fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/jbd2.h>
  25 #include <linux/slab.h>
  26 #include <linux/init.h>
  27 #include <linux/blkdev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/proc_fs.h>
  38 #include <linux/ctype.h>
  39 #include <linux/log2.h>
  40 #include <linux/crc16.h>
  41 #include <linux/cleancache.h>
  42 #include <asm/uaccess.h>
  43
  44 #include <linux/kthread.h>
  45 #include <linux/freezer.h>
  46
  47 #include "ext4.h"
  48 #include "ext4_extents.h"       /* Needed for trace points definition */
  49 #include "ext4_jbd2.h"
  50 #include "xattr.h"
  51 #include "acl.h"
  52 #include "mballoc.h"
  53
  54 #define CREATE_TRACE_POINTS
  55 #include <trace/events/ext4.h>
  56
  57 static struct proc_dir_entry *ext4_proc_root;
  58 static struct kset *ext4_kset;
  59 static struct ext4_lazy_init *ext4_li_info;
  60 static struct mutex ext4_li_mtx;
  61 static struct ext4_features *ext4_feat;
  62
  63 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  64                              unsigned long journal_devnum);
  65 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  66 static int ext4_commit_super(struct super_block *sb, int sync);
  67 static void ext4_mark_recovery_complete(struct super_block *sb,
  68                                         struct ext4_super_block *es);
  69 static void ext4_clear_journal_err(struct super_block *sb,
  70                                    struct ext4_super_block *es);
  71 static int ext4_sync_fs(struct super_block *sb, int wait);
  72 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  73 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  74 static int ext4_unfreeze(struct super_block *sb);
  75 static int ext4_freeze(struct super_block *sb);
  76 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  77                        const char *dev_name, void *data);
  78 static inline int ext2_feature_set_ok(struct super_block *sb);
  79 static inline int ext3_feature_set_ok(struct super_block *sb);
  80 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  81 static void ext4_destroy_lazyinit_thread(void);
  82 static void ext4_unregister_li_request(struct super_block *sb);
  83 static void ext4_clear_request_list(void);
  84 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
  85
  86 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  87 static struct file_system_type ext2_fs_type = {
  88         .owner          = THIS_MODULE,
  89         .name           = "ext2",
  90         .mount          = ext4_mount,
  91         .kill_sb        = kill_block_super,
  92         .fs_flags       = FS_REQUIRES_DEV,
  93 };
  94 MODULE_ALIAS_FS("ext2");
  95 MODULE_ALIAS("ext2");
  96 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  97 #else
  98 #define IS_EXT2_SB(sb) (0)
  99 #endif
 100
 101
 102 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 103 static struct file_system_type ext3_fs_type = {
 104         .owner          = THIS_MODULE,
 105         .name           = "ext3",
 106         .mount          = ext4_mount,
 107         .kill_sb        = kill_block_super,
 108         .fs_flags       = FS_REQUIRES_DEV,
 109 };
 110 MODULE_ALIAS_FS("ext3");
 111 MODULE_ALIAS("ext3");
 112 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 113 #else
 114 #define IS_EXT3_SB(sb) (0)
 115 #endif
 116
 117 static int ext4_verify_csum_type(struct super_block *sb,
 118                                  struct ext4_super_block *es)
 119 {
 120         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 121                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 122                 return 1;
 123
 124         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 125 }
 126
 127 static __le32 ext4_superblock_csum(struct super_block *sb,
 128                                    struct ext4_super_block *es)
 129 {
 130         struct ext4_sb_info *sbi = EXT4_SB(sb);
 131         int offset = offsetof(struct ext4_super_block, s_checksum);
 132         __u32 csum;
 133
 134         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 135
 136         return cpu_to_le32(csum);
 137 }
 138
 139 int ext4_superblock_csum_verify(struct super_block *sb,
 140                                 struct ext4_super_block *es)
 141 {
 142         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 143                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 144                 return 1;
 145
 146         return es->s_checksum == ext4_superblock_csum(sb, es);
 147 }
 148
 149 void ext4_superblock_csum_set(struct super_block *sb)
 150 {
 151         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 152
 153         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 154                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 155                 return;
 156
 157         es->s_checksum = ext4_superblock_csum(sb, es);
 158 }
 159
 160 void *ext4_kvmalloc(size_t size, gfp_t flags)
 161 {
 162         void *ret;
 163
 164         ret = kmalloc(size, flags);
 165         if (!ret)
 166                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 167         return ret;
 168 }
 169
 170 void *ext4_kvzalloc(size_t size, gfp_t flags)
 171 {
 172         void *ret;
 173
 174         ret = kzalloc(size, flags);
 175         if (!ret)
 176                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 177         return ret;
 178 }
 179
 180 void ext4_kvfree(void *ptr)
 181 {
 182         if (is_vmalloc_addr(ptr))
 183                 vfree(ptr);
 184         else
 185                 kfree(ptr);
 186
 187 }
 188
 189 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 190                                struct ext4_group_desc *bg)
 191 {
 192         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 193                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 194                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 195 }
 196
 197 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 198                                struct ext4_group_desc *bg)
 199 {
 200         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 201                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 202                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 203 }
 204
 205 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 206                               struct ext4_group_desc *bg)
 207 {
 208         return le32_to_cpu(bg->bg_inode_table_lo) |
 209                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 210                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 211 }
 212
 213 __u32 ext4_free_group_clusters(struct super_block *sb,
 214                                struct ext4_group_desc *bg)
 215 {
 216         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 217                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 218                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 219 }
 220
 221 __u32 ext4_free_inodes_count(struct super_block *sb,
 222                               struct ext4_group_desc *bg)
 223 {
 224         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 225                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 226                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 227 }
 228
 229 __u32 ext4_used_dirs_count(struct super_block *sb,
 230                               struct ext4_group_desc *bg)
 231 {
 232         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 233                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 234                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 235 }
 236
 237 __u32 ext4_itable_unused_count(struct super_block *sb,
 238                               struct ext4_group_desc *bg)
 239 {
 240         return le16_to_cpu(bg->bg_itable_unused_lo) |
 241                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 242                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 243 }
 244
 245 void ext4_block_bitmap_set(struct super_block *sb,
 246                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 247 {
 248         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 249         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 250                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 251 }
 252
 253 void ext4_inode_bitmap_set(struct super_block *sb,
 254                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 255 {
 256         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 257         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 258                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 259 }
 260
 261 void ext4_inode_table_set(struct super_block *sb,
 262                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 263 {
 264         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 265         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 266                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 267 }
 268
 269 void ext4_free_group_clusters_set(struct super_block *sb,
 270                                   struct ext4_group_desc *bg, __u32 count)
 271 {
 272         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 273         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 274                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 275 }
 276
 277 void ext4_free_inodes_set(struct super_block *sb,
 278                           struct ext4_group_desc *bg, __u32 count)
 279 {
 280         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 281         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 282                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 283 }
 284
 285 void ext4_used_dirs_set(struct super_block *sb,
 286                           struct ext4_group_desc *bg, __u32 count)
 287 {
 288         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 289         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 290                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 291 }
 292
 293 void ext4_itable_unused_set(struct super_block *sb,
 294                           struct ext4_group_desc *bg, __u32 count)
 295 {
 296         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 297         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 298                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 299 }
 300
 301
 302 static void __save_error_info(struct super_block *sb, const char *func,
 303                             unsigned int line)
 304 {
 305         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 306
 307         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 308         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 309         es->s_last_error_time = cpu_to_le32(get_seconds());
 310         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 311         es->s_last_error_line = cpu_to_le32(line);
 312         if (!es->s_first_error_time) {
 313                 es->s_first_error_time = es->s_last_error_time;
 314                 strncpy(es->s_first_error_func, func,
 315                         sizeof(es->s_first_error_func));
 316                 es->s_first_error_line = cpu_to_le32(line);
 317                 es->s_first_error_ino = es->s_last_error_ino;
 318                 es->s_first_error_block = es->s_last_error_block;
 319         }
 320         /*
 321          * Start the daily error reporting function if it hasn't been
 322          * started already
 323          */
 324         if (!es->s_error_count)
 325                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 326         le32_add_cpu(&es->s_error_count, 1);
 327 }
 328
 329 static void save_error_info(struct super_block *sb, const char *func,
 330                             unsigned int line)
 331 {
 332         __save_error_info(sb, func, line);
 333         ext4_commit_super(sb, 1);
 334 }
 335
 336 /*
 337  * The del_gendisk() function uninitializes the disk-specific data
 338  * structures, including the bdi structure, without telling anyone
 339  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 340  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 341  * This is a kludge to prevent these oops until we can put in a proper
 342  * hook in del_gendisk() to inform the VFS and file system layers.
 343  */
 344 static int block_device_ejected(struct super_block *sb)
 345 {
 346         struct inode *bd_inode = sb->s_bdev->bd_inode;
 347         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 348
 349         return bdi->dev == NULL;
 350 }
 351
 352 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 353 {
 354         struct super_block              *sb = journal->j_private;
 355         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 356         int                             error = is_journal_aborted(journal);
 357         struct ext4_journal_cb_entry    *jce;
 358
 359         BUG_ON(txn->t_state == T_FINISHED);
 360         spin_lock(&sbi->s_md_lock);
 361         while (!list_empty(&txn->t_private_list)) {
 362                 jce = list_entry(txn->t_private_list.next,
 363                                  struct ext4_journal_cb_entry, jce_list);
 364                 list_del_init(&jce->jce_list);
 365                 spin_unlock(&sbi->s_md_lock);
 366                 jce->jce_func(sb, jce, error);
 367                 spin_lock(&sbi->s_md_lock);
 368         }
 369         spin_unlock(&sbi->s_md_lock);
 370 }
 371
 372 /* Deal with the reporting of failure conditions on a filesystem such as
 373  * inconsistencies detected or read IO failures.
 374  *
 375  * On ext2, we can store the error state of the filesystem in the
 376  * superblock.  That is not possible on ext4, because we may have other
 377  * write ordering constraints on the superblock which prevent us from
 378  * writing it out straight away; and given that the journal is about to
 379  * be aborted, we can't rely on the current, or future, transactions to
 380  * write out the superblock safely.
 381  *
 382  * We'll just use the jbd2_journal_abort() error code to record an error in
 383  * the journal instead.  On recovery, the journal will complain about
 384  * that error until we've noted it down and cleared it.
 385  */
 386
 387 static void ext4_handle_error(struct super_block *sb)
 388 {
 389         if (sb->s_flags & MS_RDONLY)
 390                 return;
 391
 392         if (!test_opt(sb, ERRORS_CONT)) {
 393                 journal_t *journal = EXT4_SB(sb)->s_journal;
 394
 395                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 396                 if (journal)
 397                         jbd2_journal_abort(journal, -EIO);
 398         }
 399         if (test_opt(sb, ERRORS_RO)) {
 400                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 401                 sb->s_flags |= MS_RDONLY;
 402         }
 403         if (test_opt(sb, ERRORS_PANIC))
 404                 panic("EXT4-fs (device %s): panic forced after error\n",
 405                         sb->s_id);
 406 }
 407
 408 void __ext4_error(struct super_block *sb, const char *function,
 409                   unsigned int line, const char *fmt, ...)
 410 {
 411         struct va_format vaf;
 412         va_list args;
 413
 414         va_start(args, fmt);
 415         vaf.fmt = fmt;
 416         vaf.va = &args;
 417         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 418                sb->s_id, function, line, current->comm, &vaf);
 419         va_end(args);
 420         save_error_info(sb, function, line);
 421
 422         ext4_handle_error(sb);
 423 }
 424
 425 void ext4_error_inode(struct inode *inode, const char *function,
 426                       unsigned int line, ext4_fsblk_t block,
 427                       const char *fmt, ...)
 428 {
 429         va_list args;
 430         struct va_format vaf;
 431         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 432
 433         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 434         es->s_last_error_block = cpu_to_le64(block);
 435         save_error_info(inode->i_sb, function, line);
 436         va_start(args, fmt);
 437         vaf.fmt = fmt;
 438         vaf.va = &args;
 439         if (block)
 440                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 441                        "inode #%lu: block %llu: comm %s: %pV\n",
 442                        inode->i_sb->s_id, function, line, inode->i_ino,
 443                        block, current->comm, &vaf);
 444         else
 445                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 446                        "inode #%lu: comm %s: %pV\n",
 447                        inode->i_sb->s_id, function, line, inode->i_ino,
 448                        current->comm, &vaf);
 449         va_end(args);
 450
 451         ext4_handle_error(inode->i_sb);
 452 }
 453
 454 void ext4_error_file(struct file *file, const char *function,
 455                      unsigned int line, ext4_fsblk_t block,
 456                      const char *fmt, ...)
 457 {
 458         va_list args;
 459         struct va_format vaf;
 460         struct ext4_super_block *es;
 461         struct inode *inode = file_inode(file);
 462         char pathname[80], *path;
 463
 464         es = EXT4_SB(inode->i_sb)->s_es;
 465         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 466         save_error_info(inode->i_sb, function, line);
 467         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 468         if (IS_ERR(path))
 469                 path = "(unknown)";
 470         va_start(args, fmt);
 471         vaf.fmt = fmt;
 472         vaf.va = &args;
 473         if (block)
 474                 printk(KERN_CRIT
 475                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 476                        "block %llu: comm %s: path %s: %pV\n",
 477                        inode->i_sb->s_id, function, line, inode->i_ino,
 478                        block, current->comm, path, &vaf);
 479         else
 480                 printk(KERN_CRIT
 481                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 482                        "comm %s: path %s: %pV\n",
 483                        inode->i_sb->s_id, function, line, inode->i_ino,
 484                        current->comm, path, &vaf);
 485         va_end(args);
 486
 487         ext4_handle_error(inode->i_sb);
 488 }
 489
 490 const char *ext4_decode_error(struct super_block *sb, int errno,
 491                               char nbuf[16])
 492 {
 493         char *errstr = NULL;
 494
 495         switch (errno) {
 496         case -EIO:
 497                 errstr = "IO failure";
 498                 break;
 499         case -ENOMEM:
 500                 errstr = "Out of memory";
 501                 break;
 502         case -EROFS:
 503                 if (!sb || (EXT4_SB(sb)->s_journal &&
 504                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 505                         errstr = "Journal has aborted";
 506                 else
 507                         errstr = "Readonly filesystem";
 508                 break;
 509         default:
 510                 /* If the caller passed in an extra buffer for unknown
 511                  * errors, textualise them now.  Else we just return
 512                  * NULL. */
 513                 if (nbuf) {
 514                         /* Check for truncated error codes... */
 515                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 516                                 errstr = nbuf;
 517                 }
 518                 break;
 519         }
 520
 521         return errstr;
 522 }
 523
 524 /* __ext4_std_error decodes expected errors from journaling functions
 525  * automatically and invokes the appropriate error response.  */
 526
 527 void __ext4_std_error(struct super_block *sb, const char *function,
 528                       unsigned int line, int errno)
 529 {
 530         char nbuf[16];
 531         const char *errstr;
 532
 533         /* Special case: if the error is EROFS, and we're not already
 534          * inside a transaction, then there's really no point in logging
 535          * an error. */
 536         if (errno == -EROFS && journal_current_handle() == NULL &&
 537             (sb->s_flags & MS_RDONLY))
 538                 return;
 539
 540         errstr = ext4_decode_error(sb, errno, nbuf);
 541         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 542                sb->s_id, function, line, errstr);
 543         save_error_info(sb, function, line);
 544
 545         ext4_handle_error(sb);
 546 }
 547
 548 /*
 549  * ext4_abort is a much stronger failure handler than ext4_error.  The
 550  * abort function may be used to deal with unrecoverable failures such
 551  * as journal IO errors or ENOMEM at a critical moment in log management.
 552  *
 553  * We unconditionally force the filesystem into an ABORT|READONLY state,
 554  * unless the error response on the fs has been set to panic in which
 555  * case we take the easy way out and panic immediately.
 556  */
 557
 558 void __ext4_abort(struct super_block *sb, const char *function,
 559                 unsigned int line, const char *fmt, ...)
 560 {
 561         va_list args;
 562
 563         save_error_info(sb, function, line);
 564         va_start(args, fmt);
 565         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 566                function, line);
 567         vprintk(fmt, args);
 568         printk("\n");
 569         va_end(args);
 570
 571         if ((sb->s_flags & MS_RDONLY) == 0) {
 572                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 573                 sb->s_flags |= MS_RDONLY;
 574                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 575                 if (EXT4_SB(sb)->s_journal)
 576                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 577                 save_error_info(sb, function, line);
 578         }
 579         if (test_opt(sb, ERRORS_PANIC))
 580                 panic("EXT4-fs panic from previous error\n");
 581 }
 582
 583 void ext4_msg(struct super_block *sb, const char *prefix, const char *fmt, ...)
 584 {
 585         struct va_format vaf;
 586         va_list args;
 587
 588         va_start(args, fmt);
 589         vaf.fmt = fmt;
 590         vaf.va = &args;
 591         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 592         va_end(args);
 593 }
 594
 595 void __ext4_warning(struct super_block *sb, const char *function,
 596                     unsigned int line, const char *fmt, ...)
 597 {
 598         struct va_format vaf;
 599         va_list args;
 600
 601         va_start(args, fmt);
 602         vaf.fmt = fmt;
 603         vaf.va = &args;
 604         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 605                sb->s_id, function, line, &vaf);
 606         va_end(args);
 607 }
 608
 609 void __ext4_grp_locked_error(const char *function, unsigned int line,
 610                              struct super_block *sb, ext4_group_t grp,
 611                              unsigned long ino, ext4_fsblk_t block,
 612                              const char *fmt, ...)
 613 __releases(bitlock)
 614 __acquires(bitlock)
 615 {
 616         struct va_format vaf;
 617         va_list args;
 618         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 619
 620         es->s_last_error_ino = cpu_to_le32(ino);
 621         es->s_last_error_block = cpu_to_le64(block);
 622         __save_error_info(sb, function, line);
 623
 624         va_start(args, fmt);
 625
 626         vaf.fmt = fmt;
 627         vaf.va = &args;
 628         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 629                sb->s_id, function, line, grp);
 630         if (ino)
 631                 printk(KERN_CONT "inode %lu: ", ino);
 632         if (block)
 633                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 634         printk(KERN_CONT "%pV\n", &vaf);
 635         va_end(args);
 636
 637         if (test_opt(sb, ERRORS_CONT)) {
 638                 ext4_commit_super(sb, 0);
 639                 return;
 640         }
 641
 642         ext4_unlock_group(sb, grp);
 643         ext4_handle_error(sb);
 644         /*
 645          * We only get here in the ERRORS_RO case; relocking the group
 646          * may be dangerous, but nothing bad will happen since the
 647          * filesystem will have already been marked read/only and the
 648          * journal has been aborted.  We return 1 as a hint to callers
 649          * who might what to use the return value from
 650          * ext4_grp_locked_error() to distinguish between the
 651          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 652          * aggressively from the ext4 function in question, with a
 653          * more appropriate error code.
 654          */
 655         ext4_lock_group(sb, grp);
 656         return;
 657 }
 658
 659 void ext4_update_dynamic_rev(struct super_block *sb)
 660 {
 661         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 662
 663         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 664                 return;
 665
 666         ext4_warning(sb,
 667                      "updating to rev %d because of new feature flag, "
 668                      "running e2fsck is recommended",
 669                      EXT4_DYNAMIC_REV);
 670
 671         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 672         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 673         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 674         /* leave es->s_feature_*compat flags alone */
 675         /* es->s_uuid will be set by e2fsck if empty */
 676
 677         /*
 678          * The rest of the superblock fields should be zero, and if not it
 679          * means they are likely already in use, so leave them alone.  We
 680          * can leave it up to e2fsck to clean up any inconsistencies there.
