1167         {Opt_data_journal, "data=journal"},
1168         {Opt_data_ordered, "data=ordered"},
1169         {Opt_data_writeback, "data=writeback"},
1170         {Opt_data_err_abort, "data_err=abort"},
1171         {Opt_data_err_ignore, "data_err=ignore"},
1172         {Opt_offusrjquota, "usrjquota="},
1173         {Opt_usrjquota, "usrjquota=%s"},
1174         {Opt_offgrpjquota, "grpjquota="},
1175         {Opt_grpjquota, "grpjquota=%s"},
1176         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1177         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1178         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1179         {Opt_grpquota, "grpquota"},
1180         {Opt_noquota, "noquota"},
1181         {Opt_quota, "quota"},
1182         {Opt_usrquota, "usrquota"},
1183         {Opt_barrier, "barrier=%u"},
1184         {Opt_barrier, "barrier"},
1185         {Opt_nobarrier, "nobarrier"},
1186         {Opt_i_version, "i_version"},
1187         {Opt_stripe, "stripe=%u"},
1188         {Opt_delalloc, "delalloc"},
1189         {Opt_nodelalloc, "nodelalloc"},
1190         {Opt_removed, "mblk_io_submit"},
1191         {Opt_removed, "nomblk_io_submit"},
1192         {Opt_block_validity, "block_validity"},
1193         {Opt_noblock_validity, "noblock_validity"},
1194         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1195         {Opt_journal_ioprio, "journal_ioprio=%u"},
1196         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1197         {Opt_auto_da_alloc, "auto_da_alloc"},
1198         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1199         {Opt_dioread_nolock, "dioread_nolock"},
1200         {Opt_dioread_lock, "dioread_lock"},
1201         {Opt_discard, "discard"},
1202         {Opt_nodiscard, "nodiscard"},
1203         {Opt_init_itable, "init_itable=%u"},
1204         {Opt_init_itable, "init_itable"},
1205         {Opt_noinit_itable, "noinit_itable"},
1206         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1207         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1208         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1209         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1210         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1211         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1212         {Opt_err, NULL},
1213 };
1214
1215 static ext4_fsblk_t get_sb_block(void **data)
1216 {
1217         ext4_fsblk_t    sb_block;
1218         char            *options = (char *) *data;
1219
1220         if (!options || strncmp(options, "sb=", 3) != 0)
1221                 return 1;       /* Default location */
1222
1223         options += 3;
1224         /* TODO: use simple_strtoll with >32bit ext4 */
1225         sb_block = simple_strtoul(options, &options, 0);
1226         if (*options && *options != ',') {
1227                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1228                        (char *) *data);
1229                 return 1;
1230         }
1231         if (*options == ',')
1232                 options++;
1233         *data = (void *) options;
1234
1235         return sb_block;
1236 }
1237
1238 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1239 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1240         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1241
1242 #ifdef CONFIG_QUOTA
1243 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1244 {
1245         struct ext4_sb_info *sbi = EXT4_SB(sb);
1246         char *qname;
1247         int ret = -1;
1248
1249         if (sb_any_quota_loaded(sb) &&
1250                 !sbi->s_qf_names[qtype]) {
1251                 ext4_msg(sb, KERN_ERR,
1252                         "Cannot change journaled "
1253                         "quota options when quota turned on");
1254                 return -1;
1255         }
1256         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1257                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1258                          "when QUOTA feature is enabled");
1259                 return -1;
1260         }
1261         qname = match_strdup(args);
1262         if (!qname) {
1263                 ext4_msg(sb, KERN_ERR,
1264                         "Not enough memory for storing quotafile name");
1265                 return -1;
1266         }
1267         if (sbi->s_qf_names[qtype]) {
1268                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1269                         ret = 1;
1270                 else
1271                         ext4_msg(sb, KERN_ERR,
1272                                  "%s quota file already specified",
1273                                  QTYPE2NAME(qtype));
1274                 goto errout;
1275         }
1276         if (strchr(qname, '/')) {
1277                 ext4_msg(sb, KERN_ERR,
1278                         "quotafile must be on filesystem root");
1279                 goto errout;
1280         }
1281         sbi->s_qf_names[qtype] = qname;
1282         set_opt(sb, QUOTA);
1283         return 1;
1284 errout:
1285         kfree(qname);
1286         return ret;
1287 }
1288
1289 static int clear_qf_name(struct super_block *sb, int qtype)
1290 {
1291
1292         struct ext4_sb_info *sbi = EXT4_SB(sb);
1293
1294         if (sb_any_quota_loaded(sb) &&
1295                 sbi->s_qf_names[qtype]) {
1296                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1297                         " when quota turned on");
1298                 return -1;
1299         }
1300         kfree(sbi->s_qf_names[qtype]);
1301         sbi->s_qf_names[qtype] = NULL;
1302         return 1;
1303 }
1304 #endif
1305
1306 #define MOPT_SET        0x0001
1307 #define MOPT_CLEAR      0x0002
1308 #define MOPT_NOSUPPORT  0x0004
1309 #define MOPT_EXPLICIT   0x0008
1310 #define MOPT_CLEAR_ERR  0x0010
1311 #define MOPT_GTE0       0x0020
1312 #ifdef CONFIG_QUOTA
1313 #define MOPT_Q          0
1314 #define MOPT_QFMT       0x0040
1315 #else
1316 #define MOPT_Q          MOPT_NOSUPPORT
1317 #define MOPT_QFMT       MOPT_NOSUPPORT
1318 #endif
1319 #define MOPT_DATAJ      0x0080
1320 #define MOPT_NO_EXT2    0x0100
1321 #define MOPT_NO_EXT3    0x0200
1322 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1323
1324 static const struct mount_opts {
1325         int     token;
1326         int     mount_opt;
1327         int     flags;
1328 } ext4_mount_opts[] = {
1329         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1330         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1331         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1332         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1333         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1334         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1335         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1336          MOPT_EXT4_ONLY | MOPT_SET},
1337         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1338          MOPT_EXT4_ONLY | MOPT_CLEAR},
1339         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1340         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1341         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1342          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1343         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1344          MOPT_EXT4_ONLY | MOPT_CLEAR},
1345         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1346          MOPT_EXT4_ONLY | MOPT_SET},
1347         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1348                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1349          MOPT_EXT4_ONLY | MOPT_SET},
1350         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1351         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1352         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1353         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1354         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1355          MOPT_NO_EXT2 | MOPT_SET},
1356         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1357          MOPT_NO_EXT2 | MOPT_CLEAR},
1358         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1359         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1360         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1361         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1362         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1363         {Opt_commit, 0, MOPT_GTE0},
1364         {Opt_max_batch_time, 0, MOPT_GTE0},
1365         {Opt_min_batch_time, 0, MOPT_GTE0},
1366         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1367         {Opt_init_itable, 0, MOPT_GTE0},
1368         {Opt_stripe, 0, MOPT_GTE0},
1369         {Opt_resuid, 0, MOPT_GTE0},
1370         {Opt_resgid, 0, MOPT_GTE0},
1371         {Opt_journal_dev, 0, MOPT_GTE0},
1372         {Opt_journal_ioprio, 0, MOPT_GTE0},
1373         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1374         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1375         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1376          MOPT_NO_EXT2 | MOPT_DATAJ},
1377         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1378         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1379 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1380         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1381         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1382 #else
1383         {Opt_acl, 0, MOPT_NOSUPPORT},
1384         {Opt_noacl, 0, MOPT_NOSUPPORT},
1385 #endif
1386         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1387         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1388         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1389         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1390                                                         MOPT_SET | MOPT_Q},
1391         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1392                                                         MOPT_SET | MOPT_Q},
1393         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1394                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1395         {Opt_usrjquota, 0, MOPT_Q},
1396         {Opt_grpjquota, 0, MOPT_Q},
1397         {Opt_offusrjquota, 0, MOPT_Q},
1398         {Opt_offgrpjquota, 0, MOPT_Q},
1399         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1400         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1401         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1402         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1403         {Opt_err, 0, 0}
1404 };
1405
1406 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1407                             substring_t *args, unsigned long *journal_devnum,
1408                             unsigned int *journal_ioprio, int is_remount)
1409 {
1410         struct ext4_sb_info *sbi = EXT4_SB(sb);
1411         const struct mount_opts *m;
1412         kuid_t uid;
1413         kgid_t gid;
1414         int arg = 0;
1415
1416 #ifdef CONFIG_QUOTA
1417         if (token == Opt_usrjquota)
1418                 return set_qf_name(sb, USRQUOTA, &args[0]);
1419         else if (token == Opt_grpjquota)
1420                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1421         else if (token == Opt_offusrjquota)
1422                 return clear_qf_name(sb, USRQUOTA);
1423         else if (token == Opt_offgrpjquota)
1424                 return clear_qf_name(sb, GRPQUOTA);
1425 #endif
1426         switch (token) {
1427         case Opt_noacl:
1428         case Opt_nouser_xattr:
1429                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1430                 break;
1431         case Opt_sb:
1432                 return 1;       /* handled by get_sb_block() */
1433         case Opt_removed:
1434                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1435                 return 1;
1436         case Opt_abort:
1437                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1438                 return 1;
1439         case Opt_i_version:
1440                 sb->s_flags |= MS_I_VERSION;
1441                 return 1;
1442         }
1443
1444         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1445                 if (token == m->token)
1446                         break;
1447
1448         if (m->token == Opt_err) {
1449                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1450                          "or missing value", opt);
1451                 return -1;
1452         }
1453
1454         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1455                 ext4_msg(sb, KERN_ERR,
1456                          "Mount option \"%s\" incompatible with ext2", opt);
1457                 return -1;
1458         }
1459         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1460                 ext4_msg(sb, KERN_ERR,
1461                          "Mount option \"%s\" incompatible with ext3", opt);
1462                 return -1;
1463         }
1464
1465         if (args->from && match_int(args, &arg))
1466                 return -1;
1467         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1468                 return -1;
1469         if (m->flags & MOPT_EXPLICIT)
1470                 set_opt2(sb, EXPLICIT_DELALLOC);
1471         if (m->flags & MOPT_CLEAR_ERR)
1472                 clear_opt(sb, ERRORS_MASK);
1473         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1474                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1475                          "options when quota turned on");
1476                 return -1;
1477         }
1478
1479         if (m->flags & MOPT_NOSUPPORT) {
1480                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1481         } else if (token == Opt_commit) {
1482                 if (arg == 0)
1483                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1484                 sbi->s_commit_interval = HZ * arg;
1485         } else if (token == Opt_max_batch_time) {
1486                 sbi->s_max_batch_time = arg;
1487         } else if (token == Opt_min_batch_time) {
1488                 sbi->s_min_batch_time = arg;
1489         } else if (token == Opt_inode_readahead_blks) {
1490                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1491                         ext4_msg(sb, KERN_ERR,
1492                                  "EXT4-fs: inode_readahead_blks must be "
1493                                  "0 or a power of 2 smaller than 2^31");
1494                         return -1;
1495                 }
1496                 sbi->s_inode_readahead_blks = arg;
1497         } else if (token == Opt_init_itable) {
1498                 set_opt(sb, INIT_INODE_TABLE);
1499                 if (!args->from)
1500                         arg = EXT4_DEF_LI_WAIT_MULT;
1501                 sbi->s_li_wait_mult = arg;
1502         } else if (token == Opt_max_dir_size_kb) {
1503                 sbi->s_max_dir_size_kb = arg;
1504         } else if (token == Opt_stripe) {
1505                 sbi->s_stripe = arg;
1506         } else if (token == Opt_resuid) {
1507                 uid = make_kuid(current_user_ns(), arg);
1508                 if (!uid_valid(uid)) {
1509                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1510                         return -1;
1511                 }
1512                 sbi->s_resuid = uid;
1513         } else if (token == Opt_resgid) {
1514                 gid = make_kgid(current_user_ns(), arg);
1515                 if (!gid_valid(gid)) {
1516                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1517                         return -1;
1518                 }
1519                 sbi->s_resgid = gid;
1520         } else if (token == Opt_journal_dev) {
1521                 if (is_remount) {
1522                         ext4_msg(sb, KERN_ERR,
1523                                  "Cannot specify journal on remount");
1524                         return -1;
1525                 }
1526                 *journal_devnum = arg;
1527         } else if (token == Opt_journal_ioprio) {
1528                 if (arg > 7) {
1529                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1530                                  " (must be 0-7)");
1531                         return -1;
1532                 }
1533                 *journal_ioprio =
1534                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1535         } else if (m->flags & MOPT_DATAJ) {
1536                 if (is_remount) {
1537                         if (!sbi->s_journal)
1538                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1539                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1540                                 ext4_msg(sb, KERN_ERR,
1541                                          "Cannot change data mode on remount");
1542                                 return -1;
1543                         }
1544                 } else {
1545                         clear_opt(sb, DATA_FLAGS);
1546                         sbi->s_mount_opt |= m->mount_opt;
1547                 }
1548 #ifdef CONFIG_QUOTA
1549         } else if (m->flags & MOPT_QFMT) {
1550                 if (sb_any_quota_loaded(sb) &&
1551                     sbi->s_jquota_fmt != m->mount_opt) {
1552                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1553                                  "quota options when quota turned on");
1554                         return -1;
1555                 }
1556                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1557                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1558                         ext4_msg(sb, KERN_ERR,
1559                                  "Cannot set journaled quota options "
1560                                  "when QUOTA feature is enabled");
1561                         return -1;
1562                 }
1563                 sbi->s_jquota_fmt = m->mount_opt;
1564 #endif
1565         } else {
1566                 if (!args->from)
1567                         arg = 1;
1568                 if (m->flags & MOPT_CLEAR)
1569                         arg = !arg;
1570                 else if (unlikely(!(m->flags & MOPT_SET))) {
1571                         ext4_msg(sb, KERN_WARNING,
1572                                  "buggy handling of option %s", opt);
1573                         WARN_ON(1);
1574                         return -1;
1575                 }
1576                 if (arg != 0)
1577                         sbi->s_mount_opt |= m->mount_opt;
1578                 else
1579                         sbi->s_mount_opt &= ~m->mount_opt;
1580         }
1581         return 1;
1582 }
1583
1584 static int parse_options(char *options, struct super_block *sb,
1585                          unsigned long *journal_devnum,
1586                          unsigned int *journal_ioprio,
1587                          int is_remount)
1588 {
1589         struct ext4_sb_info *sbi = EXT4_SB(sb);
1590         char *p;
1591         substring_t args[MAX_OPT_ARGS];
1592         int token;
1593
1594         if (!options)
1595                 return 1;
1596
1597         while ((p = strsep(&options, ",")) != NULL) {
1598                 if (!*p)
1599                         continue;
1600                 /*
1601                  * Initialize args struct so we know whether arg was
1602                  * found; some options take optional arguments.
