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