 681          */
 682 }
 683
 684 /*
 685  * Open the external journal device
 686  */
 687 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 688 {
 689         struct block_device *bdev;
 690         char b[BDEVNAME_SIZE];
 691
 692         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 693         if (IS_ERR(bdev))
 694                 goto fail;
 695         return bdev;
 696
 697 fail:
 698         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 699                         __bdevname(dev, b), PTR_ERR(bdev));
 700         return NULL;
 701 }
 702
 703 /*
 704  * Release the journal device
 705  */
 706 static int ext4_blkdev_put(struct block_device *bdev)
 707 {
 708         return blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 709 }
 710
 711 static int ext4_blkdev_remove(struct ext4_sb_info *sbi)
 712 {
 713         struct block_device *bdev;
 714         int ret = -ENODEV;
 715
 716         bdev = sbi->journal_bdev;
 717         if (bdev) {
 718                 ret = ext4_blkdev_put(bdev);
 719                 sbi->journal_bdev = NULL;
 720         }
 721         return ret;
 722 }
 723
 724 static inline struct inode *orphan_list_entry(struct list_head *l)
 725 {
 726         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 727 }
 728
 729 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 730 {
 731         struct list_head *l;
 732
 733         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 734                  le32_to_cpu(sbi->s_es->s_last_orphan));
 735
 736         printk(KERN_ERR "sb_info orphan list:\n");
 737         list_for_each(l, &sbi->s_orphan) {
 738                 struct inode *inode = orphan_list_entry(l);
 739                 printk(KERN_ERR "  "
 740                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 741                        inode->i_sb->s_id, inode->i_ino, inode,
 742                        inode->i_mode, inode->i_nlink,
 743                        NEXT_ORPHAN(inode));
 744         }
 745 }
 746
 747 static void ext4_put_super(struct super_block *sb)
 748 {
 749         struct ext4_sb_info *sbi = EXT4_SB(sb);
 750         struct ext4_super_block *es = sbi->s_es;
 751         int i, err;
 752
 753         ext4_unregister_li_request(sb);
 754         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 755
 756         flush_workqueue(sbi->dio_unwritten_wq);
 757         destroy_workqueue(sbi->dio_unwritten_wq);
 758
 759         if (sbi->s_journal) {
 760                 err = jbd2_journal_destroy(sbi->s_journal);
 761                 sbi->s_journal = NULL;
 762                 if (err < 0)
 763                         ext4_abort(sb, "Couldn't clean up the journal");
 764         }
 765
 766         ext4_es_unregister_shrinker(sb);
 767         del_timer(&sbi->s_err_report);
 768         ext4_release_system_zone(sb);
 769         ext4_mb_release(sb);
 770         ext4_ext_release(sb);
 771         ext4_xattr_put_super(sb);
 772
 773         if (!(sb->s_flags & MS_RDONLY)) {
 774                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 775                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 776         }
 777         if (!(sb->s_flags & MS_RDONLY))
 778                 ext4_commit_super(sb, 1);
 779
 780         if (sbi->s_proc) {
 781                 remove_proc_entry("options", sbi->s_proc);
 782                 remove_proc_entry(sb->s_id, ext4_proc_root);
 783         }
 784         kobject_del(&sbi->s_kobj);
 785
 786         for (i = 0; i < sbi->s_gdb_count; i++)
 787                 brelse(sbi->s_group_desc[i]);
 788         ext4_kvfree(sbi->s_group_desc);
 789         ext4_kvfree(sbi->s_flex_groups);
 790         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 791         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 792         percpu_counter_destroy(&sbi->s_dirs_counter);
 793         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 794         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
 795         brelse(sbi->s_sbh);
 796 #ifdef CONFIG_QUOTA
 797         for (i = 0; i < MAXQUOTAS; i++)
 798                 kfree(sbi->s_qf_names[i]);
 799 #endif
 800
 801         /* Debugging code just in case the in-memory inode orphan list
 802          * isn't empty.  The on-disk one can be non-empty if we've
 803          * detected an error and taken the fs readonly, but the
 804          * in-memory list had better be clean by this point. */
 805         if (!list_empty(&sbi->s_orphan))
 806                 dump_orphan_list(sb, sbi);
 807         J_ASSERT(list_empty(&sbi->s_orphan));
 808
 809         invalidate_bdev(sb->s_bdev);
 810         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 811                 /*
 812                  * Invalidate the journal device's buffers.  We don't want them
 813                  * floating about in memory - the physical journal device may
 814                  * hotswapped, and it breaks the `ro-after' testing code.
 815                  */
 816                 sync_blockdev(sbi->journal_bdev);
 817                 invalidate_bdev(sbi->journal_bdev);
 818                 ext4_blkdev_remove(sbi);
 819         }
 820         if (sbi->s_mmp_tsk)
 821                 kthread_stop(sbi->s_mmp_tsk);
 822         sb->s_fs_info = NULL;
 823         /*
 824          * Now that we are completely done shutting down the
 825          * superblock, we need to actually destroy the kobject.
 826          */
 827         kobject_put(&sbi->s_kobj);
 828         wait_for_completion(&sbi->s_kobj_unregister);
 829         if (sbi->s_chksum_driver)
 830                 crypto_free_shash(sbi->s_chksum_driver);
 831         kfree(sbi->s_blockgroup_lock);
 832         kfree(sbi);
 833 }
 834
 835 static struct kmem_cache *ext4_inode_cachep;
 836
 837 /*
 838  * Called inside transaction, so use GFP_NOFS
 839  */
 840 static struct inode *ext4_alloc_inode(struct super_block *sb)
 841 {
 842         struct ext4_inode_info *ei;
 843
 844         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 845         if (!ei)
 846                 return NULL;
 847
 848         ei->vfs_inode.i_version = 1;
 849         INIT_LIST_HEAD(&ei->i_prealloc_list);
 850         spin_lock_init(&ei->i_prealloc_lock);
 851         ext4_es_init_tree(&ei->i_es_tree);
 852         rwlock_init(&ei->i_es_lock);
 853         INIT_LIST_HEAD(&ei->i_es_lru);
 854         ei->i_es_lru_nr = 0;
 855         ei->i_reserved_data_blocks = 0;
 856         ei->i_reserved_meta_blocks = 0;
 857         ei->i_allocated_meta_blocks = 0;
 858         ei->i_da_metadata_calc_len = 0;
 859         ei->i_da_metadata_calc_last_lblock = 0;
 860         spin_lock_init(&(ei->i_block_reservation_lock));
 861 #ifdef CONFIG_QUOTA
 862         ei->i_reserved_quota = 0;
 863 #endif
 864         ei->jinode = NULL;
 865         INIT_LIST_HEAD(&ei->i_completed_io_list);
 866         spin_lock_init(&ei->i_completed_io_lock);
 867         ei->i_sync_tid = 0;
 868         ei->i_datasync_tid = 0;
 869         atomic_set(&ei->i_ioend_count, 0);
 870         atomic_set(&ei->i_unwritten, 0);
 871         INIT_WORK(&ei->i_unwritten_work, ext4_end_io_work);
 872
 873         return &ei->vfs_inode;
 874 }
 875
 876 static int ext4_drop_inode(struct inode *inode)
 877 {
 878         int drop = generic_drop_inode(inode);
 879
 880         trace_ext4_drop_inode(inode, drop);
 881         return drop;
 882 }
 883
 884 static void ext4_i_callback(struct rcu_head *head)
 885 {
 886         struct inode *inode = container_of(head, struct inode, i_rcu);
 887         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 888 }
 889
 890 static void ext4_destroy_inode(struct inode *inode)
 891 {
 892         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 893                 ext4_msg(inode->i_sb, KERN_ERR,
 894                          "Inode %lu (%p): orphan list check failed!",
 895                          inode->i_ino, EXT4_I(inode));
 896                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 897                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 898                                 true);
 899                 dump_stack();
 900         }
 901         call_rcu(&inode->i_rcu, ext4_i_callback);
 902 }
 903
 904 static void init_once(void *foo)
 905 {
 906         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 907
 908         INIT_LIST_HEAD(&ei->i_orphan);
 909         init_rwsem(&ei->xattr_sem);
 910         init_rwsem(&ei->i_data_sem);
 911         inode_init_once(&ei->vfs_inode);
 912 }
 913
 914 static int init_inodecache(void)
 915 {
 916         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 917                                              sizeof(struct ext4_inode_info),
 918                                              0, (SLAB_RECLAIM_ACCOUNT|
 919                                                 SLAB_MEM_SPREAD),
 920                                              init_once);
 921         if (ext4_inode_cachep == NULL)
 922                 return -ENOMEM;
 923         return 0;
 924 }
 925
 926 static void destroy_inodecache(void)
 927 {
 928         /*
 929          * Make sure all delayed rcu free inodes are flushed before we
 930          * destroy cache.
 931          */
 932         rcu_barrier();
 933         kmem_cache_destroy(ext4_inode_cachep);
 934 }
 935
 936 void ext4_clear_inode(struct inode *inode)
 937 {
 938         invalidate_inode_buffers(inode);
 939         clear_inode(inode);
 940         dquot_drop(inode);
 941         ext4_discard_preallocations(inode);
 942         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 943         ext4_es_lru_del(inode);
 944         if (EXT4_I(inode)->jinode) {
 945                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 946                                                EXT4_I(inode)->jinode);
 947                 jbd2_free_inode(EXT4_I(inode)->jinode);
 948                 EXT4_I(inode)->jinode = NULL;
 949         }
 950 }
 951
 952 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 953                                         u64 ino, u32 generation)
 954 {
 955         struct inode *inode;
 956
 957         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 958                 return ERR_PTR(-ESTALE);
 959         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 960                 return ERR_PTR(-ESTALE);
 961
 962         /* iget isn't really right if the inode is currently unallocated!!
 963          *
 964          * ext4_read_inode will return a bad_inode if the inode had been
 965          * deleted, so we should be safe.
 966          *
 967          * Currently we don't know the generation for parent directory, so
 968          * a generation of 0 means "accept any"
 969          */
 970         inode = ext4_iget(sb, ino);
 971         if (IS_ERR(inode))
 972                 return ERR_CAST(inode);
 973         if (generation && inode->i_generation != generation) {
 974                 iput(inode);
 975                 return ERR_PTR(-ESTALE);
 976         }
 977
 978         return inode;
 979 }
 980
 981 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 982                                         int fh_len, int fh_type)
 983 {
 984         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 985                                     ext4_nfs_get_inode);
 986 }
 987
 988 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 989                                         int fh_len, int fh_type)
 990 {
 991         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
 992                                     ext4_nfs_get_inode);
 993 }
 994
 995 /*
 996  * Try to release metadata pages (indirect blocks, directories) which are
 997  * mapped via the block device.  Since these pages could have journal heads
 998  * which would prevent try_to_free_buffers() from freeing them, we must use
 999  * jbd2 layer's try_to_free_buffers() function to release them.
1000  */
1001 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1002                                  gfp_t wait)
1003 {
1004         journal_t *journal = EXT4_SB(sb)->s_journal;
1005
1006         WARN_ON(PageChecked(page));
1007         if (!page_has_buffers(page))
1008                 return 0;
1009         if (journal)
1010                 return jbd2_journal_try_to_free_buffers(journal, page,
1011                                                         wait & ~__GFP_WAIT);
1012         return try_to_free_buffers(page);
1013 }
1014
1015 #ifdef CONFIG_QUOTA
1016 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1017 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1018
1019 static int ext4_write_dquot(struct dquot *dquot);
1020 static int ext4_acquire_dquot(struct dquot *dquot);
1021 static int ext4_release_dquot(struct dquot *dquot);
1022 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1023 static int ext4_write_info(struct super_block *sb, int type);
1024 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1025                          struct path *path);
1026 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1027                                  int format_id);
1028 static int ext4_quota_off(struct super_block *sb, int type);
1029 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1030 static int ext4_quota_on_mount(struct super_block *sb, int type);
1031 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1032                                size_t len, loff_t off);
1033 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1034                                 const char *data, size_t len, loff_t off);
1035 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1036                              unsigned int flags);
1037 static int ext4_enable_quotas(struct super_block *sb);
1038
1039 static const struct dquot_operations ext4_quota_operations = {
1040         .get_reserved_space = ext4_get_reserved_space,
1041         .write_dquot    = ext4_write_dquot,
1042         .acquire_dquot  = ext4_acquire_dquot,
1043         .release_dquot  = ext4_release_dquot,
1044         .mark_dirty     = ext4_mark_dquot_dirty,
1045         .write_info     = ext4_write_info,
1046         .alloc_dquot    = dquot_alloc,
1047         .destroy_dquot  = dquot_destroy,
1048 };
1049
1050 static const struct quotactl_ops ext4_qctl_operations = {
1051         .quota_on       = ext4_quota_on,
1052         .quota_off      = ext4_quota_off,
1053         .quota_sync     = dquot_quota_sync,
1054         .get_info       = dquot_get_dqinfo,
1055         .set_info       = dquot_set_dqinfo,
1056         .get_dqblk      = dquot_get_dqblk,
1057         .set_dqblk      = dquot_set_dqblk
1058 };
1059
1060 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1061         .quota_on_meta  = ext4_quota_on_sysfile,
1062         .quota_off      = ext4_quota_off_sysfile,
1063         .quota_sync     = dquot_quota_sync,
1064         .get_info       = dquot_get_dqinfo,
1065         .set_info       = dquot_set_dqinfo,
1066         .get_dqblk      = dquot_get_dqblk,
1067         .set_dqblk      = dquot_set_dqblk
1068 };
1069 #endif
1070
1071 static const struct super_operations ext4_sops = {
1072         .alloc_inode    = ext4_alloc_inode,
1073         .destroy_inode  = ext4_destroy_inode,
1074         .write_inode    = ext4_write_inode,
1075         .dirty_inode    = ext4_dirty_inode,
1076         .drop_inode     = ext4_drop_inode,
1077         .evict_inode    = ext4_evict_inode,
1078         .put_super      = ext4_put_super,
1079         .sync_fs        = ext4_sync_fs,
1080         .freeze_fs      = ext4_freeze,
1081         .unfreeze_fs    = ext4_unfreeze,
1082         .statfs         = ext4_statfs,
1083         .remount_fs     = ext4_remount,
1084         .show_options   = ext4_show_options,
1085 #ifdef CONFIG_QUOTA
1086         .quota_read     = ext4_quota_read,
1087         .quota_write    = ext4_quota_write,
1088 #endif
1089         .bdev_try_to_free_page = bdev_try_to_free_page,
1090 };
1091
1092 static const struct super_operations ext4_nojournal_sops = {
1093         .alloc_inode    = ext4_alloc_inode,
1094         .destroy_inode  = ext4_destroy_inode,
1095         .write_inode    = ext4_write_inode,
1096         .dirty_inode    = ext4_dirty_inode,
1097         .drop_inode     = ext4_drop_inode,
1098         .evict_inode    = ext4_evict_inode,
1099         .put_super      = ext4_put_super,
1100         .statfs         = ext4_statfs,
1101         .remount_fs     = ext4_remount,
1102         .show_options   = ext4_show_options,
1103 #ifdef CONFIG_QUOTA
1104         .quota_read     = ext4_quota_read,
1105         .quota_write    = ext4_quota_write,
1106 #endif
1107         .bdev_try_to_free_page = bdev_try_to_free_page,
1108 };
1109
1110 static const struct export_operations ext4_export_ops = {
1111         .fh_to_dentry = ext4_fh_to_dentry,
1112         .fh_to_parent = ext4_fh_to_parent,
1113         .get_parent = ext4_get_parent,
1114 };
1115
1116 enum {
1117         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1118         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1119         Opt_nouid32, Opt_debug, Opt_removed,
1120         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1121         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1122         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1123         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1124         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1125         Opt_data_err_abort, Opt_data_err_ignore,
1126         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1127         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1128         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1129         Opt_usrquota, Opt_grpquota, Opt_i_version,
1130         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1131         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1132         Opt_inode_readahead_blks, Opt_journal_ioprio,
1133         Opt_dioread_nolock, Opt_dioread_lock,
1134         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1135         Opt_max_dir_size_kb,
1136 };
1137
1138 static const match_table_t tokens = {
1139         {Opt_bsd_df, "bsddf"},
1140         {Opt_minix_df, "minixdf"},
1141         {Opt_grpid, "grpid"},
1142         {Opt_grpid, "bsdgroups"},
1143         {Opt_nogrpid, "nogrpid"},
1144         {Opt_nogrpid, "sysvgroups"},
1145         {Opt_resgid, "resgid=%u"},
1146         {Opt_resuid, "resuid=%u"},
1147         {Opt_sb, "sb=%u"},
1148         {Opt_err_cont, "errors=continue"},
1149         {Opt_err_panic, "errors=panic"},
1150         {Opt_err_ro, "errors=remount-ro"},
1151         {Opt_nouid32, "nouid32"},
1152         {Opt_debug, "debug"},
1153         {Opt_removed, "oldalloc"},
1154         {Opt_removed, "orlov"},
1155         {Opt_user_xattr, "user_xattr"},
1156         {Opt_nouser_xattr, "nouser_xattr"},
1157         {Opt_acl, "acl"},
1158         {Opt_noacl, "noacl"},
1159         {Opt_noload, "norecovery"},
1160         {Opt_noload, "noload"},
1161         {Opt_removed, "nobh"},
1162         {Opt_removed, "bh"},
1163         {Opt_commit, "commit=%u"},
1164         {Opt_min_batch_time, "min_batch_time=%u"},
1165         {Opt_max_batch_time, "max_batch_time=%u"},
1166         {Opt_journal_dev, "journal_dev=%u"},
1167         {Opt_journal_checksum, "journal_checksum"},
1168         {Opt_journal_async_commit, "journal_async_commit"},
1169         {Opt_abort, "abort"},
1170         {Opt_data_journal, "data=journal"},
1171         {Opt_data_ordered, "data=ordered"},
1172         {Opt_data_writeback, "data=writeback"},
1173         {Opt_data_err_abort, "data_err=abort"},
1174         {Opt_data_err_ignore, "data_err=ignore"},
1175         {Opt_offusrjquota, "usrjquota="},
1176         {Opt_usrjquota, "usrjquota=%s"},
1177         {Opt_offgrpjquota, "grpjquota="},
1178         {Opt_grpjquota, "grpjquota=%s"},
1179         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1180         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1181         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1182         {Opt_grpquota, "grpquota"},
1183         {Opt_noquota, "noquota"},
1184         {Opt_quota, "quota"},
1185         {Opt_usrquota, "usrquota"},
1186         {Opt_barrier, "barrier=%u"},
1187         {Opt_barrier, "barrier"},
1188         {Opt_nobarrier, "nobarrier"},
1189         {Opt_i_version, "i_version"},
1190         {Opt_stripe, "stripe=%u"},
1191         {Opt_delalloc, "delalloc"},
1192         {Opt_nodelalloc, "nodelalloc"},
1193         {Opt_removed, "mblk_io_submit"},
1194         {Opt_removed, "nomblk_io_submit"},
1195         {Opt_block_validity, "block_validity"},
1196         {Opt_noblock_validity, "noblock_validity"},
1197         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1198         {Opt_journal_ioprio, "journal_ioprio=%u"},
1199         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1200         {Opt_auto_da_alloc, "auto_da_alloc"},
1201         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1202         {Opt_dioread_nolock, "dioread_nolock"},
1203         {Opt_dioread_lock, "dioread_lock"},
1204         {Opt_discard, "discard"},
1205         {Opt_nodiscard, "nodiscard"},
1206         {Opt_init_itable, "init_itable=%u"},
1207         {Opt_init_itable, "init_itable"},
1208         {Opt_noinit_itable, "noinit_itable"},
1209         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1210         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1211         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1212         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1213         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1214         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1215         {Opt_err, NULL},
1216 };
1217
1218 static ext4_fsblk_t get_sb_block(void **data)
1219 {
1220         ext4_fsblk_t    sb_block;
1221         char            *options = (char *) *data;
1222
1223         if (!options || strncmp(options, "sb=", 3) != 0)
1224                 return 1;       /* Default location */
1225
1226         options += 3;
1227         /* TODO: use simple_strtoll with >32bit ext4 */
1228         sb_block = simple_strtoul(options, &options, 0);
1229         if (*options && *options != ',') {
1230                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1231                        (char *) *data);
1232                 return 1;
1233         }
1234         if (*options == ',')
1235                 options++;
1236         *data = (void *) options;
1237
1238         return sb_block;
1239 }
1240
1241 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1242 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1243         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1244
1245 #ifdef CONFIG_QUOTA
1246 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1247 {
1248         struct ext4_sb_info *sbi = EXT4_SB(sb);
1249         char *qname;
1250         int ret = -1;
1251
1252         if (sb_any_quota_loaded(sb) &&
1253                 !sbi->s_qf_names[qtype]) {
1254                 ext4_msg(sb, KERN_ERR,
1255                         "Cannot change journaled "
1256                         "quota options when quota turned on");
1257                 return -1;
1258         }
1259         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1260                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1261                          "when QUOTA feature is enabled");
1262                 return -1;
1263         }
1264         qname = match_strdup(args);
1265         if (!qname) {
1266                 ext4_msg(sb, KERN_ERR,
1267                         "Not enough memory for storing quotafile name");
1268                 return -1;
1269         }
1270         if (sbi->s_qf_names[qtype]) {
1271                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1272                         ret = 1;
1273                 else
1274                         ext4_msg(sb, KERN_ERR,
1275                                  "%s quota file already specified",
1276                                  QTYPE2NAME(qtype));
1277                 goto errout;
1278         }
1279         if (strchr(qname, '/')) {
1280                 ext4_msg(sb, KERN_ERR,
1281                         "quotafile must be on filesystem root");
1282                 goto errout;
1283         }
1284         sbi->s_qf_names[qtype] = qname;
1285         set_opt(sb, QUOTA);
1286         return 1;
1287 errout:
1288         kfree(qname);
1289         return ret;
1290 }
1291
1292 static int clear_qf_name(struct super_block *sb, int qtype)
1293 {
1294
1295         struct ext4_sb_info *sbi = EXT4_SB(sb);
1296
1297         if (sb_any_quota_loaded(sb) &&
1298                 sbi->s_qf_names[qtype]) {
1299                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1300                         " when quota turned on");
1301                 return -1;
1302         }
1303         kfree(sbi->s_qf_names[qtype]);
1304         sbi->s_qf_names[qtype] = NULL;
1305         return 1;
1306 }
1307 #endif
1308
1309 #define MOPT_SET        0x0001
1310 #define MOPT_CLEAR      0x0002
1311 #define MOPT_NOSUPPORT  0x0004
1312 #define MOPT_EXPLICIT   0x0008
1313 #define MOPT_CLEAR_ERR  0x0010
1314 #define MOPT_GTE0       0x0020
1315 #ifdef CONFIG_QUOTA
1316 #define MOPT_Q          0
1317 #define MOPT_QFMT       0x0040
1318 #else
1319 #define MOPT_Q          MOPT_NOSUPPORT
1320 #define MOPT_QFMT       MOPT_NOSUPPORT
1321 #endif
1322 #define MOPT_DATAJ      0x0080
1323 #define MOPT_NO_EXT2    0x0100
1324 #define MOPT_NO_EXT3    0x0200
1325 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1326
1327 static const struct mount_opts {
1328         int     token;
1329         int     mount_opt;
1330         int     flags;
1331 } ext4_mount_opts[] = {
1332         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1333         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1334         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1335         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1336         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1337         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1338         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1339          MOPT_EXT4_ONLY | MOPT_SET},
1340         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1341          MOPT_EXT4_ONLY | MOPT_CLEAR},
1342         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1343         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1344         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1345          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1346         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1347          MOPT_EXT4_ONLY | MOPT_CLEAR | MOPT_EXPLICIT},
1348         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1349          MOPT_EXT4_ONLY | MOPT_SET},
1350         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1351                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1352          MOPT_EXT4_ONLY | MOPT_SET},
1353         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1354         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1355         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1356         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1357         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1358          MOPT_NO_EXT2 | MOPT_SET},
1359         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1360          MOPT_NO_EXT2 | MOPT_CLEAR},
1361         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1362         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1363         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1364         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1365         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1366         {Opt_commit, 0, MOPT_GTE0},
1367         {Opt_max_batch_time, 0, MOPT_GTE0},
1368         {Opt_min_batch_time, 0, MOPT_GTE0},
1369         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1370         {Opt_init_itable, 0, MOPT_GTE0},
1371         {Opt_stripe, 0, MOPT_GTE0},
1372         {Opt_resuid, 0, MOPT_GTE0},
1373         {Opt_resgid, 0, MOPT_GTE0},
1374         {Opt_journal_dev, 0, MOPT_GTE0},
1375         {Opt_journal_ioprio, 0, MOPT_GTE0},
1376         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1377         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1378         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1379          MOPT_NO_EXT2 | MOPT_DATAJ},
1380         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1381         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1382 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1383         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1384         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1385 #else
1386         {Opt_acl, 0, MOPT_NOSUPPORT},
1387         {Opt_noacl, 0, MOPT_NOSUPPORT},
1388 #endif
1389         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1390         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1391         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1392         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1393                                                         MOPT_SET | MOPT_Q},
1394         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1395                                                         MOPT_SET | MOPT_Q},
1396         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1397                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1398         {Opt_usrjquota, 0, MOPT_Q},
1399         {Opt_grpjquota, 0, MOPT_Q},
1400         {Opt_offusrjquota, 0, MOPT_Q},
1401         {Opt_offgrpjquota, 0, MOPT_Q},
1402         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1403         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1404         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1405         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1406         {Opt_err, 0, 0}