1603                  */
1604                 args[0].to = args[0].from = NULL;
1605                 token = match_token(p, tokens, args);
1606                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1607                                      journal_ioprio, is_remount) < 0)
1608                         return 0;
1609         }
1610 #ifdef CONFIG_QUOTA
1611         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1612             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1613                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1614                          "feature is enabled");
1615                 return 0;
1616         }
1617         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1618                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1619                         clear_opt(sb, USRQUOTA);
1620
1621                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1622                         clear_opt(sb, GRPQUOTA);
1623
1624                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1625                         ext4_msg(sb, KERN_ERR, "old and new quota "
1626                                         "format mixing");
1627                         return 0;
1628                 }
1629
1630                 if (!sbi->s_jquota_fmt) {
1631                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1632                                         "not specified");
1633                         return 0;
1634                 }
1635         }
1636 #endif
1637         if (test_opt(sb, DIOREAD_NOLOCK)) {
1638                 int blocksize =
1639                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1640
1641                 if (blocksize < PAGE_CACHE_SIZE) {
1642                         ext4_msg(sb, KERN_ERR, "can't mount with "
1643                                  "dioread_nolock if block size != PAGE_SIZE");
1644                         return 0;
1645                 }
1646         }
1647         return 1;
1648 }
1649
1650 static inline void ext4_show_quota_options(struct seq_file *seq,
1651                                            struct super_block *sb)
1652 {
1653 #if defined(CONFIG_QUOTA)
1654         struct ext4_sb_info *sbi = EXT4_SB(sb);
1655
1656         if (sbi->s_jquota_fmt) {
1657                 char *fmtname = "";
1658
1659                 switch (sbi->s_jquota_fmt) {
1660                 case QFMT_VFS_OLD:
1661                         fmtname = "vfsold";
1662                         break;
1663                 case QFMT_VFS_V0:
1664                         fmtname = "vfsv0";
1665                         break;
1666                 case QFMT_VFS_V1:
1667                         fmtname = "vfsv1";
1668                         break;
1669                 }
1670                 seq_printf(seq, ",jqfmt=%s", fmtname);
1671         }
1672
1673         if (sbi->s_qf_names[USRQUOTA])
1674                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1675
1676         if (sbi->s_qf_names[GRPQUOTA])
1677                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1678 #endif
1679 }
1680
1681 static const char *token2str(int token)
1682 {
1683         const struct match_token *t;
1684
1685         for (t = tokens; t->token != Opt_err; t++)
1686                 if (t->token == token && !strchr(t->pattern, '='))
1687                         break;
1688         return t->pattern;
1689 }
1690
1691 /*
1692  * Show an option if
1693  *  - it's set to a non-default value OR
1694  *  - if the per-sb default is different from the global default
1695  */
1696 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1697                               int nodefs)
1698 {
1699         struct ext4_sb_info *sbi = EXT4_SB(sb);
1700         struct ext4_super_block *es = sbi->s_es;
1701         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1702         const struct mount_opts *m;
1703         char sep = nodefs ? '\n' : ',';
1704
1705 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1706 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1707
1708         if (sbi->s_sb_block != 1)
1709                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1710
1711         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1712                 int want_set = m->flags & MOPT_SET;
1713                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1714                     (m->flags & MOPT_CLEAR_ERR))
1715                         continue;
1716                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1717                         continue; /* skip if same as the default */
1718                 if ((want_set &&
1719                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1720                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1721                         continue; /* select Opt_noFoo vs Opt_Foo */
1722                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1723         }
1724
1725         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1726             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1727                 SEQ_OPTS_PRINT("resuid=%u",
1728                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1729         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1730             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1731                 SEQ_OPTS_PRINT("resgid=%u",
1732                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1733         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1734         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1735                 SEQ_OPTS_PUTS("errors=remount-ro");
1736         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1737                 SEQ_OPTS_PUTS("errors=continue");
1738         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1739                 SEQ_OPTS_PUTS("errors=panic");
1740         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1741                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1742         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1743                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1744         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1745                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1746         if (sb->s_flags & MS_I_VERSION)
1747                 SEQ_OPTS_PUTS("i_version");
1748         if (nodefs || sbi->s_stripe)
1749                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1750         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1751                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1752                         SEQ_OPTS_PUTS("data=journal");
1753                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1754                         SEQ_OPTS_PUTS("data=ordered");
1755                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1756                         SEQ_OPTS_PUTS("data=writeback");
1757         }
1758         if (nodefs ||
1759             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1760                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1761                                sbi->s_inode_readahead_blks);
1762
1763         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1764                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1765                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1766         if (nodefs || sbi->s_max_dir_size_kb)
1767                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1768
1769         ext4_show_quota_options(seq, sb);
1770         return 0;
1771 }
1772
1773 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1774 {
1775         return _ext4_show_options(seq, root->d_sb, 0);
1776 }
1777
1778 static int options_seq_show(struct seq_file *seq, void *offset)
1779 {
1780         struct super_block *sb = seq->private;
1781         int rc;
1782
1783         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1784         rc = _ext4_show_options(seq, sb, 1);
1785         seq_puts(seq, "\n");
1786         return rc;
1787 }
1788
1789 static int options_open_fs(struct inode *inode, struct file *file)
1790 {
1791         return single_open(file, options_seq_show, PDE_DATA(inode));
1792 }
1793
1794 static const struct file_operations ext4_seq_options_fops = {
1795         .owner = THIS_MODULE,
1796         .open = options_open_fs,
1797         .read = seq_read,
1798         .llseek = seq_lseek,
1799         .release = single_release,
1800 };
1801
1802 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1803                             int read_only)
1804 {
1805         struct ext4_sb_info *sbi = EXT4_SB(sb);
1806         int res = 0;
1807
1808         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1809                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1810                          "forcing read-only mode");
1811                 res = MS_RDONLY;
1812         }
1813         if (read_only)
1814                 goto done;
1815         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1816                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1817                          "running e2fsck is recommended");
1818         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1819                 ext4_msg(sb, KERN_WARNING,
1820                          "warning: mounting fs with errors, "
1821                          "running e2fsck is recommended");
1822         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1823                  le16_to_cpu(es->s_mnt_count) >=
1824                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1825                 ext4_msg(sb, KERN_WARNING,
1826                          "warning: maximal mount count reached, "
1827                          "running e2fsck is recommended");
1828         else if (le32_to_cpu(es->s_checkinterval) &&
1829                 (le32_to_cpu(es->s_lastcheck) +
1830                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1831                 ext4_msg(sb, KERN_WARNING,
1832                          "warning: checktime reached, "
1833                          "running e2fsck is recommended");
1834         if (!sbi->s_journal)
1835                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1836         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1837                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1838         le16_add_cpu(&es->s_mnt_count, 1);
1839         es->s_mtime = cpu_to_le32(get_seconds());
1840         ext4_update_dynamic_rev(sb);
1841         if (sbi->s_journal)
1842                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1843
1844         ext4_commit_super(sb, 1);
1845 done:
1846         if (test_opt(sb, DEBUG))
1847                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1848                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1849                         sb->s_blocksize,
1850                         sbi->s_groups_count,
1851                         EXT4_BLOCKS_PER_GROUP(sb),
1852                         EXT4_INODES_PER_GROUP(sb),
1853                         sbi->s_mount_opt, sbi->s_mount_opt2);
1854
1855         cleancache_init_fs(sb);
1856         return res;
1857 }
1858
1859 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1860 {
1861         struct ext4_sb_info *sbi = EXT4_SB(sb);
1862         struct flex_groups *new_groups;
1863         int size;
1864
1865         if (!sbi->s_log_groups_per_flex)
1866                 return 0;
1867
1868         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1869         if (size <= sbi->s_flex_groups_allocated)
1870                 return 0;
1871
1872         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1873         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1874         if (!new_groups) {
1875                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1876                          size / (int) sizeof(struct flex_groups));
1877                 return -ENOMEM;
1878         }
1879
1880         if (sbi->s_flex_groups) {
1881                 memcpy(new_groups, sbi->s_flex_groups,
1882                        (sbi->s_flex_groups_allocated *
1883                         sizeof(struct flex_groups)));
1884                 ext4_kvfree(sbi->s_flex_groups);
1885         }
1886         sbi->s_flex_groups = new_groups;
1887         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1888         return 0;
1889 }
1890
1891 static int ext4_fill_flex_info(struct super_block *sb)
1892 {
1893         struct ext4_sb_info *sbi = EXT4_SB(sb);
1894         struct ext4_group_desc *gdp = NULL;
1895         ext4_group_t flex_group;
1896         unsigned int groups_per_flex = 0;
1897         int i, err;
1898
1899         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1900         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1901                 sbi->s_log_groups_per_flex = 0;
1902                 return 1;
1903         }
1904         groups_per_flex = 1U << sbi->s_log_groups_per_flex;
1905
1906         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1907         if (err)
1908                 goto failed;
1909
1910         for (i = 0; i < sbi->s_groups_count; i++) {
1911                 gdp = ext4_get_group_desc(sb, i, NULL);
1912
1913                 flex_group = ext4_flex_group(sbi, i);
1914                 atomic_add(ext4_free_inodes_count(sb, gdp),
1915                            &sbi->s_flex_groups[flex_group].free_inodes);
1916                 atomic64_add(ext4_free_group_clusters(sb, gdp),
1917                              &sbi->s_flex_groups[flex_group].free_clusters);
1918                 atomic_add(ext4_used_dirs_count(sb, gdp),
1919                            &sbi->s_flex_groups[flex_group].used_dirs);
1920         }
1921
1922         return 1;
1923 failed:
1924         return 0;
1925 }
1926
1927 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1928                                    struct ext4_group_desc *gdp)
1929 {
1930         int offset;
1931         __u16 crc = 0;
1932         __le32 le_group = cpu_to_le32(block_group);
1933
1934         if ((sbi->s_es->s_feature_ro_compat &
1935              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1936                 /* Use new metadata_csum algorithm */
1937                 __le16 save_csum;
1938                 __u32 csum32;
1939
1940                 save_csum = gdp->bg_checksum;
1941                 gdp->bg_checksum = 0;
1942                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1943                                      sizeof(le_group));
1944                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1945                                      sbi->s_desc_size);
1946                 gdp->bg_checksum = save_csum;
1947
1948                 crc = csum32 & 0xFFFF;
1949                 goto out;
1950         }
1951
1952         /* old crc16 code */
1953         offset = offsetof(struct ext4_group_desc, bg_checksum);
1954
1955         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1956         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1957         crc = crc16(crc, (__u8 *)gdp, offset);
1958         offset += sizeof(gdp->bg_checksum); /* skip checksum */
1959         /* for checksum of struct ext4_group_desc do the rest...*/
1960         if ((sbi->s_es->s_feature_incompat &
1961              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1962             offset < le16_to_cpu(sbi->s_es->s_desc_size))
1963                 crc = crc16(crc, (__u8 *)gdp + offset,
1964                             le16_to_cpu(sbi->s_es->s_desc_size) -
1965                                 offset);
1966
1967 out:
1968         return cpu_to_le16(crc);
1969 }
1970
1971 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
1972                                 struct ext4_group_desc *gdp)
1973 {
1974         if (ext4_has_group_desc_csum(sb) &&
1975             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
1976                                                       block_group, gdp)))
1977                 return 0;
1978
1979         return 1;
1980 }
1981
1982 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
1983                               struct ext4_group_desc *gdp)
1984 {
1985         if (!ext4_has_group_desc_csum(sb))
1986                 return;
1987         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
1988 }
1989
1990 /* Called at mount-time, super-block is locked */
1991 static int ext4_check_descriptors(struct super_block *sb,
1992                                   ext4_group_t *first_not_zeroed)
1993 {
1994         struct ext4_sb_info *sbi = EXT4_SB(sb);
1995         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
1996         ext4_fsblk_t last_block;
1997         ext4_fsblk_t block_bitmap;
1998         ext4_fsblk_t inode_bitmap;
1999         ext4_fsblk_t inode_table;
2000         int flexbg_flag = 0;
2001         ext4_group_t i, grp = sbi->s_groups_count;
2002
2003         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2004                 flexbg_flag = 1;
2005
2006         ext4_debug("Checking group descriptors");
2007
2008         for (i = 0; i < sbi->s_groups_count; i++) {
2009                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2010
2011                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2012                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2013                 else
2014                         last_block = first_block +
2015                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2016
2017                 if ((grp == sbi->s_groups_count) &&
2018                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2019                         grp = i;
2020
2021                 block_bitmap = ext4_block_bitmap(sb, gdp);
2022                 if (block_bitmap < first_block || block_bitmap > last_block) {
2023                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2024                                "Block bitmap for group %u not in group "
2025                                "(block %llu)!", i, block_bitmap);
2026                         return 0;
2027                 }
2028                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2029                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2030                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2031                                "Inode bitmap for group %u not in group "
2032                                "(block %llu)!", i, inode_bitmap);
2033                         return 0;
2034                 }
2035                 inode_table = ext4_inode_table(sb, gdp);
2036                 if (inode_table < first_block ||
2037                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2038                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2039                                "Inode table for group %u not in group "
2040                                "(block %llu)!", i, inode_table);
2041                         return 0;
2042                 }
2043                 ext4_lock_group(sb, i);
2044                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2045                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2046                                  "Checksum for group %u failed (%u!=%u)",
2047                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2048                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2049                         if (!(sb->s_flags & MS_RDONLY)) {
2050                                 ext4_unlock_group(sb, i);
2051                                 return 0;
2052                         }
2053                 }
2054                 ext4_unlock_group(sb, i);
2055                 if (!flexbg_flag)
2056                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2057         }
2058         if (NULL != first_not_zeroed)
2059                 *first_not_zeroed = grp;
2060
2061         ext4_free_blocks_count_set(sbi->s_es,
2062                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2063         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2064         return 1;
2065 }
2066
2067 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2068  * the superblock) which were deleted from all directories, but held open by
2069  * a process at the time of a crash.  We walk the list and try to delete these
2070  * inodes at recovery time (only with a read-write filesystem).
2071  *
2072  * In order to keep the orphan inode chain consistent during traversal (in
2073  * case of crash during recovery), we link each inode into the superblock
2074  * orphan list_head and handle it the same way as an inode deletion during
2075  * normal operation (which journals the operations for us).
2076  *
2077  * We only do an iget() and an iput() on each inode, which is very safe if we
2078  * accidentally point at an in-use or already deleted inode.  The worst that
2079  * can happen in this case is that we get a "bit already cleared" message from
2080  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2081  * e2fsck was run on this filesystem, and it must have already done the orphan
2082  * inode cleanup for us, so we can safely abort without any further action.
2083  */
2084 static void ext4_orphan_cleanup(struct super_block *sb,
2085                                 struct ext4_super_block *es)
2086 {
2087         unsigned int s_flags = sb->s_flags;
2088         int nr_orphans = 0, nr_truncates = 0;
2089 #ifdef CONFIG_QUOTA
2090         int i;
2091 #endif
2092         if (!es->s_last_orphan) {
2093                 jbd_debug(4, "no orphan inodes to clean up\n");
2094                 return;
2095         }
2096
2097         if (bdev_read_only(sb->s_bdev)) {
2098                 ext4_msg(sb, KERN_ERR, "write access "
2099                         "unavailable, skipping orphan cleanup");
2100                 return;
2101         }
2102
2103         /* Check if feature set would not allow a r/w mount */
2104         if (!ext4_feature_set_ok(sb, 0)) {
2105                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2106                          "unknown ROCOMPAT features");
2107                 return;
2108         }
2109
2110         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2111                 /* don't clear list on RO mount w/ errors */
2112                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2113                         jbd_debug(1, "Errors on filesystem, "
2114                                   "clearing orphan list.\n");
2115                         es->s_last_orphan = 0;
2116                 }
2117                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2118                 return;
2119         }
2120
2121         if (s_flags & MS_RDONLY) {
2122                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2123                 sb->s_flags &= ~MS_RDONLY;
2124         }
2125 #ifdef CONFIG_QUOTA
2126         /* Needed for iput() to work correctly and not trash data */
2127         sb->s_flags |= MS_ACTIVE;
2128         /* Turn on quotas so that they are updated correctly */
2129         for (i = 0; i < MAXQUOTAS; i++) {
2130                 if (EXT4_SB(sb)->s_qf_names[i]) {
2131                         int ret = ext4_quota_on_mount(sb, i);
2132                         if (ret < 0)
2133                                 ext4_msg(sb, KERN_ERR,
2134                                         "Cannot turn on journaled "
2135                                         "quota: error %d", ret);
2136                 }
2137         }
2138 #endif
2139
2140         while (es->s_last_orphan) {
2141                 struct inode *inode;
2142
2143                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2144                 if (IS_ERR(inode)) {
2145                         es->s_last_orphan = 0;
2146                         break;
2147                 }
2148
2149                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2150                 dquot_initialize(inode);
2151                 if (inode->i_nlink) {
2152                         ext4_msg(sb, KERN_DEBUG,
2153                                 "%s: truncating inode %lu to %lld bytes",
2154                                 __func__, inode->i_ino, inode->i_size);
2155                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2156                                   inode->i_ino, inode->i_size);
2157                         mutex_lock(&inode->i_mutex);
2158                         ext4_truncate(inode);
2159                         mutex_unlock(&inode->i_mutex);
2160                         nr_truncates++;
2161                 } else {
2162                         ext4_msg(sb, KERN_DEBUG,
2163                                 "%s: deleting unreferenced inode %lu",
2164                                 __func__, inode->i_ino);
2165                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2166                                   inode->i_ino);
2167                         nr_orphans++;
2168                 }
2169                 iput(inode);  /* The delete magic happens here! */
2170         }
2171
2172 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2173
2174         if (nr_orphans)
2175                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2176                        PLURAL(nr_orphans));
2177         if (nr_truncates)
2178                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2179                        PLURAL(nr_truncates));
2180 #ifdef CONFIG_QUOTA
2181         /* Turn quotas off */
2182         for (i = 0; i < MAXQUOTAS; i++) {
2183                 if (sb_dqopt(sb)->files[i])
2184                         dquot_quota_off(sb, i);
2185         }
2186 #endif
2187         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2188 }
2189
2190 /*
2191  * Maximal extent format file size.
