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_sync_fs_nojournal(struct super_block *sb, int wait);
  73 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  74 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  75 static int ext4_unfreeze(struct super_block *sb);
  76 static int ext4_freeze(struct super_block *sb);
  77 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  78                        const char *dev_name, void *data);
  79 static inline int ext2_feature_set_ok(struct super_block *sb);
  80 static inline int ext3_feature_set_ok(struct super_block *sb);
  81 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  82 static void ext4_destroy_lazyinit_thread(void);
  83 static void ext4_unregister_li_request(struct super_block *sb);
  84 static void ext4_clear_request_list(void);
  85 static int ext4_reserve_clusters(struct ext4_sb_info *, ext4_fsblk_t);
  86
  87 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
  88 static struct file_system_type ext2_fs_type = {
  89         .owner          = THIS_MODULE,
  90         .name           = "ext2",
  91         .mount          = ext4_mount,
  92         .kill_sb        = kill_block_super,
  93         .fs_flags       = FS_REQUIRES_DEV,
  94 };
  95 MODULE_ALIAS_FS("ext2");
  96 MODULE_ALIAS("ext2");
  97 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  98 #else
  99 #define IS_EXT2_SB(sb) (0)
 100 #endif
 101
 102
 103 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
 104 static struct file_system_type ext3_fs_type = {
 105         .owner          = THIS_MODULE,
 106         .name           = "ext3",
 107         .mount          = ext4_mount,
 108         .kill_sb        = kill_block_super,
 109         .fs_flags       = FS_REQUIRES_DEV,
 110 };
 111 MODULE_ALIAS_FS("ext3");
 112 MODULE_ALIAS("ext3");
 113 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 114 #else
 115 #define IS_EXT3_SB(sb) (0)
 116 #endif
 117
 118 static int ext4_verify_csum_type(struct super_block *sb,
 119                                  struct ext4_super_block *es)
 120 {
 121         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 122                                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 123                 return 1;
 124
 125         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 126 }
 127
 128 static __le32 ext4_superblock_csum(struct super_block *sb,
 129                                    struct ext4_super_block *es)
 130 {
 131         struct ext4_sb_info *sbi = EXT4_SB(sb);
 132         int offset = offsetof(struct ext4_super_block, s_checksum);
 133         __u32 csum;
 134
 135         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 136
 137         return cpu_to_le32(csum);
 138 }
 139
 140 int ext4_superblock_csum_verify(struct super_block *sb,
 141                                 struct ext4_super_block *es)
 142 {
 143         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 144                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 145                 return 1;
 146
 147         return es->s_checksum == ext4_superblock_csum(sb, es);
 148 }
 149
 150 void ext4_superblock_csum_set(struct super_block *sb)
 151 {
 152         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 153
 154         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb,
 155                 EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
 156                 return;
 157
 158         es->s_checksum = ext4_superblock_csum(sb, es);
 159 }
 160
 161 void *ext4_kvmalloc(size_t size, gfp_t flags)
 162 {
 163         void *ret;
 164
 165         ret = kmalloc(size, flags);
 166         if (!ret)
 167                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 168         return ret;
 169 }
 170
 171 void *ext4_kvzalloc(size_t size, gfp_t flags)
 172 {
 173         void *ret;
 174
 175         ret = kzalloc(size, flags);
 176         if (!ret)
 177                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 178         return ret;
 179 }
 180
 181 void ext4_kvfree(void *ptr)
 182 {
 183         if (is_vmalloc_addr(ptr))
 184                 vfree(ptr);
 185         else
 186                 kfree(ptr);
 187
 188 }
 189
 190 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 191                                struct ext4_group_desc *bg)
 192 {
 193         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 194                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 195                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 196 }
 197
 198 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 199                                struct ext4_group_desc *bg)
 200 {
 201         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 202                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 203                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 204 }
 205
 206 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 207                               struct ext4_group_desc *bg)
 208 {
 209         return le32_to_cpu(bg->bg_inode_table_lo) |
 210                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 211                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 212 }
 213
 214 __u32 ext4_free_group_clusters(struct super_block *sb,
 215                                struct ext4_group_desc *bg)
 216 {
 217         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 218                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 219                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 220 }
 221
 222 __u32 ext4_free_inodes_count(struct super_block *sb,
 223                               struct ext4_group_desc *bg)
 224 {
 225         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 226                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 227                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 228 }
 229
 230 __u32 ext4_used_dirs_count(struct super_block *sb,
 231                               struct ext4_group_desc *bg)
 232 {
 233         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 234                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 235                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 236 }
 237
 238 __u32 ext4_itable_unused_count(struct super_block *sb,
 239                               struct ext4_group_desc *bg)
 240 {
 241         return le16_to_cpu(bg->bg_itable_unused_lo) |
 242                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 243                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 244 }
 245
 246 void ext4_block_bitmap_set(struct super_block *sb,
 247                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 248 {
 249         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 250         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 251                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 252 }
 253
 254 void ext4_inode_bitmap_set(struct super_block *sb,
 255                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 256 {
 257         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 258         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 259                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 260 }
 261
 262 void ext4_inode_table_set(struct super_block *sb,
 263                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 264 {
 265         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 266         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 267                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 268 }
 269
 270 void ext4_free_group_clusters_set(struct super_block *sb,
 271                                   struct ext4_group_desc *bg, __u32 count)
 272 {
 273         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 274         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 275                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 276 }
 277
 278 void ext4_free_inodes_set(struct super_block *sb,
 279                           struct ext4_group_desc *bg, __u32 count)
 280 {
 281         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 282         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 283                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 284 }
 285
 286 void ext4_used_dirs_set(struct super_block *sb,
 287                           struct ext4_group_desc *bg, __u32 count)
 288 {
 289         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 290         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 291                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 292 }
 293
 294 void ext4_itable_unused_set(struct super_block *sb,
 295                           struct ext4_group_desc *bg, __u32 count)
 296 {
 297         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 298         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 299                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 300 }
 301
 302
 303 static void __save_error_info(struct super_block *sb, const char *func,
 304                             unsigned int line)
 305 {
 306         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 307
 308         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 309         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 310         es->s_last_error_time = cpu_to_le32(get_seconds());
 311         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 312         es->s_last_error_line = cpu_to_le32(line);
 313         if (!es->s_first_error_time) {
 314                 es->s_first_error_time = es->s_last_error_time;
 315                 strncpy(es->s_first_error_func, func,
 316                         sizeof(es->s_first_error_func));
 317                 es->s_first_error_line = cpu_to_le32(line);
 318                 es->s_first_error_ino = es->s_last_error_ino;
 319                 es->s_first_error_block = es->s_last_error_block;
 320         }
 321         /*
 322          * Start the daily error reporting function if it hasn't been
 323          * started already
 324          */
 325         if (!es->s_error_count)
 326                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 327         le32_add_cpu(&es->s_error_count, 1);
 328 }
 329
 330 static void save_error_info(struct super_block *sb, const char *func,
 331                             unsigned int line)
 332 {
 333         __save_error_info(sb, func, line);
 334         ext4_commit_super(sb, 1);
 335 }
 336
 337 /*
 338  * The del_gendisk() function uninitializes the disk-specific data
 339  * structures, including the bdi structure, without telling anyone
 340  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 341  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 342  * This is a kludge to prevent these oops until we can put in a proper
 343  * hook in del_gendisk() to inform the VFS and file system layers.
 344  */
 345 static int block_device_ejected(struct super_block *sb)
 346 {
 347         struct inode *bd_inode = sb->s_bdev->bd_inode;
 348         struct backing_dev_info *bdi = bd_inode->i_mapping->backing_dev_info;
 349
 350         return bdi->dev == NULL;
 351 }
 352
 353 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 354 {
 355         struct super_block              *sb = journal->j_private;
 356         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 357         int                             error = is_journal_aborted(journal);
 358         struct ext4_journal_cb_entry    *jce;
 359
 360         BUG_ON(txn->t_state == T_FINISHED);
 361         spin_lock(&sbi->s_md_lock);
 362         while (!list_empty(&txn->t_private_list)) {
 363                 jce = list_entry(txn->t_private_list.next,
 364                                  struct ext4_journal_cb_entry, jce_list);
 365                 list_del_init(&jce->jce_list);
 366                 spin_unlock(&sbi->s_md_lock);
 367                 jce->jce_func(sb, jce, error);
 368                 spin_lock(&sbi->s_md_lock);
 369         }
 370         spin_unlock(&sbi->s_md_lock);
 371 }
 372
 373 /* Deal with the reporting of failure conditions on a filesystem such as
 374  * inconsistencies detected or read IO failures.
 375  *
 376  * On ext2, we can store the error state of the filesystem in the
 377  * superblock.  That is not possible on ext4, because we may have other
 378  * write ordering constraints on the superblock which prevent us from
 379  * writing it out straight away; and given that the journal is about to
 380  * be aborted, we can't rely on the current, or future, transactions to
 381  * write out the superblock safely.
 382  *
 383  * We'll just use the jbd2_journal_abort() error code to record an error in
 384  * the journal instead.  On recovery, the journal will complain about
 385  * that error until we've noted it down and cleared it.
 386  */
 387
 388 static void ext4_handle_error(struct super_block *sb)
 389 {
 390         if (sb->s_flags & MS_RDONLY)
 391                 return;
 392
 393         if (!test_opt(sb, ERRORS_CONT)) {
 394                 journal_t *journal = EXT4_SB(sb)->s_journal;
 395
 396                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 397                 if (journal)
 398                         jbd2_journal_abort(journal, -EIO);
 399         }
 400         if (test_opt(sb, ERRORS_RO)) {
 401                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 402                 /*
 403                  * Make sure updated value of ->s_mount_flags will be visible
 404                  * before ->s_flags update
 405                  */
 406                 smp_wmb();
 407                 sb->s_flags |= MS_RDONLY;
 408         }
 409         if (test_opt(sb, ERRORS_PANIC))
 410                 panic("EXT4-fs (device %s): panic forced after error\n",
 411                         sb->s_id);
 412 }
 413
 414 void __ext4_error(struct super_block *sb, const char *function,
 415                   unsigned int line, const char *fmt, ...)
 416 {
 417         struct va_format vaf;
 418         va_list args;
 419
 420         va_start(args, fmt);
 421         vaf.fmt = fmt;
 422         vaf.va = &args;
 423         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 424                sb->s_id, function, line, current->comm, &vaf);
 425         va_end(args);
 426         save_error_info(sb, function, line);
 427
 428         ext4_handle_error(sb);
 429 }
 430
 431 void __ext4_error_inode(struct inode *inode, const char *function,
 432                         unsigned int line, ext4_fsblk_t block,
 433                         const char *fmt, ...)
 434 {
 435         va_list args;
 436         struct va_format vaf;
 437         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 438
 439         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 440         es->s_last_error_block = cpu_to_le64(block);
 441         save_error_info(inode->i_sb, function, line);
 442         va_start(args, fmt);
 443         vaf.fmt = fmt;
 444         vaf.va = &args;
 445         if (block)
 446                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 447                        "inode #%lu: block %llu: comm %s: %pV\n",
 448                        inode->i_sb->s_id, function, line, inode->i_ino,
 449                        block, current->comm, &vaf);
 450         else
 451                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 452                        "inode #%lu: comm %s: %pV\n",
 453                        inode->i_sb->s_id, function, line, inode->i_ino,
 454                        current->comm, &vaf);
 455         va_end(args);
 456
 457         ext4_handle_error(inode->i_sb);
 458 }
 459
 460 void __ext4_error_file(struct file *file, const char *function,
 461                        unsigned int line, ext4_fsblk_t block,
 462                        const char *fmt, ...)
 463 {
 464         va_list args;
 465         struct va_format vaf;
 466         struct ext4_super_block *es;
 467         struct inode *inode = file_inode(file);
 468         char pathname[80], *path;
 469
 470         es = EXT4_SB(inode->i_sb)->s_es;
 471         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 472         save_error_info(inode->i_sb, function, line);
 473         path = d_path(&(file->f_path), pathname, sizeof(pathname));
 474         if (IS_ERR(path))
 475                 path = "(unknown)";
 476         va_start(args, fmt);
 477         vaf.fmt = fmt;
 478         vaf.va = &args;
 479         if (block)
 480                 printk(KERN_CRIT
 481                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 482                        "block %llu: comm %s: path %s: %pV\n",
 483                        inode->i_sb->s_id, function, line, inode->i_ino,
 484                        block, current->comm, path, &vaf);
 485         else
 486                 printk(KERN_CRIT
 487                        "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 488                        "comm %s: path %s: %pV\n",
 489                        inode->i_sb->s_id, function, line, inode->i_ino,
 490                        current->comm, path, &vaf);
 491         va_end(args);
 492
 493         ext4_handle_error(inode->i_sb);
 494 }
 495
 496 const char *ext4_decode_error(struct super_block *sb, int errno,
 497                               char nbuf[16])
 498 {
 499         char *errstr = NULL;
 500
 501         switch (errno) {
 502         case -EIO:
 503                 errstr = "IO failure";
 504                 break;
 505         case -ENOMEM:
 506                 errstr = "Out of memory";
 507                 break;
 508         case -EROFS:
 509                 if (!sb || (EXT4_SB(sb)->s_journal &&
 510                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 511                         errstr = "Journal has aborted";
 512                 else
 513                         errstr = "Readonly filesystem";
 514                 break;
 515         default:
 516                 /* If the caller passed in an extra buffer for unknown
 517                  * errors, textualise them now.  Else we just return
 518                  * NULL. */
 519                 if (nbuf) {
 520                         /* Check for truncated error codes... */
 521                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 522                                 errstr = nbuf;
 523                 }
 524                 break;
 525         }
 526
 527         return errstr;
 528 }
 529
 530 /* __ext4_std_error decodes expected errors from journaling functions
 531  * automatically and invokes the appropriate error response.  */
 532
 533 void __ext4_std_error(struct super_block *sb, const char *function,
 534                       unsigned int line, int errno)
 535 {
 536         char nbuf[16];
 537         const char *errstr;
 538
 539         /* Special case: if the error is EROFS, and we're not already
 540          * inside a transaction, then there's really no point in logging
 541          * an error. */
 542         if (errno == -EROFS && journal_current_handle() == NULL &&
 543             (sb->s_flags & MS_RDONLY))
 544                 return;
 545
 546         errstr = ext4_decode_error(sb, errno, nbuf);
 547         printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 548                sb->s_id, function, line, errstr);
 549         save_error_info(sb, function, line);
 550
 551         ext4_handle_error(sb);
 552 }
 553
 554 /*
 555  * ext4_abort is a much stronger failure handler than ext4_error.  The
 556  * abort function may be used to deal with unrecoverable failures such
 557  * as journal IO errors or ENOMEM at a critical moment in log management.
 558  *
 559  * We unconditionally force the filesystem into an ABORT|READONLY state,
 560  * unless the error response on the fs has been set to panic in which
 561  * case we take the easy way out and panic immediately.
 562  */
 563
 564 void __ext4_abort(struct super_block *sb, const char *function,
 565                 unsigned int line, const char *fmt, ...)
 566 {
 567         va_list args;
 568
 569         save_error_info(sb, function, line);
 570         va_start(args, fmt);
 571         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 572                function, line);
 573         vprintk(fmt, args);
 574         printk("\n");
 575         va_end(args);
 576
 577         if ((sb->s_flags & MS_RDONLY) == 0) {
 578                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 579                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 580                 /*
 581                  * Make sure updated value of ->s_mount_flags will be visible
 582                  * before ->s_flags update
 583                  */
 584                 smp_wmb();
 585                 sb->s_flags |= MS_RDONLY;
 586                 if (EXT4_SB(sb)->s_journal)
 587                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 588                 save_error_info(sb, function, line);
 589         }
 590         if (test_opt(sb, ERRORS_PANIC))
 591                 panic("EXT4-fs panic from previous error\n");
 592 }
 593
 594 void __ext4_msg(struct super_block *sb,
 595                 const char *prefix, const char *fmt, ...)
 596 {
 597         struct va_format vaf;
 598         va_list args;
 599
 600         va_start(args, fmt);
 601         vaf.fmt = fmt;
 602         vaf.va = &args;
 603         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 604         va_end(args);
 605 }
 606
 607 void __ext4_warning(struct super_block *sb, const char *function,
 608                     unsigned int line, const char *fmt, ...)
 609 {
 610         struct va_format vaf;
 611         va_list args;
 612
 613         va_start(args, fmt);
 614         vaf.fmt = fmt;
 615         vaf.va = &args;
 616         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 617                sb->s_id, function, line, &vaf);
 618         va_end(args);
 619 }
 620
 621 void __ext4_grp_locked_error(const char *function, unsigned int line,
 622                              struct super_block *sb, ext4_group_t grp,
 623                              unsigned long ino, ext4_fsblk_t block,
 624                              const char *fmt, ...)
 625 __releases(bitlock)
 626 __acquires(bitlock)
 627 {
 628         struct va_format vaf;
 629         va_list args;
 630         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 631
 632         es->s_last_error_ino = cpu_to_le32(ino);
 633         es->s_last_error_block = cpu_to_le64(block);
 634         __save_error_info(sb, function, line);
 635
 636         va_start(args, fmt);
 637
 638         vaf.fmt = fmt;
 639         vaf.va = &args;
 640         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 641                sb->s_id, function, line, grp);
 642         if (ino)
 643                 printk(KERN_CONT "inode %lu: ", ino);
 644         if (block)
 645                 printk(KERN_CONT "block %llu:", (unsigned long long) block);
 646         printk(KERN_CONT "%pV\n", &vaf);
 647         va_end(args);
 648
 649         if (test_opt(sb, ERRORS_CONT)) {
 650                 ext4_commit_super(sb, 0);
 651                 return;
 652         }
 653
 654         ext4_unlock_group(sb, grp);
 655         ext4_handle_error(sb);
 656         /*
 657          * We only get here in the ERRORS_RO case; relocking the group
 658          * may be dangerous, but nothing bad will happen since the
 659          * filesystem will have already been marked read/only and the
 660          * journal has been aborted.  We return 1 as a hint to callers
 661          * who might what to use the return value from
 662          * ext4_grp_locked_error() to distinguish between the
 663          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 664          * aggressively from the ext4 function in question, with a
 665          * more appropriate error code.
 666          */
 667         ext4_lock_group(sb, grp);
 668         return;
 669 }
 670
 671 void ext4_update_dynamic_rev(struct super_block *sb)
 672 {
 673         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 674
 675         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 676                 return;
 677
 678         ext4_warning(sb,
 679                      "updating to rev %d because of new feature flag, "
 680                      "running e2fsck is recommended",
 681                      EXT4_DYNAMIC_REV);
 682
 683         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 684         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 685         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 686         /* leave es->s_feature_*compat flags alone */
 687         /* es->s_uuid will be set by e2fsck if empty */
 688
 689         /*
 690          * The rest of the superblock fields should be zero, and if not it
 691          * means they are likely already in use, so leave them alone.  We
 692          * can leave it up to e2fsck to clean up any inconsistencies there.