1407 };
1408
1409 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1410                             substring_t *args, unsigned long *journal_devnum,
1411                             unsigned int *journal_ioprio, int is_remount)
1412 {
1413         struct ext4_sb_info *sbi = EXT4_SB(sb);
1414         const struct mount_opts *m;
1415         kuid_t uid;
1416         kgid_t gid;
1417         int arg = 0;
1418
1419 #ifdef CONFIG_QUOTA
1420         if (token == Opt_usrjquota)
1421                 return set_qf_name(sb, USRQUOTA, &args[0]);
1422         else if (token == Opt_grpjquota)
1423                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1424         else if (token == Opt_offusrjquota)
1425                 return clear_qf_name(sb, USRQUOTA);
1426         else if (token == Opt_offgrpjquota)
1427                 return clear_qf_name(sb, GRPQUOTA);
1428 #endif
1429         switch (token) {
1430         case Opt_noacl:
1431         case Opt_nouser_xattr:
1432                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1433                 break;
1434         case Opt_sb:
1435                 return 1;       /* handled by get_sb_block() */
1436         case Opt_removed:
1437                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1438                 return 1;
1439         case Opt_abort:
1440                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1441                 return 1;
1442         case Opt_i_version:
1443                 sb->s_flags |= MS_I_VERSION;
1444                 return 1;
1445         }
1446
1447         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1448                 if (token == m->token)
1449                         break;
1450
1451         if (m->token == Opt_err) {
1452                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1453                          "or missing value", opt);
1454                 return -1;
1455         }
1456
1457         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1458                 ext4_msg(sb, KERN_ERR,
1459                          "Mount option \"%s\" incompatible with ext2", opt);
1460                 return -1;
1461         }
1462         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1463                 ext4_msg(sb, KERN_ERR,
1464                          "Mount option \"%s\" incompatible with ext3", opt);
1465                 return -1;
1466         }
1467
1468         if (args->from && match_int(args, &arg))
1469                 return -1;
1470         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1471                 return -1;
1472         if (m->flags & MOPT_EXPLICIT)
1473                 set_opt2(sb, EXPLICIT_DELALLOC);
1474         if (m->flags & MOPT_CLEAR_ERR)
1475                 clear_opt(sb, ERRORS_MASK);
1476         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1477                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1478                          "options when quota turned on");
1479                 return -1;
1480         }
1481
1482         if (m->flags & MOPT_NOSUPPORT) {
1483                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1484         } else if (token == Opt_commit) {
1485                 if (arg == 0)
1486                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1487                 sbi->s_commit_interval = HZ * arg;
1488         } else if (token == Opt_max_batch_time) {
1489                 if (arg == 0)
1490                         arg = EXT4_DEF_MAX_BATCH_TIME;
1491                 sbi->s_max_batch_time = arg;
1492         } else if (token == Opt_min_batch_time) {
1493                 sbi->s_min_batch_time = arg;
1494         } else if (token == Opt_inode_readahead_blks) {
1495                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1496                         ext4_msg(sb, KERN_ERR,
1497                                  "EXT4-fs: inode_readahead_blks must be "
1498                                  "0 or a power of 2 smaller than 2^31");
1499                         return -1;
1500                 }
1501                 sbi->s_inode_readahead_blks = arg;
1502         } else if (token == Opt_init_itable) {
1503                 set_opt(sb, INIT_INODE_TABLE);
1504                 if (!args->from)
1505                         arg = EXT4_DEF_LI_WAIT_MULT;
1506                 sbi->s_li_wait_mult = arg;
1507         } else if (token == Opt_max_dir_size_kb) {
1508                 sbi->s_max_dir_size_kb = arg;
1509         } else if (token == Opt_stripe) {
1510                 sbi->s_stripe = arg;
1511         } else if (token == Opt_resuid) {
1512                 uid = make_kuid(current_user_ns(), arg);
1513                 if (!uid_valid(uid)) {
1514                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1515                         return -1;
1516                 }
1517                 sbi->s_resuid = uid;
1518         } else if (token == Opt_resgid) {
1519                 gid = make_kgid(current_user_ns(), arg);
1520                 if (!gid_valid(gid)) {
1521                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1522                         return -1;
1523                 }
1524                 sbi->s_resgid = gid;
1525         } else if (token == Opt_journal_dev) {
1526                 if (is_remount) {
1527                         ext4_msg(sb, KERN_ERR,
1528                                  "Cannot specify journal on remount");
1529                         return -1;
1530                 }
1531                 *journal_devnum = arg;
1532         } else if (token == Opt_journal_ioprio) {
1533                 if (arg > 7) {
1534                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1535                                  " (must be 0-7)");
1536                         return -1;
1537                 }
1538                 *journal_ioprio =
1539                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1540         } else if (m->flags & MOPT_DATAJ) {
1541                 if (is_remount) {
1542                         if (!sbi->s_journal)
1543                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1544                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1545                                 ext4_msg(sb, KERN_ERR,
1546                                          "Cannot change data mode on remount");
1547                                 return -1;
1548                         }
1549                 } else {
1550                         clear_opt(sb, DATA_FLAGS);
1551                         sbi->s_mount_opt |= m->mount_opt;
1552                 }
1553 #ifdef CONFIG_QUOTA
1554         } else if (m->flags & MOPT_QFMT) {
1555                 if (sb_any_quota_loaded(sb) &&
1556                     sbi->s_jquota_fmt != m->mount_opt) {
1557                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1558                                  "quota options when quota turned on");
1559                         return -1;
1560                 }
1561                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1562                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1563                         ext4_msg(sb, KERN_ERR,
1564                                  "Cannot set journaled quota options "
1565                                  "when QUOTA feature is enabled");
1566                         return -1;
1567                 }
1568                 sbi->s_jquota_fmt = m->mount_opt;
1569 #endif
1570         } else {
1571                 if (!args->from)
1572                         arg = 1;
1573                 if (m->flags & MOPT_CLEAR)
1574                         arg = !arg;
1575                 else if (unlikely(!(m->flags & MOPT_SET))) {
1576                         ext4_msg(sb, KERN_WARNING,
1577                                  "buggy handling of option %s", opt);
1578                         WARN_ON(1);
1579                         return -1;
1580                 }
1581                 if (arg != 0)
1582                         sbi->s_mount_opt |= m->mount_opt;
1583                 else
1584                         sbi->s_mount_opt &= ~m->mount_opt;
1585         }
1586         return 1;
1587 }
1588
1589 static int parse_options(char *options, struct super_block *sb,
1590                          unsigned long *journal_devnum,
1591                          unsigned int *journal_ioprio,
1592                          int is_remount)
1593 {
1594         struct ext4_sb_info *sbi = EXT4_SB(sb);
1595         char *p;
1596         substring_t args[MAX_OPT_ARGS];
1597         int token;
1598
1599         if (!options)
1600                 return 1;
1601
1602         while ((p = strsep(&options, ",")) != NULL) {
1603                 if (!*p)
1604                         continue;
1605                 /*
1606                  * Initialize args struct so we know whether arg was
1607                  * found; some options take optional arguments.
1608                  */
1609                 args[0].to = args[0].from = NULL;
1610                 token = match_token(p, tokens, args);
1611                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1612                                      journal_ioprio, is_remount) < 0)
1613                         return 0;
1614         }
1615 #ifdef CONFIG_QUOTA
1616         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1617             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1618                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1619                          "feature is enabled");
1620                 return 0;
1621         }
1622         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1623                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1624                         clear_opt(sb, USRQUOTA);
1625
1626                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1627                         clear_opt(sb, GRPQUOTA);
1628
1629                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1630                         ext4_msg(sb, KERN_ERR, "old and new quota "
1631                                         "format mixing");
1632                         return 0;
1633                 }
1634
1635                 if (!sbi->s_jquota_fmt) {
1636                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1637                                         "not specified");
1638                         return 0;
1639                 }
1640         } else {
1641                 if (sbi->s_jquota_fmt) {
1642                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1643                                         "specified with no journaling "
1644                                         "enabled");
1645                         return 0;
1646                 }
1647         }
1648 #endif
1649         if (test_opt(sb, DIOREAD_NOLOCK)) {
1650                 int blocksize =
1651                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1652
1653                 if (blocksize < PAGE_CACHE_SIZE) {
1654                         ext4_msg(sb, KERN_ERR, "can't mount with "
1655                                  "dioread_nolock if block size != PAGE_SIZE");
1656                         return 0;
1657                 }
1658         }
1659         return 1;
1660 }
1661
1662 static inline void ext4_show_quota_options(struct seq_file *seq,
1663                                            struct super_block *sb)
1664 {
1665 #if defined(CONFIG_QUOTA)
1666         struct ext4_sb_info *sbi = EXT4_SB(sb);
1667
1668         if (sbi->s_jquota_fmt) {
1669                 char *fmtname = "";
1670
1671                 switch (sbi->s_jquota_fmt) {
1672                 case QFMT_VFS_OLD:
1673                         fmtname = "vfsold";
1674                         break;
1675                 case QFMT_VFS_V0:
1676                         fmtname = "vfsv0";
1677                         break;
1678                 case QFMT_VFS_V1:
1679                         fmtname = "vfsv1";
1680                         break;
1681                 }
1682                 seq_printf(seq, ",jqfmt=%s", fmtname);
1683         }
1684
1685         if (sbi->s_qf_names[USRQUOTA])
1686                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1687
1688         if (sbi->s_qf_names[GRPQUOTA])
1689                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1690
1691         if (test_opt(sb, USRQUOTA))
1692                 seq_puts(seq, ",usrquota");
1693
1694         if (test_opt(sb, GRPQUOTA))
1695                 seq_puts(seq, ",grpquota");
1696 #endif
1697 }
1698
1699 static const char *token2str(int token)
1700 {
1701         const struct match_token *t;
1702
1703         for (t = tokens; t->token != Opt_err; t++)
1704                 if (t->token == token && !strchr(t->pattern, '='))
1705                         break;
1706         return t->pattern;
1707 }
1708
1709 /*
1710  * Show an option if
1711  *  - it's set to a non-default value OR
1712  *  - if the per-sb default is different from the global default
1713  */
1714 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1715                               int nodefs)
1716 {
1717         struct ext4_sb_info *sbi = EXT4_SB(sb);
1718         struct ext4_super_block *es = sbi->s_es;
1719         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1720         const struct mount_opts *m;
1721         char sep = nodefs ? '\n' : ',';
1722
1723 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1724 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1725
1726         if (sbi->s_sb_block != 1)
1727                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1728
1729         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1730                 int want_set = m->flags & MOPT_SET;
1731                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1732                     (m->flags & MOPT_CLEAR_ERR))
1733                         continue;
1734                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1735                         continue; /* skip if same as the default */
1736                 if ((want_set &&
1737                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1738                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1739                         continue; /* select Opt_noFoo vs Opt_Foo */
1740                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1741         }
1742
1743         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1744             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1745                 SEQ_OPTS_PRINT("resuid=%u",
1746                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1747         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1748             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1749                 SEQ_OPTS_PRINT("resgid=%u",
1750                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1751         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1752         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1753                 SEQ_OPTS_PUTS("errors=remount-ro");
1754         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1755                 SEQ_OPTS_PUTS("errors=continue");
1756         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1757                 SEQ_OPTS_PUTS("errors=panic");
1758         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1759                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1760         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1761                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1762         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1763                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1764         if (sb->s_flags & MS_I_VERSION)
1765                 SEQ_OPTS_PUTS("i_version");
1766         if (nodefs || sbi->s_stripe)
1767                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1768         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1769                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1770                         SEQ_OPTS_PUTS("data=journal");
1771                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1772                         SEQ_OPTS_PUTS("data=ordered");
1773                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1774                         SEQ_OPTS_PUTS("data=writeback");
1775         }
1776         if (nodefs ||
1777             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1778                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1779                                sbi->s_inode_readahead_blks);
1780
1781         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1782                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1783                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1784         if (nodefs || sbi->s_max_dir_size_kb)
1785                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1786
1787         ext4_show_quota_options(seq, sb);
1788         return 0;
1789 }
1790
1791 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1792 {
1793         return _ext4_show_options(seq, root->d_sb, 0);
1794 }
1795
1796 static int options_seq_show(struct seq_file *seq, void *offset)
1797 {
1798         struct super_block *sb = seq->private;
1799         int rc;
1800
1801         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1802         rc = _ext4_show_options(seq, sb, 1);
1803         seq_puts(seq, "\n");
1804         return rc;
1805 }
1806
1807 static int options_open_fs(struct inode *inode, struct file *file)
1808 {
1809         return single_open(file, options_seq_show, PDE(inode)->data);
1810 }
1811
1812 static const struct file_operations ext4_seq_options_fops = {
1813         .owner = THIS_MODULE,
1814         .open = options_open_fs,
1815         .read = seq_read,
1816         .llseek = seq_lseek,
1817         .release = single_release,
1818 };
1819
1820 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1821                             int read_only)
1822 {
1823         struct ext4_sb_info *sbi = EXT4_SB(sb);
1824         int res = 0;
1825
1826         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1827                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1828                          "forcing read-only mode");
1829                 res = MS_RDONLY;
1830         }
1831         if (read_only)
1832                 goto done;
1833         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1834                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1835                          "running e2fsck is recommended");
1836         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1837                 ext4_msg(sb, KERN_WARNING,
1838                          "warning: mounting fs with errors, "
1839                          "running e2fsck is recommended");
1840         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1841                  le16_to_cpu(es->s_mnt_count) >=
1842                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1843                 ext4_msg(sb, KERN_WARNING,
1844                          "warning: maximal mount count reached, "
1845                          "running e2fsck is recommended");
1846         else if (le32_to_cpu(es->s_checkinterval) &&
1847                 (le32_to_cpu(es->s_lastcheck) +
1848                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1849                 ext4_msg(sb, KERN_WARNING,
1850                          "warning: checktime reached, "
1851                          "running e2fsck is recommended");
1852         if (!sbi->s_journal)
1853                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1854         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1855                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1856         le16_add_cpu(&es->s_mnt_count, 1);
1857         es->s_mtime = cpu_to_le32(get_seconds());
1858         ext4_update_dynamic_rev(sb);
1859         if (sbi->s_journal)
1860                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1861
1862         ext4_commit_super(sb, 1);
1863 done:
1864         if (test_opt(sb, DEBUG))
1865                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1866                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1867                         sb->s_blocksize,
1868                         sbi->s_groups_count,
1869                         EXT4_BLOCKS_PER_GROUP(sb),
1870                         EXT4_INODES_PER_GROUP(sb),
1871                         sbi->s_mount_opt, sbi->s_mount_opt2);
1872
1873         cleancache_init_fs(sb);
1874         return res;
1875 }
1876
1877 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1878 {
1879         struct ext4_sb_info *sbi = EXT4_SB(sb);
1880         struct flex_groups *new_groups;
1881         int size;
1882
1883         if (!sbi->s_log_groups_per_flex)
1884                 return 0;
1885
1886         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1887         if (size <= sbi->s_flex_groups_allocated)
1888                 return 0;
1889
1890         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1891         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1892         if (!new_groups) {
1893                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1894                          size / (int) sizeof(struct flex_groups));
1895                 return -ENOMEM;
1896         }
1897
1898         if (sbi->s_flex_groups) {
1899                 memcpy(new_groups, sbi->s_flex_groups,
1900                        (sbi->s_flex_groups_allocated *
1901                         sizeof(struct flex_groups)));
1902                 ext4_kvfree(sbi->s_flex_groups);
1903         }
1904         sbi->s_flex_groups = new_groups;
1905         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1906         return 0;
1907 }
1908
1909 static int ext4_fill_flex_info(struct super_block *sb)
1910 {
1911         struct ext4_sb_info *sbi = EXT4_SB(sb);
1912         struct ext4_group_desc *gdp = NULL;
1913         ext4_group_t flex_group;
1914         unsigned int groups_per_flex = 0;
1915         int i, err;
1916
1917         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1918         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1919                 sbi->s_log_groups_per_flex = 0;
1920                 return 1;
1921         }
1922         groups_per_flex = 1U << sbi->s_log_groups_per_flex;
1923
1924         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1925         if (err)
1926                 goto failed;
1927
1928         for (i = 0; i < sbi->s_groups_count; i++) {
1929                 gdp = ext4_get_group_desc(sb, i, NULL);
1930
1931                 flex_group = ext4_flex_group(sbi, i);
1932                 atomic_add(ext4_free_inodes_count(sb, gdp),
1933                            &sbi->s_flex_groups[flex_group].free_inodes);
1934                 atomic64_add(ext4_free_group_clusters(sb, gdp),
1935                              &sbi->s_flex_groups[flex_group].free_clusters);
1936                 atomic_add(ext4_used_dirs_count(sb, gdp),
1937                            &sbi->s_flex_groups[flex_group].used_dirs);
1938         }
1939
1940         return 1;
1941 failed:
1942         return 0;
1943 }
1944
1945 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1946                                    struct ext4_group_desc *gdp)
1947 {
1948         int offset;
1949         __u16 crc = 0;
1950         __le32 le_group = cpu_to_le32(block_group);
1951
1952         if ((sbi->s_es->s_feature_ro_compat &
1953              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1954                 /* Use new metadata_csum algorithm */
1955                 __le16 save_csum;
1956                 __u32 csum32;
1957
1958                 save_csum = gdp->bg_checksum;
1959                 gdp->bg_checksum = 0;
1960                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1961                                      sizeof(le_group));
1962                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1963                                      sbi->s_desc_size);
1964                 gdp->bg_checksum = save_csum;
1965
1966                 crc = csum32 & 0xFFFF;
1967                 goto out;
1968         }
1969
1970         /* old crc16 code */
1971         offset = offsetof(struct ext4_group_desc, bg_checksum);
1972
1973         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1974         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1975         crc = crc16(crc, (__u8 *)gdp, offset);
1976         offset += sizeof(gdp->bg_checksum); /* skip checksum */
1977         /* for checksum of struct ext4_group_desc do the rest...*/
1978         if ((sbi->s_es->s_feature_incompat &
1979              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1980             offset < le16_to_cpu(sbi->s_es->s_desc_size))
1981                 crc = crc16(crc, (__u8 *)gdp + offset,
1982                             le16_to_cpu(sbi->s_es->s_desc_size) -
1983                                 offset);
1984
1985 out:
1986         return cpu_to_le16(crc);
1987 }
1988
1989 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
1990                                 struct ext4_group_desc *gdp)
1991 {
1992         if (ext4_has_group_desc_csum(sb) &&
1993             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
1994                                                       block_group, gdp)))
1995                 return 0;
1996
1997         return 1;
1998 }
1999
2000 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2001                               struct ext4_group_desc *gdp)
2002 {
2003         if (!ext4_has_group_desc_csum(sb))
2004                 return;
2005         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2006 }
2007
2008 /* Called at mount-time, super-block is locked */
2009 static int ext4_check_descriptors(struct super_block *sb,
2010                                   ext4_group_t *first_not_zeroed)
2011 {
2012         struct ext4_sb_info *sbi = EXT4_SB(sb);
2013         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2014         ext4_fsblk_t last_block;
2015         ext4_fsblk_t block_bitmap;
2016         ext4_fsblk_t inode_bitmap;
2017         ext4_fsblk_t inode_table;
2018         int flexbg_flag = 0;
2019         ext4_group_t i, grp = sbi->s_groups_count;
2020
2021         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2022                 flexbg_flag = 1;
2023
2024         ext4_debug("Checking group descriptors");
2025
2026         for (i = 0; i < sbi->s_groups_count; i++) {
2027                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2028
2029                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2030                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2031                 else
2032                         last_block = first_block +
2033                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2034
2035                 if ((grp == sbi->s_groups_count) &&
2036                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2037                         grp = i;
2038
2039                 block_bitmap = ext4_block_bitmap(sb, gdp);
2040                 if (block_bitmap < first_block || block_bitmap > last_block) {
2041                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2042                                "Block bitmap for group %u not in group "
2043                                "(block %llu)!", i, block_bitmap);
2044                         return 0;
2045                 }
2046                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2047                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2048                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2049                                "Inode bitmap for group %u not in group "
2050                                "(block %llu)!", i, inode_bitmap);
2051                         return 0;
2052                 }
2053                 inode_table = ext4_inode_table(sb, gdp);
2054                 if (inode_table < first_block ||
2055                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2056                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2057                                "Inode table for group %u not in group "
2058                                "(block %llu)!", i, inode_table);
2059                         return 0;
2060                 }
2061                 ext4_lock_group(sb, i);
2062                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2063                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2064                                  "Checksum for group %u failed (%u!=%u)",
2065                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2066                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2067                         if (!(sb->s_flags & MS_RDONLY)) {
2068                                 ext4_unlock_group(sb, i);
2069                                 return 0;
2070                         }
2071                 }
2072                 ext4_unlock_group(sb, i);
2073                 if (!flexbg_flag)
2074                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2075         }
2076         if (NULL != first_not_zeroed)
2077                 *first_not_zeroed = grp;
2078
2079         ext4_free_blocks_count_set(sbi->s_es,
2080                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2081         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2082         return 1;
2083 }
2084
2085 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2086  * the superblock) which were deleted from all directories, but held open by
2087  * a process at the time of a crash.  We walk the list and try to delete these
2088  * inodes at recovery time (only with a read-write filesystem).