2192  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2193  * extent format containers, within a sector_t, and within i_blocks
2194  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2195  * so that won't be a limiting factor.
2196  *
2197  * However there is other limiting factor. We do store extents in the form
2198  * of starting block and length, hence the resulting length of the extent
2199  * covering maximum file size must fit into on-disk format containers as
2200  * well. Given that length is always by 1 unit bigger than max unit (because
2201  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2202  *
2203  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2204  */
2205 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2206 {
2207         loff_t res;
2208         loff_t upper_limit = MAX_LFS_FILESIZE;
2209
2210         /* small i_blocks in vfs inode? */
2211         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2212                 /*
2213                  * CONFIG_LBDAF is not enabled implies the inode
2214                  * i_block represent total blocks in 512 bytes
2215                  * 32 == size of vfs inode i_blocks * 8
2216                  */
2217                 upper_limit = (1LL << 32) - 1;
2218
2219                 /* total blocks in file system block size */
2220                 upper_limit >>= (blkbits - 9);
2221                 upper_limit <<= blkbits;
2222         }
2223
2224         /*
2225          * 32-bit extent-start container, ee_block. We lower the maxbytes
2226          * by one fs block, so ee_len can cover the extent of maximum file
2227          * size
2228          */
2229         res = (1LL << 32) - 1;
2230         res <<= blkbits;
2231
2232         /* Sanity check against vm- & vfs- imposed limits */
2233         if (res > upper_limit)
2234                 res = upper_limit;
2235
2236         return res;
2237 }
2238
2239 /*
2240  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2241  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2242  * We need to be 1 filesystem block less than the 2^48 sector limit.
2243  */
2244 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2245 {
2246         loff_t res = EXT4_NDIR_BLOCKS;
2247         int meta_blocks;
2248         loff_t upper_limit;
2249         /* This is calculated to be the largest file size for a dense, block
2250          * mapped file such that the file's total number of 512-byte sectors,
2251          * including data and all indirect blocks, does not exceed (2^48 - 1).
2252          *
2253          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2254          * number of 512-byte sectors of the file.
2255          */
2256
2257         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2258                 /*
2259                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2260                  * the inode i_block field represents total file blocks in
2261                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2262                  */
2263                 upper_limit = (1LL << 32) - 1;
2264
2265                 /* total blocks in file system block size */
2266                 upper_limit >>= (bits - 9);
2267
2268         } else {
2269                 /*
2270                  * We use 48 bit ext4_inode i_blocks
2271                  * With EXT4_HUGE_FILE_FL set the i_blocks
2272                  * represent total number of blocks in
2273                  * file system block size
2274                  */
2275                 upper_limit = (1LL << 48) - 1;
2276
2277         }
2278
2279         /* indirect blocks */
2280         meta_blocks = 1;
2281         /* double indirect blocks */
2282         meta_blocks += 1 + (1LL << (bits-2));
2283         /* tripple indirect blocks */
2284         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2285
2286         upper_limit -= meta_blocks;
2287         upper_limit <<= bits;
2288
2289         res += 1LL << (bits-2);
2290         res += 1LL << (2*(bits-2));
2291         res += 1LL << (3*(bits-2));
2292         res <<= bits;
2293         if (res > upper_limit)
2294                 res = upper_limit;
2295
2296         if (res > MAX_LFS_FILESIZE)
2297                 res = MAX_LFS_FILESIZE;
2298
2299         return res;
2300 }
2301
2302 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2303                                    ext4_fsblk_t logical_sb_block, int nr)
2304 {
2305         struct ext4_sb_info *sbi = EXT4_SB(sb);
2306         ext4_group_t bg, first_meta_bg;
2307         int has_super = 0;
2308
2309         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2310
2311         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2312             nr < first_meta_bg)
2313                 return logical_sb_block + nr + 1;
2314         bg = sbi->s_desc_per_block * nr;
2315         if (ext4_bg_has_super(sb, bg))
2316                 has_super = 1;
2317
2318         return (has_super + ext4_group_first_block_no(sb, bg));
2319 }
2320
2321 /**
2322  * ext4_get_stripe_size: Get the stripe size.
2323  * @sbi: In memory super block info
2324  *
2325  * If we have specified it via mount option, then
2326  * use the mount option value. If the value specified at mount time is
2327  * greater than the blocks per group use the super block value.
2328  * If the super block value is greater than blocks per group return 0.
2329  * Allocator needs it be less than blocks per group.
2330  *
2331  */
2332 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2333 {
2334         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2335         unsigned long stripe_width =
2336                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2337         int ret;
2338
2339         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2340                 ret = sbi->s_stripe;
2341         else if (stripe_width <= sbi->s_blocks_per_group)
2342                 ret = stripe_width;
2343         else if (stride <= sbi->s_blocks_per_group)
2344                 ret = stride;
2345         else
2346                 ret = 0;
2347
2348         /*
2349          * If the stripe width is 1, this makes no sense and
2350          * we set it to 0 to turn off stripe handling code.
2351          */
2352         if (ret <= 1)
2353                 ret = 0;
2354
2355         return ret;
2356 }
2357
2358 /* sysfs supprt */
2359
2360 struct ext4_attr {
2361         struct attribute attr;
2362         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2363         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2364                          const char *, size_t);
2365         int offset;
2366 };
2367
2368 static int parse_strtoull(const char *buf,
2369                 unsigned long long max, unsigned long long *value)
2370 {
2371         int ret;
2372
2373         ret = kstrtoull(skip_spaces(buf), 0, value);
2374         if (!ret && *value > max)
2375                 ret = -EINVAL;
2376         return ret;
2377 }
2378
2379 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2380                                               struct ext4_sb_info *sbi,
2381                                               char *buf)
2382 {
2383         return snprintf(buf, PAGE_SIZE, "%llu\n",
2384                 (s64) EXT4_C2B(sbi,
2385                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2386 }
2387
2388 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2389                                          struct ext4_sb_info *sbi, char *buf)
2390 {
2391         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2392
2393         if (!sb->s_bdev->bd_part)
2394                 return snprintf(buf, PAGE_SIZE, "0\n");
2395         return snprintf(buf, PAGE_SIZE, "%lu\n",
2396                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2397                          sbi->s_sectors_written_start) >> 1);
2398 }
2399
2400 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2401                                           struct ext4_sb_info *sbi, char *buf)
2402 {
2403         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2404
2405         if (!sb->s_bdev->bd_part)
2406                 return snprintf(buf, PAGE_SIZE, "0\n");
2407         return snprintf(buf, PAGE_SIZE, "%llu\n",
2408                         (unsigned long long)(sbi->s_kbytes_written +
2409                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2410                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2411 }
2412
2413 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2414                                           struct ext4_sb_info *sbi,
2415                                           const char *buf, size_t count)
2416 {
2417         unsigned long t;
2418         int ret;
2419
2420         ret = kstrtoul(skip_spaces(buf), 0, &t);
2421         if (ret)
2422                 return ret;
2423
2424         if (t && (!is_power_of_2(t) || t > 0x40000000))
2425                 return -EINVAL;
2426
2427         sbi->s_inode_readahead_blks = t;
2428         return count;
2429 }
2430
2431 static ssize_t sbi_ui_show(struct ext4_attr *a,
2432                            struct ext4_sb_info *sbi, char *buf)
2433 {
2434         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2435
2436         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2437 }
2438
2439 static ssize_t sbi_ui_store(struct ext4_attr *a,
2440                             struct ext4_sb_info *sbi,
2441                             const char *buf, size_t count)
2442 {
2443         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2444         unsigned long t;
2445         int ret;
2446
2447         ret = kstrtoul(skip_spaces(buf), 0, &t);
2448         if (ret)
2449                 return ret;
2450         *ui = t;
2451         return count;
2452 }
2453
2454 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2455                                   struct ext4_sb_info *sbi, char *buf)
2456 {
2457         return snprintf(buf, PAGE_SIZE, "%llu\n",
2458                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2459 }
2460
2461 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2462                                    struct ext4_sb_info *sbi,
2463                                    const char *buf, size_t count)
2464 {
2465         unsigned long long val;
2466         int ret;
2467
2468         if (parse_strtoull(buf, -1ULL, &val))
2469                 return -EINVAL;
2470         ret = ext4_reserve_clusters(sbi, val);
2471
2472         return ret ? ret : count;
2473 }
2474
2475 static ssize_t trigger_test_error(struct ext4_attr *a,
2476                                   struct ext4_sb_info *sbi,
2477                                   const char *buf, size_t count)
2478 {
2479         int len = count;
2480
2481         if (!capable(CAP_SYS_ADMIN))
2482                 return -EPERM;
2483
2484         if (len && buf[len-1] == '\n')
2485                 len--;
2486
2487         if (len)
2488                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2489         return count;
2490 }
2491
2492 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2493 static struct ext4_attr ext4_attr_##_name = {                   \
2494         .attr = {.name = __stringify(_name), .mode = _mode },   \
2495         .show   = _show,                                        \
2496         .store  = _store,                                       \
2497         .offset = offsetof(struct ext4_sb_info, _elname),       \
2498 }
2499 #define EXT4_ATTR(name, mode, show, store) \
2500 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2501
2502 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2503 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2504 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2505 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2506         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2507 #define ATTR_LIST(name) &ext4_attr_##name.attr
2508
2509 EXT4_RO_ATTR(delayed_allocation_blocks);
2510 EXT4_RO_ATTR(session_write_kbytes);
2511 EXT4_RO_ATTR(lifetime_write_kbytes);
2512 EXT4_RW_ATTR(reserved_clusters);
2513 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2514                  inode_readahead_blks_store, s_inode_readahead_blks);
2515 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2516 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2517 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2518 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2519 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2520 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2521 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2522 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2523 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2524 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2525
2526 static struct attribute *ext4_attrs[] = {
2527         ATTR_LIST(delayed_allocation_blocks),
2528         ATTR_LIST(session_write_kbytes),
2529         ATTR_LIST(lifetime_write_kbytes),
2530         ATTR_LIST(reserved_clusters),
2531         ATTR_LIST(inode_readahead_blks),
2532         ATTR_LIST(inode_goal),
2533         ATTR_LIST(mb_stats),
2534         ATTR_LIST(mb_max_to_scan),
2535         ATTR_LIST(mb_min_to_scan),
2536         ATTR_LIST(mb_order2_req),
2537         ATTR_LIST(mb_stream_req),
2538         ATTR_LIST(mb_group_prealloc),
2539         ATTR_LIST(max_writeback_mb_bump),
2540         ATTR_LIST(extent_max_zeroout_kb),
2541         ATTR_LIST(trigger_fs_error),
2542         NULL,
2543 };
2544
2545 /* Features this copy of ext4 supports */
2546 EXT4_INFO_ATTR(lazy_itable_init);
2547 EXT4_INFO_ATTR(batched_discard);
2548 EXT4_INFO_ATTR(meta_bg_resize);
2549
2550 static struct attribute *ext4_feat_attrs[] = {
2551         ATTR_LIST(lazy_itable_init),
2552         ATTR_LIST(batched_discard),
2553         ATTR_LIST(meta_bg_resize),
2554         NULL,
2555 };
2556
2557 static ssize_t ext4_attr_show(struct kobject *kobj,
2558                               struct attribute *attr, char *buf)
2559 {
2560         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2561                                                 s_kobj);
2562         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2563
2564         return a->show ? a->show(a, sbi, buf) : 0;
2565 }
2566
2567 static ssize_t ext4_attr_store(struct kobject *kobj,
2568                                struct attribute *attr,
2569                                const char *buf, size_t len)
2570 {
2571         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2572                                                 s_kobj);
2573         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2574
2575         return a->store ? a->store(a, sbi, buf, len) : 0;
2576 }
2577
2578 static void ext4_sb_release(struct kobject *kobj)
2579 {
2580         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2581                                                 s_kobj);
2582         complete(&sbi->s_kobj_unregister);
2583 }
2584
2585 static const struct sysfs_ops ext4_attr_ops = {
2586         .show   = ext4_attr_show,
2587         .store  = ext4_attr_store,
2588 };
2589
2590 static struct kobj_type ext4_ktype = {
2591         .default_attrs  = ext4_attrs,
2592         .sysfs_ops      = &ext4_attr_ops,
2593         .release        = ext4_sb_release,
2594 };
2595
2596 static void ext4_feat_release(struct kobject *kobj)
2597 {
2598         complete(&ext4_feat->f_kobj_unregister);
2599 }
2600
2601 static struct kobj_type ext4_feat_ktype = {
2602         .default_attrs  = ext4_feat_attrs,
2603         .sysfs_ops      = &ext4_attr_ops,
2604         .release        = ext4_feat_release,
2605 };
2606
2607 /*
2608  * Check whether this filesystem can be mounted based on
2609  * the features present and the RDONLY/RDWR mount requested.
2610  * Returns 1 if this filesystem can be mounted as requested,
2611  * 0 if it cannot be.
2612  */
2613 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2614 {
2615         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2616                 ext4_msg(sb, KERN_ERR,
2617                         "Couldn't mount because of "
2618                         "unsupported optional features (%x)",
2619                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2620                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2621                 return 0;
2622         }
2623
2624         if (readonly)
2625                 return 1;
2626
2627         /* Check that feature set is OK for a read-write mount */
2628         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2629                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2630                          "unsupported optional features (%x)",
2631                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2632                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2633                 return 0;
2634         }
2635         /*
2636          * Large file size enabled file system can only be mounted
2637          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2638          */
2639         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2640                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2641                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2642                                  "cannot be mounted RDWR without "
2643                                  "CONFIG_LBDAF");
2644                         return 0;
2645                 }
2646         }
2647         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2648             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2649                 ext4_msg(sb, KERN_ERR,
2650                          "Can't support bigalloc feature without "
2651                          "extents feature\n");
2652                 return 0;
2653         }
2654
2655 #ifndef CONFIG_QUOTA
2656         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2657             !readonly) {
2658                 ext4_msg(sb, KERN_ERR,
2659                          "Filesystem with quota feature cannot be mounted RDWR "
2660                          "without CONFIG_QUOTA");
2661                 return 0;
2662         }
2663 #endif  /* CONFIG_QUOTA */
2664         return 1;
2665 }
2666
2667 /*
2668  * This function is called once a day if we have errors logged
2669  * on the file system
2670  */
2671 static void print_daily_error_info(unsigned long arg)
2672 {
2673         struct super_block *sb = (struct super_block *) arg;
2674         struct ext4_sb_info *sbi;
2675         struct ext4_super_block *es;
2676
2677         sbi = EXT4_SB(sb);
2678         es = sbi->s_es;
2679
2680         if (es->s_error_count)
2681                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2682                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2683                          le32_to_cpu(es->s_error_count));
2684         if (es->s_first_error_time) {
2685                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2686                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2687                        (int) sizeof(es->s_first_error_func),
2688                        es->s_first_error_func,
2689                        le32_to_cpu(es->s_first_error_line));
2690                 if (es->s_first_error_ino)
2691                         printk(": inode %u",
2692                                le32_to_cpu(es->s_first_error_ino));
2693                 if (es->s_first_error_block)
2694                         printk(": block %llu", (unsigned long long)
2695                                le64_to_cpu(es->s_first_error_block));
2696                 printk("\n");
2697         }
2698         if (es->s_last_error_time) {
2699                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2700                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2701                        (int) sizeof(es->s_last_error_func),
2702                        es->s_last_error_func,
2703                        le32_to_cpu(es->s_last_error_line));
2704                 if (es->s_last_error_ino)
2705                         printk(": inode %u",
2706                                le32_to_cpu(es->s_last_error_ino));
2707                 if (es->s_last_error_block)
2708                         printk(": block %llu", (unsigned long long)
2709                                le64_to_cpu(es->s_last_error_block));
2710                 printk("\n");
2711         }
2712         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2713 }
2714
2715 /* Find next suitable group and run ext4_init_inode_table */
2716 static int ext4_run_li_request(struct ext4_li_request *elr)
2717 {
2718         struct ext4_group_desc *gdp = NULL;
2719         ext4_group_t group, ngroups;
2720         struct super_block *sb;
2721         unsigned long timeout = 0;
2722         int ret = 0;
2723
2724         sb = elr->lr_super;
2725         ngroups = EXT4_SB(sb)->s_groups_count;
2726
2727         sb_start_write(sb);
2728         for (group = elr->lr_next_group; group < ngroups; group++) {
2729                 gdp = ext4_get_group_desc(sb, group, NULL);
2730                 if (!gdp) {
2731                         ret = 1;
2732                         break;
2733                 }
2734
2735                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2736                         break;
2737         }
2738
2739         if (group >= ngroups)
2740                 ret = 1;
2741
2742         if (!ret) {
2743                 timeout = jiffies;
2744                 ret = ext4_init_inode_table(sb, group,
2745                                             elr->lr_timeout ? 0 : 1);
2746                 if (elr->lr_timeout == 0) {
2747                         timeout = (jiffies - timeout) *
2748                                   elr->lr_sbi->s_li_wait_mult;
2749                         elr->lr_timeout = timeout;
2750                 }
2751                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2752                 elr->lr_next_group = group + 1;
2753         }
2754         sb_end_write(sb);
2755
2756         return ret;
2757 }
2758
2759 /*
2760  * Remove lr_request from the list_request and free the
2761  * request structure. Should be called with li_list_mtx held
2762  */
2763 static void ext4_remove_li_request(struct ext4_li_request *elr)
2764 {
2765         struct ext4_sb_info *sbi;
2766
2767         if (!elr)
2768                 return;
2769
2770         sbi = elr->lr_sbi;
2771
2772         list_del(&elr->lr_request);
2773         sbi->s_li_request = NULL;
2774         kfree(elr);
2775 }
2776
2777 static void ext4_unregister_li_request(struct super_block *sb)
2778 {
2779         mutex_lock(&ext4_li_mtx);
2780         if (!ext4_li_info) {
2781                 mutex_unlock(&ext4_li_mtx);
2782                 return;
2783         }
2784
2785         mutex_lock(&ext4_li_info->li_list_mtx);
2786         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2787         mutex_unlock(&ext4_li_info->li_list_mtx);
2788         mutex_unlock(&ext4_li_mtx);
2789 }
2790
2791 static struct task_struct *ext4_lazyinit_task;
2792
2793 /*
2794  * This is the function where ext4lazyinit thread lives. It walks
2795  * through the request list searching for next scheduled filesystem.