 693          */
 694 }
 695
 696 /*
 697  * Open the external journal device
 698  */
 699 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 700 {
 701         struct block_device *bdev;
 702         char b[BDEVNAME_SIZE];
 703
 704         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 705         if (IS_ERR(bdev))
 706                 goto fail;
 707         return bdev;
 708
 709 fail:
 710         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 711                         __bdevname(dev, b), PTR_ERR(bdev));
 712         return NULL;
 713 }
 714
 715 /*
 716  * Release the journal device
 717  */
 718 static void ext4_blkdev_put(struct block_device *bdev)
 719 {
 720         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 721 }
 722
 723 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 724 {
 725         struct block_device *bdev;
 726         bdev = sbi->journal_bdev;
 727         if (bdev) {
 728                 ext4_blkdev_put(bdev);
 729                 sbi->journal_bdev = NULL;
 730         }
 731 }
 732
 733 static inline struct inode *orphan_list_entry(struct list_head *l)
 734 {
 735         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 736 }
 737
 738 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 739 {
 740         struct list_head *l;
 741
 742         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 743                  le32_to_cpu(sbi->s_es->s_last_orphan));
 744
 745         printk(KERN_ERR "sb_info orphan list:\n");
 746         list_for_each(l, &sbi->s_orphan) {
 747                 struct inode *inode = orphan_list_entry(l);
 748                 printk(KERN_ERR "  "
 749                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 750                        inode->i_sb->s_id, inode->i_ino, inode,
 751                        inode->i_mode, inode->i_nlink,
 752                        NEXT_ORPHAN(inode));
 753         }
 754 }
 755
 756 static void ext4_put_super(struct super_block *sb)
 757 {
 758         struct ext4_sb_info *sbi = EXT4_SB(sb);
 759         struct ext4_super_block *es = sbi->s_es;
 760         int i, err;
 761
 762         ext4_unregister_li_request(sb);
 763         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 764
 765         flush_workqueue(sbi->rsv_conversion_wq);
 766         destroy_workqueue(sbi->rsv_conversion_wq);
 767
 768         if (sbi->s_journal) {
 769                 err = jbd2_journal_destroy(sbi->s_journal);
 770                 sbi->s_journal = NULL;
 771                 if (err < 0)
 772                         ext4_abort(sb, "Couldn't clean up the journal");
 773         }
 774
 775         ext4_es_unregister_shrinker(sbi);
 776         del_timer(&sbi->s_err_report);
 777         ext4_release_system_zone(sb);
 778         ext4_mb_release(sb);
 779         ext4_ext_release(sb);
 780         ext4_xattr_put_super(sb);
 781
 782         if (!(sb->s_flags & MS_RDONLY)) {
 783                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 784                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 785         }
 786         if (!(sb->s_flags & MS_RDONLY))
 787                 ext4_commit_super(sb, 1);
 788
 789         if (sbi->s_proc) {
 790                 remove_proc_entry("options", sbi->s_proc);
 791                 remove_proc_entry(sb->s_id, ext4_proc_root);
 792         }
 793         kobject_del(&sbi->s_kobj);
 794
 795         for (i = 0; i < sbi->s_gdb_count; i++)
 796                 brelse(sbi->s_group_desc[i]);
 797         ext4_kvfree(sbi->s_group_desc);
 798         ext4_kvfree(sbi->s_flex_groups);
 799         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 800         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 801         percpu_counter_destroy(&sbi->s_dirs_counter);
 802         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 803         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
 804         brelse(sbi->s_sbh);
 805 #ifdef CONFIG_QUOTA
 806         for (i = 0; i < MAXQUOTAS; i++)
 807                 kfree(sbi->s_qf_names[i]);
 808 #endif
 809
 810         /* Debugging code just in case the in-memory inode orphan list
 811          * isn't empty.  The on-disk one can be non-empty if we've
 812          * detected an error and taken the fs readonly, but the
 813          * in-memory list had better be clean by this point. */
 814         if (!list_empty(&sbi->s_orphan))
 815                 dump_orphan_list(sb, sbi);
 816         J_ASSERT(list_empty(&sbi->s_orphan));
 817
 818         invalidate_bdev(sb->s_bdev);
 819         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 820                 /*
 821                  * Invalidate the journal device's buffers.  We don't want them
 822                  * floating about in memory - the physical journal device may
 823                  * hotswapped, and it breaks the `ro-after' testing code.
 824                  */
 825                 sync_blockdev(sbi->journal_bdev);
 826                 invalidate_bdev(sbi->journal_bdev);
 827                 ext4_blkdev_remove(sbi);
 828         }
 829         if (sbi->s_mmp_tsk)
 830                 kthread_stop(sbi->s_mmp_tsk);
 831         sb->s_fs_info = NULL;
 832         /*
 833          * Now that we are completely done shutting down the
 834          * superblock, we need to actually destroy the kobject.
 835          */
 836         kobject_put(&sbi->s_kobj);
 837         wait_for_completion(&sbi->s_kobj_unregister);
 838         if (sbi->s_chksum_driver)
 839                 crypto_free_shash(sbi->s_chksum_driver);
 840         kfree(sbi->s_blockgroup_lock);
 841         kfree(sbi);
 842 }
 843
 844 static struct kmem_cache *ext4_inode_cachep;
 845
 846 /*
 847  * Called inside transaction, so use GFP_NOFS
 848  */
 849 static struct inode *ext4_alloc_inode(struct super_block *sb)
 850 {
 851         struct ext4_inode_info *ei;
 852
 853         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 854         if (!ei)
 855                 return NULL;
 856
 857         ei->vfs_inode.i_version = 1;
 858         INIT_LIST_HEAD(&ei->i_prealloc_list);
 859         spin_lock_init(&ei->i_prealloc_lock);
 860         ext4_es_init_tree(&ei->i_es_tree);
 861         rwlock_init(&ei->i_es_lock);
 862         INIT_LIST_HEAD(&ei->i_es_lru);
 863         ei->i_es_lru_nr = 0;
 864         ei->i_touch_when = 0;
 865         ei->i_reserved_data_blocks = 0;
 866         ei->i_reserved_meta_blocks = 0;
 867         ei->i_allocated_meta_blocks = 0;
 868         ei->i_da_metadata_calc_len = 0;
 869         ei->i_da_metadata_calc_last_lblock = 0;
 870         spin_lock_init(&(ei->i_block_reservation_lock));
 871 #ifdef CONFIG_QUOTA
 872         ei->i_reserved_quota = 0;
 873 #endif
 874         ei->jinode = NULL;
 875         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 876         spin_lock_init(&ei->i_completed_io_lock);
 877         ei->i_sync_tid = 0;
 878         ei->i_datasync_tid = 0;
 879         atomic_set(&ei->i_ioend_count, 0);
 880         atomic_set(&ei->i_unwritten, 0);
 881         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 882
 883         return &ei->vfs_inode;
 884 }
 885
 886 static int ext4_drop_inode(struct inode *inode)
 887 {
 888         int drop = generic_drop_inode(inode);
 889
 890         trace_ext4_drop_inode(inode, drop);
 891         return drop;
 892 }
 893
 894 static void ext4_i_callback(struct rcu_head *head)
 895 {
 896         struct inode *inode = container_of(head, struct inode, i_rcu);
 897         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 898 }
 899
 900 static void ext4_destroy_inode(struct inode *inode)
 901 {
 902         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 903                 ext4_msg(inode->i_sb, KERN_ERR,
 904                          "Inode %lu (%p): orphan list check failed!",
 905                          inode->i_ino, EXT4_I(inode));
 906                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 907                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 908                                 true);
 909                 dump_stack();
 910         }
 911         call_rcu(&inode->i_rcu, ext4_i_callback);
 912 }
 913
 914 static void init_once(void *foo)
 915 {
 916         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 917
 918         INIT_LIST_HEAD(&ei->i_orphan);
 919         init_rwsem(&ei->xattr_sem);
 920         init_rwsem(&ei->i_data_sem);
 921         inode_init_once(&ei->vfs_inode);
 922 }
 923
 924 static int init_inodecache(void)
 925 {
 926         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 927                                              sizeof(struct ext4_inode_info),
 928                                              0, (SLAB_RECLAIM_ACCOUNT|
 929                                                 SLAB_MEM_SPREAD),
 930                                              init_once);
 931         if (ext4_inode_cachep == NULL)
 932                 return -ENOMEM;
 933         return 0;
 934 }
 935
 936 static void destroy_inodecache(void)
 937 {
 938         /*
 939          * Make sure all delayed rcu free inodes are flushed before we
 940          * destroy cache.
 941          */
 942         rcu_barrier();
 943         kmem_cache_destroy(ext4_inode_cachep);
 944 }
 945
 946 void ext4_clear_inode(struct inode *inode)
 947 {
 948         invalidate_inode_buffers(inode);
 949         clear_inode(inode);
 950         dquot_drop(inode);
 951         ext4_discard_preallocations(inode);
 952         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 953         ext4_es_lru_del(inode);
 954         if (EXT4_I(inode)->jinode) {
 955                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 956                                                EXT4_I(inode)->jinode);
 957                 jbd2_free_inode(EXT4_I(inode)->jinode);
 958                 EXT4_I(inode)->jinode = NULL;
 959         }
 960 }
 961
 962 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 963                                         u64 ino, u32 generation)
 964 {
 965         struct inode *inode;
 966
 967         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 968                 return ERR_PTR(-ESTALE);
 969         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 970                 return ERR_PTR(-ESTALE);
 971
 972         /* iget isn't really right if the inode is currently unallocated!!
 973          *
 974          * ext4_read_inode will return a bad_inode if the inode had been
 975          * deleted, so we should be safe.
 976          *
 977          * Currently we don't know the generation for parent directory, so
 978          * a generation of 0 means "accept any"
 979          */
 980         inode = ext4_iget(sb, ino);
 981         if (IS_ERR(inode))
 982                 return ERR_CAST(inode);
 983         if (generation && inode->i_generation != generation) {
 984                 iput(inode);
 985                 return ERR_PTR(-ESTALE);
 986         }
 987
 988         return inode;
 989 }
 990
 991 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 992                                         int fh_len, int fh_type)
 993 {
 994         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 995                                     ext4_nfs_get_inode);
 996 }
 997
 998 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
 999                                         int fh_len, int fh_type)
1000 {
1001         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1002                                     ext4_nfs_get_inode);
1003 }
1004
1005 /*
1006  * Try to release metadata pages (indirect blocks, directories) which are
1007  * mapped via the block device.  Since these pages could have journal heads
1008  * which would prevent try_to_free_buffers() from freeing them, we must use
1009  * jbd2 layer's try_to_free_buffers() function to release them.
1010  */
1011 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1012                                  gfp_t wait)
1013 {
1014         journal_t *journal = EXT4_SB(sb)->s_journal;
1015
1016         WARN_ON(PageChecked(page));
1017         if (!page_has_buffers(page))
1018                 return 0;
1019         if (journal)
1020                 return jbd2_journal_try_to_free_buffers(journal, page,
1021                                                         wait & ~__GFP_WAIT);
1022         return try_to_free_buffers(page);
1023 }
1024
1025 #ifdef CONFIG_QUOTA
1026 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1027 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1028
1029 static int ext4_write_dquot(struct dquot *dquot);
1030 static int ext4_acquire_dquot(struct dquot *dquot);
1031 static int ext4_release_dquot(struct dquot *dquot);
1032 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1033 static int ext4_write_info(struct super_block *sb, int type);
1034 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1035                          struct path *path);
1036 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1037                                  int format_id);
1038 static int ext4_quota_off(struct super_block *sb, int type);
1039 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1040 static int ext4_quota_on_mount(struct super_block *sb, int type);
1041 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1042                                size_t len, loff_t off);
1043 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1044                                 const char *data, size_t len, loff_t off);
1045 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1046                              unsigned int flags);
1047 static int ext4_enable_quotas(struct super_block *sb);
1048
1049 static const struct dquot_operations ext4_quota_operations = {
1050         .get_reserved_space = ext4_get_reserved_space,
1051         .write_dquot    = ext4_write_dquot,
1052         .acquire_dquot  = ext4_acquire_dquot,
1053         .release_dquot  = ext4_release_dquot,
1054         .mark_dirty     = ext4_mark_dquot_dirty,
1055         .write_info     = ext4_write_info,
1056         .alloc_dquot    = dquot_alloc,
1057         .destroy_dquot  = dquot_destroy,
1058 };
1059
1060 static const struct quotactl_ops ext4_qctl_operations = {
1061         .quota_on       = ext4_quota_on,
1062         .quota_off      = ext4_quota_off,
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
1070 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1071         .quota_on_meta  = ext4_quota_on_sysfile,
1072         .quota_off      = ext4_quota_off_sysfile,
1073         .quota_sync     = dquot_quota_sync,
1074         .get_info       = dquot_get_dqinfo,
1075         .set_info       = dquot_set_dqinfo,
1076         .get_dqblk      = dquot_get_dqblk,
1077         .set_dqblk      = dquot_set_dqblk
1078 };
1079 #endif
1080
1081 static const struct super_operations ext4_sops = {
1082         .alloc_inode    = ext4_alloc_inode,
1083         .destroy_inode  = ext4_destroy_inode,
1084         .write_inode    = ext4_write_inode,
1085         .dirty_inode    = ext4_dirty_inode,
1086         .drop_inode     = ext4_drop_inode,
1087         .evict_inode    = ext4_evict_inode,
1088         .put_super      = ext4_put_super,
1089         .sync_fs        = ext4_sync_fs,
1090         .freeze_fs      = ext4_freeze,
1091         .unfreeze_fs    = ext4_unfreeze,
1092         .statfs         = ext4_statfs,
1093         .remount_fs     = ext4_remount,
1094         .show_options   = ext4_show_options,
1095 #ifdef CONFIG_QUOTA
1096         .quota_read     = ext4_quota_read,
1097         .quota_write    = ext4_quota_write,
1098 #endif
1099         .bdev_try_to_free_page = bdev_try_to_free_page,
1100 };
1101
1102 static const struct super_operations ext4_nojournal_sops = {
1103         .alloc_inode    = ext4_alloc_inode,
1104         .destroy_inode  = ext4_destroy_inode,
1105         .write_inode    = ext4_write_inode,
1106         .dirty_inode    = ext4_dirty_inode,
1107         .drop_inode     = ext4_drop_inode,
1108         .evict_inode    = ext4_evict_inode,
1109         .sync_fs        = ext4_sync_fs_nojournal,
1110         .put_super      = ext4_put_super,
1111         .statfs         = ext4_statfs,
1112         .remount_fs     = ext4_remount,
1113         .show_options   = ext4_show_options,
1114 #ifdef CONFIG_QUOTA
1115         .quota_read     = ext4_quota_read,
1116         .quota_write    = ext4_quota_write,
1117 #endif
1118         .bdev_try_to_free_page = bdev_try_to_free_page,
1119 };
1120
1121 static const struct export_operations ext4_export_ops = {
1122         .fh_to_dentry = ext4_fh_to_dentry,
1123         .fh_to_parent = ext4_fh_to_parent,
1124         .get_parent = ext4_get_parent,
1125 };
1126
1127 enum {
1128         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1129         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1130         Opt_nouid32, Opt_debug, Opt_removed,
1131         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1132         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1133         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1134         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1135         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1136         Opt_data_err_abort, Opt_data_err_ignore,
1137         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1138         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1139         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1140         Opt_usrquota, Opt_grpquota, Opt_i_version,
1141         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1142         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1143         Opt_inode_readahead_blks, Opt_journal_ioprio,
1144         Opt_dioread_nolock, Opt_dioread_lock,
1145         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1146         Opt_max_dir_size_kb,
1147 };
1148
1149 static const match_table_t tokens = {
1150         {Opt_bsd_df, "bsddf"},
1151         {Opt_minix_df, "minixdf"},
1152         {Opt_grpid, "grpid"},
1153         {Opt_grpid, "bsdgroups"},
1154         {Opt_nogrpid, "nogrpid"},
1155         {Opt_nogrpid, "sysvgroups"},
1156         {Opt_resgid, "resgid=%u"},
1157         {Opt_resuid, "resuid=%u"},
1158         {Opt_sb, "sb=%u"},
1159         {Opt_err_cont, "errors=continue"},
1160         {Opt_err_panic, "errors=panic"},
1161         {Opt_err_ro, "errors=remount-ro"},
1162         {Opt_nouid32, "nouid32"},
1163         {Opt_debug, "debug"},
1164         {Opt_removed, "oldalloc"},
1165         {Opt_removed, "orlov"},
1166         {Opt_user_xattr, "user_xattr"},
1167         {Opt_nouser_xattr, "nouser_xattr"},
1168         {Opt_acl, "acl"},
1169         {Opt_noacl, "noacl"},
1170         {Opt_noload, "norecovery"},
1171         {Opt_noload, "noload"},
1172         {Opt_removed, "nobh"},
1173         {Opt_removed, "bh"},
1174         {Opt_commit, "commit=%u"},
1175         {Opt_min_batch_time, "min_batch_time=%u"},
1176         {Opt_max_batch_time, "max_batch_time=%u"},
1177         {Opt_journal_dev, "journal_dev=%u"},
1178         {Opt_journal_checksum, "journal_checksum"},
1179         {Opt_journal_async_commit, "journal_async_commit"},
1180         {Opt_abort, "abort"},
1181         {Opt_data_journal, "data=journal"},
1182         {Opt_data_ordered, "data=ordered"},
1183         {Opt_data_writeback, "data=writeback"},
1184         {Opt_data_err_abort, "data_err=abort"},
1185         {Opt_data_err_ignore, "data_err=ignore"},
1186         {Opt_offusrjquota, "usrjquota="},
1187         {Opt_usrjquota, "usrjquota=%s"},
1188         {Opt_offgrpjquota, "grpjquota="},
1189         {Opt_grpjquota, "grpjquota=%s"},
1190         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1191         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1192         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1193         {Opt_grpquota, "grpquota"},
1194         {Opt_noquota, "noquota"},
1195         {Opt_quota, "quota"},
1196         {Opt_usrquota, "usrquota"},
1197         {Opt_barrier, "barrier=%u"},
1198         {Opt_barrier, "barrier"},
1199         {Opt_nobarrier, "nobarrier"},
1200         {Opt_i_version, "i_version"},
1201         {Opt_stripe, "stripe=%u"},
1202         {Opt_delalloc, "delalloc"},
1203         {Opt_nodelalloc, "nodelalloc"},
1204         {Opt_removed, "mblk_io_submit"},
1205         {Opt_removed, "nomblk_io_submit"},
1206         {Opt_block_validity, "block_validity"},
1207         {Opt_noblock_validity, "noblock_validity"},
1208         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1209         {Opt_journal_ioprio, "journal_ioprio=%u"},
1210         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1211         {Opt_auto_da_alloc, "auto_da_alloc"},
1212         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1213         {Opt_dioread_nolock, "dioread_nolock"},
1214         {Opt_dioread_lock, "dioread_lock"},
1215         {Opt_discard, "discard"},
1216         {Opt_nodiscard, "nodiscard"},
1217         {Opt_init_itable, "init_itable=%u"},
1218         {Opt_init_itable, "init_itable"},
1219         {Opt_noinit_itable, "noinit_itable"},
1220         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1221         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1222         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1223         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1224         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1225         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1226         {Opt_err, NULL},
1227 };
1228
1229 static ext4_fsblk_t get_sb_block(void **data)
1230 {
1231         ext4_fsblk_t    sb_block;
1232         char            *options = (char *) *data;
1233
1234         if (!options || strncmp(options, "sb=", 3) != 0)
1235                 return 1;       /* Default location */
1236
1237         options += 3;
1238         /* TODO: use simple_strtoll with >32bit ext4 */
1239         sb_block = simple_strtoul(options, &options, 0);
1240         if (*options && *options != ',') {
1241                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1242                        (char *) *data);
1243                 return 1;
1244         }
1245         if (*options == ',')
1246                 options++;
1247         *data = (void *) options;
1248
1249         return sb_block;
1250 }
1251
1252 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1253 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1254         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1255
1256 #ifdef CONFIG_QUOTA
1257 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1258 {
1259         struct ext4_sb_info *sbi = EXT4_SB(sb);
1260         char *qname;
1261         int ret = -1;
1262
1263         if (sb_any_quota_loaded(sb) &&
1264                 !sbi->s_qf_names[qtype]) {
1265                 ext4_msg(sb, KERN_ERR,
1266                         "Cannot change journaled "
1267                         "quota options when quota turned on");
1268                 return -1;
1269         }
1270         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1271                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1272                          "when QUOTA feature is enabled");
1273                 return -1;
1274         }
1275         qname = match_strdup(args);
1276         if (!qname) {
1277                 ext4_msg(sb, KERN_ERR,
1278                         "Not enough memory for storing quotafile name");
1279                 return -1;
1280         }
1281         if (sbi->s_qf_names[qtype]) {
1282                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1283                         ret = 1;
1284                 else
1285                         ext4_msg(sb, KERN_ERR,
1286                                  "%s quota file already specified",
1287                                  QTYPE2NAME(qtype));
1288                 goto errout;
1289         }
1290         if (strchr(qname, '/')) {
1291                 ext4_msg(sb, KERN_ERR,
1292                         "quotafile must be on filesystem root");
1293                 goto errout;
1294         }
1295         sbi->s_qf_names[qtype] = qname;
1296         set_opt(sb, QUOTA);
1297         return 1;
1298 errout:
1299         kfree(qname);
1300         return ret;
1301 }
1302
1303 static int clear_qf_name(struct super_block *sb, int qtype)
1304 {
1305
1306         struct ext4_sb_info *sbi = EXT4_SB(sb);
1307
1308         if (sb_any_quota_loaded(sb) &&
1309                 sbi->s_qf_names[qtype]) {
1310                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1311                         " when quota turned on");
1312                 return -1;
1313         }
1314         kfree(sbi->s_qf_names[qtype]);
1315         sbi->s_qf_names[qtype] = NULL;
1316         return 1;
1317 }
1318 #endif
1319
1320 #define MOPT_SET        0x0001
1321 #define MOPT_CLEAR      0x0002
1322 #define MOPT_NOSUPPORT  0x0004
1323 #define MOPT_EXPLICIT   0x0008
1324 #define MOPT_CLEAR_ERR  0x0010
1325 #define MOPT_GTE0       0x0020
1326 #ifdef CONFIG_QUOTA
1327 #define MOPT_Q          0
1328 #define MOPT_QFMT       0x0040
1329 #else
1330 #define MOPT_Q          MOPT_NOSUPPORT
1331 #define MOPT_QFMT       MOPT_NOSUPPORT
1332 #endif
1333 #define MOPT_DATAJ      0x0080
1334 #define MOPT_NO_EXT2    0x0100
1335 #define MOPT_NO_EXT3    0x0200
1336 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1337
1338 static const struct mount_opts {
1339         int     token;
1340         int     mount_opt;
1341         int     flags;
1342 } ext4_mount_opts[] = {
1343         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1344         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1345         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1346         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1347         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1348         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1349         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1350          MOPT_EXT4_ONLY | MOPT_SET},
1351         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1352          MOPT_EXT4_ONLY | MOPT_CLEAR},
1353         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1354         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1355         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1356          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1357         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1358          MOPT_EXT4_ONLY | MOPT_CLEAR},
1359         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1360          MOPT_EXT4_ONLY | MOPT_SET},
1361         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1362                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1363          MOPT_EXT4_ONLY | MOPT_SET},
1364         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1365         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1366         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1367         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1368         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1369          MOPT_NO_EXT2 | MOPT_SET},
1370         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1371          MOPT_NO_EXT2 | MOPT_CLEAR},
1372         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1373         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1374         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1375         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1376         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1377         {Opt_commit, 0, MOPT_GTE0},
1378         {Opt_max_batch_time, 0, MOPT_GTE0},
1379         {Opt_min_batch_time, 0, MOPT_GTE0},
1380         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1381         {Opt_init_itable, 0, MOPT_GTE0},
1382         {Opt_stripe, 0, MOPT_GTE0},
1383         {Opt_resuid, 0, MOPT_GTE0},
1384         {Opt_resgid, 0, MOPT_GTE0},
1385         {Opt_journal_dev, 0, MOPT_GTE0},
1386         {Opt_journal_ioprio, 0, MOPT_GTE0},
1387         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1388         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1389         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1390          MOPT_NO_EXT2 | MOPT_DATAJ},
1391         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1392         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1393 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1394         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1395         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1396 #else
1397         {Opt_acl, 0, MOPT_NOSUPPORT},
1398         {Opt_noacl, 0, MOPT_NOSUPPORT},
1399 #endif
1400         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1401         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1402         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1403         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1404                                                         MOPT_SET | MOPT_Q},
1405         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1406                                                         MOPT_SET | MOPT_Q},
1407         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1408                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1409         {Opt_usrjquota, 0, MOPT_Q},
1410         {Opt_grpjquota, 0, MOPT_Q},
1411         {Opt_offusrjquota, 0, MOPT_Q},
1412         {Opt_offgrpjquota, 0, MOPT_Q},
1413         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1414         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1415         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1416         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1417         {Opt_err, 0, 0}