2089  *
2090  * In order to keep the orphan inode chain consistent during traversal (in
2091  * case of crash during recovery), we link each inode into the superblock
2092  * orphan list_head and handle it the same way as an inode deletion during
2093  * normal operation (which journals the operations for us).
2094  *
2095  * We only do an iget() and an iput() on each inode, which is very safe if we
2096  * accidentally point at an in-use or already deleted inode.  The worst that
2097  * can happen in this case is that we get a "bit already cleared" message from
2098  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2099  * e2fsck was run on this filesystem, and it must have already done the orphan
2100  * inode cleanup for us, so we can safely abort without any further action.
2101  */
2102 static void ext4_orphan_cleanup(struct super_block *sb,
2103                                 struct ext4_super_block *es)
2104 {
2105         unsigned int s_flags = sb->s_flags;
2106         int nr_orphans = 0, nr_truncates = 0;
2107 #ifdef CONFIG_QUOTA
2108         int i;
2109 #endif
2110         if (!es->s_last_orphan) {
2111                 jbd_debug(4, "no orphan inodes to clean up\n");
2112                 return;
2113         }
2114
2115         if (bdev_read_only(sb->s_bdev)) {
2116                 ext4_msg(sb, KERN_ERR, "write access "
2117                         "unavailable, skipping orphan cleanup");
2118                 return;
2119         }
2120
2121         /* Check if feature set would not allow a r/w mount */
2122         if (!ext4_feature_set_ok(sb, 0)) {
2123                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2124                          "unknown ROCOMPAT features");
2125                 return;
2126         }
2127
2128         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2129                 /* don't clear list on RO mount w/ errors */
2130                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2131                         jbd_debug(1, "Errors on filesystem, "
2132                                   "clearing orphan list.\n");
2133                         es->s_last_orphan = 0;
2134                 }
2135                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2136                 return;
2137         }
2138
2139         if (s_flags & MS_RDONLY) {
2140                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2141                 sb->s_flags &= ~MS_RDONLY;
2142         }
2143 #ifdef CONFIG_QUOTA
2144         /* Needed for iput() to work correctly and not trash data */
2145         sb->s_flags |= MS_ACTIVE;
2146         /* Turn on quotas so that they are updated correctly */
2147         for (i = 0; i < MAXQUOTAS; i++) {
2148                 if (EXT4_SB(sb)->s_qf_names[i]) {
2149                         int ret = ext4_quota_on_mount(sb, i);
2150                         if (ret < 0)
2151                                 ext4_msg(sb, KERN_ERR,
2152                                         "Cannot turn on journaled "
2153                                         "quota: error %d", ret);
2154                 }
2155         }
2156 #endif
2157
2158         while (es->s_last_orphan) {
2159                 struct inode *inode;
2160
2161                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2162                 if (IS_ERR(inode)) {
2163                         es->s_last_orphan = 0;
2164                         break;
2165                 }
2166
2167                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2168                 dquot_initialize(inode);
2169                 if (inode->i_nlink) {
2170                         ext4_msg(sb, KERN_DEBUG,
2171                                 "%s: truncating inode %lu to %lld bytes",
2172                                 __func__, inode->i_ino, inode->i_size);
2173                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2174                                   inode->i_ino, inode->i_size);
2175                         mutex_lock(&inode->i_mutex);
2176                         ext4_truncate(inode);
2177                         mutex_unlock(&inode->i_mutex);
2178                         nr_truncates++;
2179                 } else {
2180                         ext4_msg(sb, KERN_DEBUG,
2181                                 "%s: deleting unreferenced inode %lu",
2182                                 __func__, inode->i_ino);
2183                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2184                                   inode->i_ino);
2185                         nr_orphans++;
2186                 }
2187                 iput(inode);  /* The delete magic happens here! */
2188         }
2189
2190 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2191
2192         if (nr_orphans)
2193                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2194                        PLURAL(nr_orphans));
2195         if (nr_truncates)
2196                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2197                        PLURAL(nr_truncates));
2198 #ifdef CONFIG_QUOTA
2199         /* Turn quotas off */
2200         for (i = 0; i < MAXQUOTAS; i++) {
2201                 if (sb_dqopt(sb)->files[i])
2202                         dquot_quota_off(sb, i);
2203         }
2204 #endif
2205         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2206 }
2207
2208 /*
2209  * Maximal extent format file size.
2210  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2211  * extent format containers, within a sector_t, and within i_blocks
2212  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2213  * so that won't be a limiting factor.
2214  *
2215  * However there is other limiting factor. We do store extents in the form
2216  * of starting block and length, hence the resulting length of the extent
2217  * covering maximum file size must fit into on-disk format containers as
2218  * well. Given that length is always by 1 unit bigger than max unit (because
2219  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2220  *
2221  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2222  */
2223 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2224 {
2225         loff_t res;
2226         loff_t upper_limit = MAX_LFS_FILESIZE;
2227
2228         /* small i_blocks in vfs inode? */
2229         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2230                 /*
2231                  * CONFIG_LBDAF is not enabled implies the inode
2232                  * i_block represent total blocks in 512 bytes
2233                  * 32 == size of vfs inode i_blocks * 8
2234                  */
2235                 upper_limit = (1LL << 32) - 1;
2236
2237                 /* total blocks in file system block size */
2238                 upper_limit >>= (blkbits - 9);
2239                 upper_limit <<= blkbits;
2240         }
2241
2242         /*
2243          * 32-bit extent-start container, ee_block. We lower the maxbytes
2244          * by one fs block, so ee_len can cover the extent of maximum file
2245          * size
2246          */
2247         res = (1LL << 32) - 1;
2248         res <<= blkbits;
2249
2250         /* Sanity check against vm- & vfs- imposed limits */
2251         if (res > upper_limit)
2252                 res = upper_limit;
2253
2254         return res;
2255 }
2256
2257 /*
2258  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2259  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2260  * We need to be 1 filesystem block less than the 2^48 sector limit.
2261  */
2262 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2263 {
2264         loff_t res = EXT4_NDIR_BLOCKS;
2265         int meta_blocks;
2266         loff_t upper_limit;
2267         /* This is calculated to be the largest file size for a dense, block
2268          * mapped file such that the file's total number of 512-byte sectors,
2269          * including data and all indirect blocks, does not exceed (2^48 - 1).
2270          *
2271          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2272          * number of 512-byte sectors of the file.
2273          */
2274
2275         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2276                 /*
2277                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2278                  * the inode i_block field represents total file blocks in
2279                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2280                  */
2281                 upper_limit = (1LL << 32) - 1;
2282
2283                 /* total blocks in file system block size */
2284                 upper_limit >>= (bits - 9);
2285
2286         } else {
2287                 /*
2288                  * We use 48 bit ext4_inode i_blocks
2289                  * With EXT4_HUGE_FILE_FL set the i_blocks
2290                  * represent total number of blocks in
2291                  * file system block size
2292                  */
2293                 upper_limit = (1LL << 48) - 1;
2294
2295         }
2296
2297         /* indirect blocks */
2298         meta_blocks = 1;
2299         /* double indirect blocks */
2300         meta_blocks += 1 + (1LL << (bits-2));
2301         /* tripple indirect blocks */
2302         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2303
2304         upper_limit -= meta_blocks;
2305         upper_limit <<= bits;
2306
2307         res += 1LL << (bits-2);
2308         res += 1LL << (2*(bits-2));
2309         res += 1LL << (3*(bits-2));
2310         res <<= bits;
2311         if (res > upper_limit)
2312                 res = upper_limit;
2313
2314         if (res > MAX_LFS_FILESIZE)
2315                 res = MAX_LFS_FILESIZE;
2316
2317         return res;
2318 }
2319
2320 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2321                                    ext4_fsblk_t logical_sb_block, int nr)
2322 {
2323         struct ext4_sb_info *sbi = EXT4_SB(sb);
2324         ext4_group_t bg, first_meta_bg;
2325         int has_super = 0;
2326
2327         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2328
2329         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2330             nr < first_meta_bg)
2331                 return logical_sb_block + nr + 1;
2332         bg = sbi->s_desc_per_block * nr;
2333         if (ext4_bg_has_super(sb, bg))
2334                 has_super = 1;
2335
2336         return (has_super + ext4_group_first_block_no(sb, bg));
2337 }
2338
2339 /**
2340  * ext4_get_stripe_size: Get the stripe size.
2341  * @sbi: In memory super block info
2342  *
2343  * If we have specified it via mount option, then
2344  * use the mount option value. If the value specified at mount time is
2345  * greater than the blocks per group use the super block value.
2346  * If the super block value is greater than blocks per group return 0.
2347  * Allocator needs it be less than blocks per group.
2348  *
2349  */
2350 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2351 {
2352         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2353         unsigned long stripe_width =
2354                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2355         int ret;
2356
2357         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2358                 ret = sbi->s_stripe;
2359         else if (stripe_width <= sbi->s_blocks_per_group)
2360                 ret = stripe_width;
2361         else if (stride <= sbi->s_blocks_per_group)
2362                 ret = stride;
2363         else
2364                 ret = 0;
2365
2366         /*
2367          * If the stripe width is 1, this makes no sense and
2368          * we set it to 0 to turn off stripe handling code.
2369          */
2370         if (ret <= 1)
2371                 ret = 0;
2372
2373         return ret;
2374 }
2375
2376 /* sysfs supprt */
2377
2378 struct ext4_attr {
2379         struct attribute attr;
2380         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2381         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2382                          const char *, size_t);
2383         int offset;
2384 };
2385
2386 static int parse_strtoull(const char *buf,
2387                 unsigned long long max, unsigned long long *value)
2388 {
2389         int ret;
2390
2391         ret = kstrtoull(skip_spaces(buf), 0, value);
2392         if (!ret && *value > max)
2393                 ret = -EINVAL;
2394         return ret;
2395 }
2396
2397 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2398                                               struct ext4_sb_info *sbi,
2399                                               char *buf)
2400 {
2401         return snprintf(buf, PAGE_SIZE, "%llu\n",
2402                 (s64) EXT4_C2B(sbi,
2403                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2404 }
2405
2406 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2407                                          struct ext4_sb_info *sbi, char *buf)
2408 {
2409         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2410
2411         if (!sb->s_bdev->bd_part)
2412                 return snprintf(buf, PAGE_SIZE, "0\n");
2413         return snprintf(buf, PAGE_SIZE, "%lu\n",
2414                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2415                          sbi->s_sectors_written_start) >> 1);
2416 }
2417
2418 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2419                                           struct ext4_sb_info *sbi, char *buf)
2420 {
2421         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2422
2423         if (!sb->s_bdev->bd_part)
2424                 return snprintf(buf, PAGE_SIZE, "0\n");
2425         return snprintf(buf, PAGE_SIZE, "%llu\n",
2426                         (unsigned long long)(sbi->s_kbytes_written +
2427                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2428                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2429 }
2430
2431 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2432                                           struct ext4_sb_info *sbi,
2433                                           const char *buf, size_t count)
2434 {
2435         unsigned long t;
2436         int ret;
2437
2438         ret = kstrtoul(skip_spaces(buf), 0, &t);
2439         if (ret)
2440                 return ret;
2441
2442         if (t && (!is_power_of_2(t) || t > 0x40000000))
2443                 return -EINVAL;
2444
2445         sbi->s_inode_readahead_blks = t;
2446         return count;
2447 }
2448
2449 static ssize_t sbi_ui_show(struct ext4_attr *a,
2450                            struct ext4_sb_info *sbi, char *buf)
2451 {
2452         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2453
2454         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2455 }
2456
2457 static ssize_t sbi_ui_store(struct ext4_attr *a,
2458                             struct ext4_sb_info *sbi,
2459                             const char *buf, size_t count)
2460 {
2461         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->offset);
2462         unsigned long t;
2463         int ret;
2464
2465         ret = kstrtoul(skip_spaces(buf), 0, &t);
2466         if (ret)
2467                 return ret;
2468         *ui = t;
2469         return count;
2470 }
2471
2472 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2473                                   struct ext4_sb_info *sbi, char *buf)
2474 {
2475         return snprintf(buf, PAGE_SIZE, "%llu\n",
2476                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2477 }
2478
2479 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2480                                    struct ext4_sb_info *sbi,
2481                                    const char *buf, size_t count)
2482 {
2483         unsigned long long val;
2484         int ret;
2485
2486         if (parse_strtoull(buf, -1ULL, &val))
2487                 return -EINVAL;
2488         ret = ext4_reserve_clusters(sbi, val);
2489
2490         return ret ? ret : count;
2491 }
2492
2493 static ssize_t trigger_test_error(struct ext4_attr *a,
2494                                   struct ext4_sb_info *sbi,
2495                                   const char *buf, size_t count)
2496 {
2497         int len = count;
2498
2499         if (!capable(CAP_SYS_ADMIN))
2500                 return -EPERM;
2501
2502         if (len && buf[len-1] == '\n')
2503                 len--;
2504
2505         if (len)
2506                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2507         return count;
2508 }
2509
2510 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2511 static struct ext4_attr ext4_attr_##_name = {                   \
2512         .attr = {.name = __stringify(_name), .mode = _mode },   \
2513         .show   = _show,                                        \
2514         .store  = _store,                                       \
2515         .offset = offsetof(struct ext4_sb_info, _elname),       \
2516 }
2517 #define EXT4_ATTR(name, mode, show, store) \
2518 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2519
2520 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2521 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2522 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2523 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2524         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2525 #define ATTR_LIST(name) &ext4_attr_##name.attr
2526
2527 EXT4_RO_ATTR(delayed_allocation_blocks);
2528 EXT4_RO_ATTR(session_write_kbytes);
2529 EXT4_RO_ATTR(lifetime_write_kbytes);
2530 EXT4_RW_ATTR(reserved_clusters);
2531 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2532                  inode_readahead_blks_store, s_inode_readahead_blks);
2533 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2534 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2535 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2536 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2537 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2538 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2539 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2540 EXT4_RW_ATTR_SBI_UI(max_writeback_mb_bump, s_max_writeback_mb_bump);
2541 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2542 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2543
2544 static struct attribute *ext4_attrs[] = {
2545         ATTR_LIST(delayed_allocation_blocks),
2546         ATTR_LIST(session_write_kbytes),
2547         ATTR_LIST(lifetime_write_kbytes),
2548         ATTR_LIST(reserved_clusters),
2549         ATTR_LIST(inode_readahead_blks),
2550         ATTR_LIST(inode_goal),
2551         ATTR_LIST(mb_stats),
2552         ATTR_LIST(mb_max_to_scan),
2553         ATTR_LIST(mb_min_to_scan),
2554         ATTR_LIST(mb_order2_req),
2555         ATTR_LIST(mb_stream_req),
2556         ATTR_LIST(mb_group_prealloc),
2557         ATTR_LIST(max_writeback_mb_bump),
2558         ATTR_LIST(extent_max_zeroout_kb),
2559         ATTR_LIST(trigger_fs_error),
2560         NULL,
2561 };
2562
2563 /* Features this copy of ext4 supports */
2564 EXT4_INFO_ATTR(lazy_itable_init);
2565 EXT4_INFO_ATTR(batched_discard);
2566 EXT4_INFO_ATTR(meta_bg_resize);
2567
2568 static struct attribute *ext4_feat_attrs[] = {
2569         ATTR_LIST(lazy_itable_init),
2570         ATTR_LIST(batched_discard),
2571         ATTR_LIST(meta_bg_resize),
2572         NULL,
2573 };
2574
2575 static ssize_t ext4_attr_show(struct kobject *kobj,
2576                               struct attribute *attr, char *buf)
2577 {
2578         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2579                                                 s_kobj);
2580         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2581
2582         return a->show ? a->show(a, sbi, buf) : 0;
2583 }
2584
2585 static ssize_t ext4_attr_store(struct kobject *kobj,
2586                                struct attribute *attr,
2587                                const char *buf, size_t len)
2588 {
2589         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2590                                                 s_kobj);
2591         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2592
2593         return a->store ? a->store(a, sbi, buf, len) : 0;
2594 }
2595
2596 static void ext4_sb_release(struct kobject *kobj)
2597 {
2598         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2599                                                 s_kobj);
2600         complete(&sbi->s_kobj_unregister);
2601 }
2602
2603 static const struct sysfs_ops ext4_attr_ops = {
2604         .show   = ext4_attr_show,
2605         .store  = ext4_attr_store,
2606 };
2607
2608 static struct kobj_type ext4_ktype = {
2609         .default_attrs  = ext4_attrs,
2610         .sysfs_ops      = &ext4_attr_ops,
2611         .release        = ext4_sb_release,
2612 };
2613
2614 static void ext4_feat_release(struct kobject *kobj)
2615 {
2616         complete(&ext4_feat->f_kobj_unregister);
2617 }
2618
2619 static struct kobj_type ext4_feat_ktype = {
2620         .default_attrs  = ext4_feat_attrs,
2621         .sysfs_ops      = &ext4_attr_ops,
2622         .release        = ext4_feat_release,
2623 };
2624
2625 /*
2626  * Check whether this filesystem can be mounted based on
2627  * the features present and the RDONLY/RDWR mount requested.
2628  * Returns 1 if this filesystem can be mounted as requested,
2629  * 0 if it cannot be.