2796  * When such a fs is found, run the lazy initialization request
2797  * (ext4_rn_li_request) and keep track of the time spend in this
2798  * function. Based on that time we compute next schedule time of
2799  * the request. When walking through the list is complete, compute
2800  * next waking time and put itself into sleep.
2801  */
2802 static int ext4_lazyinit_thread(void *arg)
2803 {
2804         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2805         struct list_head *pos, *n;
2806         struct ext4_li_request *elr;
2807         unsigned long next_wakeup, cur;
2808
2809         BUG_ON(NULL == eli);
2810
2811 cont_thread:
2812         while (true) {
2813                 next_wakeup = MAX_JIFFY_OFFSET;
2814
2815                 mutex_lock(&eli->li_list_mtx);
2816                 if (list_empty(&eli->li_request_list)) {
2817                         mutex_unlock(&eli->li_list_mtx);
2818                         goto exit_thread;
2819                 }
2820
2821                 list_for_each_safe(pos, n, &eli->li_request_list) {
2822                         elr = list_entry(pos, struct ext4_li_request,
2823                                          lr_request);
2824
2825                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2826                                 if (ext4_run_li_request(elr) != 0) {
2827                                         /* error, remove the lazy_init job */
2828                                         ext4_remove_li_request(elr);
2829                                         continue;
2830                                 }
2831                         }
2832
2833                         if (time_before(elr->lr_next_sched, next_wakeup))
2834                                 next_wakeup = elr->lr_next_sched;
2835                 }
2836                 mutex_unlock(&eli->li_list_mtx);
2837
2838                 try_to_freeze();
2839
2840                 cur = jiffies;
2841                 if ((time_after_eq(cur, next_wakeup)) ||
2842                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2843                         cond_resched();
2844                         continue;
2845                 }
2846
2847                 schedule_timeout_interruptible(next_wakeup - cur);
2848
2849                 if (kthread_should_stop()) {
2850                         ext4_clear_request_list();
2851                         goto exit_thread;
2852                 }
2853         }
2854
2855 exit_thread:
2856         /*
2857          * It looks like the request list is empty, but we need
2858          * to check it under the li_list_mtx lock, to prevent any
2859          * additions into it, and of course we should lock ext4_li_mtx
2860          * to atomically free the list and ext4_li_info, because at
2861          * this point another ext4 filesystem could be registering
2862          * new one.
2863          */
2864         mutex_lock(&ext4_li_mtx);
2865         mutex_lock(&eli->li_list_mtx);
2866         if (!list_empty(&eli->li_request_list)) {
2867                 mutex_unlock(&eli->li_list_mtx);
2868                 mutex_unlock(&ext4_li_mtx);
2869                 goto cont_thread;
2870         }
2871         mutex_unlock(&eli->li_list_mtx);
2872         kfree(ext4_li_info);
2873         ext4_li_info = NULL;
2874         mutex_unlock(&ext4_li_mtx);
2875
2876         return 0;
2877 }
2878
2879 static void ext4_clear_request_list(void)
2880 {
2881         struct list_head *pos, *n;
2882         struct ext4_li_request *elr;
2883
2884         mutex_lock(&ext4_li_info->li_list_mtx);
2885         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2886                 elr = list_entry(pos, struct ext4_li_request,
2887                                  lr_request);
2888                 ext4_remove_li_request(elr);
2889         }
2890         mutex_unlock(&ext4_li_info->li_list_mtx);
2891 }
2892
2893 static int ext4_run_lazyinit_thread(void)
2894 {
2895         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2896                                          ext4_li_info, "ext4lazyinit");
2897         if (IS_ERR(ext4_lazyinit_task)) {
2898                 int err = PTR_ERR(ext4_lazyinit_task);
2899                 ext4_clear_request_list();
2900                 kfree(ext4_li_info);
2901                 ext4_li_info = NULL;
2902                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2903                                  "initialization thread\n",
2904                                  err);
2905                 return err;
2906         }
2907         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2908         return 0;
2909 }
2910
2911 /*
2912  * Check whether it make sense to run itable init. thread or not.
2913  * If there is at least one uninitialized inode table, return
2914  * corresponding group number, else the loop goes through all
2915  * groups and return total number of groups.
2916  */
2917 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2918 {
2919         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2920         struct ext4_group_desc *gdp = NULL;
2921
2922         for (group = 0; group < ngroups; group++) {
2923                 gdp = ext4_get_group_desc(sb, group, NULL);
2924                 if (!gdp)
2925                         continue;
2926
2927                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2928                         break;
2929         }
2930
2931         return group;
2932 }
2933
2934 static int ext4_li_info_new(void)
2935 {
2936         struct ext4_lazy_init *eli = NULL;
2937
2938         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2939         if (!eli)
2940                 return -ENOMEM;
2941
2942         INIT_LIST_HEAD(&eli->li_request_list);
2943         mutex_init(&eli->li_list_mtx);
2944
2945         eli->li_state |= EXT4_LAZYINIT_QUIT;
2946
2947         ext4_li_info = eli;
2948
2949         return 0;
2950 }
2951
2952 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2953                                             ext4_group_t start)
2954 {
2955         struct ext4_sb_info *sbi = EXT4_SB(sb);
2956         struct ext4_li_request *elr;
2957         unsigned long rnd;
2958
2959         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2960         if (!elr)
2961                 return NULL;
2962
2963         elr->lr_super = sb;
2964         elr->lr_sbi = sbi;
2965         elr->lr_next_group = start;
2966
2967         /*
2968          * Randomize first schedule time of the request to
2969          * spread the inode table initialization requests
2970          * better.
2971          */
2972         get_random_bytes(&rnd, sizeof(rnd));
2973         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2974                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2975
2976         return elr;
2977 }
2978
2979 int ext4_register_li_request(struct super_block *sb,
2980                              ext4_group_t first_not_zeroed)
2981 {
2982         struct ext4_sb_info *sbi = EXT4_SB(sb);
2983         struct ext4_li_request *elr = NULL;
2984         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2985         int ret = 0;
2986
2987         mutex_lock(&ext4_li_mtx);
2988         if (sbi->s_li_request != NULL) {
2989                 /*
2990                  * Reset timeout so it can be computed again, because
2991                  * s_li_wait_mult might have changed.
2992                  */
2993                 sbi->s_li_request->lr_timeout = 0;
2994                 goto out;
2995         }
2996
2997         if (first_not_zeroed == ngroups ||
2998             (sb->s_flags & MS_RDONLY) ||
2999             !test_opt(sb, INIT_INODE_TABLE))
3000                 goto out;
3001
3002         elr = ext4_li_request_new(sb, first_not_zeroed);
3003         if (!elr) {
3004                 ret = -ENOMEM;
3005                 goto out;
3006         }
3007
3008         if (NULL == ext4_li_info) {
3009                 ret = ext4_li_info_new();
3010                 if (ret)
3011                         goto out;
3012         }
3013
3014         mutex_lock(&ext4_li_info->li_list_mtx);
3015         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3016         mutex_unlock(&ext4_li_info->li_list_mtx);
3017
3018         sbi->s_li_request = elr;
3019         /*
3020          * set elr to NULL here since it has been inserted to
3021          * the request_list and the removal and free of it is
3022          * handled by ext4_clear_request_list from now on.
3023          */
3024         elr = NULL;
3025
3026         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3027                 ret = ext4_run_lazyinit_thread();
3028                 if (ret)
3029                         goto out;
3030         }
3031 out:
3032         mutex_unlock(&ext4_li_mtx);
3033         if (ret)
3034                 kfree(elr);
3035         return ret;
3036 }
3037
3038 /*
3039  * We do not need to lock anything since this is called on
3040  * module unload.
3041  */
3042 static void ext4_destroy_lazyinit_thread(void)
3043 {
3044         /*
3045          * If thread exited earlier
3046          * there's nothing to be done.
3047          */
3048         if (!ext4_li_info || !ext4_lazyinit_task)
3049                 return;
3050
3051         kthread_stop(ext4_lazyinit_task);
3052 }
3053
3054 static int set_journal_csum_feature_set(struct super_block *sb)
3055 {
3056         int ret = 1;
3057         int compat, incompat;
3058         struct ext4_sb_info *sbi = EXT4_SB(sb);
3059
3060         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3061                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3062                 /* journal checksum v2 */
3063                 compat = 0;
3064                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3065         } else {
3066                 /* journal checksum v1 */
3067                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3068                 incompat = 0;
3069         }
3070
3071         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3072                 ret = jbd2_journal_set_features(sbi->s_journal,
3073                                 compat, 0,
3074                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3075                                 incompat);
3076         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3077                 ret = jbd2_journal_set_features(sbi->s_journal,
3078                                 compat, 0,
3079                                 incompat);
3080                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3081                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3082         } else {
3083                 jbd2_journal_clear_features(sbi->s_journal,
3084                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3085                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3086                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3087         }
3088
3089         return ret;
3090 }
3091
3092 /*
3093  * Note: calculating the overhead so we can be compatible with
3094  * historical BSD practice is quite difficult in the face of
3095  * clusters/bigalloc.  This is because multiple metadata blocks from
3096  * different block group can end up in the same allocation cluster.
3097  * Calculating the exact overhead in the face of clustered allocation
3098  * requires either O(all block bitmaps) in memory or O(number of block
3099  * groups**2) in time.  We will still calculate the superblock for
3100  * older file systems --- and if we come across with a bigalloc file
3101  * system with zero in s_overhead_clusters the estimate will be close to
3102  * correct especially for very large cluster sizes --- but for newer
3103  * file systems, it's better to calculate this figure once at mkfs
3104  * time, and store it in the superblock.  If the superblock value is
3105  * present (even for non-bigalloc file systems), we will use it.
3106  */
3107 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3108                           char *buf)
3109 {
3110         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3111         struct ext4_group_desc  *gdp;
3112         ext4_fsblk_t            first_block, last_block, b;
3113         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3114         int                     s, j, count = 0;
3115
3116         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3117                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3118                         sbi->s_itb_per_group + 2);
3119
3120         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3121                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3122         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3123         for (i = 0; i < ngroups; i++) {
3124                 gdp = ext4_get_group_desc(sb, i, NULL);
3125                 b = ext4_block_bitmap(sb, gdp);
3126                 if (b >= first_block && b <= last_block) {
3127                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3128                         count++;
3129                 }
3130                 b = ext4_inode_bitmap(sb, gdp);
3131                 if (b >= first_block && b <= last_block) {
3132                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3133                         count++;
3134                 }
3135                 b = ext4_inode_table(sb, gdp);
3136                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3137                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3138                                 int c = EXT4_B2C(sbi, b - first_block);
3139                                 ext4_set_bit(c, buf);
3140                                 count++;
3141                         }
3142                 if (i != grp)
3143                         continue;
3144                 s = 0;
3145                 if (ext4_bg_has_super(sb, grp)) {
3146                         ext4_set_bit(s++, buf);
3147                         count++;
3148                 }
3149                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3150                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3151                         count++;
3152                 }
3153         }
3154         if (!count)
3155                 return 0;
3156         return EXT4_CLUSTERS_PER_GROUP(sb) -
3157                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3158 }
3159
3160 /*
3161  * Compute the overhead and stash it in sbi->s_overhead
3162  */
3163 int ext4_calculate_overhead(struct super_block *sb)
3164 {
3165         struct ext4_sb_info *sbi = EXT4_SB(sb);
3166         struct ext4_super_block *es = sbi->s_es;
3167         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3168         ext4_fsblk_t overhead = 0;
3169         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3170
3171         if (!buf)
3172                 return -ENOMEM;
3173
3174         /*
3175          * Compute the overhead (FS structures).  This is constant
3176          * for a given filesystem unless the number of block groups
3177          * changes so we cache the previous value until it does.
3178          */
3179
3180         /*
3181          * All of the blocks before first_data_block are overhead
3182          */
3183         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3184
3185         /*
3186          * Add the overhead found in each block group
3187          */
3188         for (i = 0; i < ngroups; i++) {
3189                 int blks;
3190
3191                 blks = count_overhead(sb, i, buf);
3192                 overhead += blks;
3193                 if (blks)
3194                         memset(buf, 0, PAGE_SIZE);
3195                 cond_resched();
3196         }
3197         /* Add the journal blocks as well */
3198         if (sbi->s_journal)
3199                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3200
3201         sbi->s_overhead = overhead;
3202         smp_wmb();
3203         free_page((unsigned long) buf);
3204         return 0;
3205 }
3206
3207
3208 static ext4_fsblk_t ext4_calculate_resv_clusters(struct super_block *sb)
3209 {
3210         ext4_fsblk_t resv_clusters;
3211
3212         /*
3213          * There's no need to reserve anything when we aren't using extents.
3214          * The space estimates are exact, there are no unwritten extents,
3215          * hole punching doesn't need new metadata... This is needed especially
3216          * to keep ext2/3 backward compatibility.
3217          */
3218         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS))
3219                 return 0;
3220         /*
3221          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3222          * This should cover the situations where we can not afford to run
3223          * out of space like for example punch hole, or converting
3224          * uninitialized extents in delalloc path. In most cases such
3225          * allocation would require 1, or 2 blocks, higher numbers are
3226          * very rare.