1418 };
1419
1420 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1421                             substring_t *args, unsigned long *journal_devnum,
1422                             unsigned int *journal_ioprio, int is_remount)
1423 {
1424         struct ext4_sb_info *sbi = EXT4_SB(sb);
1425         const struct mount_opts *m;
1426         kuid_t uid;
1427         kgid_t gid;
1428         int arg = 0;
1429
1430 #ifdef CONFIG_QUOTA
1431         if (token == Opt_usrjquota)
1432                 return set_qf_name(sb, USRQUOTA, &args[0]);
1433         else if (token == Opt_grpjquota)
1434                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1435         else if (token == Opt_offusrjquota)
1436                 return clear_qf_name(sb, USRQUOTA);
1437         else if (token == Opt_offgrpjquota)
1438                 return clear_qf_name(sb, GRPQUOTA);
1439 #endif
1440         switch (token) {
1441         case Opt_noacl:
1442         case Opt_nouser_xattr:
1443                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1444                 break;
1445         case Opt_sb:
1446                 return 1;       /* handled by get_sb_block() */
1447         case Opt_removed:
1448                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1449                 return 1;
1450         case Opt_abort:
1451                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1452                 return 1;
1453         case Opt_i_version:
1454                 sb->s_flags |= MS_I_VERSION;
1455                 return 1;
1456         }
1457
1458         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1459                 if (token == m->token)
1460                         break;
1461
1462         if (m->token == Opt_err) {
1463                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1464                          "or missing value", opt);
1465                 return -1;
1466         }
1467
1468         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1469                 ext4_msg(sb, KERN_ERR,
1470                          "Mount option \"%s\" incompatible with ext2", opt);
1471                 return -1;
1472         }
1473         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1474                 ext4_msg(sb, KERN_ERR,
1475                          "Mount option \"%s\" incompatible with ext3", opt);
1476                 return -1;
1477         }
1478
1479         if (args->from && match_int(args, &arg))
1480                 return -1;
1481         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1482                 return -1;
1483         if (m->flags & MOPT_EXPLICIT)
1484                 set_opt2(sb, EXPLICIT_DELALLOC);
1485         if (m->flags & MOPT_CLEAR_ERR)
1486                 clear_opt(sb, ERRORS_MASK);
1487         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1488                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1489                          "options when quota turned on");
1490                 return -1;
1491         }
1492
1493         if (m->flags & MOPT_NOSUPPORT) {
1494                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1495         } else if (token == Opt_commit) {
1496                 if (arg == 0)
1497                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1498                 sbi->s_commit_interval = HZ * arg;
1499         } else if (token == Opt_max_batch_time) {
1500                 if (arg == 0)
1501                         arg = EXT4_DEF_MAX_BATCH_TIME;
1502                 sbi->s_max_batch_time = arg;
1503         } else if (token == Opt_min_batch_time) {
1504                 sbi->s_min_batch_time = arg;
1505         } else if (token == Opt_inode_readahead_blks) {
1506                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1507                         ext4_msg(sb, KERN_ERR,
1508                                  "EXT4-fs: inode_readahead_blks must be "
1509                                  "0 or a power of 2 smaller than 2^31");
1510                         return -1;
1511                 }
1512                 sbi->s_inode_readahead_blks = arg;
1513         } else if (token == Opt_init_itable) {
1514                 set_opt(sb, INIT_INODE_TABLE);
1515                 if (!args->from)
1516                         arg = EXT4_DEF_LI_WAIT_MULT;
1517                 sbi->s_li_wait_mult = arg;
1518         } else if (token == Opt_max_dir_size_kb) {
1519                 sbi->s_max_dir_size_kb = arg;
1520         } else if (token == Opt_stripe) {
1521                 sbi->s_stripe = arg;
1522         } else if (token == Opt_resuid) {
1523                 uid = make_kuid(current_user_ns(), arg);
1524                 if (!uid_valid(uid)) {
1525                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1526                         return -1;
1527                 }
1528                 sbi->s_resuid = uid;
1529         } else if (token == Opt_resgid) {
1530                 gid = make_kgid(current_user_ns(), arg);
1531                 if (!gid_valid(gid)) {
1532                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1533                         return -1;
1534                 }
1535                 sbi->s_resgid = gid;
1536         } else if (token == Opt_journal_dev) {
1537                 if (is_remount) {
1538                         ext4_msg(sb, KERN_ERR,
1539                                  "Cannot specify journal on remount");
1540                         return -1;
1541                 }
1542                 *journal_devnum = arg;
1543         } else if (token == Opt_journal_ioprio) {
1544                 if (arg > 7) {
1545                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1546                                  " (must be 0-7)");
1547                         return -1;
1548                 }
1549                 *journal_ioprio =
1550                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1551         } else if (m->flags & MOPT_DATAJ) {
1552                 if (is_remount) {
1553                         if (!sbi->s_journal)
1554                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1555                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1556                                 ext4_msg(sb, KERN_ERR,
1557                                          "Cannot change data mode on remount");
1558                                 return -1;
1559                         }
1560                 } else {
1561                         clear_opt(sb, DATA_FLAGS);
1562                         sbi->s_mount_opt |= m->mount_opt;
1563                 }
1564 #ifdef CONFIG_QUOTA
1565         } else if (m->flags & MOPT_QFMT) {
1566                 if (sb_any_quota_loaded(sb) &&
1567                     sbi->s_jquota_fmt != m->mount_opt) {
1568                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1569                                  "quota options when quota turned on");
1570                         return -1;
1571                 }
1572                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1573                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1574                         ext4_msg(sb, KERN_ERR,
1575                                  "Cannot set journaled quota options "
1576                                  "when QUOTA feature is enabled");
1577                         return -1;
1578                 }
1579                 sbi->s_jquota_fmt = m->mount_opt;
1580 #endif
1581         } else {
1582                 if (!args->from)
1583                         arg = 1;
1584                 if (m->flags & MOPT_CLEAR)
1585                         arg = !arg;
1586                 else if (unlikely(!(m->flags & MOPT_SET))) {
1587                         ext4_msg(sb, KERN_WARNING,
1588                                  "buggy handling of option %s", opt);
1589                         WARN_ON(1);
1590                         return -1;
1591                 }
1592                 if (arg != 0)
1593                         sbi->s_mount_opt |= m->mount_opt;
1594                 else
1595                         sbi->s_mount_opt &= ~m->mount_opt;
1596         }
1597         return 1;
1598 }
1599
1600 static int parse_options(char *options, struct super_block *sb,
1601                          unsigned long *journal_devnum,
1602                          unsigned int *journal_ioprio,
1603                          int is_remount)
1604 {
1605         struct ext4_sb_info *sbi = EXT4_SB(sb);
1606         char *p;
1607         substring_t args[MAX_OPT_ARGS];
1608         int token;
1609
1610         if (!options)
1611                 return 1;
1612
1613         while ((p = strsep(&options, ",")) != NULL) {
1614                 if (!*p)
1615                         continue;
1616                 /*
1617                  * Initialize args struct so we know whether arg was
1618                  * found; some options take optional arguments.
1619                  */
1620                 args[0].to = args[0].from = NULL;
1621                 token = match_token(p, tokens, args);
1622                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1623                                      journal_ioprio, is_remount) < 0)
1624                         return 0;
1625         }
1626 #ifdef CONFIG_QUOTA
1627         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1628             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1629                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1630                          "feature is enabled");
1631                 return 0;
1632         }
1633         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1634                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1635                         clear_opt(sb, USRQUOTA);
1636
1637                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1638                         clear_opt(sb, GRPQUOTA);
1639
1640                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1641                         ext4_msg(sb, KERN_ERR, "old and new quota "
1642                                         "format mixing");
1643                         return 0;
1644                 }
1645
1646                 if (!sbi->s_jquota_fmt) {
1647                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1648                                         "not specified");
1649                         return 0;
1650                 }
1651         } else {
1652                 if (sbi->s_jquota_fmt) {
1653                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1654                                         "specified with no journaling "
1655                                         "enabled");
1656                         return 0;
1657                 }
1658         }
1659 #endif
1660         if (test_opt(sb, DIOREAD_NOLOCK)) {
1661                 int blocksize =
1662                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1663
1664                 if (blocksize < PAGE_CACHE_SIZE) {
1665                         ext4_msg(sb, KERN_ERR, "can't mount with "
1666                                  "dioread_nolock if block size != PAGE_SIZE");
1667                         return 0;
1668                 }
1669         }
1670         return 1;
1671 }
1672
1673 static inline void ext4_show_quota_options(struct seq_file *seq,
1674                                            struct super_block *sb)
1675 {
1676 #if defined(CONFIG_QUOTA)
1677         struct ext4_sb_info *sbi = EXT4_SB(sb);
1678
1679         if (sbi->s_jquota_fmt) {
1680                 char *fmtname = "";
1681
1682                 switch (sbi->s_jquota_fmt) {
1683                 case QFMT_VFS_OLD:
1684                         fmtname = "vfsold";
1685                         break;
1686                 case QFMT_VFS_V0:
1687                         fmtname = "vfsv0";
1688                         break;
1689                 case QFMT_VFS_V1:
1690                         fmtname = "vfsv1";
1691                         break;
1692                 }
1693                 seq_printf(seq, ",jqfmt=%s", fmtname);
1694         }
1695
1696         if (sbi->s_qf_names[USRQUOTA])
1697                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1698
1699         if (sbi->s_qf_names[GRPQUOTA])
1700                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1701 #endif
1702 }
1703
1704 static const char *token2str(int token)
1705 {
1706         const struct match_token *t;
1707
1708         for (t = tokens; t->token != Opt_err; t++)
1709                 if (t->token == token && !strchr(t->pattern, '='))
1710                         break;
1711         return t->pattern;
1712 }
1713
1714 /*
1715  * Show an option if
1716  *  - it's set to a non-default value OR
1717  *  - if the per-sb default is different from the global default
1718  */
1719 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1720                               int nodefs)
1721 {
1722         struct ext4_sb_info *sbi = EXT4_SB(sb);
1723         struct ext4_super_block *es = sbi->s_es;
1724         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1725         const struct mount_opts *m;
1726         char sep = nodefs ? '\n' : ',';
1727
1728 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1729 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1730
1731         if (sbi->s_sb_block != 1)
1732                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1733
1734         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1735                 int want_set = m->flags & MOPT_SET;
1736                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1737                     (m->flags & MOPT_CLEAR_ERR))
1738                         continue;
1739                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1740                         continue; /* skip if same as the default */
1741                 if ((want_set &&
1742                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1743                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1744                         continue; /* select Opt_noFoo vs Opt_Foo */
1745                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1746         }
1747
1748         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1749             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1750                 SEQ_OPTS_PRINT("resuid=%u",
1751                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1752         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1753             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1754                 SEQ_OPTS_PRINT("resgid=%u",
1755                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1756         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1757         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1758                 SEQ_OPTS_PUTS("errors=remount-ro");
1759         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1760                 SEQ_OPTS_PUTS("errors=continue");
1761         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1762                 SEQ_OPTS_PUTS("errors=panic");
1763         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1764                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1765         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1766                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1767         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1768                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1769         if (sb->s_flags & MS_I_VERSION)
1770                 SEQ_OPTS_PUTS("i_version");
1771         if (nodefs || sbi->s_stripe)
1772                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1773         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1774                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1775                         SEQ_OPTS_PUTS("data=journal");
1776                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1777                         SEQ_OPTS_PUTS("data=ordered");
1778                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1779                         SEQ_OPTS_PUTS("data=writeback");
1780         }
1781         if (nodefs ||
1782             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1783                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1784                                sbi->s_inode_readahead_blks);
1785
1786         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1787                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1788                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1789         if (nodefs || sbi->s_max_dir_size_kb)
1790                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1791
1792         ext4_show_quota_options(seq, sb);
1793         return 0;
1794 }
1795
1796 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1797 {
1798         return _ext4_show_options(seq, root->d_sb, 0);
1799 }
1800
1801 static int options_seq_show(struct seq_file *seq, void *offset)
1802 {
1803         struct super_block *sb = seq->private;
1804         int rc;
1805
1806         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1807         rc = _ext4_show_options(seq, sb, 1);
1808         seq_puts(seq, "\n");
1809         return rc;
1810 }
1811
1812 static int options_open_fs(struct inode *inode, struct file *file)
1813 {
1814         return single_open(file, options_seq_show, PDE_DATA(inode));
1815 }
1816
1817 static const struct file_operations ext4_seq_options_fops = {
1818         .owner = THIS_MODULE,
1819         .open = options_open_fs,
1820         .read = seq_read,
1821         .llseek = seq_lseek,
1822         .release = single_release,
1823 };
1824
1825 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1826                             int read_only)
1827 {
1828         struct ext4_sb_info *sbi = EXT4_SB(sb);
1829         int res = 0;
1830
1831         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1832                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1833                          "forcing read-only mode");
1834                 res = MS_RDONLY;
1835         }
1836         if (read_only)
1837                 goto done;
1838         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1839                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1840                          "running e2fsck is recommended");
1841         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1842                 ext4_msg(sb, KERN_WARNING,
1843                          "warning: mounting fs with errors, "
1844                          "running e2fsck is recommended");
1845         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1846                  le16_to_cpu(es->s_mnt_count) >=
1847                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1848                 ext4_msg(sb, KERN_WARNING,
1849                          "warning: maximal mount count reached, "
1850                          "running e2fsck is recommended");
1851         else if (le32_to_cpu(es->s_checkinterval) &&
1852                 (le32_to_cpu(es->s_lastcheck) +
1853                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1854                 ext4_msg(sb, KERN_WARNING,
1855                          "warning: checktime reached, "
1856                          "running e2fsck is recommended");
1857         if (!sbi->s_journal)
1858                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1859         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1860                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1861         le16_add_cpu(&es->s_mnt_count, 1);
1862         es->s_mtime = cpu_to_le32(get_seconds());
1863         ext4_update_dynamic_rev(sb);
1864         if (sbi->s_journal)
1865                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1866
1867         ext4_commit_super(sb, 1);
1868 done:
1869         if (test_opt(sb, DEBUG))
1870                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1871                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1872                         sb->s_blocksize,
1873                         sbi->s_groups_count,
1874                         EXT4_BLOCKS_PER_GROUP(sb),
1875                         EXT4_INODES_PER_GROUP(sb),
1876                         sbi->s_mount_opt, sbi->s_mount_opt2);
1877
1878         cleancache_init_fs(sb);
1879         return res;
1880 }
1881
1882 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1883 {
1884         struct ext4_sb_info *sbi = EXT4_SB(sb);
1885         struct flex_groups *new_groups;
1886         int size;
1887
1888         if (!sbi->s_log_groups_per_flex)
1889                 return 0;
1890
1891         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1892         if (size <= sbi->s_flex_groups_allocated)
1893                 return 0;
1894
1895         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1896         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1897         if (!new_groups) {
1898                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1899                          size / (int) sizeof(struct flex_groups));
1900                 return -ENOMEM;
1901         }
1902
1903         if (sbi->s_flex_groups) {
1904                 memcpy(new_groups, sbi->s_flex_groups,
1905                        (sbi->s_flex_groups_allocated *
1906                         sizeof(struct flex_groups)));
1907                 ext4_kvfree(sbi->s_flex_groups);
1908         }
1909         sbi->s_flex_groups = new_groups;
1910         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1911         return 0;
1912 }
1913
1914 static int ext4_fill_flex_info(struct super_block *sb)
1915 {
1916         struct ext4_sb_info *sbi = EXT4_SB(sb);
1917         struct ext4_group_desc *gdp = NULL;
1918         ext4_group_t flex_group;
1919         int i, err;
1920
1921         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1922         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1923                 sbi->s_log_groups_per_flex = 0;
1924                 return 1;
1925         }
1926
1927         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1928         if (err)
1929                 goto failed;
1930
1931         for (i = 0; i < sbi->s_groups_count; i++) {
1932                 gdp = ext4_get_group_desc(sb, i, NULL);
1933
1934                 flex_group = ext4_flex_group(sbi, i);
1935                 atomic_add(ext4_free_inodes_count(sb, gdp),
1936                            &sbi->s_flex_groups[flex_group].free_inodes);
1937                 atomic64_add(ext4_free_group_clusters(sb, gdp),
1938                              &sbi->s_flex_groups[flex_group].free_clusters);
1939                 atomic_add(ext4_used_dirs_count(sb, gdp),
1940                            &sbi->s_flex_groups[flex_group].used_dirs);
1941         }
1942
1943         return 1;
1944 failed:
1945         return 0;
1946 }
1947
1948 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1949                                    struct ext4_group_desc *gdp)
1950 {
1951         int offset;
1952         __u16 crc = 0;
1953         __le32 le_group = cpu_to_le32(block_group);
1954
1955         if ((sbi->s_es->s_feature_ro_compat &
1956              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1957                 /* Use new metadata_csum algorithm */
1958                 __le16 save_csum;
1959                 __u32 csum32;
1960
1961                 save_csum = gdp->bg_checksum;
1962                 gdp->bg_checksum = 0;
1963                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1964                                      sizeof(le_group));
1965                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1966                                      sbi->s_desc_size);
1967                 gdp->bg_checksum = save_csum;
1968
1969                 crc = csum32 & 0xFFFF;
1970                 goto out;
1971         }
1972
1973         /* old crc16 code */
1974         offset = offsetof(struct ext4_group_desc, bg_checksum);
1975
1976         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1977         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1978         crc = crc16(crc, (__u8 *)gdp, offset);
1979         offset += sizeof(gdp->bg_checksum); /* skip checksum */
1980         /* for checksum of struct ext4_group_desc do the rest...*/
1981         if ((sbi->s_es->s_feature_incompat &
1982              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1983             offset < le16_to_cpu(sbi->s_es->s_desc_size))
1984                 crc = crc16(crc, (__u8 *)gdp + offset,
1985                             le16_to_cpu(sbi->s_es->s_desc_size) -
1986                                 offset);
1987
1988 out:
1989         return cpu_to_le16(crc);
1990 }
1991
1992 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
1993                                 struct ext4_group_desc *gdp)
1994 {
1995         if (ext4_has_group_desc_csum(sb) &&
1996             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
1997                                                       block_group, gdp)))
1998                 return 0;
1999
2000         return 1;
2001 }
2002
2003 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2004                               struct ext4_group_desc *gdp)
2005 {
2006         if (!ext4_has_group_desc_csum(sb))
2007                 return;
2008         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2009 }
2010
2011 /* Called at mount-time, super-block is locked */
2012 static int ext4_check_descriptors(struct super_block *sb,
2013                                   ext4_group_t *first_not_zeroed)
2014 {
2015         struct ext4_sb_info *sbi = EXT4_SB(sb);
2016         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2017         ext4_fsblk_t last_block;
2018         ext4_fsblk_t block_bitmap;
2019         ext4_fsblk_t inode_bitmap;
2020         ext4_fsblk_t inode_table;
2021         int flexbg_flag = 0;
2022         ext4_group_t i, grp = sbi->s_groups_count;
2023
2024         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2025                 flexbg_flag = 1;
2026
2027         ext4_debug("Checking group descriptors");
2028
2029         for (i = 0; i < sbi->s_groups_count; i++) {
2030                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2031
2032                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2033                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2034                 else
2035                         last_block = first_block +
2036                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2037
2038                 if ((grp == sbi->s_groups_count) &&
2039                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2040                         grp = i;
2041
2042                 block_bitmap = ext4_block_bitmap(sb, gdp);
2043                 if (block_bitmap < first_block || block_bitmap > last_block) {
2044                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2045                                "Block bitmap for group %u not in group "
2046                                "(block %llu)!", i, block_bitmap);
2047                         return 0;
2048                 }
2049                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2050                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2051                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2052                                "Inode bitmap for group %u not in group "
2053                                "(block %llu)!", i, inode_bitmap);
2054                         return 0;
2055                 }
2056                 inode_table = ext4_inode_table(sb, gdp);
2057                 if (inode_table < first_block ||
2058                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2059                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2060                                "Inode table for group %u not in group "
2061                                "(block %llu)!", i, inode_table);
2062                         return 0;
2063                 }
2064                 ext4_lock_group(sb, i);
2065                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2066                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2067                                  "Checksum for group %u failed (%u!=%u)",
2068                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2069                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2070                         if (!(sb->s_flags & MS_RDONLY)) {
2071                                 ext4_unlock_group(sb, i);
2072                                 return 0;
2073                         }
2074                 }
2075                 ext4_unlock_group(sb, i);
2076                 if (!flexbg_flag)
2077                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2078         }
2079         if (NULL != first_not_zeroed)
2080                 *first_not_zeroed = grp;
2081
2082         ext4_free_blocks_count_set(sbi->s_es,
2083                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2084         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2085         return 1;
2086 }
2087
2088 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2089  * the superblock) which were deleted from all directories, but held open by
2090  * a process at the time of a crash.  We walk the list and try to delete these
2091  * inodes at recovery time (only with a read-write filesystem).