2630  */
2631 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2632 {
2633         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2634                 ext4_msg(sb, KERN_ERR,
2635                         "Couldn't mount because of "
2636                         "unsupported optional features (%x)",
2637                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2638                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2639                 return 0;
2640         }
2641
2642         if (readonly)
2643                 return 1;
2644
2645         /* Check that feature set is OK for a read-write mount */
2646         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2647                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2648                          "unsupported optional features (%x)",
2649                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2650                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2651                 return 0;
2652         }
2653         /*
2654          * Large file size enabled file system can only be mounted
2655          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2656          */
2657         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2658                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2659                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2660                                  "cannot be mounted RDWR without "
2661                                  "CONFIG_LBDAF");
2662                         return 0;
2663                 }
2664         }
2665         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2666             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2667                 ext4_msg(sb, KERN_ERR,
2668                          "Can't support bigalloc feature without "
2669                          "extents feature\n");
2670                 return 0;
2671         }
2672
2673 #ifndef CONFIG_QUOTA
2674         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2675             !readonly) {
2676                 ext4_msg(sb, KERN_ERR,
2677                          "Filesystem with quota feature cannot be mounted RDWR "
2678                          "without CONFIG_QUOTA");
2679                 return 0;
2680         }
2681 #endif  /* CONFIG_QUOTA */
2682         return 1;
2683 }
2684
2685 /*
2686  * This function is called once a day if we have errors logged
2687  * on the file system
2688  */
2689 static void print_daily_error_info(unsigned long arg)
2690 {
2691         struct super_block *sb = (struct super_block *) arg;
2692         struct ext4_sb_info *sbi;
2693         struct ext4_super_block *es;
2694
2695         sbi = EXT4_SB(sb);
2696         es = sbi->s_es;
2697
2698         if (es->s_error_count)
2699                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2700                          le32_to_cpu(es->s_error_count));
2701         if (es->s_first_error_time) {
2702                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2703                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2704                        (int) sizeof(es->s_first_error_func),
2705                        es->s_first_error_func,
2706                        le32_to_cpu(es->s_first_error_line));
2707                 if (es->s_first_error_ino)
2708                         printk(": inode %u",
2709                                le32_to_cpu(es->s_first_error_ino));
2710                 if (es->s_first_error_block)
2711                         printk(": block %llu", (unsigned long long)
2712                                le64_to_cpu(es->s_first_error_block));
2713                 printk("\n");
2714         }
2715         if (es->s_last_error_time) {
2716                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2717                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2718                        (int) sizeof(es->s_last_error_func),
2719                        es->s_last_error_func,
2720                        le32_to_cpu(es->s_last_error_line));
2721                 if (es->s_last_error_ino)
2722                         printk(": inode %u",
2723                                le32_to_cpu(es->s_last_error_ino));
2724                 if (es->s_last_error_block)
2725                         printk(": block %llu", (unsigned long long)
2726                                le64_to_cpu(es->s_last_error_block));
2727                 printk("\n");
2728         }
2729         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2730 }
2731
2732 /* Find next suitable group and run ext4_init_inode_table */
2733 static int ext4_run_li_request(struct ext4_li_request *elr)
2734 {
2735         struct ext4_group_desc *gdp = NULL;
2736         ext4_group_t group, ngroups;
2737         struct super_block *sb;
2738         unsigned long timeout = 0;
2739         int ret = 0;
2740
2741         sb = elr->lr_super;
2742         ngroups = EXT4_SB(sb)->s_groups_count;
2743
2744         sb_start_write(sb);
2745         for (group = elr->lr_next_group; group < ngroups; group++) {
2746                 gdp = ext4_get_group_desc(sb, group, NULL);
2747                 if (!gdp) {
2748                         ret = 1;
2749                         break;
2750                 }
2751
2752                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2753                         break;
2754         }
2755
2756         if (group >= ngroups)
2757                 ret = 1;
2758
2759         if (!ret) {
2760                 timeout = jiffies;
2761                 ret = ext4_init_inode_table(sb, group,
2762                                             elr->lr_timeout ? 0 : 1);
2763                 if (elr->lr_timeout == 0) {
2764                         timeout = (jiffies - timeout) *
2765                                   elr->lr_sbi->s_li_wait_mult;
2766                         elr->lr_timeout = timeout;
2767                 }
2768                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2769                 elr->lr_next_group = group + 1;
2770         }
2771         sb_end_write(sb);
2772
2773         return ret;
2774 }
2775
2776 /*
2777  * Remove lr_request from the list_request and free the
2778  * request structure. Should be called with li_list_mtx held
2779  */
2780 static void ext4_remove_li_request(struct ext4_li_request *elr)
2781 {
2782         struct ext4_sb_info *sbi;
2783
2784         if (!elr)
2785                 return;
2786
2787         sbi = elr->lr_sbi;
2788
2789         list_del(&elr->lr_request);
2790         sbi->s_li_request = NULL;
2791         kfree(elr);
2792 }
2793
2794 static void ext4_unregister_li_request(struct super_block *sb)
2795 {
2796         mutex_lock(&ext4_li_mtx);
2797         if (!ext4_li_info) {
2798                 mutex_unlock(&ext4_li_mtx);
2799                 return;
2800         }
2801
2802         mutex_lock(&ext4_li_info->li_list_mtx);
2803         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2804         mutex_unlock(&ext4_li_info->li_list_mtx);
2805         mutex_unlock(&ext4_li_mtx);
2806 }
2807
2808 static struct task_struct *ext4_lazyinit_task;
2809
2810 /*
2811  * This is the function where ext4lazyinit thread lives. It walks
2812  * through the request list searching for next scheduled filesystem.
2813  * When such a fs is found, run the lazy initialization request
2814  * (ext4_rn_li_request) and keep track of the time spend in this
2815  * function. Based on that time we compute next schedule time of
2816  * the request. When walking through the list is complete, compute
2817  * next waking time and put itself into sleep.
2818  */
2819 static int ext4_lazyinit_thread(void *arg)
2820 {
2821         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2822         struct list_head *pos, *n;
2823         struct ext4_li_request *elr;
2824         unsigned long next_wakeup, cur;
2825
2826         BUG_ON(NULL == eli);
2827
2828 cont_thread:
2829         while (true) {
2830                 next_wakeup = MAX_JIFFY_OFFSET;
2831
2832                 mutex_lock(&eli->li_list_mtx);
2833                 if (list_empty(&eli->li_request_list)) {
2834                         mutex_unlock(&eli->li_list_mtx);
2835                         goto exit_thread;
2836                 }
2837
2838                 list_for_each_safe(pos, n, &eli->li_request_list) {
2839                         elr = list_entry(pos, struct ext4_li_request,
2840                                          lr_request);
2841
2842                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2843                                 if (ext4_run_li_request(elr) != 0) {
2844                                         /* error, remove the lazy_init job */
2845                                         ext4_remove_li_request(elr);
2846                                         continue;
2847                                 }
2848                         }
2849
2850                         if (time_before(elr->lr_next_sched, next_wakeup))
2851                                 next_wakeup = elr->lr_next_sched;
2852                 }
2853                 mutex_unlock(&eli->li_list_mtx);
2854
2855                 try_to_freeze();
2856
2857                 cur = jiffies;
2858                 if ((time_after_eq(cur, next_wakeup)) ||
2859                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2860                         cond_resched();
2861                         continue;
2862                 }
2863
2864                 schedule_timeout_interruptible(next_wakeup - cur);
2865
2866                 if (kthread_should_stop()) {
2867                         ext4_clear_request_list();
2868                         goto exit_thread;
2869                 }
2870         }
2871
2872 exit_thread:
2873         /*
2874          * It looks like the request list is empty, but we need
2875          * to check it under the li_list_mtx lock, to prevent any
2876          * additions into it, and of course we should lock ext4_li_mtx
2877          * to atomically free the list and ext4_li_info, because at
2878          * this point another ext4 filesystem could be registering
2879          * new one.
2880          */
2881         mutex_lock(&ext4_li_mtx);
2882         mutex_lock(&eli->li_list_mtx);
2883         if (!list_empty(&eli->li_request_list)) {
2884                 mutex_unlock(&eli->li_list_mtx);
2885                 mutex_unlock(&ext4_li_mtx);
2886                 goto cont_thread;
2887         }
2888         mutex_unlock(&eli->li_list_mtx);
2889         kfree(ext4_li_info);
2890         ext4_li_info = NULL;
2891         mutex_unlock(&ext4_li_mtx);
2892
2893         return 0;
2894 }
2895
2896 static void ext4_clear_request_list(void)
2897 {
2898         struct list_head *pos, *n;
2899         struct ext4_li_request *elr;
2900
2901         mutex_lock(&ext4_li_info->li_list_mtx);
2902         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2903                 elr = list_entry(pos, struct ext4_li_request,
2904                                  lr_request);
2905                 ext4_remove_li_request(elr);
2906         }
2907         mutex_unlock(&ext4_li_info->li_list_mtx);
2908 }
2909
2910 static int ext4_run_lazyinit_thread(void)
2911 {
2912         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2913                                          ext4_li_info, "ext4lazyinit");
2914         if (IS_ERR(ext4_lazyinit_task)) {
2915                 int err = PTR_ERR(ext4_lazyinit_task);
2916                 ext4_clear_request_list();
2917                 kfree(ext4_li_info);
2918                 ext4_li_info = NULL;
2919                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2920                                  "initialization thread\n",
2921                                  err);
2922                 return err;
2923         }
2924         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2925         return 0;
2926 }
2927
2928 /*
2929  * Check whether it make sense to run itable init. thread or not.
2930  * If there is at least one uninitialized inode table, return
2931  * corresponding group number, else the loop goes through all
2932  * groups and return total number of groups.
2933  */
2934 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2935 {
2936         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2937         struct ext4_group_desc *gdp = NULL;
2938
2939         for (group = 0; group < ngroups; group++) {
2940                 gdp = ext4_get_group_desc(sb, group, NULL);
2941                 if (!gdp)
2942                         continue;
2943
2944                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2945                         break;
2946         }
2947
2948         return group;
2949 }
2950
2951 static int ext4_li_info_new(void)
2952 {
2953         struct ext4_lazy_init *eli = NULL;
2954
2955         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2956         if (!eli)
2957                 return -ENOMEM;
2958
2959         INIT_LIST_HEAD(&eli->li_request_list);
2960         mutex_init(&eli->li_list_mtx);
2961
2962         eli->li_state |= EXT4_LAZYINIT_QUIT;
2963
2964         ext4_li_info = eli;
2965
2966         return 0;
2967 }
2968
2969 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2970                                             ext4_group_t start)
2971 {
2972         struct ext4_sb_info *sbi = EXT4_SB(sb);
2973         struct ext4_li_request *elr;
2974         unsigned long rnd;
2975
2976         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2977         if (!elr)
2978                 return NULL;
2979
2980         elr->lr_super = sb;
2981         elr->lr_sbi = sbi;
2982         elr->lr_next_group = start;
2983
2984         /*
2985          * Randomize first schedule time of the request to
2986          * spread the inode table initialization requests
2987          * better.
2988          */
2989         get_random_bytes(&rnd, sizeof(rnd));
2990         elr->lr_next_sched = jiffies + (unsigned long)rnd %
2991                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
2992
2993         return elr;
2994 }
2995
2996 int ext4_register_li_request(struct super_block *sb,
2997                              ext4_group_t first_not_zeroed)
2998 {
2999         struct ext4_sb_info *sbi = EXT4_SB(sb);
3000         struct ext4_li_request *elr = NULL;
3001         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3002         int ret = 0;
3003
3004         mutex_lock(&ext4_li_mtx);
3005         if (sbi->s_li_request != NULL) {
3006                 /*
3007                  * Reset timeout so it can be computed again, because
3008                  * s_li_wait_mult might have changed.
3009                  */
3010                 sbi->s_li_request->lr_timeout = 0;
3011                 goto out;
3012         }
3013
3014         if (first_not_zeroed == ngroups ||
3015             (sb->s_flags & MS_RDONLY) ||
3016             !test_opt(sb, INIT_INODE_TABLE))
3017                 goto out;
3018
3019         elr = ext4_li_request_new(sb, first_not_zeroed);
3020         if (!elr) {
3021                 ret = -ENOMEM;
3022                 goto out;
3023         }
3024
3025         if (NULL == ext4_li_info) {
3026                 ret = ext4_li_info_new();
3027                 if (ret)
3028                         goto out;
3029         }
3030
3031         mutex_lock(&ext4_li_info->li_list_mtx);
3032         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3033         mutex_unlock(&ext4_li_info->li_list_mtx);
3034
3035         sbi->s_li_request = elr;
3036         /*
3037          * set elr to NULL here since it has been inserted to
3038          * the request_list and the removal and free of it is
3039          * handled by ext4_clear_request_list from now on.
3040          */
3041         elr = NULL;
3042
3043         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3044                 ret = ext4_run_lazyinit_thread();
3045                 if (ret)
3046                         goto out;
3047         }
3048 out:
3049         mutex_unlock(&ext4_li_mtx);
3050         if (ret)
3051                 kfree(elr);
3052         return ret;
3053 }
3054
3055 /*
3056  * We do not need to lock anything since this is called on
3057  * module unload.
3058  */
3059 static void ext4_destroy_lazyinit_thread(void)
3060 {
3061         /*
3062          * If thread exited earlier
3063          * there's nothing to be done.
3064          */
3065         if (!ext4_li_info || !ext4_lazyinit_task)
3066                 return;
3067
3068         kthread_stop(ext4_lazyinit_task);
3069 }
3070
3071 static int set_journal_csum_feature_set(struct super_block *sb)
3072 {
3073         int ret = 1;
3074         int compat, incompat;
3075         struct ext4_sb_info *sbi = EXT4_SB(sb);
3076
3077         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3078                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3079                 /* journal checksum v2 */
3080                 compat = 0;
3081                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3082         } else {
3083                 /* journal checksum v1 */
3084                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3085                 incompat = 0;
3086         }
3087
3088         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3089                 ret = jbd2_journal_set_features(sbi->s_journal,
3090                                 compat, 0,
3091                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3092                                 incompat);
3093         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3094                 ret = jbd2_journal_set_features(sbi->s_journal,
3095                                 compat, 0,
3096                                 incompat);
3097                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3098                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3099         } else {
3100                 jbd2_journal_clear_features(sbi->s_journal,
3101                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3102                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3103                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3104         }
3105
3106         return ret;
3107 }
3108
3109 /*
3110  * Note: calculating the overhead so we can be compatible with
3111  * historical BSD practice is quite difficult in the face of
3112  * clusters/bigalloc.  This is because multiple metadata blocks from
3113  * different block group can end up in the same allocation cluster.
3114  * Calculating the exact overhead in the face of clustered allocation
3115  * requires either O(all block bitmaps) in memory or O(number of block
3116  * groups**2) in time.  We will still calculate the superblock for
3117  * older file systems --- and if we come across with a bigalloc file
3118  * system with zero in s_overhead_clusters the estimate will be close to
3119  * correct especially for very large cluster sizes --- but for newer
3120  * file systems, it's better to calculate this figure once at mkfs
3121  * time, and store it in the superblock.  If the superblock value is
3122  * present (even for non-bigalloc file systems), we will use it.
3123  */
3124 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3125                           char *buf)
3126 {
3127         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3128         struct ext4_group_desc  *gdp;
3129         ext4_fsblk_t            first_block, last_block, b;
3130         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3131         int                     s, j, count = 0;
3132
3133         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3134                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3135                         sbi->s_itb_per_group + 2);
3136
3137         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3138                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3139         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3140         for (i = 0; i < ngroups; i++) {
3141                 gdp = ext4_get_group_desc(sb, i, NULL);
3142                 b = ext4_block_bitmap(sb, gdp);
3143                 if (b >= first_block && b <= last_block) {
3144                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3145                         count++;
3146                 }
3147                 b = ext4_inode_bitmap(sb, gdp);
3148                 if (b >= first_block && b <= last_block) {
3149                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3150                         count++;
3151                 }
3152                 b = ext4_inode_table(sb, gdp);
3153                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3154                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3155                                 int c = EXT4_B2C(sbi, b - first_block);
3156                                 ext4_set_bit(c, buf);
3157                                 count++;
3158                         }
3159                 if (i != grp)
3160                         continue;
3161                 s = 0;
3162                 if (ext4_bg_has_super(sb, grp)) {
3163                         ext4_set_bit(s++, buf);
3164                         count++;
3165                 }
3166                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3167                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3168                         count++;
3169                 }
3170         }
3171         if (!count)
3172                 return 0;
3173         return EXT4_CLUSTERS_PER_GROUP(sb) -
3174                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3175 }
3176
3177 /*
3178  * Compute the overhead and stash it in sbi->s_overhead
3179  */
3180 int ext4_calculate_overhead(struct super_block *sb)
3181 {
3182         struct ext4_sb_info *sbi = EXT4_SB(sb);
3183         struct ext4_super_block *es = sbi->s_es;
3184         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3185         ext4_fsblk_t overhead = 0;
3186         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3187
3188         if (!buf)
3189                 return -ENOMEM;
3190
3191         /*
3192          * Compute the overhead (FS structures).  This is constant
3193          * for a given filesystem unless the number of block groups
3194          * changes so we cache the previous value until it does.
3195          */
3196
3197         /*
3198          * All of the blocks before first_data_block are overhead
3199          */
3200         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3201
3202         /*
3203          * Add the overhead found in each block group
3204          */
3205         for (i = 0; i < ngroups; i++) {
3206                 int blks;
3207
3208                 blks = count_overhead(sb, i, buf);
3209                 overhead += blks;
3210                 if (blks)
3211                         memset(buf, 0, PAGE_SIZE);
3212                 cond_resched();
3213         }
3214         /* Add the journal blocks as well */
3215         if (sbi->s_journal)
3216                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3217
3218         sbi->s_overhead = overhead;
3219         smp_wmb();
3220         free_page((unsigned long) buf);
3221         return 0;
3222 }
3223
3224
3225 static ext4_fsblk_t ext4_calculate_resv_clusters(struct ext4_sb_info *sbi)
3226 {
3227         ext4_fsblk_t resv_clusters;
3228
3229         /*
3230          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3231          * This should cover the situations where we can not afford to run
3232          * out of space like for example punch hole, or converting
3233          * uninitialized extents in delalloc path. In most cases such
3234          * allocation would require 1, or 2 blocks, higher numbers are
3235          * very rare.
3236          */
3237         resv_clusters = ext4_blocks_count(sbi->s_es) >> sbi->s_cluster_bits;
3238
3239         do_div(resv_clusters, 50);
3240         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3241
3242         return resv_clusters;
3243 }
3244
3245
3246 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3247 {
3248         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3249                                 sbi->s_cluster_bits;
3250
3251         if (count >= clusters)
3252                 return -EINVAL;
3253
3254         atomic64_set(&sbi->s_resv_clusters, count);
3255         return 0;
3256 }
3257
3258 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3259 {
3260         char *orig_data = kstrdup(data, GFP_KERNEL);
3261         struct buffer_head *bh;
3262         struct ext4_super_block *es = NULL;
3263         struct ext4_sb_info *sbi;
3264         ext4_fsblk_t block;
3265         ext4_fsblk_t sb_block = get_sb_block(&data);
3266         ext4_fsblk_t logical_sb_block;
3267         unsigned long offset = 0;
3268         unsigned long journal_devnum = 0;
3269         unsigned long def_mount_opts;
3270         struct inode *root;
3271         char *cp;
3272         const char *descr;
3273         int ret = -ENOMEM;
3274         int blocksize, clustersize;
3275         unsigned int db_count;
3276         unsigned int i;
3277         int needs_recovery, has_huge_files, has_bigalloc;
3278         __u64 blocks_count;
3279         int err = 0;
3280         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3281         ext4_group_t first_not_zeroed;
3282
3283         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3284         if (!sbi)
3285                 goto out_free_orig;
3286
3287         sbi->s_blockgroup_lock =
3288                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3289         if (!sbi->s_blockgroup_lock) {
3290                 kfree(sbi);
3291                 goto out_free_orig;
3292         }
3293         sb->s_fs_info = sbi;
3294         sbi->s_sb = sb;
3295         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3296         sbi->s_sb_block = sb_block;
3297         if (sb->s_bdev->bd_part)
3298                 sbi->s_sectors_written_start =
3299                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3300
3301         /* Cleanup superblock name */
3302         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3303                 *cp = '!';
3304
3305         /* -EINVAL is default */
3306         ret = -EINVAL;
3307         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3308         if (!blocksize) {
3309                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3310                 goto out_fail;
3311         }
3312
3313         /*
3314          * The ext4 superblock will not be buffer aligned for other than 1kB
3315          * block sizes.  We need to calculate the offset from buffer start.