3227          */
3228         resv_clusters = ext4_blocks_count(EXT4_SB(sb)->s_es) >>
3229                         EXT4_SB(sb)->s_cluster_bits;
3230
3231         do_div(resv_clusters, 50);
3232         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3233
3234         return resv_clusters;
3235 }
3236
3237
3238 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3239 {
3240         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3241                                 sbi->s_cluster_bits;
3242
3243         if (count >= clusters)
3244                 return -EINVAL;
3245
3246         atomic64_set(&sbi->s_resv_clusters, count);
3247         return 0;
3248 }
3249
3250 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3251 {
3252         char *orig_data = kstrdup(data, GFP_KERNEL);
3253         struct buffer_head *bh;
3254         struct ext4_super_block *es = NULL;
3255         struct ext4_sb_info *sbi;
3256         ext4_fsblk_t block;
3257         ext4_fsblk_t sb_block = get_sb_block(&data);
3258         ext4_fsblk_t logical_sb_block;
3259         unsigned long offset = 0;
3260         unsigned long journal_devnum = 0;
3261         unsigned long def_mount_opts;
3262         struct inode *root;
3263         char *cp;
3264         const char *descr;
3265         int ret = -ENOMEM;
3266         int blocksize, clustersize;
3267         unsigned int db_count;
3268         unsigned int i;
3269         int needs_recovery, has_huge_files, has_bigalloc;
3270         __u64 blocks_count;
3271         int err = 0;
3272         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3273         ext4_group_t first_not_zeroed;
3274
3275         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3276         if (!sbi)
3277                 goto out_free_orig;
3278
3279         sbi->s_blockgroup_lock =
3280                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3281         if (!sbi->s_blockgroup_lock) {
3282                 kfree(sbi);
3283                 goto out_free_orig;
3284         }
3285         sb->s_fs_info = sbi;
3286         sbi->s_sb = sb;
3287         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3288         sbi->s_sb_block = sb_block;
3289         if (sb->s_bdev->bd_part)
3290                 sbi->s_sectors_written_start =
3291                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3292
3293         /* Cleanup superblock name */
3294         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3295                 *cp = '!';
3296
3297         /* -EINVAL is default */
3298         ret = -EINVAL;
3299         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3300         if (!blocksize) {
3301                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3302                 goto out_fail;
3303         }
3304
3305         /*
3306          * The ext4 superblock will not be buffer aligned for other than 1kB
3307          * block sizes.  We need to calculate the offset from buffer start.
3308          */
3309         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3310                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3311                 offset = do_div(logical_sb_block, blocksize);
3312         } else {
3313                 logical_sb_block = sb_block;
3314         }
3315
3316         if (!(bh = sb_bread(sb, logical_sb_block))) {
3317                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3318                 goto out_fail;
3319         }
3320         /*
3321          * Note: s_es must be initialized as soon as possible because
3322          *       some ext4 macro-instructions depend on its value
3323          */
3324         es = (struct ext4_super_block *) (bh->b_data + offset);
3325         sbi->s_es = es;
3326         sb->s_magic = le16_to_cpu(es->s_magic);
3327         if (sb->s_magic != EXT4_SUPER_MAGIC)
3328                 goto cantfind_ext4;
3329         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3330
3331         /* Warn if metadata_csum and gdt_csum are both set. */
3332         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3333                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3334             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3335                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3336                              "redundant flags; please run fsck.");
3337
3338         /* Check for a known checksum algorithm */
3339         if (!ext4_verify_csum_type(sb, es)) {
3340                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3341                          "unknown checksum algorithm.");
3342                 silent = 1;
3343                 goto cantfind_ext4;
3344         }
3345
3346         /* Load the checksum driver */
3347         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3348                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3349                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3350                 if (IS_ERR(sbi->s_chksum_driver)) {
3351                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3352                         ret = PTR_ERR(sbi->s_chksum_driver);
3353                         sbi->s_chksum_driver = NULL;
3354                         goto failed_mount;
3355                 }
3356         }
3357
3358         /* Check superblock checksum */
3359         if (!ext4_superblock_csum_verify(sb, es)) {
3360                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3361                          "invalid superblock checksum.  Run e2fsck?");
3362                 silent = 1;
3363                 goto cantfind_ext4;
3364         }
3365
3366         /* Precompute checksum seed for all metadata */
3367         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3368                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3369                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3370                                                sizeof(es->s_uuid));
3371
3372         /* Set defaults before we parse the mount options */
3373         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3374         set_opt(sb, INIT_INODE_TABLE);
3375         if (def_mount_opts & EXT4_DEFM_DEBUG)
3376                 set_opt(sb, DEBUG);
3377         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3378                 set_opt(sb, GRPID);
3379         if (def_mount_opts & EXT4_DEFM_UID16)
3380                 set_opt(sb, NO_UID32);
3381         /* xattr user namespace & acls are now defaulted on */
3382         set_opt(sb, XATTR_USER);
3383 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3384         set_opt(sb, POSIX_ACL);
3385 #endif
3386         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3387                 set_opt(sb, JOURNAL_DATA);
3388         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3389                 set_opt(sb, ORDERED_DATA);
3390         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3391                 set_opt(sb, WRITEBACK_DATA);
3392
3393         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3394                 set_opt(sb, ERRORS_PANIC);
3395         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3396                 set_opt(sb, ERRORS_CONT);
3397         else
3398                 set_opt(sb, ERRORS_RO);
3399         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3400                 set_opt(sb, BLOCK_VALIDITY);
3401         if (def_mount_opts & EXT4_DEFM_DISCARD)
3402                 set_opt(sb, DISCARD);
3403
3404         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3405         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3406         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3407         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3408         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3409
3410         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3411                 set_opt(sb, BARRIER);
3412
3413         /*
3414          * enable delayed allocation by default
3415          * Use -o nodelalloc to turn it off
3416          */
3417         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3418             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3419                 set_opt(sb, DELALLOC);
3420
3421         /*
3422          * set default s_li_wait_mult for lazyinit, for the case there is
3423          * no mount option specified.
3424          */
3425         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3426
3427         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3428                            &journal_devnum, &journal_ioprio, 0)) {
3429                 ext4_msg(sb, KERN_WARNING,
3430                          "failed to parse options in superblock: %s",
3431                          sbi->s_es->s_mount_opts);
3432         }
3433         sbi->s_def_mount_opt = sbi->s_mount_opt;
3434         if (!parse_options((char *) data, sb, &journal_devnum,
3435                            &journal_ioprio, 0))
3436                 goto failed_mount;
3437
3438         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3439                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3440                             "with data=journal disables delayed "
3441                             "allocation and O_DIRECT support!\n");
3442                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3443                         ext4_msg(sb, KERN_ERR, "can't mount with "
3444                                  "both data=journal and delalloc");
3445                         goto failed_mount;
3446                 }
3447                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3448                         ext4_msg(sb, KERN_ERR, "can't mount with "
3449                                  "both data=journal and dioread_nolock");
3450                         goto failed_mount;
3451                 }
3452                 if (test_opt(sb, DELALLOC))
3453                         clear_opt(sb, DELALLOC);
3454         }
3455
3456         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3457                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3458
3459         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3460             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3461              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3462              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3463                 ext4_msg(sb, KERN_WARNING,
3464                        "feature flags set on rev 0 fs, "
3465                        "running e2fsck is recommended");
3466
3467         if (IS_EXT2_SB(sb)) {
3468                 if (ext2_feature_set_ok(sb))
3469                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3470                                  "using the ext4 subsystem");
3471                 else {
3472                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3473                                  "to feature incompatibilities");
3474                         goto failed_mount;
3475                 }
3476         }
3477
3478         if (IS_EXT3_SB(sb)) {
3479                 if (ext3_feature_set_ok(sb))
3480                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3481                                  "using the ext4 subsystem");
3482                 else {
3483                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3484                                  "to feature incompatibilities");
3485                         goto failed_mount;
3486                 }
3487         }
3488
3489         /*
3490          * Check feature flags regardless of the revision level, since we
3491          * previously didn't change the revision level when setting the flags,
3492          * so there is a chance incompat flags are set on a rev 0 filesystem.
3493          */
3494         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3495                 goto failed_mount;
3496
3497         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3498         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3499             blocksize > EXT4_MAX_BLOCK_SIZE) {
3500                 ext4_msg(sb, KERN_ERR,
3501                        "Unsupported filesystem blocksize %d", blocksize);
3502                 goto failed_mount;
3503         }
3504
3505         if (sb->s_blocksize != blocksize) {
3506                 /* Validate the filesystem blocksize */
3507                 if (!sb_set_blocksize(sb, blocksize)) {
3508                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3509                                         blocksize);
3510                         goto failed_mount;
3511                 }
3512
3513                 brelse(bh);
3514                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3515                 offset = do_div(logical_sb_block, blocksize);
3516                 bh = sb_bread(sb, logical_sb_block);
3517                 if (!bh) {
3518                         ext4_msg(sb, KERN_ERR,
3519                                "Can't read superblock on 2nd try");
3520                         goto failed_mount;
3521                 }
3522                 es = (struct ext4_super_block *)(bh->b_data + offset);
3523                 sbi->s_es = es;
3524                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3525                         ext4_msg(sb, KERN_ERR,
3526                                "Magic mismatch, very weird!");
3527                         goto failed_mount;
3528                 }
3529         }
3530
3531         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3532                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3533         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3534                                                       has_huge_files);
3535         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3536
3537         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3538                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3539                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3540         } else {
3541                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3542                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3543                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3544                     (!is_power_of_2(sbi->s_inode_size)) ||
3545                     (sbi->s_inode_size > blocksize)) {
3546                         ext4_msg(sb, KERN_ERR,
3547                                "unsupported inode size: %d",
3548                                sbi->s_inode_size);
3549                         goto failed_mount;
3550                 }
3551                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3552                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3553         }
3554
3555         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3556         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3557                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3558                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3559                     !is_power_of_2(sbi->s_desc_size)) {
3560                         ext4_msg(sb, KERN_ERR,
3561                                "unsupported descriptor size %lu",
3562                                sbi->s_desc_size);
3563                         goto failed_mount;
3564                 }
3565         } else
3566                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3567
3568         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3569         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3570         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3571                 goto cantfind_ext4;
3572
3573         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3574         if (sbi->s_inodes_per_block == 0)
3575                 goto cantfind_ext4;
3576         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3577                                         sbi->s_inodes_per_block;
3578         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3579         sbi->s_sbh = bh;
3580         sbi->s_mount_state = le16_to_cpu(es->s_state);
3581         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3582         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3583
3584         for (i = 0; i < 4; i++)
3585                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3586         sbi->s_def_hash_version = es->s_def_hash_version;
3587         if (EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_DIR_INDEX)) {
3588                 i = le32_to_cpu(es->s_flags);
3589                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3590                         sbi->s_hash_unsigned = 3;
3591                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3592 #ifdef __CHAR_UNSIGNED__
3593                         if (!(sb->s_flags & MS_RDONLY))
3594                                 es->s_flags |=
3595                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3596                         sbi->s_hash_unsigned = 3;
3597 #else
3598                         if (!(sb->s_flags & MS_RDONLY))
3599                                 es->s_flags |=
3600                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3601 #endif
3602                 }
3603         }
3604
3605         /* Handle clustersize */
3606         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3607         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3608                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3609         if (has_bigalloc) {
3610                 if (clustersize < blocksize) {
3611                         ext4_msg(sb, KERN_ERR,
3612                                  "cluster size (%d) smaller than "
3613                                  "block size (%d)", clustersize, blocksize);
3614                         goto failed_mount;
3615                 }
3616                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3617                         le32_to_cpu(es->s_log_block_size);
3618                 sbi->s_clusters_per_group =
3619                         le32_to_cpu(es->s_clusters_per_group);
3620                 if (sbi->s_clusters_per_group > blocksize * 8) {
3621                         ext4_msg(sb, KERN_ERR,
3622                                  "#clusters per group too big: %lu",
3623                                  sbi->s_clusters_per_group);
3624                         goto failed_mount;
3625                 }
3626                 if (sbi->s_blocks_per_group !=
3627                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3628                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3629                                  "clusters per group (%lu) inconsistent",
3630                                  sbi->s_blocks_per_group,
3631                                  sbi->s_clusters_per_group);
3632                         goto failed_mount;
3633                 }
3634         } else {
3635                 if (clustersize != blocksize) {
3636                         ext4_warning(sb, "fragment/cluster size (%d) != "
3637                                      "block size (%d)", clustersize,
3638                                      blocksize);
3639                         clustersize = blocksize;
3640                 }
3641                 if (sbi->s_blocks_per_group > blocksize * 8) {
3642                         ext4_msg(sb, KERN_ERR,
3643                                  "#blocks per group too big: %lu",
3644                                  sbi->s_blocks_per_group);
3645                         goto failed_mount;
3646                 }
3647                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3648                 sbi->s_cluster_bits = 0;
3649         }
3650         sbi->s_cluster_ratio = clustersize / blocksize;
3651
3652         if (sbi->s_inodes_per_group > blocksize * 8) {
3653                 ext4_msg(sb, KERN_ERR,
3654                        "#inodes per group too big: %lu",
3655                        sbi->s_inodes_per_group);
3656                 goto failed_mount;
3657         }
3658
3659         /* Do we have standard group size of clustersize * 8 blocks ? */
3660         if (sbi->s_blocks_per_group == clustersize << 3)
3661                 set_opt2(sb, STD_GROUP_SIZE);
3662
3663         /*
3664          * Test whether we have more sectors than will fit in sector_t,
3665          * and whether the max offset is addressable by the page cache.
3666          */
3667         err = generic_check_addressable(sb->s_blocksize_bits,
3668                                         ext4_blocks_count(es));
3669         if (err) {
3670                 ext4_msg(sb, KERN_ERR, "filesystem"
3671                          " too large to mount safely on this system");
3672                 if (sizeof(sector_t) < 8)
3673                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3674                 goto failed_mount;
3675         }
3676
3677         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3678                 goto cantfind_ext4;
3679
3680         /* check blocks count against device size */
3681         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3682         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3683                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3684                        "exceeds size of device (%llu blocks)",
3685                        ext4_blocks_count(es), blocks_count);
3686                 goto failed_mount;
3687         }
3688
3689         /*
3690          * It makes no sense for the first data block to be beyond the end
3691          * of the filesystem.
3692          */
3693         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3694                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3695                          "block %u is beyond end of filesystem (%llu)",
3696                          le32_to_cpu(es->s_first_data_block),
3697                          ext4_blocks_count(es));
3698                 goto failed_mount;
3699         }
3700         blocks_count = (ext4_blocks_count(es) -
3701                         le32_to_cpu(es->s_first_data_block) +
3702                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3703         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3704         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3705                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3706                        "(block count %llu, first data block %u, "
3707                        "blocks per group %lu)", sbi->s_groups_count,
3708                        ext4_blocks_count(es),
3709                        le32_to_cpu(es->s_first_data_block),
3710                        EXT4_BLOCKS_PER_GROUP(sb));
3711                 goto failed_mount;
3712         }
3713         sbi->s_groups_count = blocks_count;
3714         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3715                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3716         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3717                    EXT4_DESC_PER_BLOCK(sb);
3718         sbi->s_group_desc = ext4_kvmalloc(db_count *
3719                                           sizeof(struct buffer_head *),
3720                                           GFP_KERNEL);
3721         if (sbi->s_group_desc == NULL) {
3722                 ext4_msg(sb, KERN_ERR, "not enough memory");
3723                 ret = -ENOMEM;
3724                 goto failed_mount;
3725         }
3726
3727         if (ext4_proc_root)
3728                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3729
3730         if (sbi->s_proc)
3731                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3732                                  &ext4_seq_options_fops, sb);
3733
3734         bgl_lock_init(sbi->s_blockgroup_lock);
3735
3736         for (i = 0; i < db_count; i++) {
3737                 block = descriptor_loc(sb, logical_sb_block, i);
3738                 sbi->s_group_desc[i] = sb_bread(sb, block);
3739                 if (!sbi->s_group_desc[i]) {
3740                         ext4_msg(sb, KERN_ERR,
3741                                "can't read group descriptor %d", i);
3742                         db_count = i;
3743                         goto failed_mount2;
3744                 }
3745         }
3746         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3747                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3748                 goto failed_mount2;
3749         }
3750         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3751                 if (!ext4_fill_flex_info(sb)) {
3752                         ext4_msg(sb, KERN_ERR,
3753                                "unable to initialize "
3754                                "flex_bg meta info!");
3755                         goto failed_mount2;
3756                 }
3757
3758         sbi->s_gdb_count = db_count;
3759         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3760         spin_lock_init(&sbi->s_next_gen_lock);
3761
3762         init_timer(&sbi->s_err_report);
3763         sbi->s_err_report.function = print_daily_error_info;
3764         sbi->s_err_report.data = (unsigned long) sb;
3765
3766         /* Register extent status tree shrinker */
3767         ext4_es_register_shrinker(sb);
3768
3769         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3770                         ext4_count_free_clusters(sb));
3771         if (!err) {
3772                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3773                                 ext4_count_free_inodes(sb));
3774         }
3775         if (!err) {
3776                 err = percpu_counter_init(&sbi->s_dirs_counter,
3777                                 ext4_count_dirs(sb));
3778         }
3779         if (!err) {
3780                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3781         }
3782         if (!err) {
3783                 err = percpu_counter_init(&sbi->s_extent_cache_cnt, 0);
3784         }
3785         if (err) {
3786                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3787                 goto failed_mount3;
3788         }
3789
3790         sbi->s_stripe = ext4_get_stripe_size(sbi);
3791         sbi->s_max_writeback_mb_bump = 128;
3792         sbi->s_extent_max_zeroout_kb = 32;
3793
3794         /*
3795          * set up enough so that it can read an inode
3796          */
3797         if (!test_opt(sb, NOLOAD) &&
3798             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3799                 sb->s_op = &ext4_sops;
3800         else
3801                 sb->s_op = &ext4_nojournal_sops;
3802         sb->s_export_op = &ext4_export_ops;
3803         sb->s_xattr = ext4_xattr_handlers;
3804 #ifdef CONFIG_QUOTA
3805         sb->dq_op = &ext4_quota_operations;
3806         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3807                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3808         else
3809                 sb->s_qcop = &ext4_qctl_operations;
3810 #endif
3811         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3812
3813         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3814         mutex_init(&sbi->s_orphan_lock);
3815
3816         sb->s_root = NULL;
3817
3818         needs_recovery = (es->s_last_orphan != 0 ||
3819                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3820                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3821
3822         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3823             !(sb->s_flags & MS_RDONLY))
3824                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3825                         goto failed_mount3;
3826
3827         /*
3828          * The first inode we look at is the journal inode.  Don't try
3829          * root first: it may be modified in the journal!