2092  *
2093  * In order to keep the orphan inode chain consistent during traversal (in
2094  * case of crash during recovery), we link each inode into the superblock
2095  * orphan list_head and handle it the same way as an inode deletion during
2096  * normal operation (which journals the operations for us).
2097  *
2098  * We only do an iget() and an iput() on each inode, which is very safe if we
2099  * accidentally point at an in-use or already deleted inode.  The worst that
2100  * can happen in this case is that we get a "bit already cleared" message from
2101  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2102  * e2fsck was run on this filesystem, and it must have already done the orphan
2103  * inode cleanup for us, so we can safely abort without any further action.
2104  */
2105 static void ext4_orphan_cleanup(struct super_block *sb,
2106                                 struct ext4_super_block *es)
2107 {
2108         unsigned int s_flags = sb->s_flags;
2109         int nr_orphans = 0, nr_truncates = 0;
2110 #ifdef CONFIG_QUOTA
2111         int i;
2112 #endif
2113         if (!es->s_last_orphan) {
2114                 jbd_debug(4, "no orphan inodes to clean up\n");
2115                 return;
2116         }
2117
2118         if (bdev_read_only(sb->s_bdev)) {
2119                 ext4_msg(sb, KERN_ERR, "write access "
2120                         "unavailable, skipping orphan cleanup");
2121                 return;
2122         }
2123
2124         /* Check if feature set would not allow a r/w mount */
2125         if (!ext4_feature_set_ok(sb, 0)) {
2126                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2127                          "unknown ROCOMPAT features");
2128                 return;
2129         }
2130
2131         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2132                 /* don't clear list on RO mount w/ errors */
2133                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2134                         jbd_debug(1, "Errors on filesystem, "
2135                                   "clearing orphan list.\n");
2136                         es->s_last_orphan = 0;
2137                 }
2138                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2139                 return;
2140         }
2141
2142         if (s_flags & MS_RDONLY) {
2143                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2144                 sb->s_flags &= ~MS_RDONLY;
2145         }
2146 #ifdef CONFIG_QUOTA
2147         /* Needed for iput() to work correctly and not trash data */
2148         sb->s_flags |= MS_ACTIVE;
2149         /* Turn on quotas so that they are updated correctly */
2150         for (i = 0; i < MAXQUOTAS; i++) {
2151                 if (EXT4_SB(sb)->s_qf_names[i]) {
2152                         int ret = ext4_quota_on_mount(sb, i);
2153                         if (ret < 0)
2154                                 ext4_msg(sb, KERN_ERR,
2155                                         "Cannot turn on journaled "
2156                                         "quota: error %d", ret);
2157                 }
2158         }
2159 #endif
2160
2161         while (es->s_last_orphan) {
2162                 struct inode *inode;
2163
2164                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2165                 if (IS_ERR(inode)) {
2166                         es->s_last_orphan = 0;
2167                         break;
2168                 }
2169
2170                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2171                 dquot_initialize(inode);
2172                 if (inode->i_nlink) {
2173                         if (test_opt(sb, DEBUG))
2174                                 ext4_msg(sb, KERN_DEBUG,
2175                                         "%s: truncating inode %lu to %lld bytes",
2176                                         __func__, inode->i_ino, inode->i_size);
2177                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2178                                   inode->i_ino, inode->i_size);
2179                         mutex_lock(&inode->i_mutex);
2180                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2181                         ext4_truncate(inode);
2182                         mutex_unlock(&inode->i_mutex);
2183                         nr_truncates++;
2184                 } else {
2185                         if (test_opt(sb, DEBUG))
2186                                 ext4_msg(sb, KERN_DEBUG,
2187                                         "%s: deleting unreferenced inode %lu",
2188                                         __func__, inode->i_ino);
2189                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2190                                   inode->i_ino);
2191                         nr_orphans++;
2192                 }
2193                 iput(inode);  /* The delete magic happens here! */
2194         }
2195
2196 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2197
2198         if (nr_orphans)
2199                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2200                        PLURAL(nr_orphans));
2201         if (nr_truncates)
2202                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2203                        PLURAL(nr_truncates));
2204 #ifdef CONFIG_QUOTA
2205         /* Turn quotas off */
2206         for (i = 0; i < MAXQUOTAS; i++) {
2207                 if (sb_dqopt(sb)->files[i])
2208                         dquot_quota_off(sb, i);
2209         }
2210 #endif
2211         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2212 }
2213
2214 /*
2215  * Maximal extent format file size.
2216  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2217  * extent format containers, within a sector_t, and within i_blocks
2218  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2219  * so that won't be a limiting factor.
2220  *
2221  * However there is other limiting factor. We do store extents in the form
2222  * of starting block and length, hence the resulting length of the extent
2223  * covering maximum file size must fit into on-disk format containers as
2224  * well. Given that length is always by 1 unit bigger than max unit (because
2225  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2226  *
2227  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2228  */
2229 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2230 {
2231         loff_t res;
2232         loff_t upper_limit = MAX_LFS_FILESIZE;
2233
2234         /* small i_blocks in vfs inode? */
2235         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2236                 /*
2237                  * CONFIG_LBDAF is not enabled implies the inode
2238                  * i_block represent total blocks in 512 bytes
2239                  * 32 == size of vfs inode i_blocks * 8
2240                  */
2241                 upper_limit = (1LL << 32) - 1;
2242
2243                 /* total blocks in file system block size */
2244                 upper_limit >>= (blkbits - 9);
2245                 upper_limit <<= blkbits;
2246         }
2247
2248         /*
2249          * 32-bit extent-start container, ee_block. We lower the maxbytes
2250          * by one fs block, so ee_len can cover the extent of maximum file
2251          * size
2252          */
2253         res = (1LL << 32) - 1;
2254         res <<= blkbits;
2255
2256         /* Sanity check against vm- & vfs- imposed limits */
2257         if (res > upper_limit)
2258                 res = upper_limit;
2259
2260         return res;
2261 }
2262
2263 /*
2264  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2265  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2266  * We need to be 1 filesystem block less than the 2^48 sector limit.
2267  */
2268 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2269 {
2270         loff_t res = EXT4_NDIR_BLOCKS;
2271         int meta_blocks;
2272         loff_t upper_limit;
2273         /* This is calculated to be the largest file size for a dense, block
2274          * mapped file such that the file's total number of 512-byte sectors,
2275          * including data and all indirect blocks, does not exceed (2^48 - 1).
2276          *
2277          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2278          * number of 512-byte sectors of the file.
2279          */
2280
2281         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2282                 /*
2283                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2284                  * the inode i_block field represents total file blocks in
2285                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2286                  */
2287                 upper_limit = (1LL << 32) - 1;
2288
2289                 /* total blocks in file system block size */
2290                 upper_limit >>= (bits - 9);
2291
2292         } else {
2293                 /*
2294                  * We use 48 bit ext4_inode i_blocks
2295                  * With EXT4_HUGE_FILE_FL set the i_blocks
2296                  * represent total number of blocks in
2297                  * file system block size
2298                  */
2299                 upper_limit = (1LL << 48) - 1;
2300
2301         }
2302
2303         /* indirect blocks */
2304         meta_blocks = 1;
2305         /* double indirect blocks */
2306         meta_blocks += 1 + (1LL << (bits-2));
2307         /* tripple indirect blocks */
2308         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2309
2310         upper_limit -= meta_blocks;
2311         upper_limit <<= bits;
2312
2313         res += 1LL << (bits-2);
2314         res += 1LL << (2*(bits-2));
2315         res += 1LL << (3*(bits-2));
2316         res <<= bits;
2317         if (res > upper_limit)
2318                 res = upper_limit;
2319
2320         if (res > MAX_LFS_FILESIZE)
2321                 res = MAX_LFS_FILESIZE;
2322
2323         return res;
2324 }
2325
2326 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2327                                    ext4_fsblk_t logical_sb_block, int nr)
2328 {
2329         struct ext4_sb_info *sbi = EXT4_SB(sb);
2330         ext4_group_t bg, first_meta_bg;
2331         int has_super = 0;
2332
2333         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2334
2335         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2336             nr < first_meta_bg)
2337                 return logical_sb_block + nr + 1;
2338         bg = sbi->s_desc_per_block * nr;
2339         if (ext4_bg_has_super(sb, bg))
2340                 has_super = 1;
2341
2342         return (has_super + ext4_group_first_block_no(sb, bg));
2343 }
2344
2345 /**
2346  * ext4_get_stripe_size: Get the stripe size.
2347  * @sbi: In memory super block info
2348  *
2349  * If we have specified it via mount option, then
2350  * use the mount option value. If the value specified at mount time is
2351  * greater than the blocks per group use the super block value.
2352  * If the super block value is greater than blocks per group return 0.
2353  * Allocator needs it be less than blocks per group.
2354  *
2355  */
2356 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2357 {
2358         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2359         unsigned long stripe_width =
2360                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2361         int ret;
2362
2363         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2364                 ret = sbi->s_stripe;
2365         else if (stripe_width <= sbi->s_blocks_per_group)
2366                 ret = stripe_width;
2367         else if (stride <= sbi->s_blocks_per_group)
2368                 ret = stride;
2369         else
2370                 ret = 0;
2371
2372         /*
2373          * If the stripe width is 1, this makes no sense and
2374          * we set it to 0 to turn off stripe handling code.
2375          */
2376         if (ret <= 1)
2377                 ret = 0;
2378
2379         return ret;
2380 }
2381
2382 /* sysfs supprt */
2383
2384 struct ext4_attr {
2385         struct attribute attr;
2386         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2387         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2388                          const char *, size_t);
2389         union {
2390                 int offset;
2391                 int deprecated_val;
2392         } u;
2393 };
2394
2395 static int parse_strtoull(const char *buf,
2396                 unsigned long long max, unsigned long long *value)
2397 {
2398         int ret;
2399
2400         ret = kstrtoull(skip_spaces(buf), 0, value);
2401         if (!ret && *value > max)
2402                 ret = -EINVAL;
2403         return ret;
2404 }
2405
2406 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2407                                               struct ext4_sb_info *sbi,
2408                                               char *buf)
2409 {
2410         return snprintf(buf, PAGE_SIZE, "%llu\n",
2411                 (s64) EXT4_C2B(sbi,
2412                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2413 }
2414
2415 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2416                                          struct ext4_sb_info *sbi, char *buf)
2417 {
2418         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2419
2420         if (!sb->s_bdev->bd_part)
2421                 return snprintf(buf, PAGE_SIZE, "0\n");
2422         return snprintf(buf, PAGE_SIZE, "%lu\n",
2423                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2424                          sbi->s_sectors_written_start) >> 1);
2425 }
2426
2427 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2428                                           struct ext4_sb_info *sbi, char *buf)
2429 {
2430         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2431
2432         if (!sb->s_bdev->bd_part)
2433                 return snprintf(buf, PAGE_SIZE, "0\n");
2434         return snprintf(buf, PAGE_SIZE, "%llu\n",
2435                         (unsigned long long)(sbi->s_kbytes_written +
2436                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2437                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2438 }
2439
2440 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2441                                           struct ext4_sb_info *sbi,
2442                                           const char *buf, size_t count)
2443 {
2444         unsigned long t;
2445         int ret;
2446
2447         ret = kstrtoul(skip_spaces(buf), 0, &t);
2448         if (ret)
2449                 return ret;
2450
2451         if (t && (!is_power_of_2(t) || t > 0x40000000))
2452                 return -EINVAL;
2453
2454         sbi->s_inode_readahead_blks = t;
2455         return count;
2456 }
2457
2458 static ssize_t sbi_ui_show(struct ext4_attr *a,
2459                            struct ext4_sb_info *sbi, char *buf)
2460 {
2461         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2462
2463         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2464 }
2465
2466 static ssize_t sbi_ui_store(struct ext4_attr *a,
2467                             struct ext4_sb_info *sbi,
2468                             const char *buf, size_t count)
2469 {
2470         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2471         unsigned long t;
2472         int ret;
2473
2474         ret = kstrtoul(skip_spaces(buf), 0, &t);
2475         if (ret)
2476                 return ret;
2477         *ui = t;
2478         return count;
2479 }
2480
2481 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2482                                   struct ext4_sb_info *sbi, char *buf)
2483 {
2484         return snprintf(buf, PAGE_SIZE, "%llu\n",
2485                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2486 }
2487
2488 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2489                                    struct ext4_sb_info *sbi,
2490                                    const char *buf, size_t count)
2491 {
2492         unsigned long long val;
2493         int ret;
2494
2495         if (parse_strtoull(buf, -1ULL, &val))
2496                 return -EINVAL;
2497         ret = ext4_reserve_clusters(sbi, val);
2498
2499         return ret ? ret : count;
2500 }
2501
2502 static ssize_t trigger_test_error(struct ext4_attr *a,
2503                                   struct ext4_sb_info *sbi,
2504                                   const char *buf, size_t count)
2505 {
2506         int len = count;
2507
2508         if (!capable(CAP_SYS_ADMIN))
2509                 return -EPERM;
2510
2511         if (len && buf[len-1] == '\n')
2512                 len--;
2513
2514         if (len)
2515                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2516         return count;
2517 }
2518
2519 static ssize_t sbi_deprecated_show(struct ext4_attr *a,
2520                                    struct ext4_sb_info *sbi, char *buf)
2521 {
2522         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
2523 }
2524
2525 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2526 static struct ext4_attr ext4_attr_##_name = {                   \
2527         .attr = {.name = __stringify(_name), .mode = _mode },   \
2528         .show   = _show,                                        \
2529         .store  = _store,                                       \
2530         .u = {                                                  \
2531                 .offset = offsetof(struct ext4_sb_info, _elname),\
2532         },                                                      \
2533 }
2534 #define EXT4_ATTR(name, mode, show, store) \
2535 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2536
2537 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2538 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2539 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2540 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2541         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2542 #define ATTR_LIST(name) &ext4_attr_##name.attr
2543 #define EXT4_DEPRECATED_ATTR(_name, _val)       \
2544 static struct ext4_attr ext4_attr_##_name = {                   \
2545         .attr = {.name = __stringify(_name), .mode = 0444 },    \
2546         .show   = sbi_deprecated_show,                          \
2547         .u = {                                                  \
2548                 .deprecated_val = _val,                         \
2549         },                                                      \
2550 }
2551
2552 EXT4_RO_ATTR(delayed_allocation_blocks);
2553 EXT4_RO_ATTR(session_write_kbytes);
2554 EXT4_RO_ATTR(lifetime_write_kbytes);
2555 EXT4_RW_ATTR(reserved_clusters);
2556 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2557                  inode_readahead_blks_store, s_inode_readahead_blks);
2558 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2559 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2560 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2561 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2562 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2563 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2564 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2565 EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
2566 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2567 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2568
2569 static struct attribute *ext4_attrs[] = {
2570         ATTR_LIST(delayed_allocation_blocks),
2571         ATTR_LIST(session_write_kbytes),
2572         ATTR_LIST(lifetime_write_kbytes),
2573         ATTR_LIST(reserved_clusters),
2574         ATTR_LIST(inode_readahead_blks),
2575         ATTR_LIST(inode_goal),
2576         ATTR_LIST(mb_stats),
2577         ATTR_LIST(mb_max_to_scan),
2578         ATTR_LIST(mb_min_to_scan),
2579         ATTR_LIST(mb_order2_req),
2580         ATTR_LIST(mb_stream_req),
2581         ATTR_LIST(mb_group_prealloc),
2582         ATTR_LIST(max_writeback_mb_bump),
2583         ATTR_LIST(extent_max_zeroout_kb),
2584         ATTR_LIST(trigger_fs_error),
2585         NULL,
2586 };
2587
2588 /* Features this copy of ext4 supports */
2589 EXT4_INFO_ATTR(lazy_itable_init);
2590 EXT4_INFO_ATTR(batched_discard);
2591 EXT4_INFO_ATTR(meta_bg_resize);
2592
2593 static struct attribute *ext4_feat_attrs[] = {
2594         ATTR_LIST(lazy_itable_init),
2595         ATTR_LIST(batched_discard),
2596         ATTR_LIST(meta_bg_resize),
2597         NULL,
2598 };
2599
2600 static ssize_t ext4_attr_show(struct kobject *kobj,
2601                               struct attribute *attr, char *buf)
2602 {
2603         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2604                                                 s_kobj);
2605         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2606
2607         return a->show ? a->show(a, sbi, buf) : 0;
2608 }
2609
2610 static ssize_t ext4_attr_store(struct kobject *kobj,
2611                                struct attribute *attr,
2612                                const char *buf, size_t len)
2613 {
2614         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2615                                                 s_kobj);
2616         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2617
2618         return a->store ? a->store(a, sbi, buf, len) : 0;
2619 }
2620
2621 static void ext4_sb_release(struct kobject *kobj)
2622 {
2623         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2624                                                 s_kobj);
2625         complete(&sbi->s_kobj_unregister);
2626 }
2627
2628 static const struct sysfs_ops ext4_attr_ops = {
2629         .show   = ext4_attr_show,
2630         .store  = ext4_attr_store,
2631 };
2632
2633 static struct kobj_type ext4_ktype = {
2634         .default_attrs  = ext4_attrs,
2635         .sysfs_ops      = &ext4_attr_ops,
2636         .release        = ext4_sb_release,
2637 };
2638
2639 static void ext4_feat_release(struct kobject *kobj)
2640 {
2641         complete(&ext4_feat->f_kobj_unregister);
2642 }
2643
2644 static struct kobj_type ext4_feat_ktype = {
2645         .default_attrs  = ext4_feat_attrs,
2646         .sysfs_ops      = &ext4_attr_ops,
2647         .release        = ext4_feat_release,
2648 };
2649
2650 /*
2651  * Check whether this filesystem can be mounted based on
2652  * the features present and the RDONLY/RDWR mount requested.
2653  * Returns 1 if this filesystem can be mounted as requested,
2654  * 0 if it cannot be.