3316          */
3317         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3318                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3319                 offset = do_div(logical_sb_block, blocksize);
3320         } else {
3321                 logical_sb_block = sb_block;
3322         }
3323
3324         if (!(bh = sb_bread(sb, logical_sb_block))) {
3325                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3326                 goto out_fail;
3327         }
3328         /*
3329          * Note: s_es must be initialized as soon as possible because
3330          *       some ext4 macro-instructions depend on its value
3331          */
3332         es = (struct ext4_super_block *) (bh->b_data + offset);
3333         sbi->s_es = es;
3334         sb->s_magic = le16_to_cpu(es->s_magic);
3335         if (sb->s_magic != EXT4_SUPER_MAGIC)
3336                 goto cantfind_ext4;
3337         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3338
3339         /* Warn if metadata_csum and gdt_csum are both set. */
3340         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3341                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3342             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3343                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3344                              "redundant flags; please run fsck.");
3345
3346         /* Check for a known checksum algorithm */
3347         if (!ext4_verify_csum_type(sb, es)) {
3348                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3349                          "unknown checksum algorithm.");
3350                 silent = 1;
3351                 goto cantfind_ext4;
3352         }
3353
3354         /* Load the checksum driver */
3355         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3356                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3357                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3358                 if (IS_ERR(sbi->s_chksum_driver)) {
3359                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3360                         ret = PTR_ERR(sbi->s_chksum_driver);
3361                         sbi->s_chksum_driver = NULL;
3362                         goto failed_mount;
3363                 }
3364         }
3365
3366         /* Check superblock checksum */
3367         if (!ext4_superblock_csum_verify(sb, es)) {
3368                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3369                          "invalid superblock checksum.  Run e2fsck?");
3370                 silent = 1;
3371                 goto cantfind_ext4;
3372         }
3373
3374         /* Precompute checksum seed for all metadata */
3375         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3376                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3377                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3378                                                sizeof(es->s_uuid));
3379
3380         /* Set defaults before we parse the mount options */
3381         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3382         set_opt(sb, INIT_INODE_TABLE);
3383         if (def_mount_opts & EXT4_DEFM_DEBUG)
3384                 set_opt(sb, DEBUG);
3385         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3386                 set_opt(sb, GRPID);
3387         if (def_mount_opts & EXT4_DEFM_UID16)
3388                 set_opt(sb, NO_UID32);
3389         /* xattr user namespace & acls are now defaulted on */
3390         set_opt(sb, XATTR_USER);
3391 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3392         set_opt(sb, POSIX_ACL);
3393 #endif
3394         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3395                 set_opt(sb, JOURNAL_DATA);
3396         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3397                 set_opt(sb, ORDERED_DATA);
3398         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3399                 set_opt(sb, WRITEBACK_DATA);
3400
3401         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3402                 set_opt(sb, ERRORS_PANIC);
3403         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3404                 set_opt(sb, ERRORS_CONT);
3405         else
3406                 set_opt(sb, ERRORS_RO);
3407         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3408                 set_opt(sb, BLOCK_VALIDITY);
3409         if (def_mount_opts & EXT4_DEFM_DISCARD)
3410                 set_opt(sb, DISCARD);
3411
3412         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3413         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3414         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3415         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3416         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3417
3418         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3419                 set_opt(sb, BARRIER);
3420
3421         /*
3422          * enable delayed allocation by default
3423          * Use -o nodelalloc to turn it off
3424          */
3425         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3426             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3427                 set_opt(sb, DELALLOC);
3428
3429         /*
3430          * set default s_li_wait_mult for lazyinit, for the case there is
3431          * no mount option specified.
3432          */
3433         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3434
3435         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3436                            &journal_devnum, &journal_ioprio, 0)) {
3437                 ext4_msg(sb, KERN_WARNING,
3438                          "failed to parse options in superblock: %s",
3439                          sbi->s_es->s_mount_opts);
3440         }
3441         sbi->s_def_mount_opt = sbi->s_mount_opt;
3442         if (!parse_options((char *) data, sb, &journal_devnum,
3443                            &journal_ioprio, 0))
3444                 goto failed_mount;
3445
3446         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3447                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3448                             "with data=journal disables delayed "
3449                             "allocation and O_DIRECT support!\n");
3450                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3451                         ext4_msg(sb, KERN_ERR, "can't mount with "
3452                                  "both data=journal and delalloc");
3453                         goto failed_mount;
3454                 }
3455                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3456                         ext4_msg(sb, KERN_ERR, "can't mount with "
3457                                  "both data=journal and delalloc");
3458                         goto failed_mount;
3459                 }
3460                 if (test_opt(sb, DELALLOC))
3461                         clear_opt(sb, DELALLOC);
3462         }
3463
3464         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3465                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3466
3467         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3468             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3469              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3470              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3471                 ext4_msg(sb, KERN_WARNING,
3472                        "feature flags set on rev 0 fs, "
3473                        "running e2fsck is recommended");
3474
3475         if (IS_EXT2_SB(sb)) {
3476                 if (ext2_feature_set_ok(sb))
3477                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3478                                  "using the ext4 subsystem");
3479                 else {
3480                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3481                                  "to feature incompatibilities");
3482                         goto failed_mount;
3483                 }
3484         }
3485
3486         if (IS_EXT3_SB(sb)) {
3487                 if (ext3_feature_set_ok(sb))
3488                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3489                                  "using the ext4 subsystem");
3490                 else {
3491                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3492                                  "to feature incompatibilities");
3493                         goto failed_mount;
3494                 }
3495         }
3496
3497         /*
3498          * Check feature flags regardless of the revision level, since we
3499          * previously didn't change the revision level when setting the flags,
3500          * so there is a chance incompat flags are set on a rev 0 filesystem.
3501          */
3502         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3503                 goto failed_mount;
3504
3505         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3506         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3507             blocksize > EXT4_MAX_BLOCK_SIZE) {
3508                 ext4_msg(sb, KERN_ERR,
3509                        "Unsupported filesystem blocksize %d", blocksize);
3510                 goto failed_mount;
3511         }
3512
3513         if (sb->s_blocksize != blocksize) {
3514                 /* Validate the filesystem blocksize */
3515                 if (!sb_set_blocksize(sb, blocksize)) {
3516                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3517                                         blocksize);
3518                         goto failed_mount;
3519                 }
3520
3521                 brelse(bh);
3522                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3523                 offset = do_div(logical_sb_block, blocksize);
3524                 bh = sb_bread(sb, logical_sb_block);
3525                 if (!bh) {
3526                         ext4_msg(sb, KERN_ERR,
3527                                "Can't read superblock on 2nd try");
3528                         goto failed_mount;
3529                 }
3530                 es = (struct ext4_super_block *)(bh->b_data + offset);
3531                 sbi->s_es = es;
3532                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3533                         ext4_msg(sb, KERN_ERR,
3534                                "Magic mismatch, very weird!");
3535                         goto failed_mount;
3536                 }
3537         }
3538
3539         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3540                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3541         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3542                                                       has_huge_files);
3543         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3544
3545         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3546                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3547                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3548         } else {
3549                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3550                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3551                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3552                     (!is_power_of_2(sbi->s_inode_size)) ||
3553                     (sbi->s_inode_size > blocksize)) {
3554                         ext4_msg(sb, KERN_ERR,
3555                                "unsupported inode size: %d",
3556                                sbi->s_inode_size);
3557                         goto failed_mount;
3558                 }
3559                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3560                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3561         }
3562
3563         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3564         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3565                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3566                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3567                     !is_power_of_2(sbi->s_desc_size)) {
3568                         ext4_msg(sb, KERN_ERR,
3569                                "unsupported descriptor size %lu",
3570                                sbi->s_desc_size);
3571                         goto failed_mount;
3572                 }
3573         } else
3574                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3575
3576         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3577         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3578         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3579                 goto cantfind_ext4;
3580
3581         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3582         if (sbi->s_inodes_per_block == 0)
3583                 goto cantfind_ext4;
3584         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3585                                         sbi->s_inodes_per_block;
3586         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3587         sbi->s_sbh = bh;
3588         sbi->s_mount_state = le16_to_cpu(es->s_state);
3589         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3590         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3591
3592         /* Do we have standard group size of blocksize * 8 blocks ? */
3593         if (sbi->s_blocks_per_group == blocksize << 3)
3594                 set_opt2(sb, STD_GROUP_SIZE);
3595
3596         for (i = 0; i < 4; i++)
3597                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3598         sbi->s_def_hash_version = es->s_def_hash_version;
3599         i = le32_to_cpu(es->s_flags);
3600         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3601                 sbi->s_hash_unsigned = 3;
3602         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3603 #ifdef __CHAR_UNSIGNED__
3604                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3605                 sbi->s_hash_unsigned = 3;
3606 #else
3607                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3608 #endif
3609         }
3610
3611         /* Handle clustersize */
3612         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3613         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3614                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3615         if (has_bigalloc) {
3616                 if (clustersize < blocksize) {
3617                         ext4_msg(sb, KERN_ERR,
3618                                  "cluster size (%d) smaller than "
3619                                  "block size (%d)", clustersize, blocksize);
3620                         goto failed_mount;
3621                 }
3622                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3623                         le32_to_cpu(es->s_log_block_size);
3624                 sbi->s_clusters_per_group =
3625                         le32_to_cpu(es->s_clusters_per_group);
3626                 if (sbi->s_clusters_per_group > blocksize * 8) {
3627                         ext4_msg(sb, KERN_ERR,
3628                                  "#clusters per group too big: %lu",
3629                                  sbi->s_clusters_per_group);
3630                         goto failed_mount;
3631                 }
3632                 if (sbi->s_blocks_per_group !=
3633                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3634                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3635                                  "clusters per group (%lu) inconsistent",
3636                                  sbi->s_blocks_per_group,
3637                                  sbi->s_clusters_per_group);
3638                         goto failed_mount;
3639                 }
3640         } else {
3641                 if (clustersize != blocksize) {
3642                         ext4_warning(sb, "fragment/cluster size (%d) != "
3643                                      "block size (%d)", clustersize,
3644                                      blocksize);
3645                         clustersize = blocksize;
3646                 }
3647                 if (sbi->s_blocks_per_group > blocksize * 8) {
3648                         ext4_msg(sb, KERN_ERR,
3649                                  "#blocks per group too big: %lu",
3650                                  sbi->s_blocks_per_group);
3651                         goto failed_mount;
3652                 }
3653                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3654                 sbi->s_cluster_bits = 0;
3655         }
3656         sbi->s_cluster_ratio = clustersize / blocksize;
3657
3658         if (sbi->s_inodes_per_group > blocksize * 8) {
3659                 ext4_msg(sb, KERN_ERR,
3660                        "#inodes per group too big: %lu",
3661                        sbi->s_inodes_per_group);
3662                 goto failed_mount;
3663         }
3664
3665         /*
3666          * Test whether we have more sectors than will fit in sector_t,
3667          * and whether the max offset is addressable by the page cache.
3668          */
3669         err = generic_check_addressable(sb->s_blocksize_bits,
3670                                         ext4_blocks_count(es));
3671         if (err) {
3672                 ext4_msg(sb, KERN_ERR, "filesystem"
3673                          " too large to mount safely on this system");
3674                 if (sizeof(sector_t) < 8)
3675                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3676                 goto failed_mount;
3677         }
3678
3679         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3680                 goto cantfind_ext4;
3681
3682         /* check blocks count against device size */
3683         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3684         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3685                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3686                        "exceeds size of device (%llu blocks)",
3687                        ext4_blocks_count(es), blocks_count);
3688                 goto failed_mount;
3689         }
3690
3691         /*
3692          * It makes no sense for the first data block to be beyond the end
3693          * of the filesystem.
3694          */
3695         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3696                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3697                          "block %u is beyond end of filesystem (%llu)",
3698                          le32_to_cpu(es->s_first_data_block),
3699                          ext4_blocks_count(es));
3700                 goto failed_mount;
3701         }
3702         blocks_count = (ext4_blocks_count(es) -
3703                         le32_to_cpu(es->s_first_data_block) +
3704                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3705         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3706         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3707                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3708                        "(block count %llu, first data block %u, "
3709                        "blocks per group %lu)", sbi->s_groups_count,
3710                        ext4_blocks_count(es),
3711                        le32_to_cpu(es->s_first_data_block),
3712                        EXT4_BLOCKS_PER_GROUP(sb));
3713                 goto failed_mount;
3714         }
3715         sbi->s_groups_count = blocks_count;
3716         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3717                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3718         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3719                    EXT4_DESC_PER_BLOCK(sb);
3720         sbi->s_group_desc = ext4_kvmalloc(db_count *
3721                                           sizeof(struct buffer_head *),
3722                                           GFP_KERNEL);
3723         if (sbi->s_group_desc == NULL) {
3724                 ext4_msg(sb, KERN_ERR, "not enough memory");
3725                 ret = -ENOMEM;
3726                 goto failed_mount;
3727         }
3728
3729         if (ext4_proc_root)
3730                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3731
3732         if (sbi->s_proc)
3733                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3734                                  &ext4_seq_options_fops, sb);
3735
3736         bgl_lock_init(sbi->s_blockgroup_lock);
3737
3738         for (i = 0; i < db_count; i++) {
3739                 block = descriptor_loc(sb, logical_sb_block, i);
3740                 sbi->s_group_desc[i] = sb_bread(sb, block);
3741                 if (!sbi->s_group_desc[i]) {
3742                         ext4_msg(sb, KERN_ERR,
3743                                "can't read group descriptor %d", i);
3744                         db_count = i;
3745                         goto failed_mount2;
3746                 }
3747         }
3748         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3749                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3750                 goto failed_mount2;
3751         }
3752         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3753                 if (!ext4_fill_flex_info(sb)) {
3754                         ext4_msg(sb, KERN_ERR,
3755                                "unable to initialize "
3756                                "flex_bg meta info!");
3757                         goto failed_mount2;
3758                 }
3759
3760         sbi->s_gdb_count = db_count;
3761         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3762         spin_lock_init(&sbi->s_next_gen_lock);
3763
3764         init_timer(&sbi->s_err_report);
3765         sbi->s_err_report.function = print_daily_error_info;
3766         sbi->s_err_report.data = (unsigned long) sb;
3767
3768         /* Register extent status tree shrinker */
3769         ext4_es_register_shrinker(sb);
3770
3771         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3772                         ext4_count_free_clusters(sb));
3773         if (!err) {
3774                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3775                                 ext4_count_free_inodes(sb));
3776         }
3777         if (!err) {
3778                 err = percpu_counter_init(&sbi->s_dirs_counter,
3779                                 ext4_count_dirs(sb));
3780         }
3781         if (!err) {
3782                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3783         }
3784         if (!err) {
3785                 err = percpu_counter_init(&sbi->s_extent_cache_cnt, 0);
3786         }
3787         if (err) {
3788                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3789                 goto failed_mount3;
3790         }
3791
3792         sbi->s_stripe = ext4_get_stripe_size(sbi);
3793         sbi->s_max_writeback_mb_bump = 128;
3794         sbi->s_extent_max_zeroout_kb = 32;
3795
3796         /*
3797          * set up enough so that it can read an inode
3798          */
3799         if (!test_opt(sb, NOLOAD) &&
3800             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3801                 sb->s_op = &ext4_sops;
3802         else
3803                 sb->s_op = &ext4_nojournal_sops;
3804         sb->s_export_op = &ext4_export_ops;
3805         sb->s_xattr = ext4_xattr_handlers;
3806 #ifdef CONFIG_QUOTA
3807         sb->dq_op = &ext4_quota_operations;
3808         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3809                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3810         else
3811                 sb->s_qcop = &ext4_qctl_operations;
3812 #endif
3813         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3814
3815         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3816         mutex_init(&sbi->s_orphan_lock);
3817
3818         sb->s_root = NULL;
3819
3820         needs_recovery = (es->s_last_orphan != 0 ||
3821                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3822                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3823
3824         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3825             !(sb->s_flags & MS_RDONLY))
3826                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3827                         goto failed_mount3;
3828
3829         /*
3830          * The first inode we look at is the journal inode.  Don't try
3831          * root first: it may be modified in the journal!
3832          */
3833         if (!test_opt(sb, NOLOAD) &&
3834             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3835                 if (ext4_load_journal(sb, es, journal_devnum))
3836                         goto failed_mount3;
3837         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3838               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3839                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3840                        "suppressed and not mounted read-only");
3841                 goto failed_mount_wq;
3842         } else {
3843                 clear_opt(sb, DATA_FLAGS);
3844                 sbi->s_journal = NULL;
3845                 needs_recovery = 0;
3846                 goto no_journal;
3847         }
3848
3849         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3850             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3851                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3852                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3853                 goto failed_mount_wq;
3854         }
3855
3856         if (!set_journal_csum_feature_set(sb)) {
3857                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3858                          "feature set");
3859                 goto failed_mount_wq;
3860         }
3861
3862         /* We have now updated the journal if required, so we can
3863          * validate the data journaling mode. */
3864         switch (test_opt(sb, DATA_FLAGS)) {
3865         case 0:
3866                 /* No mode set, assume a default based on the journal
3867                  * capabilities: ORDERED_DATA if the journal can
3868                  * cope, else JOURNAL_DATA
3869                  */
3870                 if (jbd2_journal_check_available_features
3871                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3872                         set_opt(sb, ORDERED_DATA);
3873                 else
3874                         set_opt(sb, JOURNAL_DATA);
3875                 break;
3876
3877         case EXT4_MOUNT_ORDERED_DATA:
3878         case EXT4_MOUNT_WRITEBACK_DATA:
3879                 if (!jbd2_journal_check_available_features
3880                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3881                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3882                                "requested data journaling mode");
3883                         goto failed_mount_wq;
3884                 }
3885         default:
3886                 break;
3887         }
3888         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3889
3890         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3891
3892         /*
3893          * The journal may have updated the bg summary counts, so we
3894          * need to update the global counters.
3895          */
3896         percpu_counter_set(&sbi->s_freeclusters_counter,
3897                            ext4_count_free_clusters(sb));
3898         percpu_counter_set(&sbi->s_freeinodes_counter,
3899                            ext4_count_free_inodes(sb));
3900         percpu_counter_set(&sbi->s_dirs_counter,
3901                            ext4_count_dirs(sb));
3902         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3903
3904 no_journal:
3905         /*
3906          * Get the # of file system overhead blocks from the
3907          * superblock if present.
3908          */
3909         if (es->s_overhead_clusters)
3910                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3911         else {
3912                 err = ext4_calculate_overhead(sb);
3913                 if (err)
3914                         goto failed_mount_wq;
3915         }
3916
3917         /*
3918          * The maximum number of concurrent works can be high and
3919          * concurrency isn't really necessary.  Limit it to 1.
3920          */
3921         EXT4_SB(sb)->dio_unwritten_wq =
3922                 alloc_workqueue("ext4-dio-unwritten", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3923         if (!EXT4_SB(sb)->dio_unwritten_wq) {
3924                 printk(KERN_ERR "EXT4-fs: failed to create DIO workqueue\n");
3925                 ret = -ENOMEM;
3926                 goto failed_mount_wq;
3927         }
3928
3929         /*
3930          * The jbd2_journal_load will have done any necessary log recovery,
3931          * so we can safely mount the rest of the filesystem now.