3830          */
3831         if (!test_opt(sb, NOLOAD) &&
3832             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3833                 if (ext4_load_journal(sb, es, journal_devnum))
3834                         goto failed_mount3;
3835         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3836               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3837                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3838                        "suppressed and not mounted read-only");
3839                 goto failed_mount_wq;
3840         } else {
3841                 clear_opt(sb, DATA_FLAGS);
3842                 sbi->s_journal = NULL;
3843                 needs_recovery = 0;
3844                 goto no_journal;
3845         }
3846
3847         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3848             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3849                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3850                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3851                 goto failed_mount_wq;
3852         }
3853
3854         if (!set_journal_csum_feature_set(sb)) {
3855                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3856                          "feature set");
3857                 goto failed_mount_wq;
3858         }
3859
3860         /* We have now updated the journal if required, so we can
3861          * validate the data journaling mode. */
3862         switch (test_opt(sb, DATA_FLAGS)) {
3863         case 0:
3864                 /* No mode set, assume a default based on the journal
3865                  * capabilities: ORDERED_DATA if the journal can
3866                  * cope, else JOURNAL_DATA
3867                  */
3868                 if (jbd2_journal_check_available_features
3869                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3870                         set_opt(sb, ORDERED_DATA);
3871                 else
3872                         set_opt(sb, JOURNAL_DATA);
3873                 break;
3874
3875         case EXT4_MOUNT_ORDERED_DATA:
3876         case EXT4_MOUNT_WRITEBACK_DATA:
3877                 if (!jbd2_journal_check_available_features
3878                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3879                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3880                                "requested data journaling mode");
3881                         goto failed_mount_wq;
3882                 }
3883         default:
3884                 break;
3885         }
3886         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3887
3888         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3889
3890         /*
3891          * The journal may have updated the bg summary counts, so we
3892          * need to update the global counters.
3893          */
3894         percpu_counter_set(&sbi->s_freeclusters_counter,
3895                            ext4_count_free_clusters(sb));
3896         percpu_counter_set(&sbi->s_freeinodes_counter,
3897                            ext4_count_free_inodes(sb));
3898         percpu_counter_set(&sbi->s_dirs_counter,
3899                            ext4_count_dirs(sb));
3900         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3901
3902 no_journal:
3903         /*
3904          * Get the # of file system overhead blocks from the
3905          * superblock if present.
3906          */
3907         if (es->s_overhead_clusters)
3908                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3909         else {
3910                 err = ext4_calculate_overhead(sb);
3911                 if (err)
3912                         goto failed_mount_wq;
3913         }
3914
3915         /*
3916          * The maximum number of concurrent works can be high and
3917          * concurrency isn't really necessary.  Limit it to 1.
3918          */
3919         EXT4_SB(sb)->dio_unwritten_wq =
3920                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3921         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3922                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3923                 ret = -ENOMEM;
3924                 goto failed_mount_wq;
3925         }
3926
3927         /*
3928          * The jbd2_journal_load will have done any necessary log recovery,
3929          * so we can safely mount the rest of the filesystem now.
3930          */
3931
3932         root = ext4_iget(sb, EXT4_ROOT_INO);
3933         if (IS_ERR(root)) {
3934                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3935                 ret = PTR_ERR(root);
3936                 root = NULL;
3937                 goto failed_mount4;
3938         }
3939         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3940                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3941                 iput(root);
3942                 goto failed_mount4;
3943         }
3944         sb->s_root = d_make_root(root);
3945         if (!sb->s_root) {
3946                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3947                 ret = -ENOMEM;
3948                 goto failed_mount4;
3949         }
3950
3951         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3952                 sb->s_flags |= MS_RDONLY;
3953
3954         /* determine the minimum size of new large inodes, if present */
3955         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3956                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3957                                                      EXT4_GOOD_OLD_INODE_SIZE;
3958                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3959                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3960                         if (sbi->s_want_extra_isize <
3961                             le16_to_cpu(es->s_want_extra_isize))
3962                                 sbi->s_want_extra_isize =
3963                                         le16_to_cpu(es->s_want_extra_isize);
3964                         if (sbi->s_want_extra_isize <
3965                             le16_to_cpu(es->s_min_extra_isize))
3966                                 sbi->s_want_extra_isize =
3967                                         le16_to_cpu(es->s_min_extra_isize);
3968                 }
3969         }
3970         /* Check if enough inode space is available */
3971         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3972                                                         sbi->s_inode_size) {
3973                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3974                                                        EXT4_GOOD_OLD_INODE_SIZE;
3975                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3976                          "available");
3977         }
3978
3979         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sb));
3980         if (err) {
3981                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
3982                          "reserved pool", ext4_calculate_resv_clusters(sb));
3983                 goto failed_mount4a;
3984         }
3985
3986         err = ext4_setup_system_zone(sb);
3987         if (err) {
3988                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3989                          "zone (%d)", err);
3990                 goto failed_mount4a;
3991         }
3992
3993         ext4_ext_init(sb);
3994         err = ext4_mb_init(sb);
3995         if (err) {
3996                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3997                          err);
3998                 goto failed_mount5;
3999         }
4000
4001         err = ext4_register_li_request(sb, first_not_zeroed);
4002         if (err)
4003                 goto failed_mount6;
4004
4005         sbi->s_kobj.kset = ext4_kset;
4006         init_completion(&sbi->s_kobj_unregister);
4007         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4008                                    "%s", sb->s_id);
4009         if (err)
4010                 goto failed_mount7;
4011
4012 #ifdef CONFIG_QUOTA
4013         /* Enable quota usage during mount. */
4014         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4015             !(sb->s_flags & MS_RDONLY)) {
4016                 err = ext4_enable_quotas(sb);
4017                 if (err)
4018                         goto failed_mount8;
4019         }
4020 #endif  /* CONFIG_QUOTA */
4021
4022         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4023         ext4_orphan_cleanup(sb, es);
4024         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4025         if (needs_recovery) {
4026                 ext4_msg(sb, KERN_INFO, "recovery complete");
4027                 ext4_mark_recovery_complete(sb, es);
4028         }
4029         if (EXT4_SB(sb)->s_journal) {
4030                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4031                         descr = " journalled data mode";
4032                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4033                         descr = " ordered data mode";
4034                 else
4035                         descr = " writeback data mode";
4036         } else
4037                 descr = "out journal";
4038
4039         if (test_opt(sb, DISCARD)) {
4040                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4041                 if (!blk_queue_discard(q))
4042                         ext4_msg(sb, KERN_WARNING,
4043                                  "mounting with \"discard\" option, but "
4044                                  "the device does not support discard");
4045         }
4046
4047         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4048                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4049                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4050
4051         if (es->s_error_count)
4052                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4053
4054         kfree(orig_data);
4055         return 0;
4056
4057 cantfind_ext4:
4058         if (!silent)
4059                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4060         goto failed_mount;
4061
4062 #ifdef CONFIG_QUOTA
4063 failed_mount8:
4064         kobject_del(&sbi->s_kobj);
4065 #endif
4066 failed_mount7:
4067         ext4_unregister_li_request(sb);
4068 failed_mount6:
4069         ext4_mb_release(sb);
4070 failed_mount5:
4071         ext4_ext_release(sb);
4072         ext4_release_system_zone(sb);
4073 failed_mount4a:
4074         dput(sb->s_root);
4075         sb->s_root = NULL;
4076 failed_mount4:
4077         ext4_msg(sb, KERN_ERR, "mount failed");
4078         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4079 failed_mount_wq:
4080         if (sbi->s_journal) {
4081                 jbd2_journal_destroy(sbi->s_journal);
4082                 sbi->s_journal = NULL;
4083         }
4084 failed_mount3:
4085         ext4_es_unregister_shrinker(sb);
4086         del_timer(&sbi->s_err_report);
4087         if (sbi->s_flex_groups)
4088                 ext4_kvfree(sbi->s_flex_groups);
4089         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4090         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4091         percpu_counter_destroy(&sbi->s_dirs_counter);
4092         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4093         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
4094         if (sbi->s_mmp_tsk)
4095                 kthread_stop(sbi->s_mmp_tsk);
4096 failed_mount2:
4097         for (i = 0; i < db_count; i++)
4098                 brelse(sbi->s_group_desc[i]);
4099         ext4_kvfree(sbi->s_group_desc);
4100 failed_mount:
4101         if (sbi->s_chksum_driver)
4102                 crypto_free_shash(sbi->s_chksum_driver);
4103         if (sbi->s_proc) {
4104                 remove_proc_entry("options", sbi->s_proc);
4105                 remove_proc_entry(sb->s_id, ext4_proc_root);
4106         }
4107 #ifdef CONFIG_QUOTA
4108         for (i = 0; i < MAXQUOTAS; i++)
4109                 kfree(sbi->s_qf_names[i]);
4110 #endif
4111         ext4_blkdev_remove(sbi);
4112         brelse(bh);
4113 out_fail:
4114         sb->s_fs_info = NULL;
4115         kfree(sbi->s_blockgroup_lock);
4116         kfree(sbi);
4117 out_free_orig:
4118         kfree(orig_data);
4119         return err ? err : ret;
4120 }
4121
4122 /*
4123  * Setup any per-fs journal parameters now.  We'll do this both on
4124  * initial mount, once the journal has been initialised but before we've
4125  * done any recovery; and again on any subsequent remount.
4126  */
4127 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4128 {
4129         struct ext4_sb_info *sbi = EXT4_SB(sb);
4130
4131         journal->j_commit_interval = sbi->s_commit_interval;
4132         journal->j_min_batch_time = sbi->s_min_batch_time;
4133         journal->j_max_batch_time = sbi->s_max_batch_time;
4134
4135         write_lock(&journal->j_state_lock);
4136         if (test_opt(sb, BARRIER))
4137                 journal->j_flags |= JBD2_BARRIER;
4138         else
4139                 journal->j_flags &= ~JBD2_BARRIER;
4140         if (test_opt(sb, DATA_ERR_ABORT))
4141                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4142         else
4143                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4144         write_unlock(&journal->j_state_lock);
4145 }
4146
4147 static journal_t *ext4_get_journal(struct super_block *sb,
4148                                    unsigned int journal_inum)
4149 {
4150         struct inode *journal_inode;
4151         journal_t *journal;
4152
4153         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4154
4155         /* First, test for the existence of a valid inode on disk.  Bad
4156          * things happen if we iget() an unused inode, as the subsequent
4157          * iput() will try to delete it. */
4158
4159         journal_inode = ext4_iget(sb, journal_inum);
4160         if (IS_ERR(journal_inode)) {
4161                 ext4_msg(sb, KERN_ERR, "no journal found");
4162                 return NULL;
4163         }
4164         if (!journal_inode->i_nlink) {
4165                 make_bad_inode(journal_inode);
4166                 iput(journal_inode);
4167                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4168                 return NULL;
4169         }
4170
4171         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4172                   journal_inode, journal_inode->i_size);
4173         if (!S_ISREG(journal_inode->i_mode)) {
4174                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4175                 iput(journal_inode);
4176                 return NULL;
4177         }
4178
4179         journal = jbd2_journal_init_inode(journal_inode);
4180         if (!journal) {
4181                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4182                 iput(journal_inode);
4183                 return NULL;
4184         }
4185         journal->j_private = sb;
4186         ext4_init_journal_params(sb, journal);
4187         return journal;
4188 }
4189
4190 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4191                                        dev_t j_dev)
4192 {
4193         struct buffer_head *bh;
4194         journal_t *journal;
4195         ext4_fsblk_t start;
4196         ext4_fsblk_t len;
4197         int hblock, blocksize;
4198         ext4_fsblk_t sb_block;
4199         unsigned long offset;
4200         struct ext4_super_block *es;
4201         struct block_device *bdev;
4202
4203         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4204
4205         bdev = ext4_blkdev_get(j_dev, sb);
4206         if (bdev == NULL)
4207                 return NULL;
4208
4209         blocksize = sb->s_blocksize;
4210         hblock = bdev_logical_block_size(bdev);
4211         if (blocksize < hblock) {
4212                 ext4_msg(sb, KERN_ERR,
4213                         "blocksize too small for journal device");
4214                 goto out_bdev;
4215         }
4216
4217         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4218         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4219         set_blocksize(bdev, blocksize);
4220         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4221                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4222                        "external journal");
4223                 goto out_bdev;
4224         }
4225
4226         es = (struct ext4_super_block *) (bh->b_data + offset);
4227         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4228             !(le32_to_cpu(es->s_feature_incompat) &
4229               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4230                 ext4_msg(sb, KERN_ERR, "external journal has "
4231                                         "bad superblock");
4232                 brelse(bh);
4233                 goto out_bdev;
4234         }
4235
4236         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4237                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4238                 brelse(bh);
4239                 goto out_bdev;
4240         }
4241
4242         len = ext4_blocks_count(es);
4243         start = sb_block + 1;
4244         brelse(bh);     /* we're done with the superblock */
4245
4246         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4247                                         start, len, blocksize);
4248         if (!journal) {
4249                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4250                 goto out_bdev;
4251         }
4252         journal->j_private = sb;
4253         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4254         wait_on_buffer(journal->j_sb_buffer);
4255         if (!buffer_uptodate(journal->j_sb_buffer)) {
4256                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4257                 goto out_journal;
4258         }
4259         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4260                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4261                                         "user (unsupported) - %d",
4262                         be32_to_cpu(journal->j_superblock->s_nr_users));
4263                 goto out_journal;
4264         }
4265         EXT4_SB(sb)->journal_bdev = bdev;
4266         ext4_init_journal_params(sb, journal);
4267         return journal;
4268
4269 out_journal:
4270         jbd2_journal_destroy(journal);
4271 out_bdev:
4272         ext4_blkdev_put(bdev);
4273         return NULL;
4274 }
4275
4276 static int ext4_load_journal(struct super_block *sb,
4277                              struct ext4_super_block *es,
4278                              unsigned long journal_devnum)
4279 {
4280         journal_t *journal;
4281         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4282         dev_t journal_dev;
4283         int err = 0;
4284         int really_read_only;
4285
4286         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4287
4288         if (journal_devnum &&
4289             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4290                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4291                         "numbers have changed");
4292                 journal_dev = new_decode_dev(journal_devnum);
4293         } else
4294                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4295
4296         really_read_only = bdev_read_only(sb->s_bdev);
4297
4298         /*
4299          * Are we loading a blank journal or performing recovery after a
4300          * crash?  For recovery, we need to check in advance whether we
4301          * can get read-write access to the device.