2655  */
2656 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2657 {
2658         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2659                 ext4_msg(sb, KERN_ERR,
2660                         "Couldn't mount because of "
2661                         "unsupported optional features (%x)",
2662                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2663                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2664                 return 0;
2665         }
2666
2667         if (readonly)
2668                 return 1;
2669
2670         /* Check that feature set is OK for a read-write mount */
2671         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2672                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2673                          "unsupported optional features (%x)",
2674                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2675                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2676                 return 0;
2677         }
2678         /*
2679          * Large file size enabled file system can only be mounted
2680          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2681          */
2682         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2683                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2684                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2685                                  "cannot be mounted RDWR without "
2686                                  "CONFIG_LBDAF");
2687                         return 0;
2688                 }
2689         }
2690         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2691             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2692                 ext4_msg(sb, KERN_ERR,
2693                          "Can't support bigalloc feature without "
2694                          "extents feature\n");
2695                 return 0;
2696         }
2697
2698 #ifndef CONFIG_QUOTA
2699         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2700             !readonly) {
2701                 ext4_msg(sb, KERN_ERR,
2702                          "Filesystem with quota feature cannot be mounted RDWR "
2703                          "without CONFIG_QUOTA");
2704                 return 0;
2705         }
2706 #endif  /* CONFIG_QUOTA */
2707         return 1;
2708 }
2709
2710 /*
2711  * This function is called once a day if we have errors logged
2712  * on the file system
2713  */
2714 static void print_daily_error_info(unsigned long arg)
2715 {
2716         struct super_block *sb = (struct super_block *) arg;
2717         struct ext4_sb_info *sbi;
2718         struct ext4_super_block *es;
2719
2720         sbi = EXT4_SB(sb);
2721         es = sbi->s_es;
2722
2723         if (es->s_error_count)
2724                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2725                          le32_to_cpu(es->s_error_count));
2726         if (es->s_first_error_time) {
2727                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2728                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2729                        (int) sizeof(es->s_first_error_func),
2730                        es->s_first_error_func,
2731                        le32_to_cpu(es->s_first_error_line));
2732                 if (es->s_first_error_ino)
2733                         printk(": inode %u",
2734                                le32_to_cpu(es->s_first_error_ino));
2735                 if (es->s_first_error_block)
2736                         printk(": block %llu", (unsigned long long)
2737                                le64_to_cpu(es->s_first_error_block));
2738                 printk("\n");
2739         }
2740         if (es->s_last_error_time) {
2741                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2742                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2743                        (int) sizeof(es->s_last_error_func),
2744                        es->s_last_error_func,
2745                        le32_to_cpu(es->s_last_error_line));
2746                 if (es->s_last_error_ino)
2747                         printk(": inode %u",
2748                                le32_to_cpu(es->s_last_error_ino));
2749                 if (es->s_last_error_block)
2750                         printk(": block %llu", (unsigned long long)
2751                                le64_to_cpu(es->s_last_error_block));
2752                 printk("\n");
2753         }
2754         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2755 }
2756
2757 /* Find next suitable group and run ext4_init_inode_table */
2758 static int ext4_run_li_request(struct ext4_li_request *elr)
2759 {
2760         struct ext4_group_desc *gdp = NULL;
2761         ext4_group_t group, ngroups;
2762         struct super_block *sb;
2763         unsigned long timeout = 0;
2764         int ret = 0;
2765
2766         sb = elr->lr_super;
2767         ngroups = EXT4_SB(sb)->s_groups_count;
2768
2769         sb_start_write(sb);
2770         for (group = elr->lr_next_group; group < ngroups; group++) {
2771                 gdp = ext4_get_group_desc(sb, group, NULL);
2772                 if (!gdp) {
2773                         ret = 1;
2774                         break;
2775                 }
2776
2777                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2778                         break;
2779         }
2780
2781         if (group >= ngroups)
2782                 ret = 1;
2783
2784         if (!ret) {
2785                 timeout = jiffies;
2786                 ret = ext4_init_inode_table(sb, group,
2787                                             elr->lr_timeout ? 0 : 1);
2788                 if (elr->lr_timeout == 0) {
2789                         timeout = (jiffies - timeout) *
2790                                   elr->lr_sbi->s_li_wait_mult;
2791                         elr->lr_timeout = timeout;
2792                 }
2793                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2794                 elr->lr_next_group = group + 1;
2795         }
2796         sb_end_write(sb);
2797
2798         return ret;
2799 }
2800
2801 /*
2802  * Remove lr_request from the list_request and free the
2803  * request structure. Should be called with li_list_mtx held
2804  */
2805 static void ext4_remove_li_request(struct ext4_li_request *elr)
2806 {
2807         struct ext4_sb_info *sbi;
2808
2809         if (!elr)
2810                 return;
2811
2812         sbi = elr->lr_sbi;
2813
2814         list_del(&elr->lr_request);
2815         sbi->s_li_request = NULL;
2816         kfree(elr);
2817 }
2818
2819 static void ext4_unregister_li_request(struct super_block *sb)
2820 {
2821         mutex_lock(&ext4_li_mtx);
2822         if (!ext4_li_info) {
2823                 mutex_unlock(&ext4_li_mtx);
2824                 return;
2825         }
2826
2827         mutex_lock(&ext4_li_info->li_list_mtx);
2828         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2829         mutex_unlock(&ext4_li_info->li_list_mtx);
2830         mutex_unlock(&ext4_li_mtx);
2831 }
2832
2833 static struct task_struct *ext4_lazyinit_task;
2834
2835 /*
2836  * This is the function where ext4lazyinit thread lives. It walks
2837  * through the request list searching for next scheduled filesystem.
2838  * When such a fs is found, run the lazy initialization request
2839  * (ext4_rn_li_request) and keep track of the time spend in this
2840  * function. Based on that time we compute next schedule time of
2841  * the request. When walking through the list is complete, compute
2842  * next waking time and put itself into sleep.
2843  */
2844 static int ext4_lazyinit_thread(void *arg)
2845 {
2846         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2847         struct list_head *pos, *n;
2848         struct ext4_li_request *elr;
2849         unsigned long next_wakeup, cur;
2850
2851         BUG_ON(NULL == eli);
2852
2853 cont_thread:
2854         while (true) {
2855                 next_wakeup = MAX_JIFFY_OFFSET;
2856
2857                 mutex_lock(&eli->li_list_mtx);
2858                 if (list_empty(&eli->li_request_list)) {
2859                         mutex_unlock(&eli->li_list_mtx);
2860                         goto exit_thread;
2861                 }
2862
2863                 list_for_each_safe(pos, n, &eli->li_request_list) {
2864                         elr = list_entry(pos, struct ext4_li_request,
2865                                          lr_request);
2866
2867                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2868                                 if (ext4_run_li_request(elr) != 0) {
2869                                         /* error, remove the lazy_init job */
2870                                         ext4_remove_li_request(elr);
2871                                         continue;
2872                                 }
2873                         }
2874
2875                         if (time_before(elr->lr_next_sched, next_wakeup))
2876                                 next_wakeup = elr->lr_next_sched;
2877                 }
2878                 mutex_unlock(&eli->li_list_mtx);
2879
2880                 try_to_freeze();
2881
2882                 cur = jiffies;
2883                 if ((time_after_eq(cur, next_wakeup)) ||
2884                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2885                         cond_resched();
2886                         continue;
2887                 }
2888
2889                 schedule_timeout_interruptible(next_wakeup - cur);
2890
2891                 if (kthread_should_stop()) {
2892                         ext4_clear_request_list();
2893                         goto exit_thread;
2894                 }
2895         }
2896
2897 exit_thread:
2898         /*
2899          * It looks like the request list is empty, but we need
2900          * to check it under the li_list_mtx lock, to prevent any
2901          * additions into it, and of course we should lock ext4_li_mtx
2902          * to atomically free the list and ext4_li_info, because at
2903          * this point another ext4 filesystem could be registering
2904          * new one.
2905          */
2906         mutex_lock(&ext4_li_mtx);
2907         mutex_lock(&eli->li_list_mtx);
2908         if (!list_empty(&eli->li_request_list)) {
2909                 mutex_unlock(&eli->li_list_mtx);
2910                 mutex_unlock(&ext4_li_mtx);
2911                 goto cont_thread;
2912         }
2913         mutex_unlock(&eli->li_list_mtx);
2914         kfree(ext4_li_info);
2915         ext4_li_info = NULL;
2916         mutex_unlock(&ext4_li_mtx);
2917
2918         return 0;
2919 }
2920
2921 static void ext4_clear_request_list(void)
2922 {
2923         struct list_head *pos, *n;
2924         struct ext4_li_request *elr;
2925
2926         mutex_lock(&ext4_li_info->li_list_mtx);
2927         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2928                 elr = list_entry(pos, struct ext4_li_request,
2929                                  lr_request);
2930                 ext4_remove_li_request(elr);
2931         }
2932         mutex_unlock(&ext4_li_info->li_list_mtx);
2933 }
2934
2935 static int ext4_run_lazyinit_thread(void)
2936 {
2937         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2938                                          ext4_li_info, "ext4lazyinit");
2939         if (IS_ERR(ext4_lazyinit_task)) {
2940                 int err = PTR_ERR(ext4_lazyinit_task);
2941                 ext4_clear_request_list();
2942                 kfree(ext4_li_info);
2943                 ext4_li_info = NULL;
2944                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2945                                  "initialization thread\n",
2946                                  err);
2947                 return err;
2948         }
2949         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2950         return 0;
2951 }
2952
2953 /*
2954  * Check whether it make sense to run itable init. thread or not.
2955  * If there is at least one uninitialized inode table, return
2956  * corresponding group number, else the loop goes through all
2957  * groups and return total number of groups.
2958  */
2959 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2960 {
2961         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2962         struct ext4_group_desc *gdp = NULL;
2963
2964         for (group = 0; group < ngroups; group++) {
2965                 gdp = ext4_get_group_desc(sb, group, NULL);
2966                 if (!gdp)
2967                         continue;
2968
2969                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2970                         break;
2971         }
2972
2973         return group;
2974 }
2975
2976 static int ext4_li_info_new(void)
2977 {
2978         struct ext4_lazy_init *eli = NULL;
2979
2980         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2981         if (!eli)
2982                 return -ENOMEM;
2983
2984         INIT_LIST_HEAD(&eli->li_request_list);
2985         mutex_init(&eli->li_list_mtx);
2986
2987         eli->li_state |= EXT4_LAZYINIT_QUIT;
2988
2989         ext4_li_info = eli;
2990
2991         return 0;
2992 }
2993
2994 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2995                                             ext4_group_t start)
2996 {
2997         struct ext4_sb_info *sbi = EXT4_SB(sb);
2998         struct ext4_li_request *elr;
2999         unsigned long rnd;
3000
3001         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3002         if (!elr)
3003                 return NULL;
3004
3005         elr->lr_super = sb;
3006         elr->lr_sbi = sbi;
3007         elr->lr_next_group = start;
3008
3009         /*
3010          * Randomize first schedule time of the request to
3011          * spread the inode table initialization requests
3012          * better.
3013          */
3014         get_random_bytes(&rnd, sizeof(rnd));
3015         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3016                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3017
3018         return elr;
3019 }
3020
3021 int ext4_register_li_request(struct super_block *sb,
3022                              ext4_group_t first_not_zeroed)
3023 {
3024         struct ext4_sb_info *sbi = EXT4_SB(sb);
3025         struct ext4_li_request *elr = NULL;
3026         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3027         int ret = 0;
3028
3029         mutex_lock(&ext4_li_mtx);
3030         if (sbi->s_li_request != NULL) {
3031                 /*
3032                  * Reset timeout so it can be computed again, because
3033                  * s_li_wait_mult might have changed.
3034                  */
3035                 sbi->s_li_request->lr_timeout = 0;
3036                 goto out;
3037         }
3038
3039         if (first_not_zeroed == ngroups ||
3040             (sb->s_flags & MS_RDONLY) ||
3041             !test_opt(sb, INIT_INODE_TABLE))
3042                 goto out;
3043
3044         elr = ext4_li_request_new(sb, first_not_zeroed);
3045         if (!elr) {
3046                 ret = -ENOMEM;
3047                 goto out;
3048         }
3049
3050         if (NULL == ext4_li_info) {
3051                 ret = ext4_li_info_new();
3052                 if (ret)
3053                         goto out;
3054         }
3055
3056         mutex_lock(&ext4_li_info->li_list_mtx);
3057         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3058         mutex_unlock(&ext4_li_info->li_list_mtx);
3059
3060         sbi->s_li_request = elr;
3061         /*
3062          * set elr to NULL here since it has been inserted to
3063          * the request_list and the removal and free of it is
3064          * handled by ext4_clear_request_list from now on.
3065          */
3066         elr = NULL;
3067
3068         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3069                 ret = ext4_run_lazyinit_thread();
3070                 if (ret)
3071                         goto out;
3072         }
3073 out:
3074         mutex_unlock(&ext4_li_mtx);
3075         if (ret)
3076                 kfree(elr);
3077         return ret;
3078 }
3079
3080 /*
3081  * We do not need to lock anything since this is called on
3082  * module unload.
3083  */
3084 static void ext4_destroy_lazyinit_thread(void)
3085 {
3086         /*
3087          * If thread exited earlier
3088          * there's nothing to be done.
3089          */
3090         if (!ext4_li_info || !ext4_lazyinit_task)
3091                 return;
3092
3093         kthread_stop(ext4_lazyinit_task);
3094 }
3095
3096 static int set_journal_csum_feature_set(struct super_block *sb)
3097 {
3098         int ret = 1;
3099         int compat, incompat;
3100         struct ext4_sb_info *sbi = EXT4_SB(sb);
3101
3102         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3103                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3104                 /* journal checksum v2 */
3105                 compat = 0;
3106                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3107         } else {
3108                 /* journal checksum v1 */
3109                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3110                 incompat = 0;
3111         }
3112
3113         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3114                 ret = jbd2_journal_set_features(sbi->s_journal,
3115                                 compat, 0,
3116                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3117                                 incompat);
3118         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3119                 ret = jbd2_journal_set_features(sbi->s_journal,
3120                                 compat, 0,
3121                                 incompat);
3122                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3123                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3124         } else {
3125                 jbd2_journal_clear_features(sbi->s_journal,
3126                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3127                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3128                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3129         }
3130
3131         return ret;
3132 }
3133
3134 /*
3135  * Note: calculating the overhead so we can be compatible with
3136  * historical BSD practice is quite difficult in the face of
3137  * clusters/bigalloc.  This is because multiple metadata blocks from
3138  * different block group can end up in the same allocation cluster.
3139  * Calculating the exact overhead in the face of clustered allocation
3140  * requires either O(all block bitmaps) in memory or O(number of block
3141  * groups**2) in time.  We will still calculate the superblock for
3142  * older file systems --- and if we come across with a bigalloc file
3143  * system with zero in s_overhead_clusters the estimate will be close to
3144  * correct especially for very large cluster sizes --- but for newer
3145  * file systems, it's better to calculate this figure once at mkfs
3146  * time, and store it in the superblock.  If the superblock value is
3147  * present (even for non-bigalloc file systems), we will use it.
3148  */
3149 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3150                           char *buf)
3151 {
3152         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3153         struct ext4_group_desc  *gdp;
3154         ext4_fsblk_t            first_block, last_block, b;
3155         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3156         int                     s, j, count = 0;
3157
3158         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3159                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3160                         sbi->s_itb_per_group + 2);
3161
3162         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3163                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3164         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3165         for (i = 0; i < ngroups; i++) {
3166                 gdp = ext4_get_group_desc(sb, i, NULL);
3167                 b = ext4_block_bitmap(sb, gdp);
3168                 if (b >= first_block && b <= last_block) {
3169                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3170                         count++;
3171                 }
3172                 b = ext4_inode_bitmap(sb, gdp);
3173                 if (b >= first_block && b <= last_block) {
3174                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3175                         count++;
3176                 }
3177                 b = ext4_inode_table(sb, gdp);
3178                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3179                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3180                                 int c = EXT4_B2C(sbi, b - first_block);
3181                                 ext4_set_bit(c, buf);
3182                                 count++;
3183                         }
3184                 if (i != grp)
3185                         continue;
3186                 s = 0;
3187                 if (ext4_bg_has_super(sb, grp)) {
3188                         ext4_set_bit(s++, buf);
3189                         count++;
3190                 }
3191                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3192                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3193                         count++;
3194                 }
3195         }
3196         if (!count)
3197                 return 0;
3198         return EXT4_CLUSTERS_PER_GROUP(sb) -
3199                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3200 }
3201
3202 /*
3203  * Compute the overhead and stash it in sbi->s_overhead
3204  */
3205 int ext4_calculate_overhead(struct super_block *sb)
3206 {
3207         struct ext4_sb_info *sbi = EXT4_SB(sb);
3208         struct ext4_super_block *es = sbi->s_es;
3209         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3210         ext4_fsblk_t overhead = 0;
3211         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3212
3213         if (!buf)
3214                 return -ENOMEM;
3215
3216         /*
3217          * Compute the overhead (FS structures).  This is constant
3218          * for a given filesystem unless the number of block groups
3219          * changes so we cache the previous value until it does.
3220          */
3221
3222         /*
3223          * All of the blocks before first_data_block are overhead
3224          */
3225         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3226
3227         /*
3228          * Add the overhead found in each block group
3229          */
3230         for (i = 0; i < ngroups; i++) {
3231                 int blks;
3232
3233                 blks = count_overhead(sb, i, buf);
3234                 overhead += blks;
3235                 if (blks)
3236                         memset(buf, 0, PAGE_SIZE);
3237                 cond_resched();
3238         }
3239         /* Add the journal blocks as well */
3240         if (sbi->s_journal)
3241                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3242
3243         sbi->s_overhead = overhead;
3244         smp_wmb();
3245         free_page((unsigned long) buf);
3246         return 0;
3247 }
3248
3249
3250 static ext4_fsblk_t ext4_calculate_resv_clusters(struct ext4_sb_info *sbi)
3251 {
3252         ext4_fsblk_t resv_clusters;
3253
3254         /*
3255          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3256          * This should cover the situations where we can not afford to run
3257          * out of space like for example punch hole, or converting
3258          * uninitialized extents in delalloc path. In most cases such
3259          * allocation would require 1, or 2 blocks, higher numbers are
3260          * very rare.
3261          */
3262         resv_clusters = ext4_blocks_count(sbi->s_es) >> sbi->s_cluster_bits;
3263
3264         do_div(resv_clusters, 50);
3265         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3266
3267         return resv_clusters;
3268 }
3269
3270
3271 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3272 {
3273         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3274                                 sbi->s_cluster_bits;
3275
3276         if (count >= clusters)
3277                 return -EINVAL;
3278
3279         atomic64_set(&sbi->s_resv_clusters, count);
3280         return 0;
3281 }
3282
3283 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3284 {
3285         char *orig_data = kstrdup(data, GFP_KERNEL);
3286         struct buffer_head *bh;
3287         struct ext4_super_block *es = NULL;
3288         struct ext4_sb_info *sbi;
3289         ext4_fsblk_t block;
3290         ext4_fsblk_t sb_block = get_sb_block(&data);
3291         ext4_fsblk_t logical_sb_block;
3292         unsigned long offset = 0;
3293         unsigned long journal_devnum = 0;
3294         unsigned long def_mount_opts;
3295         struct inode *root;
3296         char *cp;
3297         const char *descr;
3298         int ret = -ENOMEM;
3299         int blocksize, clustersize;
3300         unsigned int db_count;
3301         unsigned int i;
3302         int needs_recovery, has_huge_files, has_bigalloc;
3303         __u64 blocks_count;
3304         int err = 0;
3305         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3306         ext4_group_t first_not_zeroed;
3307
3308         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3309         if (!sbi)
3310                 goto out_free_orig;
3311
3312         sbi->s_blockgroup_lock =
3313                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3314         if (!sbi->s_blockgroup_lock) {
3315                 kfree(sbi);
3316                 goto out_free_orig;
3317         }
3318         sb->s_fs_info = sbi;
3319         sbi->s_sb = sb;
3320         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3321         sbi->s_sb_block = sb_block;
3322         if (sb->s_bdev->bd_part)
3323                 sbi->s_sectors_written_start =
3324                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3325
3326         /* Cleanup superblock name */
3327         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3328                 *cp = '!';
3329
3330         /* -EINVAL is default */
3331         ret = -EINVAL;
3332         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3333         if (!blocksize) {
3334                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3335                 goto out_fail;
3336         }
3337
3338         /*
3339          * The ext4 superblock will not be buffer aligned for other than 1kB
3340          * block sizes.  We need to calculate the offset from buffer start.