3932          */
3933
3934         root = ext4_iget(sb, EXT4_ROOT_INO);
3935         if (IS_ERR(root)) {
3936                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3937                 ret = PTR_ERR(root);
3938                 root = NULL;
3939                 goto failed_mount4;
3940         }
3941         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3942                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3943                 iput(root);
3944                 goto failed_mount4;
3945         }
3946         sb->s_root = d_make_root(root);
3947         if (!sb->s_root) {
3948                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3949                 ret = -ENOMEM;
3950                 goto failed_mount4;
3951         }
3952
3953         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3954                 sb->s_flags |= MS_RDONLY;
3955
3956         /* determine the minimum size of new large inodes, if present */
3957         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3958                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3959                                                      EXT4_GOOD_OLD_INODE_SIZE;
3960                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3961                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3962                         if (sbi->s_want_extra_isize <
3963                             le16_to_cpu(es->s_want_extra_isize))
3964                                 sbi->s_want_extra_isize =
3965                                         le16_to_cpu(es->s_want_extra_isize);
3966                         if (sbi->s_want_extra_isize <
3967                             le16_to_cpu(es->s_min_extra_isize))
3968                                 sbi->s_want_extra_isize =
3969                                         le16_to_cpu(es->s_min_extra_isize);
3970                 }
3971         }
3972         /* Check if enough inode space is available */
3973         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3974                                                         sbi->s_inode_size) {
3975                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3976                                                        EXT4_GOOD_OLD_INODE_SIZE;
3977                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3978                          "available");
3979         }
3980
3981         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sbi));
3982         if (err) {
3983                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
3984                          "reserved pool", ext4_calculate_resv_clusters(sbi));
3985                 goto failed_mount4a;
3986         }
3987
3988         err = ext4_setup_system_zone(sb);
3989         if (err) {
3990                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3991                          "zone (%d)", err);
3992                 goto failed_mount4a;
3993         }
3994
3995         ext4_ext_init(sb);
3996         err = ext4_mb_init(sb);
3997         if (err) {
3998                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3999                          err);
4000                 goto failed_mount5;
4001         }
4002
4003         err = ext4_register_li_request(sb, first_not_zeroed);
4004         if (err)
4005                 goto failed_mount6;
4006
4007         sbi->s_kobj.kset = ext4_kset;
4008         init_completion(&sbi->s_kobj_unregister);
4009         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4010                                    "%s", sb->s_id);
4011         if (err)
4012                 goto failed_mount7;
4013
4014 #ifdef CONFIG_QUOTA
4015         /* Enable quota usage during mount. */
4016         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4017             !(sb->s_flags & MS_RDONLY)) {
4018                 err = ext4_enable_quotas(sb);
4019                 if (err)
4020                         goto failed_mount8;
4021         }
4022 #endif  /* CONFIG_QUOTA */
4023
4024         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4025         ext4_orphan_cleanup(sb, es);
4026         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4027         if (needs_recovery) {
4028                 ext4_msg(sb, KERN_INFO, "recovery complete");
4029                 ext4_mark_recovery_complete(sb, es);
4030         }
4031         if (EXT4_SB(sb)->s_journal) {
4032                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4033                         descr = " journalled data mode";
4034                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4035                         descr = " ordered data mode";
4036                 else
4037                         descr = " writeback data mode";
4038         } else
4039                 descr = "out journal";
4040
4041         if (test_opt(sb, DISCARD)) {
4042                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4043                 if (!blk_queue_discard(q))
4044                         ext4_msg(sb, KERN_WARNING,
4045                                  "mounting with \"discard\" option, but "
4046                                  "the device does not support discard");
4047         }
4048
4049         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4050                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4051                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4052
4053         if (es->s_error_count)
4054                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4055
4056         kfree(orig_data);
4057         return 0;
4058
4059 cantfind_ext4:
4060         if (!silent)
4061                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4062         goto failed_mount;
4063
4064 #ifdef CONFIG_QUOTA
4065 failed_mount8:
4066         kobject_del(&sbi->s_kobj);
4067 #endif
4068 failed_mount7:
4069         ext4_unregister_li_request(sb);
4070 failed_mount6:
4071         ext4_mb_release(sb);
4072 failed_mount5:
4073         ext4_ext_release(sb);
4074         ext4_release_system_zone(sb);
4075 failed_mount4a:
4076         dput(sb->s_root);
4077         sb->s_root = NULL;
4078 failed_mount4:
4079         ext4_msg(sb, KERN_ERR, "mount failed");
4080         destroy_workqueue(EXT4_SB(sb)->dio_unwritten_wq);
4081 failed_mount_wq:
4082         if (sbi->s_journal) {
4083                 jbd2_journal_destroy(sbi->s_journal);
4084                 sbi->s_journal = NULL;
4085         }
4086 failed_mount3:
4087         ext4_es_unregister_shrinker(sb);
4088         del_timer(&sbi->s_err_report);
4089         if (sbi->s_flex_groups)
4090                 ext4_kvfree(sbi->s_flex_groups);
4091         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4092         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4093         percpu_counter_destroy(&sbi->s_dirs_counter);
4094         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4095         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
4096         if (sbi->s_mmp_tsk)
4097                 kthread_stop(sbi->s_mmp_tsk);
4098 failed_mount2:
4099         for (i = 0; i < db_count; i++)
4100                 brelse(sbi->s_group_desc[i]);
4101         ext4_kvfree(sbi->s_group_desc);
4102 failed_mount:
4103         if (sbi->s_chksum_driver)
4104                 crypto_free_shash(sbi->s_chksum_driver);
4105         if (sbi->s_proc) {
4106                 remove_proc_entry("options", sbi->s_proc);
4107                 remove_proc_entry(sb->s_id, ext4_proc_root);
4108         }
4109 #ifdef CONFIG_QUOTA
4110         for (i = 0; i < MAXQUOTAS; i++)
4111                 kfree(sbi->s_qf_names[i]);
4112 #endif
4113         ext4_blkdev_remove(sbi);
4114         brelse(bh);
4115 out_fail:
4116         sb->s_fs_info = NULL;
4117         kfree(sbi->s_blockgroup_lock);
4118         kfree(sbi);
4119 out_free_orig:
4120         kfree(orig_data);
4121         return err ? err : ret;
4122 }
4123
4124 /*
4125  * Setup any per-fs journal parameters now.  We'll do this both on
4126  * initial mount, once the journal has been initialised but before we've
4127  * done any recovery; and again on any subsequent remount.
4128  */
4129 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4130 {
4131         struct ext4_sb_info *sbi = EXT4_SB(sb);
4132
4133         journal->j_commit_interval = sbi->s_commit_interval;
4134         journal->j_min_batch_time = sbi->s_min_batch_time;
4135         journal->j_max_batch_time = sbi->s_max_batch_time;
4136
4137         write_lock(&journal->j_state_lock);
4138         if (test_opt(sb, BARRIER))
4139                 journal->j_flags |= JBD2_BARRIER;
4140         else
4141                 journal->j_flags &= ~JBD2_BARRIER;
4142         if (test_opt(sb, DATA_ERR_ABORT))
4143                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4144         else
4145                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4146         write_unlock(&journal->j_state_lock);
4147 }
4148
4149 static journal_t *ext4_get_journal(struct super_block *sb,
4150                                    unsigned int journal_inum)
4151 {
4152         struct inode *journal_inode;
4153         journal_t *journal;
4154
4155         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4156
4157         /* First, test for the existence of a valid inode on disk.  Bad
4158          * things happen if we iget() an unused inode, as the subsequent
4159          * iput() will try to delete it. */
4160
4161         journal_inode = ext4_iget(sb, journal_inum);
4162         if (IS_ERR(journal_inode)) {
4163                 ext4_msg(sb, KERN_ERR, "no journal found");
4164                 return NULL;
4165         }
4166         if (!journal_inode->i_nlink) {
4167                 make_bad_inode(journal_inode);
4168                 iput(journal_inode);
4169                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4170                 return NULL;
4171         }
4172
4173         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4174                   journal_inode, journal_inode->i_size);
4175         if (!S_ISREG(journal_inode->i_mode)) {
4176                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4177                 iput(journal_inode);
4178                 return NULL;
4179         }
4180
4181         journal = jbd2_journal_init_inode(journal_inode);
4182         if (!journal) {
4183                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4184                 iput(journal_inode);
4185                 return NULL;
4186         }
4187         journal->j_private = sb;
4188         ext4_init_journal_params(sb, journal);
4189         return journal;
4190 }
4191
4192 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4193                                        dev_t j_dev)
4194 {
4195         struct buffer_head *bh;
4196         journal_t *journal;
4197         ext4_fsblk_t start;
4198         ext4_fsblk_t len;
4199         int hblock, blocksize;
4200         ext4_fsblk_t sb_block;
4201         unsigned long offset;
4202         struct ext4_super_block *es;
4203         struct block_device *bdev;
4204
4205         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4206
4207         bdev = ext4_blkdev_get(j_dev, sb);
4208         if (bdev == NULL)
4209                 return NULL;
4210
4211         blocksize = sb->s_blocksize;
4212         hblock = bdev_logical_block_size(bdev);
4213         if (blocksize < hblock) {
4214                 ext4_msg(sb, KERN_ERR,
4215                         "blocksize too small for journal device");
4216                 goto out_bdev;
4217         }
4218
4219         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4220         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4221         set_blocksize(bdev, blocksize);
4222         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4223                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4224                        "external journal");
4225                 goto out_bdev;
4226         }
4227
4228         es = (struct ext4_super_block *) (bh->b_data + offset);
4229         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4230             !(le32_to_cpu(es->s_feature_incompat) &
4231               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4232                 ext4_msg(sb, KERN_ERR, "external journal has "
4233                                         "bad superblock");
4234                 brelse(bh);
4235                 goto out_bdev;
4236         }
4237
4238         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4239                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4240                 brelse(bh);
4241                 goto out_bdev;
4242         }
4243
4244         len = ext4_blocks_count(es);
4245         start = sb_block + 1;
4246         brelse(bh);     /* we're done with the superblock */
4247
4248         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4249                                         start, len, blocksize);
4250         if (!journal) {
4251                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4252                 goto out_bdev;
4253         }
4254         journal->j_private = sb;
4255         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4256         wait_on_buffer(journal->j_sb_buffer);
4257         if (!buffer_uptodate(journal->j_sb_buffer)) {
4258                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4259                 goto out_journal;
4260         }
4261         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4262                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4263                                         "user (unsupported) - %d",
4264                         be32_to_cpu(journal->j_superblock->s_nr_users));
4265                 goto out_journal;
4266         }
4267         EXT4_SB(sb)->journal_bdev = bdev;
4268         ext4_init_journal_params(sb, journal);
4269         return journal;
4270
4271 out_journal:
4272         jbd2_journal_destroy(journal);
4273 out_bdev:
4274         ext4_blkdev_put(bdev);
4275         return NULL;
4276 }
4277
4278 static int ext4_load_journal(struct super_block *sb,
4279                              struct ext4_super_block *es,
4280                              unsigned long journal_devnum)
4281 {
4282         journal_t *journal;
4283         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4284         dev_t journal_dev;
4285         int err = 0;
4286         int really_read_only;
4287
4288         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4289
4290         if (journal_devnum &&
4291             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4292                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4293                         "numbers have changed");
4294                 journal_dev = new_decode_dev(journal_devnum);
4295         } else
4296                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4297
4298         really_read_only = bdev_read_only(sb->s_bdev);
4299
4300         /*
4301          * Are we loading a blank journal or performing recovery after a
4302          * crash?  For recovery, we need to check in advance whether we
4303          * can get read-write access to the device.
4304          */
4305         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4306                 if (sb->s_flags & MS_RDONLY) {
4307                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4308                                         "required on readonly filesystem");
4309                         if (really_read_only) {
4310                                 ext4_msg(sb, KERN_ERR, "write access "
4311                                         "unavailable, cannot proceed");
4312                                 return -EROFS;
4313                         }
4314                         ext4_msg(sb, KERN_INFO, "write access will "
4315                                "be enabled during recovery");
4316                 }
4317         }
4318
4319         if (journal_inum && journal_dev) {
4320                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4321                        "and inode journals!");
4322                 return -EINVAL;
4323         }
4324
4325         if (journal_inum) {
4326                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4327                         return -EINVAL;
4328         } else {
4329                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4330                         return -EINVAL;
4331         }
4332
4333         if (!(journal->j_flags & JBD2_BARRIER))
4334                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4335
4336         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4337                 err = jbd2_journal_wipe(journal, !really_read_only);
4338         if (!err) {
4339                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4340                 if (save)
4341                         memcpy(save, ((char *) es) +
4342                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4343                 err = jbd2_journal_load(journal);
4344                 if (save)
4345                         memcpy(((char *) es) + EXT4_S_ERR_START,
4346                                save, EXT4_S_ERR_LEN);
4347                 kfree(save);
4348         }
4349
4350         if (err) {
4351                 ext4_msg(sb, KERN_ERR, "error loading journal");
4352                 jbd2_journal_destroy(journal);
4353                 return err;
4354         }
4355
4356         EXT4_SB(sb)->s_journal = journal;
4357         ext4_clear_journal_err(sb, es);
4358
4359         if (!really_read_only && journal_devnum &&
4360             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4361                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4362
4363                 /* Make sure we flush the recovery flag to disk. */
4364                 ext4_commit_super(sb, 1);
4365         }
4366
4367         return 0;
4368 }
4369
4370 static int ext4_commit_super(struct super_block *sb, int sync)
4371 {
4372         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4373         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4374         int error = 0;
4375
4376         if (!sbh || block_device_ejected(sb))
4377                 return error;
4378         if (buffer_write_io_error(sbh)) {
4379                 /*
4380                  * Oh, dear.  A previous attempt to write the
4381                  * superblock failed.  This could happen because the
4382                  * USB device was yanked out.  Or it could happen to
4383                  * be a transient write error and maybe the block will
4384                  * be remapped.  Nothing we can do but to retry the
4385                  * write and hope for the best.
4386                  */
4387                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4388                        "superblock detected");
4389                 clear_buffer_write_io_error(sbh);
4390                 set_buffer_uptodate(sbh);
4391         }
4392         /*
4393          * If the file system is mounted read-only, don't update the
4394          * superblock write time.  This avoids updating the superblock
4395          * write time when we are mounting the root file system
4396          * read/only but we need to replay the journal; at that point,
4397          * for people who are east of GMT and who make their clock
4398          * tick in localtime for Windows bug-for-bug compatibility,
4399          * the clock is set in the future, and this will cause e2fsck
4400          * to complain and force a full file system check.
4401          */
4402         if (!(sb->s_flags & MS_RDONLY))
4403                 es->s_wtime = cpu_to_le32(get_seconds());
4404         if (sb->s_bdev->bd_part)
4405                 es->s_kbytes_written =
4406                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4407                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4408                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4409         else
4410                 es->s_kbytes_written =
4411                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4412         ext4_free_blocks_count_set(es,
4413                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4414                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4415         es->s_free_inodes_count =
4416                 cpu_to_le32(percpu_counter_sum_positive(
4417                                 &EXT4_SB(sb)->s_freeinodes_counter));
4418         BUFFER_TRACE(sbh, "marking dirty");
4419         ext4_superblock_csum_set(sb);
4420         mark_buffer_dirty(sbh);
4421         if (sync) {
4422                 error = sync_dirty_buffer(sbh);
4423                 if (error)
4424                         return error;
4425
4426                 error = buffer_write_io_error(sbh);
4427                 if (error) {
4428                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4429                                "superblock");
4430                         clear_buffer_write_io_error(sbh);
4431                         set_buffer_uptodate(sbh);
4432                 }
4433         }
4434         return error;
4435 }
4436
4437 /*
4438  * Have we just finished recovery?  If so, and if we are mounting (or
4439  * remounting) the filesystem readonly, then we will end up with a
4440  * consistent fs on disk.  Record that fact.
4441  */
4442 static void ext4_mark_recovery_complete(struct super_block *sb,
4443                                         struct ext4_super_block *es)
4444 {
4445         journal_t *journal = EXT4_SB(sb)->s_journal;
4446
4447         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4448                 BUG_ON(journal != NULL);
4449                 return;
4450         }
4451         jbd2_journal_lock_updates(journal);
4452         if (jbd2_journal_flush(journal) < 0)
4453                 goto out;
4454
4455         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4456             sb->s_flags & MS_RDONLY) {
4457                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4458                 ext4_commit_super(sb, 1);
4459         }
4460
4461 out:
4462         jbd2_journal_unlock_updates(journal);
4463 }
4464
4465 /*
4466  * If we are mounting (or read-write remounting) a filesystem whose journal
4467  * has recorded an error from a previous lifetime, move that error to the
4468  * main filesystem now.
4469  */
4470 static void ext4_clear_journal_err(struct super_block *sb,
4471                                    struct ext4_super_block *es)
4472 {
4473         journal_t *journal;
4474         int j_errno;
4475         const char *errstr;
4476
4477         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4478
4479         journal = EXT4_SB(sb)->s_journal;
4480
4481         /*
4482          * Now check for any error status which may have been recorded in the
4483          * journal by a prior ext4_error() or ext4_abort()
4484          */
4485
4486         j_errno = jbd2_journal_errno(journal);
4487         if (j_errno) {
4488                 char nbuf[16];
4489
4490                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4491                 ext4_warning(sb, "Filesystem error recorded "
4492                              "from previous mount: %s", errstr);
4493                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4494
4495                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4496                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4497                 ext4_commit_super(sb, 1);
4498
4499                 jbd2_journal_clear_err(journal);
4500                 jbd2_journal_update_sb_errno(journal);
4501         }
4502 }
4503
4504 /*
4505  * Force the running and committing transactions to commit,
4506  * and wait on the commit.
4507  */
4508 int ext4_force_commit(struct super_block *sb)
4509 {
4510         journal_t *journal;
4511
4512         if (sb->s_flags & MS_RDONLY)
4513                 return 0;
4514
4515         journal = EXT4_SB(sb)->s_journal;
4516         return ext4_journal_force_commit(journal);
4517 }
4518
4519 static int ext4_sync_fs(struct super_block *sb, int wait)
4520 {
4521         int ret = 0;
4522         tid_t target;
4523         struct ext4_sb_info *sbi = EXT4_SB(sb);
4524
4525         trace_ext4_sync_fs(sb, wait);
4526         flush_workqueue(sbi->dio_unwritten_wq);
4527         /*
4528          * Writeback quota in non-journalled quota case - journalled quota has
4529          * no dirty dquots
4530          */
4531         dquot_writeback_dquots(sb, -1);
4532         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4533                 if (wait)
4534                         jbd2_log_wait_commit(sbi->s_journal, target);
4535         }
4536         return ret;
4537 }
4538
4539 /*
4540  * LVM calls this function before a (read-only) snapshot is created.  This
4541  * gives us a chance to flush the journal completely and mark the fs clean.
4542  *
4543  * Note that only this function cannot bring a filesystem to be in a clean
4544  * state independently. It relies on upper layer to stop all data & metadata
4545  * modifications.
4546  */
4547 static int ext4_freeze(struct super_block *sb)
4548 {
4549         int error = 0;
4550         journal_t *journal;
4551
4552         if (sb->s_flags & MS_RDONLY)
4553                 return 0;
4554
4555         journal = EXT4_SB(sb)->s_journal;
4556
4557         /* Now we set up the journal barrier. */
4558         jbd2_journal_lock_updates(journal);
4559
4560         /*
4561          * Don't clear the needs_recovery flag if we failed to flush
4562          * the journal.
4563          */
4564         error = jbd2_journal_flush(journal);
4565         if (error < 0)
4566                 goto out;
4567
4568         /* Journal blocked and flushed, clear needs_recovery flag. */
4569         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4570         error = ext4_commit_super(sb, 1);
4571 out:
4572         /* we rely on upper layer to stop further updates */
4573         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4574         return error;
4575 }
4576
4577 /*
4578  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4579  * flag here, even though the filesystem is not technically dirty yet.
4580  */
4581 static int ext4_unfreeze(struct super_block *sb)
4582 {
4583         if (sb->s_flags & MS_RDONLY)
4584                 return 0;
4585
4586         /* Reset the needs_recovery flag before the fs is unlocked. */
4587         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4588         ext4_commit_super(sb, 1);
4589         return 0;
4590 }
4591
4592 /*
4593  * Structure to save mount options for ext4_remount's benefit
4594  */
4595 struct ext4_mount_options {
4596         unsigned long s_mount_opt;
4597         unsigned long s_mount_opt2;
4598         kuid_t s_resuid;
4599         kgid_t s_resgid;
4600         unsigned long s_commit_interval;
4601         u32 s_min_batch_time, s_max_batch_time;
4602 #ifdef CONFIG_QUOTA
4603         int s_jquota_fmt;
4604         char *s_qf_names[MAXQUOTAS];
4605 #endif
4606 };
4607
4608 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4609 {
4610         struct ext4_super_block *es;
4611         struct ext4_sb_info *sbi = EXT4_SB(sb);
4612         unsigned long old_sb_flags;
4613         struct ext4_mount_options old_opts;
4614         int enable_quota = 0;
4615         ext4_group_t g;
4616         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4617         int err = 0;
4618 #ifdef CONFIG_QUOTA
4619         int i, j;
4620 #endif
4621         char *orig_data = kstrdup(data, GFP_KERNEL);
4622
4623         /* Store the original options */
4624         old_sb_flags = sb->s_flags;
4625         old_opts.s_mount_opt = sbi->s_mount_opt;
4626         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4627         old_opts.s_resuid = sbi->s_resuid;
4628         old_opts.s_resgid = sbi->s_resgid;
4629         old_opts.s_commit_interval = sbi->s_commit_interval;
4630         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4631         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4632 #ifdef CONFIG_QUOTA
4633         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4634         for (i = 0; i < MAXQUOTAS; i++)
4635                 if (sbi->s_qf_names[i]) {
4636                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4637                                                          GFP_KERNEL);
4638                         if (!old_opts.s_qf_names[i]) {
4639                                 for (j = 0; j < i; j++)
4640                                         kfree(old_opts.s_qf_names[j]);
4641                                 kfree(orig_data);
4642                                 return -ENOMEM;
4643                         }
4644                 } else
4645                         old_opts.s_qf_names[i] = NULL;
4646 #endif
4647         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4648                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4649
4650         /*
4651          * Allow the "check" option to be passed as a remount option.
4652          */
4653         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4654                 err = -EINVAL;
4655                 goto restore_opts;
4656         }
4657
4658         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4659                 ext4_abort(sb, "Abort forced by user");
4660
4661         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4662                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4663
4664         es = sbi->s_es;
4665
4666         if (sbi->s_journal) {
4667                 ext4_init_journal_params(sb, sbi->s_journal);
4668                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4669         }
4670
4671         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4672                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4673                         err = -EROFS;
4674                         goto restore_opts;
4675                 }
4676
4677                 if (*flags & MS_RDONLY) {
4678                         err = dquot_suspend(sb, -1);
4679                         if (err < 0)
4680                                 goto restore_opts;
4681
4682                         /*
4683                          * First of all, the unconditional stuff we have to do
4684                          * to disable replay of the journal when we next remount
4685                          */
4686                         sb->s_flags |= MS_RDONLY;
4687
4688                         /*
4689                          * OK, test if we are remounting a valid rw partition
4690                          * readonly, and if so set the rdonly flag and then
4691                          * mark the partition as valid again.