4302          */
4303         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4304                 if (sb->s_flags & MS_RDONLY) {
4305                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4306                                         "required on readonly filesystem");
4307                         if (really_read_only) {
4308                                 ext4_msg(sb, KERN_ERR, "write access "
4309                                         "unavailable, cannot proceed");
4310                                 return -EROFS;
4311                         }
4312                         ext4_msg(sb, KERN_INFO, "write access will "
4313                                "be enabled during recovery");
4314                 }
4315         }
4316
4317         if (journal_inum && journal_dev) {
4318                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4319                        "and inode journals!");
4320                 return -EINVAL;
4321         }
4322
4323         if (journal_inum) {
4324                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4325                         return -EINVAL;
4326         } else {
4327                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4328                         return -EINVAL;
4329         }
4330
4331         if (!(journal->j_flags & JBD2_BARRIER))
4332                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4333
4334         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4335                 err = jbd2_journal_wipe(journal, !really_read_only);
4336         if (!err) {
4337                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4338                 if (save)
4339                         memcpy(save, ((char *) es) +
4340                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4341                 err = jbd2_journal_load(journal);
4342                 if (save)
4343                         memcpy(((char *) es) + EXT4_S_ERR_START,
4344                                save, EXT4_S_ERR_LEN);
4345                 kfree(save);
4346         }
4347
4348         if (err) {
4349                 ext4_msg(sb, KERN_ERR, "error loading journal");
4350                 jbd2_journal_destroy(journal);
4351                 return err;
4352         }
4353
4354         EXT4_SB(sb)->s_journal = journal;
4355         ext4_clear_journal_err(sb, es);
4356
4357         if (!really_read_only && journal_devnum &&
4358             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4359                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4360
4361                 /* Make sure we flush the recovery flag to disk. */
4362                 ext4_commit_super(sb, 1);
4363         }
4364
4365         return 0;
4366 }
4367
4368 static int ext4_commit_super(struct super_block *sb, int sync)
4369 {
4370         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4371         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4372         int error = 0;
4373
4374         if (!sbh || block_device_ejected(sb))
4375                 return error;
4376         if (buffer_write_io_error(sbh)) {
4377                 /*
4378                  * Oh, dear.  A previous attempt to write the
4379                  * superblock failed.  This could happen because the
4380                  * USB device was yanked out.  Or it could happen to
4381                  * be a transient write error and maybe the block will
4382                  * be remapped.  Nothing we can do but to retry the
4383                  * write and hope for the best.
4384                  */
4385                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4386                        "superblock detected");
4387                 clear_buffer_write_io_error(sbh);
4388                 set_buffer_uptodate(sbh);
4389         }
4390         /*
4391          * If the file system is mounted read-only, don't update the
4392          * superblock write time.  This avoids updating the superblock
4393          * write time when we are mounting the root file system
4394          * read/only but we need to replay the journal; at that point,
4395          * for people who are east of GMT and who make their clock
4396          * tick in localtime for Windows bug-for-bug compatibility,
4397          * the clock is set in the future, and this will cause e2fsck
4398          * to complain and force a full file system check.
4399          */
4400         if (!(sb->s_flags & MS_RDONLY))
4401                 es->s_wtime = cpu_to_le32(get_seconds());
4402         if (sb->s_bdev->bd_part)
4403                 es->s_kbytes_written =
4404                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4405                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4406                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4407         else
4408                 es->s_kbytes_written =
4409                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4410         ext4_free_blocks_count_set(es,
4411                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4412                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4413         es->s_free_inodes_count =
4414                 cpu_to_le32(percpu_counter_sum_positive(
4415                                 &EXT4_SB(sb)->s_freeinodes_counter));
4416         BUFFER_TRACE(sbh, "marking dirty");
4417         ext4_superblock_csum_set(sb);
4418         mark_buffer_dirty(sbh);
4419         if (sync) {
4420                 error = sync_dirty_buffer(sbh);
4421                 if (error)
4422                         return error;
4423
4424                 error = buffer_write_io_error(sbh);
4425                 if (error) {
4426                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4427                                "superblock");
4428                         clear_buffer_write_io_error(sbh);
4429                         set_buffer_uptodate(sbh);
4430                 }
4431         }
4432         return error;
4433 }
4434
4435 /*
4436  * Have we just finished recovery?  If so, and if we are mounting (or
4437  * remounting) the filesystem readonly, then we will end up with a
4438  * consistent fs on disk.  Record that fact.
4439  */
4440 static void ext4_mark_recovery_complete(struct super_block *sb,
4441                                         struct ext4_super_block *es)
4442 {
4443         journal_t *journal = EXT4_SB(sb)->s_journal;
4444
4445         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4446                 BUG_ON(journal != NULL);
4447                 return;
4448         }
4449         jbd2_journal_lock_updates(journal);
4450         if (jbd2_journal_flush(journal) < 0)
4451                 goto out;
4452
4453         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4454             sb->s_flags & MS_RDONLY) {
4455                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4456                 ext4_commit_super(sb, 1);
4457         }
4458
4459 out:
4460         jbd2_journal_unlock_updates(journal);
4461 }
4462
4463 /*
4464  * If we are mounting (or read-write remounting) a filesystem whose journal
4465  * has recorded an error from a previous lifetime, move that error to the
4466  * main filesystem now.
4467  */
4468 static void ext4_clear_journal_err(struct super_block *sb,
4469                                    struct ext4_super_block *es)
4470 {
4471         journal_t *journal;
4472         int j_errno;
4473         const char *errstr;
4474
4475         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4476
4477         journal = EXT4_SB(sb)->s_journal;
4478
4479         /*
4480          * Now check for any error status which may have been recorded in the
4481          * journal by a prior ext4_error() or ext4_abort()
4482          */
4483
4484         j_errno = jbd2_journal_errno(journal);
4485         if (j_errno) {
4486                 char nbuf[16];
4487
4488                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4489                 ext4_warning(sb, "Filesystem error recorded "
4490                              "from previous mount: %s", errstr);
4491                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4492
4493                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4494                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4495                 ext4_commit_super(sb, 1);
4496
4497                 jbd2_journal_clear_err(journal);
4498                 jbd2_journal_update_sb_errno(journal);
4499         }
4500 }
4501
4502 /*
4503  * Force the running and committing transactions to commit,
4504  * and wait on the commit.
4505  */
4506 int ext4_force_commit(struct super_block *sb)
4507 {
4508         journal_t *journal;
4509
4510         if (sb->s_flags & MS_RDONLY)
4511                 return 0;
4512
4513         journal = EXT4_SB(sb)->s_journal;
4514         return ext4_journal_force_commit(journal);
4515 }
4516
4517 static int ext4_sync_fs(struct super_block *sb, int wait)
4518 {
4519         int ret = 0;
4520         tid_t target;
4521         struct ext4_sb_info *sbi = EXT4_SB(sb);
4522
4523         trace_ext4_sync_fs(sb, wait);
4524         flush_workqueue(sbi->dio_unwritten_wq);
4525         /*
4526          * Writeback quota in non-journalled quota case - journalled quota has
4527          * no dirty dquots
4528          */
4529         dquot_writeback_dquots(sb, -1);
4530         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4531                 if (wait)
4532                         jbd2_log_wait_commit(sbi->s_journal, target);
4533         }
4534         return ret;
4535 }
4536
4537 /*
4538  * LVM calls this function before a (read-only) snapshot is created.  This
4539  * gives us a chance to flush the journal completely and mark the fs clean.
4540  *
4541  * Note that only this function cannot bring a filesystem to be in a clean
4542  * state independently. It relies on upper layer to stop all data & metadata
4543  * modifications.
4544  */
4545 static int ext4_freeze(struct super_block *sb)
4546 {
4547         int error = 0;
4548         journal_t *journal;
4549
4550         if (sb->s_flags & MS_RDONLY)
4551                 return 0;
4552
4553         journal = EXT4_SB(sb)->s_journal;
4554
4555         /* Now we set up the journal barrier. */
4556         jbd2_journal_lock_updates(journal);
4557
4558         /*
4559          * Don't clear the needs_recovery flag if we failed to flush
4560          * the journal.
4561          */
4562         error = jbd2_journal_flush(journal);
4563         if (error < 0)
4564                 goto out;
4565
4566         /* Journal blocked and flushed, clear needs_recovery flag. */
4567         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4568         error = ext4_commit_super(sb, 1);
4569 out:
4570         /* we rely on upper layer to stop further updates */
4571         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4572         return error;
4573 }
4574
4575 /*
4576  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4577  * flag here, even though the filesystem is not technically dirty yet.
4578  */
4579 static int ext4_unfreeze(struct super_block *sb)
4580 {
4581         if (sb->s_flags & MS_RDONLY)
4582                 return 0;
4583
4584         /* Reset the needs_recovery flag before the fs is unlocked. */
4585         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4586         ext4_commit_super(sb, 1);
4587         return 0;
4588 }
4589
4590 /*
4591  * Structure to save mount options for ext4_remount's benefit
4592  */
4593 struct ext4_mount_options {
4594         unsigned long s_mount_opt;
4595         unsigned long s_mount_opt2;
4596         kuid_t s_resuid;
4597         kgid_t s_resgid;
4598         unsigned long s_commit_interval;
4599         u32 s_min_batch_time, s_max_batch_time;
4600 #ifdef CONFIG_QUOTA
4601         int s_jquota_fmt;
4602         char *s_qf_names[MAXQUOTAS];
4603 #endif
4604 };
4605
4606 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4607 {
4608         struct ext4_super_block *es;
4609         struct ext4_sb_info *sbi = EXT4_SB(sb);
4610         unsigned long old_sb_flags;
4611         struct ext4_mount_options old_opts;
4612         int enable_quota = 0;
4613         ext4_group_t g;
4614         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4615         int err = 0;
4616 #ifdef CONFIG_QUOTA
4617         int i, j;
4618 #endif
4619         char *orig_data = kstrdup(data, GFP_KERNEL);
4620
4621         /* Store the original options */
4622         old_sb_flags = sb->s_flags;
4623         old_opts.s_mount_opt = sbi->s_mount_opt;
4624         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4625         old_opts.s_resuid = sbi->s_resuid;
4626         old_opts.s_resgid = sbi->s_resgid;
4627         old_opts.s_commit_interval = sbi->s_commit_interval;
4628         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4629         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4630 #ifdef CONFIG_QUOTA
4631         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4632         for (i = 0; i < MAXQUOTAS; i++)
4633                 if (sbi->s_qf_names[i]) {
4634                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4635                                                          GFP_KERNEL);
4636                         if (!old_opts.s_qf_names[i]) {
4637                                 for (j = 0; j < i; j++)
4638                                         kfree(old_opts.s_qf_names[j]);
4639                                 kfree(orig_data);
4640                                 return -ENOMEM;
4641                         }
4642                 } else
4643                         old_opts.s_qf_names[i] = NULL;
4644 #endif
4645         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4646                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4647
4648         /*
4649          * Allow the "check" option to be passed as a remount option.
4650          */
4651         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4652                 err = -EINVAL;
4653                 goto restore_opts;
4654         }
4655
4656         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4657                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4658                         ext4_msg(sb, KERN_ERR, "can't mount with "
4659                                  "both data=journal and delalloc");
4660                         err = -EINVAL;
4661                         goto restore_opts;
4662                 }
4663                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4664                         ext4_msg(sb, KERN_ERR, "can't mount with "
4665                                  "both data=journal and dioread_nolock");
4666                         err = -EINVAL;
4667                         goto restore_opts;
4668                 }
4669         }
4670
4671         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4672                 ext4_abort(sb, "Abort forced by user");
4673
4674         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4675                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4676
4677         es = sbi->s_es;
4678
4679         if (sbi->s_journal) {
4680                 ext4_init_journal_params(sb, sbi->s_journal);
4681                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4682         }
4683
4684         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4685                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4686                         err = -EROFS;
4687                         goto restore_opts;
4688                 }
4689
4690                 if (*flags & MS_RDONLY) {
4691                         err = dquot_suspend(sb, -1);
4692                         if (err < 0)
4693                                 goto restore_opts;
4694
4695                         /*
4696                          * First of all, the unconditional stuff we have to do
4697                          * to disable replay of the journal when we next remount
4698                          */
4699                         sb->s_flags |= MS_RDONLY;
4700
4701                         /*
4702                          * OK, test if we are remounting a valid rw partition
4703                          * readonly, and if so set the rdonly flag and then
4704                          * mark the partition as valid again.
4705                          */
4706                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4707                             (sbi->s_mount_state & EXT4_VALID_FS))
4708                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4709
4710                         if (sbi->s_journal)
4711                                 ext4_mark_recovery_complete(sb, es);
4712                 } else {
4713                         /* Make sure we can mount this feature set readwrite */
4714                         if (!ext4_feature_set_ok(sb, 0)) {
4715                                 err = -EROFS;
4716                                 goto restore_opts;
4717                         }
4718                         /*
4719                          * Make sure the group descriptor checksums
4720                          * are sane.  If they aren't, refuse to remount r/w.
4721                          */
4722                         for (g = 0; g < sbi->s_groups_count; g++) {
4723                                 struct ext4_group_desc *gdp =
4724                                         ext4_get_group_desc(sb, g, NULL);
4725
4726                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4727                                         ext4_msg(sb, KERN_ERR,
4728                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4729                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4730                                                le16_to_cpu(gdp->bg_checksum));
4731                                         err = -EINVAL;
4732                                         goto restore_opts;
4733                                 }
4734                         }
4735
4736                         /*
4737                          * If we have an unprocessed orphan list hanging
4738                          * around from a previously readonly bdev mount,
4739                          * require a full umount/remount for now.
4740                          */
4741                         if (es->s_last_orphan) {
4742                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4743                                        "remount RDWR because of unprocessed "
4744                                        "orphan inode list.  Please "
4745                                        "umount/remount instead");
4746                                 err = -EINVAL;
4747                                 goto restore_opts;
4748                         }
4749
4750                         /*
4751                          * Mounting a RDONLY partition read-write, so reread
4752                          * and store the current valid flag.  (It may have
4753                          * been changed by e2fsck since we originally mounted
4754                          * the partition.)