3341          */
3342         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3343                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3344                 offset = do_div(logical_sb_block, blocksize);
3345         } else {
3346                 logical_sb_block = sb_block;
3347         }
3348
3349         if (!(bh = sb_bread(sb, logical_sb_block))) {
3350                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3351                 goto out_fail;
3352         }
3353         /*
3354          * Note: s_es must be initialized as soon as possible because
3355          *       some ext4 macro-instructions depend on its value
3356          */
3357         es = (struct ext4_super_block *) (bh->b_data + offset);
3358         sbi->s_es = es;
3359         sb->s_magic = le16_to_cpu(es->s_magic);
3360         if (sb->s_magic != EXT4_SUPER_MAGIC)
3361                 goto cantfind_ext4;
3362         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3363
3364         /* Warn if metadata_csum and gdt_csum are both set. */
3365         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3366                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3367             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3368                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3369                              "redundant flags; please run fsck.");
3370
3371         /* Check for a known checksum algorithm */
3372         if (!ext4_verify_csum_type(sb, es)) {
3373                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3374                          "unknown checksum algorithm.");
3375                 silent = 1;
3376                 goto cantfind_ext4;
3377         }
3378
3379         /* Load the checksum driver */
3380         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3381                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3382                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3383                 if (IS_ERR(sbi->s_chksum_driver)) {
3384                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3385                         ret = PTR_ERR(sbi->s_chksum_driver);
3386                         sbi->s_chksum_driver = NULL;
3387                         goto failed_mount;
3388                 }
3389         }
3390
3391         /* Check superblock checksum */
3392         if (!ext4_superblock_csum_verify(sb, es)) {
3393                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3394                          "invalid superblock checksum.  Run e2fsck?");
3395                 silent = 1;
3396                 goto cantfind_ext4;
3397         }
3398
3399         /* Precompute checksum seed for all metadata */
3400         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3401                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3402                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3403                                                sizeof(es->s_uuid));
3404
3405         /* Set defaults before we parse the mount options */
3406         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3407         set_opt(sb, INIT_INODE_TABLE);
3408         if (def_mount_opts & EXT4_DEFM_DEBUG)
3409                 set_opt(sb, DEBUG);
3410         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3411                 set_opt(sb, GRPID);
3412         if (def_mount_opts & EXT4_DEFM_UID16)
3413                 set_opt(sb, NO_UID32);
3414         /* xattr user namespace & acls are now defaulted on */
3415         set_opt(sb, XATTR_USER);
3416 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3417         set_opt(sb, POSIX_ACL);
3418 #endif
3419         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3420                 set_opt(sb, JOURNAL_DATA);
3421         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3422                 set_opt(sb, ORDERED_DATA);
3423         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3424                 set_opt(sb, WRITEBACK_DATA);
3425
3426         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3427                 set_opt(sb, ERRORS_PANIC);
3428         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3429                 set_opt(sb, ERRORS_CONT);
3430         else
3431                 set_opt(sb, ERRORS_RO);
3432         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3433                 set_opt(sb, BLOCK_VALIDITY);
3434         if (def_mount_opts & EXT4_DEFM_DISCARD)
3435                 set_opt(sb, DISCARD);
3436
3437         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3438         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3439         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3440         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3441         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3442
3443         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3444                 set_opt(sb, BARRIER);
3445
3446         /*
3447          * enable delayed allocation by default
3448          * Use -o nodelalloc to turn it off
3449          */
3450         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3451             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3452                 set_opt(sb, DELALLOC);
3453
3454         /*
3455          * set default s_li_wait_mult for lazyinit, for the case there is
3456          * no mount option specified.
3457          */
3458         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3459
3460         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3461                            &journal_devnum, &journal_ioprio, 0)) {
3462                 ext4_msg(sb, KERN_WARNING,
3463                          "failed to parse options in superblock: %s",
3464                          sbi->s_es->s_mount_opts);
3465         }
3466         sbi->s_def_mount_opt = sbi->s_mount_opt;
3467         if (!parse_options((char *) data, sb, &journal_devnum,
3468                            &journal_ioprio, 0))
3469                 goto failed_mount;
3470
3471         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3472                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3473                             "with data=journal disables delayed "
3474                             "allocation and O_DIRECT support!\n");
3475                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3476                         ext4_msg(sb, KERN_ERR, "can't mount with "
3477                                  "both data=journal and delalloc");
3478                         goto failed_mount;
3479                 }
3480                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3481                         ext4_msg(sb, KERN_ERR, "can't mount with "
3482                                  "both data=journal and dioread_nolock");
3483                         goto failed_mount;
3484                 }
3485                 if (test_opt(sb, DELALLOC))
3486                         clear_opt(sb, DELALLOC);
3487         }
3488
3489         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3490                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3491
3492         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3493             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3494              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3495              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3496                 ext4_msg(sb, KERN_WARNING,
3497                        "feature flags set on rev 0 fs, "
3498                        "running e2fsck is recommended");
3499
3500         if (IS_EXT2_SB(sb)) {
3501                 if (ext2_feature_set_ok(sb))
3502                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3503                                  "using the ext4 subsystem");
3504                 else {
3505                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3506                                  "to feature incompatibilities");
3507                         goto failed_mount;
3508                 }
3509         }
3510
3511         if (IS_EXT3_SB(sb)) {
3512                 if (ext3_feature_set_ok(sb))
3513                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3514                                  "using the ext4 subsystem");
3515                 else {
3516                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3517                                  "to feature incompatibilities");
3518                         goto failed_mount;
3519                 }
3520         }
3521
3522         /*
3523          * Check feature flags regardless of the revision level, since we
3524          * previously didn't change the revision level when setting the flags,
3525          * so there is a chance incompat flags are set on a rev 0 filesystem.
3526          */
3527         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3528                 goto failed_mount;
3529
3530         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3531         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3532             blocksize > EXT4_MAX_BLOCK_SIZE) {
3533                 ext4_msg(sb, KERN_ERR,
3534                        "Unsupported filesystem blocksize %d", blocksize);
3535                 goto failed_mount;
3536         }
3537
3538         if (sb->s_blocksize != blocksize) {
3539                 /* Validate the filesystem blocksize */
3540                 if (!sb_set_blocksize(sb, blocksize)) {
3541                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3542                                         blocksize);
3543                         goto failed_mount;
3544                 }
3545
3546                 brelse(bh);
3547                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3548                 offset = do_div(logical_sb_block, blocksize);
3549                 bh = sb_bread(sb, logical_sb_block);
3550                 if (!bh) {
3551                         ext4_msg(sb, KERN_ERR,
3552                                "Can't read superblock on 2nd try");
3553                         goto failed_mount;
3554                 }
3555                 es = (struct ext4_super_block *)(bh->b_data + offset);
3556                 sbi->s_es = es;
3557                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3558                         ext4_msg(sb, KERN_ERR,
3559                                "Magic mismatch, very weird!");
3560                         goto failed_mount;
3561                 }
3562         }
3563
3564         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3565                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3566         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3567                                                       has_huge_files);
3568         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3569
3570         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3571                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3572                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3573         } else {
3574                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3575                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3576                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3577                     (!is_power_of_2(sbi->s_inode_size)) ||
3578                     (sbi->s_inode_size > blocksize)) {
3579                         ext4_msg(sb, KERN_ERR,
3580                                "unsupported inode size: %d",
3581                                sbi->s_inode_size);
3582                         goto failed_mount;
3583                 }
3584                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3585                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3586         }
3587
3588         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3589         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3590                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3591                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3592                     !is_power_of_2(sbi->s_desc_size)) {
3593                         ext4_msg(sb, KERN_ERR,
3594                                "unsupported descriptor size %lu",
3595                                sbi->s_desc_size);
3596                         goto failed_mount;
3597                 }
3598         } else
3599                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3600
3601         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3602         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3603         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3604                 goto cantfind_ext4;
3605
3606         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3607         if (sbi->s_inodes_per_block == 0)
3608                 goto cantfind_ext4;
3609         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3610                                         sbi->s_inodes_per_block;
3611         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3612         sbi->s_sbh = bh;
3613         sbi->s_mount_state = le16_to_cpu(es->s_state);
3614         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3615         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3616
3617         for (i = 0; i < 4; i++)
3618                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3619         sbi->s_def_hash_version = es->s_def_hash_version;
3620         i = le32_to_cpu(es->s_flags);
3621         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3622                 sbi->s_hash_unsigned = 3;
3623         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3624 #ifdef __CHAR_UNSIGNED__
3625                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3626                 sbi->s_hash_unsigned = 3;
3627 #else
3628                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3629 #endif
3630         }
3631
3632         /* Handle clustersize */
3633         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3634         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3635                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3636         if (has_bigalloc) {
3637                 if (clustersize < blocksize) {
3638                         ext4_msg(sb, KERN_ERR,
3639                                  "cluster size (%d) smaller than "
3640                                  "block size (%d)", clustersize, blocksize);
3641                         goto failed_mount;
3642                 }
3643                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3644                         le32_to_cpu(es->s_log_block_size);
3645                 sbi->s_clusters_per_group =
3646                         le32_to_cpu(es->s_clusters_per_group);
3647                 if (sbi->s_clusters_per_group > blocksize * 8) {
3648                         ext4_msg(sb, KERN_ERR,
3649                                  "#clusters per group too big: %lu",
3650                                  sbi->s_clusters_per_group);
3651                         goto failed_mount;
3652                 }
3653                 if (sbi->s_blocks_per_group !=
3654                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3655                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3656                                  "clusters per group (%lu) inconsistent",
3657                                  sbi->s_blocks_per_group,
3658                                  sbi->s_clusters_per_group);
3659                         goto failed_mount;
3660                 }
3661         } else {
3662                 if (clustersize != blocksize) {
3663                         ext4_warning(sb, "fragment/cluster size (%d) != "
3664                                      "block size (%d)", clustersize,
3665                                      blocksize);
3666                         clustersize = blocksize;
3667                 }
3668                 if (sbi->s_blocks_per_group > blocksize * 8) {
3669                         ext4_msg(sb, KERN_ERR,
3670                                  "#blocks per group too big: %lu",
3671                                  sbi->s_blocks_per_group);
3672                         goto failed_mount;
3673                 }
3674                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3675                 sbi->s_cluster_bits = 0;
3676         }
3677         sbi->s_cluster_ratio = clustersize / blocksize;
3678
3679         if (sbi->s_inodes_per_group > blocksize * 8) {
3680                 ext4_msg(sb, KERN_ERR,
3681                        "#inodes per group too big: %lu",
3682                        sbi->s_inodes_per_group);
3683                 goto failed_mount;
3684         }
3685
3686         /* Do we have standard group size of clustersize * 8 blocks ? */
3687         if (sbi->s_blocks_per_group == clustersize << 3)
3688                 set_opt2(sb, STD_GROUP_SIZE);
3689
3690         /*
3691          * Test whether we have more sectors than will fit in sector_t,
3692          * and whether the max offset is addressable by the page cache.
3693          */
3694         err = generic_check_addressable(sb->s_blocksize_bits,
3695                                         ext4_blocks_count(es));
3696         if (err) {
3697                 ext4_msg(sb, KERN_ERR, "filesystem"
3698                          " too large to mount safely on this system");
3699                 if (sizeof(sector_t) < 8)
3700                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3701                 goto failed_mount;
3702         }
3703
3704         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3705                 goto cantfind_ext4;
3706
3707         /* check blocks count against device size */
3708         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3709         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3710                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3711                        "exceeds size of device (%llu blocks)",
3712                        ext4_blocks_count(es), blocks_count);
3713                 goto failed_mount;
3714         }
3715
3716         /*
3717          * It makes no sense for the first data block to be beyond the end
3718          * of the filesystem.
3719          */
3720         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3721                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3722                          "block %u is beyond end of filesystem (%llu)",
3723                          le32_to_cpu(es->s_first_data_block),
3724                          ext4_blocks_count(es));
3725                 goto failed_mount;
3726         }
3727         blocks_count = (ext4_blocks_count(es) -
3728                         le32_to_cpu(es->s_first_data_block) +
3729                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3730         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3731         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3732                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3733                        "(block count %llu, first data block %u, "
3734                        "blocks per group %lu)", sbi->s_groups_count,
3735                        ext4_blocks_count(es),
3736                        le32_to_cpu(es->s_first_data_block),
3737                        EXT4_BLOCKS_PER_GROUP(sb));
3738                 goto failed_mount;
3739         }
3740         sbi->s_groups_count = blocks_count;
3741         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3742                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3743         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3744                    EXT4_DESC_PER_BLOCK(sb);
3745         sbi->s_group_desc = ext4_kvmalloc(db_count *
3746                                           sizeof(struct buffer_head *),
3747                                           GFP_KERNEL);
3748         if (sbi->s_group_desc == NULL) {
3749                 ext4_msg(sb, KERN_ERR, "not enough memory");
3750                 ret = -ENOMEM;
3751                 goto failed_mount;
3752         }
3753
3754         if (ext4_proc_root)
3755                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3756
3757         if (sbi->s_proc)
3758                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3759                                  &ext4_seq_options_fops, sb);
3760
3761         bgl_lock_init(sbi->s_blockgroup_lock);
3762
3763         for (i = 0; i < db_count; i++) {
3764                 block = descriptor_loc(sb, logical_sb_block, i);
3765                 sbi->s_group_desc[i] = sb_bread(sb, block);
3766                 if (!sbi->s_group_desc[i]) {
3767                         ext4_msg(sb, KERN_ERR,
3768                                "can't read group descriptor %d", i);
3769                         db_count = i;
3770                         goto failed_mount2;
3771                 }
3772         }
3773         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3774                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3775                 goto failed_mount2;
3776         }
3777         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3778                 if (!ext4_fill_flex_info(sb)) {
3779                         ext4_msg(sb, KERN_ERR,
3780                                "unable to initialize "
3781                                "flex_bg meta info!");
3782                         goto failed_mount2;
3783                 }
3784
3785         sbi->s_gdb_count = db_count;
3786         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3787         spin_lock_init(&sbi->s_next_gen_lock);
3788
3789         init_timer(&sbi->s_err_report);
3790         sbi->s_err_report.function = print_daily_error_info;
3791         sbi->s_err_report.data = (unsigned long) sb;
3792
3793         /* Register extent status tree shrinker */
3794         ext4_es_register_shrinker(sbi);
3795
3796         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3797                         ext4_count_free_clusters(sb));
3798         if (!err) {
3799                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3800                                 ext4_count_free_inodes(sb));
3801         }
3802         if (!err) {
3803                 err = percpu_counter_init(&sbi->s_dirs_counter,
3804                                 ext4_count_dirs(sb));
3805         }
3806         if (!err) {
3807                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3808         }
3809         if (!err) {
3810                 err = percpu_counter_init(&sbi->s_extent_cache_cnt, 0);
3811         }
3812         if (err) {
3813                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3814                 goto failed_mount3;
3815         }
3816
3817         sbi->s_stripe = ext4_get_stripe_size(sbi);
3818         sbi->s_extent_max_zeroout_kb = 32;
3819
3820         /*
3821          * set up enough so that it can read an inode
3822          */
3823         if (!test_opt(sb, NOLOAD) &&
3824             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3825                 sb->s_op = &ext4_sops;
3826         else
3827                 sb->s_op = &ext4_nojournal_sops;
3828         sb->s_export_op = &ext4_export_ops;
3829         sb->s_xattr = ext4_xattr_handlers;
3830 #ifdef CONFIG_QUOTA
3831         sb->dq_op = &ext4_quota_operations;
3832         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3833                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3834         else
3835                 sb->s_qcop = &ext4_qctl_operations;
3836 #endif
3837         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3838
3839         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3840         mutex_init(&sbi->s_orphan_lock);
3841
3842         sb->s_root = NULL;
3843
3844         needs_recovery = (es->s_last_orphan != 0 ||
3845                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3846                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3847
3848         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3849             !(sb->s_flags & MS_RDONLY))
3850                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3851                         goto failed_mount3;
3852
3853         /*
3854          * The first inode we look at is the journal inode.  Don't try
3855          * root first: it may be modified in the journal!
3856          */
3857         if (!test_opt(sb, NOLOAD) &&
3858             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3859                 if (ext4_load_journal(sb, es, journal_devnum))
3860                         goto failed_mount3;
3861         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3862               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3863                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3864                        "suppressed and not mounted read-only");
3865                 goto failed_mount_wq;
3866         } else {
3867                 clear_opt(sb, DATA_FLAGS);
3868                 sbi->s_journal = NULL;
3869                 needs_recovery = 0;
3870                 goto no_journal;
3871         }
3872
3873         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3874             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3875                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3876                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3877                 goto failed_mount_wq;
3878         }
3879
3880         if (!set_journal_csum_feature_set(sb)) {
3881                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3882                          "feature set");
3883                 goto failed_mount_wq;
3884         }
3885
3886         /* We have now updated the journal if required, so we can
3887          * validate the data journaling mode. */
3888         switch (test_opt(sb, DATA_FLAGS)) {
3889         case 0:
3890                 /* No mode set, assume a default based on the journal
3891                  * capabilities: ORDERED_DATA if the journal can
3892                  * cope, else JOURNAL_DATA
3893                  */
3894                 if (jbd2_journal_check_available_features
3895                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3896                         set_opt(sb, ORDERED_DATA);
3897                 else
3898                         set_opt(sb, JOURNAL_DATA);
3899                 break;
3900
3901         case EXT4_MOUNT_ORDERED_DATA:
3902         case EXT4_MOUNT_WRITEBACK_DATA:
3903                 if (!jbd2_journal_check_available_features
3904                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3905                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3906                                "requested data journaling mode");
3907                         goto failed_mount_wq;
3908                 }
3909         default:
3910                 break;
3911         }
3912         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3913
3914         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3915
3916         /*
3917          * The journal may have updated the bg summary counts, so we
3918          * need to update the global counters.
3919          */
3920         percpu_counter_set(&sbi->s_freeclusters_counter,
3921                            ext4_count_free_clusters(sb));
3922         percpu_counter_set(&sbi->s_freeinodes_counter,
3923                            ext4_count_free_inodes(sb));
3924         percpu_counter_set(&sbi->s_dirs_counter,
3925                            ext4_count_dirs(sb));
3926         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3927
3928 no_journal:
3929         /*
3930          * Get the # of file system overhead blocks from the
3931          * superblock if present.
3932          */
3933         if (es->s_overhead_clusters)
3934                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3935         else {
3936                 err = ext4_calculate_overhead(sb);
3937                 if (err)
3938                         goto failed_mount_wq;
3939         }
3940
3941         /*
3942          * The maximum number of concurrent works can be high and
3943          * concurrency isn't really necessary.  Limit it to 1.
3944          */
3945         EXT4_SB(sb)->rsv_conversion_wq =
3946                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3947         if (!EXT4_SB(sb)->rsv_conversion_wq) {
3948                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3949                 ret = -ENOMEM;
3950                 goto failed_mount4;
3951         }
3952
3953         /*
3954          * The jbd2_journal_load will have done any necessary log recovery,
3955          * so we can safely mount the rest of the filesystem now.
3956          */
3957
3958         root = ext4_iget(sb, EXT4_ROOT_INO);
3959         if (IS_ERR(root)) {
3960                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3961                 ret = PTR_ERR(root);
3962                 root = NULL;
3963                 goto failed_mount4;
3964         }
3965         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3966                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3967                 iput(root);
3968                 goto failed_mount4;
3969         }
3970         sb->s_root = d_make_root(root);
3971         if (!sb->s_root) {
3972                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3973                 ret = -ENOMEM;
3974                 goto failed_mount4;
3975         }
3976
3977         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3978                 sb->s_flags |= MS_RDONLY;
3979
3980         /* determine the minimum size of new large inodes, if present */
3981         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3982                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3983                                                      EXT4_GOOD_OLD_INODE_SIZE;
3984                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3985                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
3986                         if (sbi->s_want_extra_isize <
3987                             le16_to_cpu(es->s_want_extra_isize))
3988                                 sbi->s_want_extra_isize =
3989                                         le16_to_cpu(es->s_want_extra_isize);
3990                         if (sbi->s_want_extra_isize <
3991                             le16_to_cpu(es->s_min_extra_isize))
3992                                 sbi->s_want_extra_isize =
3993                                         le16_to_cpu(es->s_min_extra_isize);
3994                 }
3995         }
3996         /* Check if enough inode space is available */
3997         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3998                                                         sbi->s_inode_size) {
3999                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4000                                                        EXT4_GOOD_OLD_INODE_SIZE;
4001                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4002                          "available");
4003         }
4004
4005         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sbi));
4006         if (err) {
4007                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
4008                          "reserved pool", ext4_calculate_resv_clusters(sbi));
4009                 goto failed_mount4a;
4010         }
4011
4012         err = ext4_setup_system_zone(sb);
4013         if (err) {
4014                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4015                          "zone (%d)", err);
4016                 goto failed_mount4a;
4017         }
4018
4019         ext4_ext_init(sb);
4020         err = ext4_mb_init(sb);
4021         if (err) {
4022                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4023                          err);
4024                 goto failed_mount5;
4025         }
4026
4027         err = ext4_register_li_request(sb, first_not_zeroed);
4028         if (err)
4029                 goto failed_mount6;
4030
4031         sbi->s_kobj.kset = ext4_kset;
4032         init_completion(&sbi->s_kobj_unregister);
4033         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4034                                    "%s", sb->s_id);
4035         if (err)
4036                 goto failed_mount7;
4037
4038 #ifdef CONFIG_QUOTA
4039         /* Enable quota usage during mount. */
4040         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4041             !(sb->s_flags & MS_RDONLY)) {
4042                 err = ext4_enable_quotas(sb);
4043                 if (err)
4044                         goto failed_mount8;
4045         }
4046 #endif  /* CONFIG_QUOTA */
4047
4048         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4049         ext4_orphan_cleanup(sb, es);
4050         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4051         if (needs_recovery) {
4052                 ext4_msg(sb, KERN_INFO, "recovery complete");
4053                 ext4_mark_recovery_complete(sb, es);
4054         }
4055         if (EXT4_SB(sb)->s_journal) {
4056                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4057                         descr = " journalled data mode";
4058                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4059                         descr = " ordered data mode";
4060                 else
4061                         descr = " writeback data mode";
4062         } else
4063                 descr = "out journal";
4064
4065         if (test_opt(sb, DISCARD)) {
4066                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4067                 if (!blk_queue_discard(q))
4068                         ext4_msg(sb, KERN_WARNING,
4069                                  "mounting with \"discard\" option, but "
4070                                  "the device does not support discard");
4071         }
4072
4073         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4074                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4075                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4076
4077         if (es->s_error_count)
4078                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4079
4080         kfree(orig_data);
4081         return 0;
4082
4083 cantfind_ext4:
4084         if (!silent)
4085                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4086         goto failed_mount;
4087
4088 #ifdef CONFIG_QUOTA
4089 failed_mount8:
4090         kobject_del(&sbi->s_kobj);
4091 #endif
4092 failed_mount7:
4093         ext4_unregister_li_request(sb);
4094 failed_mount6:
4095         ext4_mb_release(sb);
4096 failed_mount5:
4097         ext4_ext_release(sb);
4098         ext4_release_system_zone(sb);
4099 failed_mount4a:
4100         dput(sb->s_root);
4101         sb->s_root = NULL;
4102 failed_mount4:
4103         ext4_msg(sb, KERN_ERR, "mount failed");
4104         if (EXT4_SB(sb)->rsv_conversion_wq)
4105                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4106 failed_mount_wq:
4107         if (sbi->s_journal) {
4108                 jbd2_journal_destroy(sbi->s_journal);
4109                 sbi->s_journal = NULL;
4110         }
4111 failed_mount3:
4112         ext4_es_unregister_shrinker(sbi);
4113         del_timer(&sbi->s_err_report);
4114         if (sbi->s_flex_groups)
4115                 ext4_kvfree(sbi->s_flex_groups);
4116         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4117         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4118         percpu_counter_destroy(&sbi->s_dirs_counter);
4119         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4120         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
4121         if (sbi->s_mmp_tsk)
4122                 kthread_stop(sbi->s_mmp_tsk);
4123 failed_mount2:
4124         for (i = 0; i < db_count; i++)
4125                 brelse(sbi->s_group_desc[i]);
4126         ext4_kvfree(sbi->s_group_desc);
4127 failed_mount:
4128         if (sbi->s_chksum_driver)
4129                 crypto_free_shash(sbi->s_chksum_driver);
4130         if (sbi->s_proc) {
4131                 remove_proc_entry("options", sbi->s_proc);
4132                 remove_proc_entry(sb->s_id, ext4_proc_root);
4133         }
4134 #ifdef CONFIG_QUOTA
4135         for (i = 0; i < MAXQUOTAS; i++)
4136                 kfree(sbi->s_qf_names[i]);
4137 #endif
4138         ext4_blkdev_remove(sbi);
4139         brelse(bh);
4140 out_fail:
4141         sb->s_fs_info = NULL;
4142         kfree(sbi->s_blockgroup_lock);
4143         kfree(sbi);
4144 out_free_orig:
4145         kfree(orig_data);
4146         return err ? err : ret;
4147 }
4148
4149 /*
4150  * Setup any per-fs journal parameters now.  We'll do this both on
4151  * initial mount, once the journal has been initialised but before we've
4152  * done any recovery; and again on any subsequent remount.