4692                          */
4693                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4694                             (sbi->s_mount_state & EXT4_VALID_FS))
4695                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4696
4697                         if (sbi->s_journal)
4698                                 ext4_mark_recovery_complete(sb, es);
4699                 } else {
4700                         /* Make sure we can mount this feature set readwrite */
4701                         if (!ext4_feature_set_ok(sb, 0)) {
4702                                 err = -EROFS;
4703                                 goto restore_opts;
4704                         }
4705                         /*
4706                          * Make sure the group descriptor checksums
4707                          * are sane.  If they aren't, refuse to remount r/w.
4708                          */
4709                         for (g = 0; g < sbi->s_groups_count; g++) {
4710                                 struct ext4_group_desc *gdp =
4711                                         ext4_get_group_desc(sb, g, NULL);
4712
4713                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4714                                         ext4_msg(sb, KERN_ERR,
4715                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4716                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4717                                                le16_to_cpu(gdp->bg_checksum));
4718                                         err = -EINVAL;
4719                                         goto restore_opts;
4720                                 }
4721                         }
4722
4723                         /*
4724                          * If we have an unprocessed orphan list hanging
4725                          * around from a previously readonly bdev mount,
4726                          * require a full umount/remount for now.
4727                          */
4728                         if (es->s_last_orphan) {
4729                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4730                                        "remount RDWR because of unprocessed "
4731                                        "orphan inode list.  Please "
4732                                        "umount/remount instead");
4733                                 err = -EINVAL;
4734                                 goto restore_opts;
4735                         }
4736
4737                         /*
4738                          * Mounting a RDONLY partition read-write, so reread
4739                          * and store the current valid flag.  (It may have
4740                          * been changed by e2fsck since we originally mounted
4741                          * the partition.)
4742                          */
4743                         if (sbi->s_journal)
4744                                 ext4_clear_journal_err(sb, es);
4745                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4746                         if (!ext4_setup_super(sb, es, 0))
4747                                 sb->s_flags &= ~MS_RDONLY;
4748                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4749                                                      EXT4_FEATURE_INCOMPAT_MMP))
4750                                 if (ext4_multi_mount_protect(sb,
4751                                                 le64_to_cpu(es->s_mmp_block))) {
4752                                         err = -EROFS;
4753                                         goto restore_opts;
4754                                 }
4755                         enable_quota = 1;
4756                 }
4757         }
4758
4759         /*
4760          * Reinitialize lazy itable initialization thread based on
4761          * current settings
4762          */
4763         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4764                 ext4_unregister_li_request(sb);
4765         else {
4766                 ext4_group_t first_not_zeroed;
4767                 first_not_zeroed = ext4_has_uninit_itable(sb);
4768                 ext4_register_li_request(sb, first_not_zeroed);
4769         }
4770
4771         ext4_setup_system_zone(sb);
4772         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4773                 ext4_commit_super(sb, 1);
4774
4775 #ifdef CONFIG_QUOTA
4776         /* Release old quota file names */
4777         for (i = 0; i < MAXQUOTAS; i++)
4778                 kfree(old_opts.s_qf_names[i]);
4779         if (enable_quota) {
4780                 if (sb_any_quota_suspended(sb))
4781                         dquot_resume(sb, -1);
4782                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4783                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4784                         err = ext4_enable_quotas(sb);
4785                         if (err)
4786                                 goto restore_opts;
4787                 }
4788         }
4789 #endif
4790
4791         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4792         kfree(orig_data);
4793         return 0;
4794
4795 restore_opts:
4796         sb->s_flags = old_sb_flags;
4797         sbi->s_mount_opt = old_opts.s_mount_opt;
4798         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4799         sbi->s_resuid = old_opts.s_resuid;
4800         sbi->s_resgid = old_opts.s_resgid;
4801         sbi->s_commit_interval = old_opts.s_commit_interval;
4802         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4803         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4804 #ifdef CONFIG_QUOTA
4805         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4806         for (i = 0; i < MAXQUOTAS; i++) {
4807                 kfree(sbi->s_qf_names[i]);
4808                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4809         }
4810 #endif
4811         kfree(orig_data);
4812         return err;
4813 }
4814
4815 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4816 {
4817         struct super_block *sb = dentry->d_sb;
4818         struct ext4_sb_info *sbi = EXT4_SB(sb);
4819         struct ext4_super_block *es = sbi->s_es;
4820         ext4_fsblk_t overhead = 0, resv_blocks;
4821         u64 fsid;
4822         s64 bfree;
4823         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4824
4825         if (!test_opt(sb, MINIX_DF))
4826                 overhead = sbi->s_overhead;
4827
4828         buf->f_type = EXT4_SUPER_MAGIC;
4829         buf->f_bsize = sb->s_blocksize;
4830         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4831         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4832                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4833         /* prevent underflow in case that few free space is available */
4834         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4835         buf->f_bavail = buf->f_bfree -
4836                         (ext4_r_blocks_count(es) + resv_blocks);
4837         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4838                 buf->f_bavail = 0;
4839         buf->f_files = le32_to_cpu(es->s_inodes_count);
4840         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4841         buf->f_namelen = EXT4_NAME_LEN;
4842         fsid = le64_to_cpup((void *)es->s_uuid) ^
4843                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4844         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4845         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4846
4847         return 0;
4848 }
4849
4850 /* Helper function for writing quotas on sync - we need to start transaction
4851  * before quota file is locked for write. Otherwise the are possible deadlocks:
4852  * Process 1                         Process 2
4853  * ext4_create()                     quota_sync()
4854  *   jbd2_journal_start()                  write_dquot()
4855  *   dquot_initialize()                         down(dqio_mutex)
4856  *     down(dqio_mutex)                    jbd2_journal_start()
4857  *
4858  */
4859
4860 #ifdef CONFIG_QUOTA
4861
4862 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4863 {
4864         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4865 }
4866
4867 static int ext4_write_dquot(struct dquot *dquot)
4868 {
4869         int ret, err;
4870         handle_t *handle;
4871         struct inode *inode;
4872
4873         inode = dquot_to_inode(dquot);
4874         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4875                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4876         if (IS_ERR(handle))
4877                 return PTR_ERR(handle);
4878         ret = dquot_commit(dquot);
4879         err = ext4_journal_stop(handle);
4880         if (!ret)
4881                 ret = err;
4882         return ret;
4883 }
4884
4885 static int ext4_acquire_dquot(struct dquot *dquot)
4886 {
4887         int ret, err;
4888         handle_t *handle;
4889
4890         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4891                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4892         if (IS_ERR(handle))
4893                 return PTR_ERR(handle);
4894         ret = dquot_acquire(dquot);
4895         err = ext4_journal_stop(handle);
4896         if (!ret)
4897                 ret = err;
4898         return ret;
4899 }
4900
4901 static int ext4_release_dquot(struct dquot *dquot)
4902 {
4903         int ret, err;
4904         handle_t *handle;
4905
4906         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4907                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4908         if (IS_ERR(handle)) {
4909                 /* Release dquot anyway to avoid endless cycle in dqput() */
4910                 dquot_release(dquot);
4911                 return PTR_ERR(handle);
4912         }
4913         ret = dquot_release(dquot);
4914         err = ext4_journal_stop(handle);
4915         if (!ret)
4916                 ret = err;
4917         return ret;
4918 }
4919
4920 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4921 {
4922         struct super_block *sb = dquot->dq_sb;
4923         struct ext4_sb_info *sbi = EXT4_SB(sb);
4924
4925         /* Are we journaling quotas? */
4926         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
4927             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4928                 dquot_mark_dquot_dirty(dquot);
4929                 return ext4_write_dquot(dquot);
4930         } else {
4931                 return dquot_mark_dquot_dirty(dquot);
4932         }
4933 }
4934
4935 static int ext4_write_info(struct super_block *sb, int type)
4936 {
4937         int ret, err;
4938         handle_t *handle;
4939
4940         /* Data block + inode block */
4941         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
4942         if (IS_ERR(handle))
4943                 return PTR_ERR(handle);
4944         ret = dquot_commit_info(sb, type);
4945         err = ext4_journal_stop(handle);
4946         if (!ret)
4947                 ret = err;
4948         return ret;
4949 }
4950
4951 /*
4952  * Turn on quotas during mount time - we need to find
4953  * the quota file and such...
4954  */
4955 static int ext4_quota_on_mount(struct super_block *sb, int type)
4956 {
4957         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4958                                         EXT4_SB(sb)->s_jquota_fmt, type);
4959 }
4960
4961 /*
4962  * Standard function to be called on quota_on
4963  */
4964 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4965                          struct path *path)
4966 {
4967         int err;
4968
4969         if (!test_opt(sb, QUOTA))
4970                 return -EINVAL;
4971
4972         /* Quotafile not on the same filesystem? */
4973         if (path->dentry->d_sb != sb)
4974                 return -EXDEV;
4975         /* Journaling quota? */
4976         if (EXT4_SB(sb)->s_qf_names[type]) {
4977                 /* Quotafile not in fs root? */
4978                 if (path->dentry->d_parent != sb->s_root)
4979                         ext4_msg(sb, KERN_WARNING,
4980                                 "Quota file not on filesystem root. "
4981                                 "Journaled quota will not work");
4982         }
4983
4984         /*
4985          * When we journal data on quota file, we have to flush journal to see
4986          * all updates to the file when we bypass pagecache...
4987          */
4988         if (EXT4_SB(sb)->s_journal &&
4989             ext4_should_journal_data(path->dentry->d_inode)) {
4990                 /*
4991                  * We don't need to lock updates but journal_flush() could
4992                  * otherwise be livelocked...
4993                  */
4994                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4995                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4996                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4997                 if (err)
4998                         return err;
4999         }
5000
5001         return dquot_quota_on(sb, type, format_id, path);
5002 }
5003
5004 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5005                              unsigned int flags)
5006 {
5007         int err;
5008         struct inode *qf_inode;
5009         unsigned long qf_inums[MAXQUOTAS] = {
5010                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5011                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5012         };
5013
5014         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5015
5016         if (!qf_inums[type])
5017                 return -EPERM;
5018
5019         qf_inode = ext4_iget(sb, qf_inums[type]);
5020         if (IS_ERR(qf_inode)) {
5021                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5022                 return PTR_ERR(qf_inode);
5023         }
5024
5025         /* Don't account quota for quota files to avoid recursion */
5026         qf_inode->i_flags |= S_NOQUOTA;
5027         err = dquot_enable(qf_inode, type, format_id, flags);
5028         iput(qf_inode);
5029
5030         return err;
5031 }
5032
5033 /* Enable usage tracking for all quota types. */
5034 static int ext4_enable_quotas(struct super_block *sb)
5035 {
5036         int type, err = 0;
5037         unsigned long qf_inums[MAXQUOTAS] = {
5038                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5039                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5040         };
5041
5042         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5043         for (type = 0; type < MAXQUOTAS; type++) {
5044                 if (qf_inums[type]) {
5045                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5046                                                 DQUOT_USAGE_ENABLED);
5047                         if (err) {
5048                                 ext4_warning(sb,
5049                                         "Failed to enable quota tracking "
5050                                         "(type=%d, err=%d). Please run "
5051                                         "e2fsck to fix.", type, err);
5052                                 return err;
5053                         }
5054                 }
5055         }
5056         return 0;
5057 }
5058
5059 /*
5060  * quota_on function that is used when QUOTA feature is set.
5061  */
5062 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5063                                  int format_id)
5064 {
5065         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5066                 return -EINVAL;
5067
5068         /*
5069          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5070          */
5071         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5072 }
5073
5074 static int ext4_quota_off(struct super_block *sb, int type)
5075 {
5076         struct inode *inode = sb_dqopt(sb)->files[type];
5077         handle_t *handle;
5078
5079         /* Force all delayed allocation blocks to be allocated.
5080          * Caller already holds s_umount sem */
5081         if (test_opt(sb, DELALLOC))
5082                 sync_filesystem(sb);
5083
5084         if (!inode)
5085                 goto out;
5086
5087         /* Update modification times of quota files when userspace can
5088          * start looking at them */
5089         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5090         if (IS_ERR(handle))
5091                 goto out;
5092         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5093         ext4_mark_inode_dirty(handle, inode);
5094         ext4_journal_stop(handle);
5095
5096 out:
5097         return dquot_quota_off(sb, type);
5098 }
5099
5100 /*
5101  * quota_off function that is used when QUOTA feature is set.
5102  */
5103 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5104 {
5105         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5106                 return -EINVAL;
5107
5108         /* Disable only the limits. */
5109         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5110 }
5111
5112 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5113  * acquiring the locks... As quota files are never truncated and quota code
5114  * itself serializes the operations (and no one else should touch the files)
5115  * we don't have to be afraid of races */
5116 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5117                                size_t len, loff_t off)
5118 {
5119         struct inode *inode = sb_dqopt(sb)->files[type];
5120         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5121         int err = 0;
5122         int offset = off & (sb->s_blocksize - 1);
5123         int tocopy;
5124         size_t toread;
5125         struct buffer_head *bh;
5126         loff_t i_size = i_size_read(inode);
5127
5128         if (off > i_size)
5129                 return 0;
5130         if (off+len > i_size)
5131                 len = i_size-off;
5132         toread = len;
5133         while (toread > 0) {
5134                 tocopy = sb->s_blocksize - offset < toread ?
5135                                 sb->s_blocksize - offset : toread;
5136                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5137                 if (err)
5138                         return err;
5139                 if (!bh)        /* A hole? */
5140                         memset(data, 0, tocopy);
5141                 else
5142                         memcpy(data, bh->b_data+offset, tocopy);
5143                 brelse(bh);
5144                 offset = 0;
5145                 toread -= tocopy;
5146                 data += tocopy;
5147                 blk++;
5148         }
5149         return len;
5150 }
5151
5152 /* Write to quotafile (we know the transaction is already started and has
5153  * enough credits) */
5154 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5155                                 const char *data, size_t len, loff_t off)
5156 {
5157         struct inode *inode = sb_dqopt(sb)->files[type];
5158         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5159         int err = 0;
5160         int offset = off & (sb->s_blocksize - 1);
5161         struct buffer_head *bh;
5162         handle_t *handle = journal_current_handle();
5163
5164         if (EXT4_SB(sb)->s_journal && !handle) {
5165                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5166                         " cancelled because transaction is not started",
5167                         (unsigned long long)off, (unsigned long long)len);
5168                 return -EIO;
5169         }
5170         /*
5171          * Since we account only one data block in transaction credits,
5172          * then it is impossible to cross a block boundary.
5173          */
5174         if (sb->s_blocksize - offset < len) {
5175                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5176                         " cancelled because not block aligned",
5177                         (unsigned long long)off, (unsigned long long)len);
5178                 return -EIO;
5179         }
5180
5181         bh = ext4_bread(handle, inode, blk, 1, &err);
5182         if (!bh)
5183                 goto out;
5184         err = ext4_journal_get_write_access(handle, bh);
5185         if (err) {
5186                 brelse(bh);
5187                 goto out;
5188         }
5189         lock_buffer(bh);
5190         memcpy(bh->b_data+offset, data, len);
5191         flush_dcache_page(bh->b_page);
5192         unlock_buffer(bh);
5193         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5194         brelse(bh);
5195 out:
5196         if (err)
5197                 return err;
5198         if (inode->i_size < off + len) {
5199                 i_size_write(inode, off + len);
5200                 EXT4_I(inode)->i_disksize = inode->i_size;
5201                 ext4_mark_inode_dirty(handle, inode);
5202         }
5203         return len;
5204 }
5205
5206 #endif
5207
5208 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5209                        const char *dev_name, void *data)
5210 {
5211         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5212 }
5213
5214 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5215 static inline void register_as_ext2(void)
5216 {
5217         int err = register_filesystem(&ext2_fs_type);
5218         if (err)
5219                 printk(KERN_WARNING
5220                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5221 }
5222
5223 static inline void unregister_as_ext2(void)
5224 {
5225         unregister_filesystem(&ext2_fs_type);
5226 }
5227
5228 static inline int ext2_feature_set_ok(struct super_block *sb)
5229 {
5230         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5231                 return 0;
5232         if (sb->s_flags & MS_RDONLY)
5233                 return 1;
5234         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5235                 return 0;
5236         return 1;
5237 }
5238 #else
5239 static inline void register_as_ext2(void) { }
5240 static inline void unregister_as_ext2(void) { }
5241 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5242 #endif
5243
5244 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5245 static inline void register_as_ext3(void)
5246 {
5247         int err = register_filesystem(&ext3_fs_type);
5248         if (err)
5249                 printk(KERN_WARNING
5250                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5251 }
5252
5253 static inline void unregister_as_ext3(void)
5254 {
5255         unregister_filesystem(&ext3_fs_type);
5256 }
5257
5258 static inline int ext3_feature_set_ok(struct super_block *sb)
5259 {
5260         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5261                 return 0;
5262         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5263                 return 0;
5264         if (sb->s_flags & MS_RDONLY)
5265                 return 1;
5266         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5267                 return 0;
5268         return 1;
5269 }
5270 #else
5271 static inline void register_as_ext3(void) { }
5272 static inline void unregister_as_ext3(void) { }
5273 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5274 #endif
5275
5276 static struct file_system_type ext4_fs_type = {
5277         .owner          = THIS_MODULE,
5278         .name           = "ext4",
5279         .mount          = ext4_mount,
5280         .kill_sb        = kill_block_super,
5281         .fs_flags       = FS_REQUIRES_DEV,
5282 };
5283 MODULE_ALIAS_FS("ext4");
5284
5285 static int __init ext4_init_feat_adverts(void)
5286 {
5287         struct ext4_features *ef;
5288         int ret = -ENOMEM;
5289
5290         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5291         if (!ef)
5292                 goto out;
5293
5294         ef->f_kobj.kset = ext4_kset;
5295         init_completion(&ef->f_kobj_unregister);
5296         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5297                                    "features");
5298         if (ret) {
5299                 kfree(ef);
5300                 goto out;
5301         }
5302
5303         ext4_feat = ef;
5304         ret = 0;
5305 out:
5306         return ret;
5307 }
5308
5309 static void ext4_exit_feat_adverts(void)
5310 {
5311         kobject_put(&ext4_feat->f_kobj);
5312         wait_for_completion(&ext4_feat->f_kobj_unregister);
5313         kfree(ext4_feat);
5314 }
5315
5316 /* Shared across all ext4 file systems */
5317 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5318 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5319
5320 static int __init ext4_init_fs(void)
5321 {
5322         int i, err;
5323
5324         ext4_li_info = NULL;
5325         mutex_init(&ext4_li_mtx);
5326
5327         /* Build-time check for flags consistency */
5328         ext4_check_flag_values();
5329
5330         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5331                 mutex_init(&ext4__aio_mutex[i]);
5332                 init_waitqueue_head(&ext4__ioend_wq[i]);
5333         }
5334
5335         err = ext4_init_es();
5336         if (err)
5337                 return err;
5338
5339         err = ext4_init_pageio();
5340         if (err)
5341                 goto out7;
5342
5343         err = ext4_init_system_zone();
5344         if (err)
5345                 goto out6;
5346         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5347         if (!ext4_kset) {
5348                 err = -ENOMEM;
5349                 goto out5;
5350         }
5351         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5352
5353         err = ext4_init_feat_adverts();
5354         if (err)
5355                 goto out4;
5356
5357         err = ext4_init_mballoc();
5358         if (err)
5359                 goto out3;
5360
5361         err = ext4_init_xattr();
5362         if (err)
5363                 goto out2;
5364         err = init_inodecache();
5365         if (err)
5366                 goto out1;
5367         register_as_ext3();
5368         register_as_ext2();
5369         err = register_filesystem(&ext4_fs_type);
5370         if (err)
5371                 goto out;
5372
5373         return 0;
5374 out:
5375         unregister_as_ext2();
5376         unregister_as_ext3();
5377         destroy_inodecache();
5378 out1:
5379         ext4_exit_xattr();
5380 out2:
5381         ext4_exit_mballoc();
5382 out3:
5383         ext4_exit_feat_adverts();
5384 out4:
5385         if (ext4_proc_root)
5386                 remove_proc_entry("fs/ext4", NULL);
5387         kset_unregister(ext4_kset);
5388 out5:
5389         ext4_exit_system_zone();
5390 out6:
5391         ext4_exit_pageio();
5392 out7:
5393         ext4_exit_es();
5394
5395         return err;
5396 }
5397
5398 static void __exit ext4_exit_fs(void)
5399 {
5400         ext4_destroy_lazyinit_thread();
5401         unregister_as_ext2();
5402         unregister_as_ext3();
5403         unregister_filesystem(&ext4_fs_type);
5404         destroy_inodecache();
5405         ext4_exit_xattr();
5406         ext4_exit_mballoc();
5407         ext4_exit_feat_adverts();
5408         remove_proc_entry("fs/ext4", NULL);
5409         kset_unregister(ext4_kset);
5410         ext4_exit_system_zone();
5411         ext4_exit_pageio();
5412 }
5413
5414 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5415 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5416 MODULE_LICENSE("GPL");
5417 module_init(ext4_init_fs)
5418 module_exit(ext4_exit_fs)