4755                          */
4756                         if (sbi->s_journal)
4757                                 ext4_clear_journal_err(sb, es);
4758                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4759                         if (!ext4_setup_super(sb, es, 0))
4760                                 sb->s_flags &= ~MS_RDONLY;
4761                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4762                                                      EXT4_FEATURE_INCOMPAT_MMP))
4763                                 if (ext4_multi_mount_protect(sb,
4764                                                 le64_to_cpu(es->s_mmp_block))) {
4765                                         err = -EROFS;
4766                                         goto restore_opts;
4767                                 }
4768                         enable_quota = 1;
4769                 }
4770         }
4771
4772         /*
4773          * Reinitialize lazy itable initialization thread based on
4774          * current settings
4775          */
4776         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4777                 ext4_unregister_li_request(sb);
4778         else {
4779                 ext4_group_t first_not_zeroed;
4780                 first_not_zeroed = ext4_has_uninit_itable(sb);
4781                 ext4_register_li_request(sb, first_not_zeroed);
4782         }
4783
4784         ext4_setup_system_zone(sb);
4785         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4786                 ext4_commit_super(sb, 1);
4787
4788 #ifdef CONFIG_QUOTA
4789         /* Release old quota file names */
4790         for (i = 0; i < MAXQUOTAS; i++)
4791                 kfree(old_opts.s_qf_names[i]);
4792         if (enable_quota) {
4793                 if (sb_any_quota_suspended(sb))
4794                         dquot_resume(sb, -1);
4795                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4796                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4797                         err = ext4_enable_quotas(sb);
4798                         if (err)
4799                                 goto restore_opts;
4800                 }
4801         }
4802 #endif
4803
4804         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4805         kfree(orig_data);
4806         return 0;
4807
4808 restore_opts:
4809         sb->s_flags = old_sb_flags;
4810         sbi->s_mount_opt = old_opts.s_mount_opt;
4811         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4812         sbi->s_resuid = old_opts.s_resuid;
4813         sbi->s_resgid = old_opts.s_resgid;
4814         sbi->s_commit_interval = old_opts.s_commit_interval;
4815         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4816         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4817 #ifdef CONFIG_QUOTA
4818         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4819         for (i = 0; i < MAXQUOTAS; i++) {
4820                 kfree(sbi->s_qf_names[i]);
4821                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4822         }
4823 #endif
4824         kfree(orig_data);
4825         return err;
4826 }
4827
4828 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4829 {
4830         struct super_block *sb = dentry->d_sb;
4831         struct ext4_sb_info *sbi = EXT4_SB(sb);
4832         struct ext4_super_block *es = sbi->s_es;
4833         ext4_fsblk_t overhead = 0, resv_blocks;
4834         u64 fsid;
4835         s64 bfree;
4836         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4837
4838         if (!test_opt(sb, MINIX_DF))
4839                 overhead = sbi->s_overhead;
4840
4841         buf->f_type = EXT4_SUPER_MAGIC;
4842         buf->f_bsize = sb->s_blocksize;
4843         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4844         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4845                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4846         /* prevent underflow in case that few free space is available */
4847         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4848         buf->f_bavail = buf->f_bfree -
4849                         (ext4_r_blocks_count(es) + resv_blocks);
4850         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4851                 buf->f_bavail = 0;
4852         buf->f_files = le32_to_cpu(es->s_inodes_count);
4853         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4854         buf->f_namelen = EXT4_NAME_LEN;
4855         fsid = le64_to_cpup((void *)es->s_uuid) ^
4856                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4857         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4858         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4859
4860         return 0;
4861 }
4862
4863 /* Helper function for writing quotas on sync - we need to start transaction
4864  * before quota file is locked for write. Otherwise the are possible deadlocks:
4865  * Process 1                         Process 2
4866  * ext4_create()                     quota_sync()
4867  *   jbd2_journal_start()                  write_dquot()
4868  *   dquot_initialize()                         down(dqio_mutex)
4869  *     down(dqio_mutex)                    jbd2_journal_start()
4870  *
4871  */
4872
4873 #ifdef CONFIG_QUOTA
4874
4875 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4876 {
4877         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4878 }
4879
4880 static int ext4_write_dquot(struct dquot *dquot)
4881 {
4882         int ret, err;
4883         handle_t *handle;
4884         struct inode *inode;
4885
4886         inode = dquot_to_inode(dquot);
4887         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4888                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4889         if (IS_ERR(handle))
4890                 return PTR_ERR(handle);
4891         ret = dquot_commit(dquot);
4892         err = ext4_journal_stop(handle);
4893         if (!ret)
4894                 ret = err;
4895         return ret;
4896 }
4897
4898 static int ext4_acquire_dquot(struct dquot *dquot)
4899 {
4900         int ret, err;
4901         handle_t *handle;
4902
4903         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4904                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4905         if (IS_ERR(handle))
4906                 return PTR_ERR(handle);
4907         ret = dquot_acquire(dquot);
4908         err = ext4_journal_stop(handle);
4909         if (!ret)
4910                 ret = err;
4911         return ret;
4912 }
4913
4914 static int ext4_release_dquot(struct dquot *dquot)
4915 {
4916         int ret, err;
4917         handle_t *handle;
4918
4919         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4920                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4921         if (IS_ERR(handle)) {
4922                 /* Release dquot anyway to avoid endless cycle in dqput() */
4923                 dquot_release(dquot);
4924                 return PTR_ERR(handle);
4925         }
4926         ret = dquot_release(dquot);
4927         err = ext4_journal_stop(handle);
4928         if (!ret)
4929                 ret = err;
4930         return ret;
4931 }
4932
4933 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4934 {
4935         struct super_block *sb = dquot->dq_sb;
4936         struct ext4_sb_info *sbi = EXT4_SB(sb);
4937
4938         /* Are we journaling quotas? */
4939         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
4940             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4941                 dquot_mark_dquot_dirty(dquot);
4942                 return ext4_write_dquot(dquot);
4943         } else {
4944                 return dquot_mark_dquot_dirty(dquot);
4945         }
4946 }
4947
4948 static int ext4_write_info(struct super_block *sb, int type)
4949 {
4950         int ret, err;
4951         handle_t *handle;
4952
4953         /* Data block + inode block */
4954         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
4955         if (IS_ERR(handle))
4956                 return PTR_ERR(handle);
4957         ret = dquot_commit_info(sb, type);
4958         err = ext4_journal_stop(handle);
4959         if (!ret)
4960                 ret = err;
4961         return ret;
4962 }
4963
4964 /*
4965  * Turn on quotas during mount time - we need to find
4966  * the quota file and such...
4967  */
4968 static int ext4_quota_on_mount(struct super_block *sb, int type)
4969 {
4970         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4971                                         EXT4_SB(sb)->s_jquota_fmt, type);
4972 }
4973
4974 /*
4975  * Standard function to be called on quota_on
4976  */
4977 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4978                          struct path *path)
4979 {
4980         int err;
4981
4982         if (!test_opt(sb, QUOTA))
4983                 return -EINVAL;
4984
4985         /* Quotafile not on the same filesystem? */
4986         if (path->dentry->d_sb != sb)
4987                 return -EXDEV;
4988         /* Journaling quota? */
4989         if (EXT4_SB(sb)->s_qf_names[type]) {
4990                 /* Quotafile not in fs root? */
4991                 if (path->dentry->d_parent != sb->s_root)
4992                         ext4_msg(sb, KERN_WARNING,
4993                                 "Quota file not on filesystem root. "
4994                                 "Journaled quota will not work");
4995         }
4996
4997         /*
4998          * When we journal data on quota file, we have to flush journal to see
4999          * all updates to the file when we bypass pagecache...
5000          */
5001         if (EXT4_SB(sb)->s_journal &&
5002             ext4_should_journal_data(path->dentry->d_inode)) {
5003                 /*
5004                  * We don't need to lock updates but journal_flush() could
5005                  * otherwise be livelocked...
5006                  */
5007                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5008                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5009                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5010                 if (err)
5011                         return err;
5012         }
5013
5014         return dquot_quota_on(sb, type, format_id, path);
5015 }
5016
5017 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5018                              unsigned int flags)
5019 {
5020         int err;
5021         struct inode *qf_inode;
5022         unsigned long qf_inums[MAXQUOTAS] = {
5023                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5024                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5025         };
5026
5027         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5028
5029         if (!qf_inums[type])
5030                 return -EPERM;
5031
5032         qf_inode = ext4_iget(sb, qf_inums[type]);
5033         if (IS_ERR(qf_inode)) {
5034                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5035                 return PTR_ERR(qf_inode);
5036         }
5037
5038         /* Don't account quota for quota files to avoid recursion */
5039         qf_inode->i_flags |= S_NOQUOTA;
5040         err = dquot_enable(qf_inode, type, format_id, flags);
5041         iput(qf_inode);
5042
5043         return err;
5044 }
5045
5046 /* Enable usage tracking for all quota types. */
5047 static int ext4_enable_quotas(struct super_block *sb)
5048 {
5049         int type, err = 0;
5050         unsigned long qf_inums[MAXQUOTAS] = {
5051                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5052                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5053         };
5054
5055         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5056         for (type = 0; type < MAXQUOTAS; type++) {
5057                 if (qf_inums[type]) {
5058                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5059                                                 DQUOT_USAGE_ENABLED);
5060                         if (err) {
5061                                 ext4_warning(sb,
5062                                         "Failed to enable quota tracking "
5063                                         "(type=%d, err=%d). Please run "
5064                                         "e2fsck to fix.", type, err);
5065                                 return err;
5066                         }
5067                 }
5068         }
5069         return 0;
5070 }
5071
5072 /*
5073  * quota_on function that is used when QUOTA feature is set.
5074  */
5075 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5076                                  int format_id)
5077 {
5078         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5079                 return -EINVAL;
5080
5081         /*
5082          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5083          */
5084         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5085 }
5086
5087 static int ext4_quota_off(struct super_block *sb, int type)
5088 {
5089         struct inode *inode = sb_dqopt(sb)->files[type];
5090         handle_t *handle;
5091
5092         /* Force all delayed allocation blocks to be allocated.
5093          * Caller already holds s_umount sem */
5094         if (test_opt(sb, DELALLOC))
5095                 sync_filesystem(sb);
5096
5097         if (!inode)
5098                 goto out;
5099
5100         /* Update modification times of quota files when userspace can
5101          * start looking at them */
5102         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5103         if (IS_ERR(handle))
5104                 goto out;
5105         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5106         ext4_mark_inode_dirty(handle, inode);
5107         ext4_journal_stop(handle);
5108
5109 out:
5110         return dquot_quota_off(sb, type);
5111 }
5112
5113 /*
5114  * quota_off function that is used when QUOTA feature is set.
5115  */
5116 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5117 {
5118         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5119                 return -EINVAL;
5120
5121         /* Disable only the limits. */
5122         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5123 }
5124
5125 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5126  * acquiring the locks... As quota files are never truncated and quota code
5127  * itself serializes the operations (and no one else should touch the files)
5128  * we don't have to be afraid of races */
5129 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5130                                size_t len, loff_t off)
5131 {
5132         struct inode *inode = sb_dqopt(sb)->files[type];
5133         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5134         int err = 0;
5135         int offset = off & (sb->s_blocksize - 1);
5136         int tocopy;
5137         size_t toread;
5138         struct buffer_head *bh;
5139         loff_t i_size = i_size_read(inode);
5140
5141         if (off > i_size)
5142                 return 0;
5143         if (off+len > i_size)
5144                 len = i_size-off;
5145         toread = len;
5146         while (toread > 0) {
5147                 tocopy = sb->s_blocksize - offset < toread ?
5148                                 sb->s_blocksize - offset : toread;
5149                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5150                 if (err)
5151                         return err;
5152                 if (!bh)        /* A hole? */
5153                         memset(data, 0, tocopy);
5154                 else
5155                         memcpy(data, bh->b_data+offset, tocopy);
5156                 brelse(bh);
5157                 offset = 0;
5158                 toread -= tocopy;
5159                 data += tocopy;
5160                 blk++;
5161         }
5162         return len;
5163 }
5164
5165 /* Write to quotafile (we know the transaction is already started and has
5166  * enough credits) */
5167 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5168                                 const char *data, size_t len, loff_t off)
5169 {
5170         struct inode *inode = sb_dqopt(sb)->files[type];
5171         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5172         int err = 0;
5173         int offset = off & (sb->s_blocksize - 1);
5174         struct buffer_head *bh;
5175         handle_t *handle = journal_current_handle();
5176
5177         if (EXT4_SB(sb)->s_journal && !handle) {
5178                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5179                         " cancelled because transaction is not started",
5180                         (unsigned long long)off, (unsigned long long)len);
5181                 return -EIO;
5182         }
5183         /*
5184          * Since we account only one data block in transaction credits,
5185          * then it is impossible to cross a block boundary.
5186          */
5187         if (sb->s_blocksize - offset < len) {
5188                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5189                         " cancelled because not block aligned",
5190                         (unsigned long long)off, (unsigned long long)len);
5191                 return -EIO;
5192         }
5193
5194         bh = ext4_bread(handle, inode, blk, 1, &err);
5195         if (!bh)
5196                 goto out;
5197         err = ext4_journal_get_write_access(handle, bh);
5198         if (err) {
5199                 brelse(bh);
5200                 goto out;
5201         }
5202         lock_buffer(bh);
5203         memcpy(bh->b_data+offset, data, len);
5204         flush_dcache_page(bh->b_page);
5205         unlock_buffer(bh);
5206         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5207         brelse(bh);
5208 out:
5209         if (err)
5210                 return err;
5211         if (inode->i_size < off + len) {
5212                 i_size_write(inode, off + len);
5213                 EXT4_I(inode)->i_disksize = inode->i_size;
5214                 ext4_mark_inode_dirty(handle, inode);
5215         }
5216         return len;
5217 }
5218
5219 #endif
5220
5221 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5222                        const char *dev_name, void *data)
5223 {
5224         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5225 }
5226
5227 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5228 static inline void register_as_ext2(void)
5229 {
5230         int err = register_filesystem(&ext2_fs_type);
5231         if (err)
5232                 printk(KERN_WARNING
5233                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5234 }
5235
5236 static inline void unregister_as_ext2(void)
5237 {
5238         unregister_filesystem(&ext2_fs_type);
5239 }
5240
5241 static inline int ext2_feature_set_ok(struct super_block *sb)
5242 {
5243         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5244                 return 0;
5245         if (sb->s_flags & MS_RDONLY)
5246                 return 1;
5247         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5248                 return 0;
5249         return 1;
5250 }
5251 #else
5252 static inline void register_as_ext2(void) { }
5253 static inline void unregister_as_ext2(void) { }
5254 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5255 #endif
5256
5257 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5258 static inline void register_as_ext3(void)
5259 {
5260         int err = register_filesystem(&ext3_fs_type);
5261         if (err)
5262                 printk(KERN_WARNING
5263                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5264 }
5265
5266 static inline void unregister_as_ext3(void)
5267 {
5268         unregister_filesystem(&ext3_fs_type);
5269 }
5270
5271 static inline int ext3_feature_set_ok(struct super_block *sb)
5272 {
5273         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5274                 return 0;
5275         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5276                 return 0;
5277         if (sb->s_flags & MS_RDONLY)
5278                 return 1;
5279         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5280                 return 0;
5281         return 1;
5282 }
5283 #else
5284 static inline void register_as_ext3(void) { }
5285 static inline void unregister_as_ext3(void) { }
5286 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5287 #endif
5288
5289 static struct file_system_type ext4_fs_type = {
5290         .owner          = THIS_MODULE,
5291         .name           = "ext4",
5292         .mount          = ext4_mount,
5293         .kill_sb        = kill_block_super,
5294         .fs_flags       = FS_REQUIRES_DEV,
5295 };
5296 MODULE_ALIAS_FS("ext4");
5297
5298 static int __init ext4_init_feat_adverts(void)
5299 {
5300         struct ext4_features *ef;
5301         int ret = -ENOMEM;
5302
5303         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5304         if (!ef)
5305                 goto out;
5306
5307         ef->f_kobj.kset = ext4_kset;
5308         init_completion(&ef->f_kobj_unregister);
5309         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5310                                    "features");
5311         if (ret) {
5312                 kfree(ef);
5313                 goto out;
5314         }
5315
5316         ext4_feat = ef;
5317         ret = 0;
5318 out:
5319         return ret;
5320 }
5321
5322 static void ext4_exit_feat_adverts(void)
5323 {
5324         kobject_put(&ext4_feat->f_kobj);
5325         wait_for_completion(&ext4_feat->f_kobj_unregister);
5326         kfree(ext4_feat);
5327 }
5328
5329 /* Shared across all ext4 file systems */
5330 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5331 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5332
5333 static int __init ext4_init_fs(void)
5334 {
5335         int i, err;
5336
5337         ext4_li_info = NULL;
5338         mutex_init(&ext4_li_mtx);
5339
5340         /* Build-time check for flags consistency */
5341         ext4_check_flag_values();
5342
5343         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5344                 mutex_init(&ext4__aio_mutex[i]);
5345                 init_waitqueue_head(&ext4__ioend_wq[i]);
5346         }
5347
5348         err = ext4_init_es();
5349         if (err)
5350                 return err;
5351
5352         err = ext4_init_pageio();
5353         if (err)
5354                 goto out7;
5355
5356         err = ext4_init_system_zone();
5357         if (err)
5358                 goto out6;
5359         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5360         if (!ext4_kset) {
5361                 err = -ENOMEM;
5362                 goto out5;
5363         }
5364         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5365
5366         err = ext4_init_feat_adverts();
5367         if (err)
5368                 goto out4;
5369
5370         err = ext4_init_mballoc();
5371         if (err)
5372                 goto out3;
5373
5374         err = ext4_init_xattr();
5375         if (err)
5376                 goto out2;
5377         err = init_inodecache();
5378         if (err)
5379                 goto out1;
5380         register_as_ext3();
5381         register_as_ext2();
5382         err = register_filesystem(&ext4_fs_type);
5383         if (err)
5384                 goto out;
5385
5386         return 0;
5387 out:
5388         unregister_as_ext2();
5389         unregister_as_ext3();
5390         destroy_inodecache();
5391 out1:
5392         ext4_exit_xattr();
5393 out2:
5394         ext4_exit_mballoc();
5395 out3:
5396         ext4_exit_feat_adverts();
5397 out4:
5398         if (ext4_proc_root)
5399                 remove_proc_entry("fs/ext4", NULL);
5400         kset_unregister(ext4_kset);
5401 out5:
5402         ext4_exit_system_zone();
5403 out6:
5404         ext4_exit_pageio();
5405 out7:
5406         ext4_exit_es();
5407
5408         return err;
5409 }
5410
5411 static void __exit ext4_exit_fs(void)
5412 {
5413         ext4_destroy_lazyinit_thread();
5414         unregister_as_ext2();
5415         unregister_as_ext3();
5416         unregister_filesystem(&ext4_fs_type);
5417         destroy_inodecache();
5418         ext4_exit_xattr();
5419         ext4_exit_mballoc();
5420         ext4_exit_feat_adverts();
5421         remove_proc_entry("fs/ext4", NULL);
5422         kset_unregister(ext4_kset);
5423         ext4_exit_system_zone();
5424         ext4_exit_pageio();
5425         ext4_exit_es();
5426 }
5427
5428 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5429 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5430 MODULE_LICENSE("GPL");
5431 module_init(ext4_init_fs)
5432 module_exit(ext4_exit_fs)