4153  */
4154 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4155 {
4156         struct ext4_sb_info *sbi = EXT4_SB(sb);
4157
4158         journal->j_commit_interval = sbi->s_commit_interval;
4159         journal->j_min_batch_time = sbi->s_min_batch_time;
4160         journal->j_max_batch_time = sbi->s_max_batch_time;
4161
4162         write_lock(&journal->j_state_lock);
4163         if (test_opt(sb, BARRIER))
4164                 journal->j_flags |= JBD2_BARRIER;
4165         else
4166                 journal->j_flags &= ~JBD2_BARRIER;
4167         if (test_opt(sb, DATA_ERR_ABORT))
4168                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4169         else
4170                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4171         write_unlock(&journal->j_state_lock);
4172 }
4173
4174 static journal_t *ext4_get_journal(struct super_block *sb,
4175                                    unsigned int journal_inum)
4176 {
4177         struct inode *journal_inode;
4178         journal_t *journal;
4179
4180         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4181
4182         /* First, test for the existence of a valid inode on disk.  Bad
4183          * things happen if we iget() an unused inode, as the subsequent
4184          * iput() will try to delete it. */
4185
4186         journal_inode = ext4_iget(sb, journal_inum);
4187         if (IS_ERR(journal_inode)) {
4188                 ext4_msg(sb, KERN_ERR, "no journal found");
4189                 return NULL;
4190         }
4191         if (!journal_inode->i_nlink) {
4192                 make_bad_inode(journal_inode);
4193                 iput(journal_inode);
4194                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4195                 return NULL;
4196         }
4197
4198         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4199                   journal_inode, journal_inode->i_size);
4200         if (!S_ISREG(journal_inode->i_mode)) {
4201                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4202                 iput(journal_inode);
4203                 return NULL;
4204         }
4205
4206         journal = jbd2_journal_init_inode(journal_inode);
4207         if (!journal) {
4208                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4209                 iput(journal_inode);
4210                 return NULL;
4211         }
4212         journal->j_private = sb;
4213         ext4_init_journal_params(sb, journal);
4214         return journal;
4215 }
4216
4217 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4218                                        dev_t j_dev)
4219 {
4220         struct buffer_head *bh;
4221         journal_t *journal;
4222         ext4_fsblk_t start;
4223         ext4_fsblk_t len;
4224         int hblock, blocksize;
4225         ext4_fsblk_t sb_block;
4226         unsigned long offset;
4227         struct ext4_super_block *es;
4228         struct block_device *bdev;
4229
4230         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4231
4232         bdev = ext4_blkdev_get(j_dev, sb);
4233         if (bdev == NULL)
4234                 return NULL;
4235
4236         blocksize = sb->s_blocksize;
4237         hblock = bdev_logical_block_size(bdev);
4238         if (blocksize < hblock) {
4239                 ext4_msg(sb, KERN_ERR,
4240                         "blocksize too small for journal device");
4241                 goto out_bdev;
4242         }
4243
4244         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4245         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4246         set_blocksize(bdev, blocksize);
4247         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4248                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4249                        "external journal");
4250                 goto out_bdev;
4251         }
4252
4253         es = (struct ext4_super_block *) (bh->b_data + offset);
4254         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4255             !(le32_to_cpu(es->s_feature_incompat) &
4256               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4257                 ext4_msg(sb, KERN_ERR, "external journal has "
4258                                         "bad superblock");
4259                 brelse(bh);
4260                 goto out_bdev;
4261         }
4262
4263         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4264                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4265                 brelse(bh);
4266                 goto out_bdev;
4267         }
4268
4269         len = ext4_blocks_count(es);
4270         start = sb_block + 1;
4271         brelse(bh);     /* we're done with the superblock */
4272
4273         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4274                                         start, len, blocksize);
4275         if (!journal) {
4276                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4277                 goto out_bdev;
4278         }
4279         journal->j_private = sb;
4280         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4281         wait_on_buffer(journal->j_sb_buffer);
4282         if (!buffer_uptodate(journal->j_sb_buffer)) {
4283                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4284                 goto out_journal;
4285         }
4286         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4287                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4288                                         "user (unsupported) - %d",
4289                         be32_to_cpu(journal->j_superblock->s_nr_users));
4290                 goto out_journal;
4291         }
4292         EXT4_SB(sb)->journal_bdev = bdev;
4293         ext4_init_journal_params(sb, journal);
4294         return journal;
4295
4296 out_journal:
4297         jbd2_journal_destroy(journal);
4298 out_bdev:
4299         ext4_blkdev_put(bdev);
4300         return NULL;
4301 }
4302
4303 static int ext4_load_journal(struct super_block *sb,
4304                              struct ext4_super_block *es,
4305                              unsigned long journal_devnum)
4306 {
4307         journal_t *journal;
4308         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4309         dev_t journal_dev;
4310         int err = 0;
4311         int really_read_only;
4312
4313         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4314
4315         if (journal_devnum &&
4316             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4317                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4318                         "numbers have changed");
4319                 journal_dev = new_decode_dev(journal_devnum);
4320         } else
4321                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4322
4323         really_read_only = bdev_read_only(sb->s_bdev);
4324
4325         /*
4326          * Are we loading a blank journal or performing recovery after a
4327          * crash?  For recovery, we need to check in advance whether we
4328          * can get read-write access to the device.
4329          */
4330         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4331                 if (sb->s_flags & MS_RDONLY) {
4332                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4333                                         "required on readonly filesystem");
4334                         if (really_read_only) {
4335                                 ext4_msg(sb, KERN_ERR, "write access "
4336                                         "unavailable, cannot proceed");
4337                                 return -EROFS;
4338                         }
4339                         ext4_msg(sb, KERN_INFO, "write access will "
4340                                "be enabled during recovery");
4341                 }
4342         }
4343
4344         if (journal_inum && journal_dev) {
4345                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4346                        "and inode journals!");
4347                 return -EINVAL;
4348         }
4349
4350         if (journal_inum) {
4351                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4352                         return -EINVAL;
4353         } else {
4354                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4355                         return -EINVAL;
4356         }
4357
4358         if (!(journal->j_flags & JBD2_BARRIER))
4359                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4360
4361         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4362                 err = jbd2_journal_wipe(journal, !really_read_only);
4363         if (!err) {
4364                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4365                 if (save)
4366                         memcpy(save, ((char *) es) +
4367                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4368                 err = jbd2_journal_load(journal);
4369                 if (save)
4370                         memcpy(((char *) es) + EXT4_S_ERR_START,
4371                                save, EXT4_S_ERR_LEN);
4372                 kfree(save);
4373         }
4374
4375         if (err) {
4376                 ext4_msg(sb, KERN_ERR, "error loading journal");
4377                 jbd2_journal_destroy(journal);
4378                 return err;
4379         }
4380
4381         EXT4_SB(sb)->s_journal = journal;
4382         ext4_clear_journal_err(sb, es);
4383
4384         if (!really_read_only && journal_devnum &&
4385             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4386                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4387
4388                 /* Make sure we flush the recovery flag to disk. */
4389                 ext4_commit_super(sb, 1);
4390         }
4391
4392         return 0;
4393 }
4394
4395 static int ext4_commit_super(struct super_block *sb, int sync)
4396 {
4397         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4398         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4399         int error = 0;
4400
4401         if (!sbh || block_device_ejected(sb))
4402                 return error;
4403         if (buffer_write_io_error(sbh)) {
4404                 /*
4405                  * Oh, dear.  A previous attempt to write the
4406                  * superblock failed.  This could happen because the
4407                  * USB device was yanked out.  Or it could happen to
4408                  * be a transient write error and maybe the block will
4409                  * be remapped.  Nothing we can do but to retry the
4410                  * write and hope for the best.
4411                  */
4412                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4413                        "superblock detected");
4414                 clear_buffer_write_io_error(sbh);
4415                 set_buffer_uptodate(sbh);
4416         }
4417         /*
4418          * If the file system is mounted read-only, don't update the
4419          * superblock write time.  This avoids updating the superblock
4420          * write time when we are mounting the root file system
4421          * read/only but we need to replay the journal; at that point,
4422          * for people who are east of GMT and who make their clock
4423          * tick in localtime for Windows bug-for-bug compatibility,
4424          * the clock is set in the future, and this will cause e2fsck
4425          * to complain and force a full file system check.
4426          */
4427         if (!(sb->s_flags & MS_RDONLY))
4428                 es->s_wtime = cpu_to_le32(get_seconds());
4429         if (sb->s_bdev->bd_part)
4430                 es->s_kbytes_written =
4431                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4432                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4433                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4434         else
4435                 es->s_kbytes_written =
4436                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4437         ext4_free_blocks_count_set(es,
4438                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4439                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4440         es->s_free_inodes_count =
4441                 cpu_to_le32(percpu_counter_sum_positive(
4442                                 &EXT4_SB(sb)->s_freeinodes_counter));
4443         BUFFER_TRACE(sbh, "marking dirty");
4444         ext4_superblock_csum_set(sb);
4445         mark_buffer_dirty(sbh);
4446         if (sync) {
4447                 error = sync_dirty_buffer(sbh);
4448                 if (error)
4449                         return error;
4450
4451                 error = buffer_write_io_error(sbh);
4452                 if (error) {
4453                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4454                                "superblock");
4455                         clear_buffer_write_io_error(sbh);
4456                         set_buffer_uptodate(sbh);
4457                 }
4458         }
4459         return error;
4460 }
4461
4462 /*
4463  * Have we just finished recovery?  If so, and if we are mounting (or
4464  * remounting) the filesystem readonly, then we will end up with a
4465  * consistent fs on disk.  Record that fact.
4466  */
4467 static void ext4_mark_recovery_complete(struct super_block *sb,
4468                                         struct ext4_super_block *es)
4469 {
4470         journal_t *journal = EXT4_SB(sb)->s_journal;
4471
4472         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4473                 BUG_ON(journal != NULL);
4474                 return;
4475         }
4476         jbd2_journal_lock_updates(journal);
4477         if (jbd2_journal_flush(journal) < 0)
4478                 goto out;
4479
4480         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4481             sb->s_flags & MS_RDONLY) {
4482                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4483                 ext4_commit_super(sb, 1);
4484         }
4485
4486 out:
4487         jbd2_journal_unlock_updates(journal);
4488 }
4489
4490 /*
4491  * If we are mounting (or read-write remounting) a filesystem whose journal
4492  * has recorded an error from a previous lifetime, move that error to the
4493  * main filesystem now.
4494  */
4495 static void ext4_clear_journal_err(struct super_block *sb,
4496                                    struct ext4_super_block *es)
4497 {
4498         journal_t *journal;
4499         int j_errno;
4500         const char *errstr;
4501
4502         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4503
4504         journal = EXT4_SB(sb)->s_journal;
4505
4506         /*
4507          * Now check for any error status which may have been recorded in the
4508          * journal by a prior ext4_error() or ext4_abort()
4509          */
4510
4511         j_errno = jbd2_journal_errno(journal);
4512         if (j_errno) {
4513                 char nbuf[16];
4514
4515                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4516                 ext4_warning(sb, "Filesystem error recorded "
4517                              "from previous mount: %s", errstr);
4518                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4519
4520                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4521                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4522                 ext4_commit_super(sb, 1);
4523
4524                 jbd2_journal_clear_err(journal);
4525                 jbd2_journal_update_sb_errno(journal);
4526         }
4527 }
4528
4529 /*
4530  * Force the running and committing transactions to commit,
4531  * and wait on the commit.
4532  */
4533 int ext4_force_commit(struct super_block *sb)
4534 {
4535         journal_t *journal;
4536
4537         if (sb->s_flags & MS_RDONLY)
4538                 return 0;
4539
4540         journal = EXT4_SB(sb)->s_journal;
4541         return ext4_journal_force_commit(journal);
4542 }
4543
4544 static int ext4_sync_fs(struct super_block *sb, int wait)
4545 {
4546         int ret = 0;
4547         tid_t target;
4548         bool needs_barrier = false;
4549         struct ext4_sb_info *sbi = EXT4_SB(sb);
4550
4551         trace_ext4_sync_fs(sb, wait);
4552         flush_workqueue(sbi->rsv_conversion_wq);
4553         /*
4554          * Writeback quota in non-journalled quota case - journalled quota has
4555          * no dirty dquots
4556          */
4557         dquot_writeback_dquots(sb, -1);
4558         /*
4559          * Data writeback is possible w/o journal transaction, so barrier must
4560          * being sent at the end of the function. But we can skip it if
4561          * transaction_commit will do it for us.
4562          */
4563         target = jbd2_get_latest_transaction(sbi->s_journal);
4564         if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4565             !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4566                 needs_barrier = true;
4567
4568         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4569                 if (wait)
4570                         ret = jbd2_log_wait_commit(sbi->s_journal, target);
4571         }
4572         if (needs_barrier) {
4573                 int err;
4574                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4575                 if (!ret)
4576                         ret = err;
4577         }
4578
4579         return ret;
4580 }
4581
4582 static int ext4_sync_fs_nojournal(struct super_block *sb, int wait)
4583 {
4584         int ret = 0;
4585
4586         trace_ext4_sync_fs(sb, wait);
4587         flush_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4588         dquot_writeback_dquots(sb, -1);
4589         if (wait && test_opt(sb, BARRIER))
4590                 ret = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4591
4592         return ret;
4593 }
4594
4595 /*
4596  * LVM calls this function before a (read-only) snapshot is created.  This
4597  * gives us a chance to flush the journal completely and mark the fs clean.
4598  *
4599  * Note that only this function cannot bring a filesystem to be in a clean
4600  * state independently. It relies on upper layer to stop all data & metadata
4601  * modifications.
4602  */
4603 static int ext4_freeze(struct super_block *sb)
4604 {
4605         int error = 0;
4606         journal_t *journal;
4607
4608         if (sb->s_flags & MS_RDONLY)
4609                 return 0;
4610
4611         journal = EXT4_SB(sb)->s_journal;
4612
4613         /* Now we set up the journal barrier. */
4614         jbd2_journal_lock_updates(journal);
4615
4616         /*
4617          * Don't clear the needs_recovery flag if we failed to flush
4618          * the journal.
4619          */
4620         error = jbd2_journal_flush(journal);
4621         if (error < 0)
4622                 goto out;
4623
4624         /* Journal blocked and flushed, clear needs_recovery flag. */
4625         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4626         error = ext4_commit_super(sb, 1);
4627 out:
4628         /* we rely on upper layer to stop further updates */
4629         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4630         return error;
4631 }
4632
4633 /*
4634  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4635  * flag here, even though the filesystem is not technically dirty yet.
4636  */
4637 static int ext4_unfreeze(struct super_block *sb)
4638 {
4639         if (sb->s_flags & MS_RDONLY)
4640                 return 0;
4641
4642         /* Reset the needs_recovery flag before the fs is unlocked. */
4643         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4644         ext4_commit_super(sb, 1);
4645         return 0;
4646 }
4647
4648 /*
4649  * Structure to save mount options for ext4_remount's benefit
4650  */
4651 struct ext4_mount_options {
4652         unsigned long s_mount_opt;
4653         unsigned long s_mount_opt2;
4654         kuid_t s_resuid;
4655         kgid_t s_resgid;
4656         unsigned long s_commit_interval;
4657         u32 s_min_batch_time, s_max_batch_time;
4658 #ifdef CONFIG_QUOTA
4659         int s_jquota_fmt;
4660         char *s_qf_names[MAXQUOTAS];
4661 #endif
4662 };
4663
4664 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4665 {
4666         struct ext4_super_block *es;
4667         struct ext4_sb_info *sbi = EXT4_SB(sb);
4668         unsigned long old_sb_flags;
4669         struct ext4_mount_options old_opts;
4670         int enable_quota = 0;
4671         ext4_group_t g;
4672         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4673         int err = 0;
4674 #ifdef CONFIG_QUOTA
4675         int i, j;
4676 #endif
4677         char *orig_data = kstrdup(data, GFP_KERNEL);
4678
4679         /* Store the original options */
4680         old_sb_flags = sb->s_flags;
4681         old_opts.s_mount_opt = sbi->s_mount_opt;
4682         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4683         old_opts.s_resuid = sbi->s_resuid;
4684         old_opts.s_resgid = sbi->s_resgid;
4685         old_opts.s_commit_interval = sbi->s_commit_interval;
4686         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4687         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4688 #ifdef CONFIG_QUOTA
4689         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4690         for (i = 0; i < MAXQUOTAS; i++)
4691                 if (sbi->s_qf_names[i]) {
4692                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4693                                                          GFP_KERNEL);
4694                         if (!old_opts.s_qf_names[i]) {
4695                                 for (j = 0; j < i; j++)
4696                                         kfree(old_opts.s_qf_names[j]);
4697                                 kfree(orig_data);
4698                                 return -ENOMEM;
4699                         }
4700                 } else
4701                         old_opts.s_qf_names[i] = NULL;
4702 #endif
4703         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4704                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4705
4706         /*
4707          * Allow the "check" option to be passed as a remount option.
4708          */
4709         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4710                 err = -EINVAL;
4711                 goto restore_opts;
4712         }
4713
4714         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4715                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4716                         ext4_msg(sb, KERN_ERR, "can't mount with "
4717                                  "both data=journal and delalloc");
4718                         err = -EINVAL;
4719                         goto restore_opts;
4720                 }
4721                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4722                         ext4_msg(sb, KERN_ERR, "can't mount with "
4723                                  "both data=journal and dioread_nolock");
4724                         err = -EINVAL;
4725                         goto restore_opts;
4726                 }
4727         }
4728
4729         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4730                 ext4_abort(sb, "Abort forced by user");
4731
4732         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4733                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4734
4735         es = sbi->s_es;
4736
4737         if (sbi->s_journal) {
4738                 ext4_init_journal_params(sb, sbi->s_journal);
4739                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4740         }
4741
4742         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4743                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4744                         err = -EROFS;
4745                         goto restore_opts;
4746                 }
4747
4748                 if (*flags & MS_RDONLY) {
4749                         err = dquot_suspend(sb, -1);
4750                         if (err < 0)
4751                                 goto restore_opts;
4752
4753                         /*
4754                          * First of all, the unconditional stuff we have to do
4755                          * to disable replay of the journal when we next remount
4756                          */
4757                         sb->s_flags |= MS_RDONLY;
4758
4759                         /*
4760                          * OK, test if we are remounting a valid rw partition
4761                          * readonly, and if so set the rdonly flag and then
4762                          * mark the partition as valid again.
4763                          */
4764                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4765                             (sbi->s_mount_state & EXT4_VALID_FS))
4766                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4767
4768                         if (sbi->s_journal)
4769                                 ext4_mark_recovery_complete(sb, es);
4770                 } else {
4771                         /* Make sure we can mount this feature set readwrite */
4772                         if (!ext4_feature_set_ok(sb, 0)) {
4773                                 err = -EROFS;
4774                                 goto restore_opts;
4775                         }
4776                         /*
4777                          * Make sure the group descriptor checksums
4778                          * are sane.  If they aren't, refuse to remount r/w.
4779                          */
4780                         for (g = 0; g < sbi->s_groups_count; g++) {
4781                                 struct ext4_group_desc *gdp =
4782                                         ext4_get_group_desc(sb, g, NULL);
4783
4784                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4785                                         ext4_msg(sb, KERN_ERR,
4786                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4787                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4788                                                le16_to_cpu(gdp->bg_checksum));
4789                                         err = -EINVAL;
4790                                         goto restore_opts;
4791                                 }
4792                         }
4793
4794                         /*
4795                          * If we have an unprocessed orphan list hanging
4796                          * around from a previously readonly bdev mount,
4797                          * require a full umount/remount for now.
4798                          */
4799                         if (es->s_last_orphan) {
4800                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4801                                        "remount RDWR because of unprocessed "
4802                                        "orphan inode list.  Please "
4803                                        "umount/remount instead");
4804                                 err = -EINVAL;
4805                                 goto restore_opts;
4806                         }
4807
4808                         /*
4809                          * Mounting a RDONLY partition read-write, so reread
4810                          * and store the current valid flag.  (It may have
4811                          * been changed by e2fsck since we originally mounted
4812                          * the partition.)
4813                          */
4814                         if (sbi->s_journal)
4815                                 ext4_clear_journal_err(sb, es);
4816                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4817                         if (!ext4_setup_super(sb, es, 0))
4818                                 sb->s_flags &= ~MS_RDONLY;
4819                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4820                                                      EXT4_FEATURE_INCOMPAT_MMP))
4821                                 if (ext4_multi_mount_protect(sb,
4822                                                 le64_to_cpu(es->s_mmp_block))) {
4823                                         err = -EROFS;
4824                                         goto restore_opts;
4825                                 }
4826                         enable_quota = 1;
4827                 }
4828         }
4829
4830         /*
4831          * Reinitialize lazy itable initialization thread based on
4832          * current settings
4833          */
4834         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4835                 ext4_unregister_li_request(sb);
4836         else {
4837                 ext4_group_t first_not_zeroed;
4838                 first_not_zeroed = ext4_has_uninit_itable(sb);
4839                 ext4_register_li_request(sb, first_not_zeroed);
4840         }
4841
4842         ext4_setup_system_zone(sb);
4843         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4844                 ext4_commit_super(sb, 1);
4845
4846 #ifdef CONFIG_QUOTA
4847         /* Release old quota file names */
4848         for (i = 0; i < MAXQUOTAS; i++)
4849                 kfree(old_opts.s_qf_names[i]);
4850         if (enable_quota) {
4851                 if (sb_any_quota_suspended(sb))
4852                         dquot_resume(sb, -1);
4853                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4854                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4855                         err = ext4_enable_quotas(sb);
4856                         if (err)
4857                                 goto restore_opts;
4858                 }
4859         }
4860 #endif
4861
4862         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4863         kfree(orig_data);
4864         return 0;
4865
4866 restore_opts:
4867         sb->s_flags = old_sb_flags;
4868         sbi->s_mount_opt = old_opts.s_mount_opt;
4869         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4870         sbi->s_resuid = old_opts.s_resuid;
4871         sbi->s_resgid = old_opts.s_resgid;
4872         sbi->s_commit_interval = old_opts.s_commit_interval;
4873         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4874         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4875 #ifdef CONFIG_QUOTA
4876         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4877         for (i = 0; i < MAXQUOTAS; i++) {
4878                 kfree(sbi->s_qf_names[i]);
4879                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4880         }
4881 #endif
4882         kfree(orig_data);
4883         return err;
4884 }
4885
4886 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4887 {
4888         struct super_block *sb = dentry->d_sb;
4889         struct ext4_sb_info *sbi = EXT4_SB(sb);
4890         struct ext4_super_block *es = sbi->s_es;
4891         ext4_fsblk_t overhead = 0, resv_blocks;
4892         u64 fsid;
4893         s64 bfree;
4894         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4895
4896         if (!test_opt(sb, MINIX_DF))
4897                 overhead = sbi->s_overhead;
4898
4899         buf->f_type = EXT4_SUPER_MAGIC;
4900         buf->f_bsize = sb->s_blocksize;
4901         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4902         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4903                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4904         /* prevent underflow in case that few free space is available */
4905         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4906         buf->f_bavail = buf->f_bfree -
4907                         (ext4_r_blocks_count(es) + resv_blocks);
4908         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4909                 buf->f_bavail = 0;
4910         buf->f_files = le32_to_cpu(es->s_inodes_count);
4911         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4912         buf->f_namelen = EXT4_NAME_LEN;
4913         fsid = le64_to_cpup((void *)es->s_uuid) ^
4914                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4915         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4916         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4917
4918         return 0;
4919 }
4920
4921 /* Helper function for writing quotas on sync - we need to start transaction
4922  * before quota file is locked for write. Otherwise the are possible deadlocks:
4923  * Process 1                         Process 2
4924  * ext4_create()                     quota_sync()
4925  *   jbd2_journal_start()                  write_dquot()
4926  *   dquot_initialize()                         down(dqio_mutex)
4927  *     down(dqio_mutex)                    jbd2_journal_start()
4928  *
4929  */
4930
4931 #ifdef CONFIG_QUOTA
4932
4933 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4934 {
4935         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4936 }
4937
4938 static int ext4_write_dquot(struct dquot *dquot)
4939 {
4940         int ret, err;
4941         handle_t *handle;
4942         struct inode *inode;
4943
4944         inode = dquot_to_inode(dquot);
4945         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4946                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4947         if (IS_ERR(handle))
4948                 return PTR_ERR(handle);
4949         ret = dquot_commit(dquot);
4950         err = ext4_journal_stop(handle);
4951         if (!ret)
4952                 ret = err;
4953         return ret;
4954 }
4955
4956 static int ext4_acquire_dquot(struct dquot *dquot)
4957 {
4958         int ret, err;
4959         handle_t *handle;
4960
4961         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4962                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4963         if (IS_ERR(handle))
4964                 return PTR_ERR(handle);
4965         ret = dquot_acquire(dquot);
4966         err = ext4_journal_stop(handle);
4967         if (!ret)
4968                 ret = err;
4969         return ret;
4970 }
4971
4972 static int ext4_release_dquot(struct dquot *dquot)
4973 {
4974         int ret, err;
4975         handle_t *handle;
4976
4977         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4978                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4979         if (IS_ERR(handle)) {
4980                 /* Release dquot anyway to avoid endless cycle in dqput() */
4981                 dquot_release(dquot);
4982                 return PTR_ERR(handle);
4983         }
4984         ret = dquot_release(dquot);
4985         err = ext4_journal_stop(handle);
4986         if (!ret)
4987                 ret = err;
4988         return ret;
4989 }
4990
4991 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4992 {
4993         struct super_block *sb = dquot->dq_sb;
4994         struct ext4_sb_info *sbi = EXT4_SB(sb);
4995
4996         /* Are we journaling quotas? */
4997         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
4998             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4999                 dquot_mark_dquot_dirty(dquot);
5000                 return ext4_write_dquot(dquot);
5001         } else {
5002                 return dquot_mark_dquot_dirty(dquot);
5003         }
5004 }
5005
5006 static int ext4_write_info(struct super_block *sb, int type)
5007 {
5008         int ret, err;
5009         handle_t *handle;
5010
5011         /* Data block + inode block */
5012         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
5013         if (IS_ERR(handle))
5014                 return PTR_ERR(handle);
5015         ret = dquot_commit_info(sb, type);
5016         err = ext4_journal_stop(handle);
5017         if (!ret)
5018                 ret = err;
5019         return ret;
5020 }
5021
5022 /*
5023  * Turn on quotas during mount time - we need to find
5024  * the quota file and such...
5025  */
5026 static int ext4_quota_on_mount(struct super_block *sb, int type)
5027 {
5028         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5029                                         EXT4_SB(sb)->s_jquota_fmt, type);
5030 }
5031
5032 /*
5033  * Standard function to be called on quota_on
5034  */
5035 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5036                          struct path *path)
5037 {
5038         int err;
5039
5040         if (!test_opt(sb, QUOTA))
5041                 return -EINVAL;
5042
5043         /* Quotafile not on the same filesystem? */
5044         if (path->dentry->d_sb != sb)
5045                 return -EXDEV;
5046         /* Journaling quota? */
5047         if (EXT4_SB(sb)->s_qf_names[type]) {
5048                 /* Quotafile not in fs root? */
5049                 if (path->dentry->d_parent != sb->s_root)
5050                         ext4_msg(sb, KERN_WARNING,
5051                                 "Quota file not on filesystem root. "
5052                                 "Journaled quota will not work");
5053         }
5054
5055         /*
5056          * When we journal data on quota file, we have to flush journal to see
5057          * all updates to the file when we bypass pagecache...
5058          */
5059         if (EXT4_SB(sb)->s_journal &&
5060             ext4_should_journal_data(path->dentry->d_inode)) {
5061                 /*
5062                  * We don't need to lock updates but journal_flush() could
5063                  * otherwise be livelocked...
5064                  */
5065                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5066                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5067                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5068                 if (err)
5069                         return err;
5070         }
5071
5072         return dquot_quota_on(sb, type, format_id, path);
5073 }
5074
5075 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5076                              unsigned int flags)
5077 {
5078         int err;
5079         struct inode *qf_inode;
5080         unsigned long qf_inums[MAXQUOTAS] = {
5081                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5082                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5083         };
5084
5085         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5086
5087         if (!qf_inums[type])
5088                 return -EPERM;
5089
5090         qf_inode = ext4_iget(sb, qf_inums[type]);
5091         if (IS_ERR(qf_inode)) {
5092                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5093                 return PTR_ERR(qf_inode);
5094         }
5095
5096         /* Don't account quota for quota files to avoid recursion */
5097         qf_inode->i_flags |= S_NOQUOTA;
5098         err = dquot_enable(qf_inode, type, format_id, flags);
5099         iput(qf_inode);
5100
5101         return err;
5102 }
5103
5104 /* Enable usage tracking for all quota types. */
5105 static int ext4_enable_quotas(struct super_block *sb)
5106 {
5107         int type, err = 0;
5108         unsigned long qf_inums[MAXQUOTAS] = {
5109                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5110                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5111         };
5112
5113         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5114         for (type = 0; type < MAXQUOTAS; type++) {
5115                 if (qf_inums[type]) {
5116                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5117                                                 DQUOT_USAGE_ENABLED);
5118                         if (err) {
5119                                 ext4_warning(sb,
5120                                         "Failed to enable quota tracking "
5121                                         "(type=%d, err=%d). Please run "
5122                                         "e2fsck to fix.", type, err);
5123                                 return err;
5124                         }
5125                 }
5126         }
5127         return 0;
5128 }
5129
5130 /*
5131  * quota_on function that is used when QUOTA feature is set.
5132  */
5133 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5134                                  int format_id)
5135 {
5136         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5137                 return -EINVAL;
5138
5139         /*
5140          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5141          */
5142         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5143 }
5144
5145 static int ext4_quota_off(struct super_block *sb, int type)
5146 {
5147         struct inode *inode = sb_dqopt(sb)->files[type];
5148         handle_t *handle;
5149
5150         /* Force all delayed allocation blocks to be allocated.
5151          * Caller already holds s_umount sem */
5152         if (test_opt(sb, DELALLOC))
5153                 sync_filesystem(sb);
5154
5155         if (!inode)
5156                 goto out;
5157
5158         /* Update modification times of quota files when userspace can
5159          * start looking at them */
5160         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5161         if (IS_ERR(handle))
5162                 goto out;
5163         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5164         ext4_mark_inode_dirty(handle, inode);
5165         ext4_journal_stop(handle);
5166
5167 out:
5168         return dquot_quota_off(sb, type);
5169 }
5170
5171 /*
5172  * quota_off function that is used when QUOTA feature is set.
5173  */
5174 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5175 {
5176         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5177                 return -EINVAL;
5178
5179         /* Disable only the limits. */
5180         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5181 }
5182
5183 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5184  * acquiring the locks... As quota files are never truncated and quota code
5185  * itself serializes the operations (and no one else should touch the files)
5186  * we don't have to be afraid of races */
5187 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5188                                size_t len, loff_t off)
5189 {
5190         struct inode *inode = sb_dqopt(sb)->files[type];
5191         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5192         int err = 0;
5193         int offset = off & (sb->s_blocksize - 1);
5194         int tocopy;
5195         size_t toread;
5196         struct buffer_head *bh;
5197         loff_t i_size = i_size_read(inode);
5198
5199         if (off > i_size)
5200                 return 0;
5201         if (off+len > i_size)
5202                 len = i_size-off;
5203         toread = len;
5204         while (toread > 0) {
5205                 tocopy = sb->s_blocksize - offset < toread ?
5206                                 sb->s_blocksize - offset : toread;
5207                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5208                 if (err)
5209                         return err;
5210                 if (!bh)        /* A hole? */
5211                         memset(data, 0, tocopy);
5212                 else
5213                         memcpy(data, bh->b_data+offset, tocopy);
5214                 brelse(bh);
5215                 offset = 0;
5216                 toread -= tocopy;
5217                 data += tocopy;
5218                 blk++;
5219         }
5220         return len;
5221 }
5222
5223 /* Write to quotafile (we know the transaction is already started and has
5224  * enough credits) */
5225 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5226                                 const char *data, size_t len, loff_t off)
5227 {
5228         struct inode *inode = sb_dqopt(sb)->files[type];
5229         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5230         int err = 0;
5231         int offset = off & (sb->s_blocksize - 1);
5232         struct buffer_head *bh;
5233         handle_t *handle = journal_current_handle();
5234
5235         if (EXT4_SB(sb)->s_journal && !handle) {
5236                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5237                         " cancelled because transaction is not started",
5238                         (unsigned long long)off, (unsigned long long)len);
5239                 return -EIO;
5240         }
5241         /*
5242          * Since we account only one data block in transaction credits,
5243          * then it is impossible to cross a block boundary.
5244          */
5245         if (sb->s_blocksize - offset < len) {
5246                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5247                         " cancelled because not block aligned",
5248                         (unsigned long long)off, (unsigned long long)len);
5249                 return -EIO;
5250         }
5251
5252         bh = ext4_bread(handle, inode, blk, 1, &err);
5253         if (!bh)
5254                 goto out;
5255         err = ext4_journal_get_write_access(handle, bh);
5256         if (err) {
5257                 brelse(bh);
5258                 goto out;
5259         }
5260         lock_buffer(bh);
5261         memcpy(bh->b_data+offset, data, len);
5262         flush_dcache_page(bh->b_page);
5263         unlock_buffer(bh);
5264         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5265         brelse(bh);
5266 out:
5267         if (err)
5268                 return err;
5269         if (inode->i_size < off + len) {
5270                 i_size_write(inode, off + len);
5271                 EXT4_I(inode)->i_disksize = inode->i_size;
5272                 ext4_mark_inode_dirty(handle, inode);
5273         }
5274         return len;
5275 }
5276
5277 #endif
5278
5279 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5280                        const char *dev_name, void *data)
5281 {
5282         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5283 }
5284
5285 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5286 static inline void register_as_ext2(void)
5287 {
5288         int err = register_filesystem(&ext2_fs_type);
5289         if (err)
5290                 printk(KERN_WARNING
5291                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5292 }
5293
5294 static inline void unregister_as_ext2(void)
5295 {
5296         unregister_filesystem(&ext2_fs_type);
5297 }
5298
5299 static inline int ext2_feature_set_ok(struct super_block *sb)
5300 {
5301         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5302                 return 0;
5303         if (sb->s_flags & MS_RDONLY)
5304                 return 1;
5305         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5306                 return 0;
5307         return 1;
5308 }
5309 #else
5310 static inline void register_as_ext2(void) { }
5311 static inline void unregister_as_ext2(void) { }
5312 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5313 #endif
5314
5315 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5316 static inline void register_as_ext3(void)
5317 {
5318         int err = register_filesystem(&ext3_fs_type);
5319         if (err)
5320                 printk(KERN_WARNING
5321                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5322 }
5323
5324 static inline void unregister_as_ext3(void)
5325 {
5326         unregister_filesystem(&ext3_fs_type);
5327 }
5328
5329 static inline int ext3_feature_set_ok(struct super_block *sb)
5330 {
5331         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5332                 return 0;
5333         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5334                 return 0;
5335         if (sb->s_flags & MS_RDONLY)
5336                 return 1;
5337         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5338                 return 0;
5339         return 1;
5340 }
5341 #else
5342 static inline void register_as_ext3(void) { }
5343 static inline void unregister_as_ext3(void) { }
5344 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5345 #endif
5346
5347 static struct file_system_type ext4_fs_type = {
5348         .owner          = THIS_MODULE,
5349         .name           = "ext4",
5350         .mount          = ext4_mount,
5351         .kill_sb        = kill_block_super,
5352         .fs_flags       = FS_REQUIRES_DEV,
5353 };
5354 MODULE_ALIAS_FS("ext4");
5355
5356 static int __init ext4_init_feat_adverts(void)
5357 {
5358         struct ext4_features *ef;
5359         int ret = -ENOMEM;
5360
5361         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5362         if (!ef)
5363                 goto out;
5364
5365         ef->f_kobj.kset = ext4_kset;
5366         init_completion(&ef->f_kobj_unregister);
5367         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5368                                    "features");
5369         if (ret) {
5370                 kfree(ef);
5371                 goto out;
5372         }
5373
5374         ext4_feat = ef;
5375         ret = 0;
5376 out:
5377         return ret;
5378 }
5379
5380 static void ext4_exit_feat_adverts(void)
5381 {
5382         kobject_put(&ext4_feat->f_kobj);
5383         wait_for_completion(&ext4_feat->f_kobj_unregister);
5384         kfree(ext4_feat);
5385 }
5386
5387 /* Shared across all ext4 file systems */
5388 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5389 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5390
5391 static int __init ext4_init_fs(void)
5392 {
5393         int i, err;
5394
5395         ext4_li_info = NULL;
5396         mutex_init(&ext4_li_mtx);
5397
5398         /* Build-time check for flags consistency */
5399         ext4_check_flag_values();
5400
5401         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5402                 mutex_init(&ext4__aio_mutex[i]);
5403                 init_waitqueue_head(&ext4__ioend_wq[i]);
5404         }
5405
5406         err = ext4_init_es();
5407         if (err)
5408                 return err;
5409
5410         err = ext4_init_pageio();
5411         if (err)
5412                 goto out7;
5413
5414         err = ext4_init_system_zone();
5415         if (err)
5416                 goto out6;
5417         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5418         if (!ext4_kset) {
5419                 err = -ENOMEM;
5420                 goto out5;
5421         }
5422         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5423
5424         err = ext4_init_feat_adverts();
5425         if (err)
5426                 goto out4;
5427
5428         err = ext4_init_mballoc();
5429         if (err)
5430                 goto out3;
5431
5432         err = ext4_init_xattr();
5433         if (err)
5434                 goto out2;
5435         err = init_inodecache();
5436         if (err)
5437                 goto out1;
5438         register_as_ext3();
5439         register_as_ext2();
5440         err = register_filesystem(&ext4_fs_type);
5441         if (err)
5442                 goto out;
5443
5444         return 0;
5445 out:
5446         unregister_as_ext2();
5447         unregister_as_ext3();
5448         destroy_inodecache();
5449 out1:
5450         ext4_exit_xattr();
5451 out2:
5452         ext4_exit_mballoc();
5453 out3:
5454         ext4_exit_feat_adverts();
5455 out4:
5456         if (ext4_proc_root)
5457                 remove_proc_entry("fs/ext4", NULL);
5458         kset_unregister(ext4_kset);
5459 out5:
5460         ext4_exit_system_zone();
5461 out6:
5462         ext4_exit_pageio();
5463 out7:
5464         ext4_exit_es();
5465
5466         return err;
5467 }
5468
5469 static void __exit ext4_exit_fs(void)
5470 {
5471         ext4_destroy_lazyinit_thread();
5472         unregister_as_ext2();
5473         unregister_as_ext3();
5474         unregister_filesystem(&ext4_fs_type);
5475         destroy_inodecache();
5476         ext4_exit_xattr();
5477         ext4_exit_mballoc();
5478         ext4_exit_feat_adverts();
5479         remove_proc_entry("fs/ext4", NULL);
5480         kset_unregister(ext4_kset);
5481         ext4_exit_system_zone();
5482         ext4_exit_pageio();
5483         ext4_exit_es();
5484 }
5485
5486 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5487 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5488 MODULE_LICENSE("GPL");
5489 module_init(ext4_init_fs)
5490 module_exit(ext4_exit_fs)