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->unrsv_conversion_wq);
 766         flush_workqueue(sbi->rsv_conversion_wq);
 767         destroy_workqueue(sbi->unrsv_conversion_wq);
 768         destroy_workqueue(sbi->rsv_conversion_wq);
 769
 770         if (sbi->s_journal) {
 771                 err = jbd2_journal_destroy(sbi->s_journal);
 772                 sbi->s_journal = NULL;
 773                 if (err < 0)
 774                         ext4_abort(sb, "Couldn't clean up the journal");
 775         }
 776
 777         ext4_es_unregister_shrinker(sbi);
 778         del_timer(&sbi->s_err_report);
 779         ext4_release_system_zone(sb);
 780         ext4_mb_release(sb);
 781         ext4_ext_release(sb);
 782         ext4_xattr_put_super(sb);
 783
 784         if (!(sb->s_flags & MS_RDONLY)) {
 785                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
 786                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 787         }
 788         if (!(sb->s_flags & MS_RDONLY))
 789                 ext4_commit_super(sb, 1);
 790
 791         if (sbi->s_proc) {
 792                 remove_proc_entry("options", sbi->s_proc);
 793                 remove_proc_entry(sb->s_id, ext4_proc_root);
 794         }
 795         kobject_del(&sbi->s_kobj);
 796
 797         for (i = 0; i < sbi->s_gdb_count; i++)
 798                 brelse(sbi->s_group_desc[i]);
 799         ext4_kvfree(sbi->s_group_desc);
 800         ext4_kvfree(sbi->s_flex_groups);
 801         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 802         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 803         percpu_counter_destroy(&sbi->s_dirs_counter);
 804         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 805         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
 806         brelse(sbi->s_sbh);
 807 #ifdef CONFIG_QUOTA
 808         for (i = 0; i < MAXQUOTAS; i++)
 809                 kfree(sbi->s_qf_names[i]);
 810 #endif
 811
 812         /* Debugging code just in case the in-memory inode orphan list
 813          * isn't empty.  The on-disk one can be non-empty if we've
 814          * detected an error and taken the fs readonly, but the
 815          * in-memory list had better be clean by this point. */
 816         if (!list_empty(&sbi->s_orphan))
 817                 dump_orphan_list(sb, sbi);
 818         J_ASSERT(list_empty(&sbi->s_orphan));
 819
 820         invalidate_bdev(sb->s_bdev);
 821         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 822                 /*
 823                  * Invalidate the journal device's buffers.  We don't want them
 824                  * floating about in memory - the physical journal device may
 825                  * hotswapped, and it breaks the `ro-after' testing code.
 826                  */
 827                 sync_blockdev(sbi->journal_bdev);
 828                 invalidate_bdev(sbi->journal_bdev);
 829                 ext4_blkdev_remove(sbi);
 830         }
 831         if (sbi->s_mmp_tsk)
 832                 kthread_stop(sbi->s_mmp_tsk);
 833         sb->s_fs_info = NULL;
 834         /*
 835          * Now that we are completely done shutting down the
 836          * superblock, we need to actually destroy the kobject.
 837          */
 838         kobject_put(&sbi->s_kobj);
 839         wait_for_completion(&sbi->s_kobj_unregister);
 840         if (sbi->s_chksum_driver)
 841                 crypto_free_shash(sbi->s_chksum_driver);
 842         kfree(sbi->s_blockgroup_lock);
 843         kfree(sbi);
 844 }
 845
 846 static struct kmem_cache *ext4_inode_cachep;
 847
 848 /*
 849  * Called inside transaction, so use GFP_NOFS
 850  */
 851 static struct inode *ext4_alloc_inode(struct super_block *sb)
 852 {
 853         struct ext4_inode_info *ei;
 854
 855         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 856         if (!ei)
 857                 return NULL;
 858
 859         ei->vfs_inode.i_version = 1;
 860         INIT_LIST_HEAD(&ei->i_prealloc_list);
 861         spin_lock_init(&ei->i_prealloc_lock);
 862         ext4_es_init_tree(&ei->i_es_tree);
 863         rwlock_init(&ei->i_es_lock);
 864         INIT_LIST_HEAD(&ei->i_es_lru);
 865         ei->i_es_lru_nr = 0;
 866         ei->i_touch_when = 0;
 867         ei->i_reserved_data_blocks = 0;
 868         ei->i_reserved_meta_blocks = 0;
 869         ei->i_allocated_meta_blocks = 0;
 870         ei->i_da_metadata_calc_len = 0;
 871         ei->i_da_metadata_calc_last_lblock = 0;
 872         spin_lock_init(&(ei->i_block_reservation_lock));
 873 #ifdef CONFIG_QUOTA
 874         ei->i_reserved_quota = 0;
 875 #endif
 876         ei->jinode = NULL;
 877         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 878         INIT_LIST_HEAD(&ei->i_unrsv_conversion_list);
 879         spin_lock_init(&ei->i_completed_io_lock);
 880         ei->i_sync_tid = 0;
 881         ei->i_datasync_tid = 0;
 882         atomic_set(&ei->i_ioend_count, 0);
 883         atomic_set(&ei->i_unwritten, 0);
 884         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 885         INIT_WORK(&ei->i_unrsv_conversion_work, ext4_end_io_unrsv_work);
 886
 887         return &ei->vfs_inode;
 888 }
 889
 890 static int ext4_drop_inode(struct inode *inode)
 891 {
 892         int drop = generic_drop_inode(inode);
 893
 894         trace_ext4_drop_inode(inode, drop);
 895         return drop;
 896 }
 897
 898 static void ext4_i_callback(struct rcu_head *head)
 899 {
 900         struct inode *inode = container_of(head, struct inode, i_rcu);
 901         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 902 }
 903
 904 static void ext4_destroy_inode(struct inode *inode)
 905 {
 906         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 907                 ext4_msg(inode->i_sb, KERN_ERR,
 908                          "Inode %lu (%p): orphan list check failed!",
 909                          inode->i_ino, EXT4_I(inode));
 910                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 911                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 912                                 true);
 913                 dump_stack();
 914         }
 915         call_rcu(&inode->i_rcu, ext4_i_callback);
 916 }
 917
 918 static void init_once(void *foo)
 919 {
 920         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 921
 922         INIT_LIST_HEAD(&ei->i_orphan);
 923         init_rwsem(&ei->xattr_sem);
 924         init_rwsem(&ei->i_data_sem);
 925         inode_init_once(&ei->vfs_inode);
 926 }
 927
 928 static int init_inodecache(void)
 929 {
 930         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 931                                              sizeof(struct ext4_inode_info),
 932                                              0, (SLAB_RECLAIM_ACCOUNT|
 933                                                 SLAB_MEM_SPREAD),
 934                                              init_once);
 935         if (ext4_inode_cachep == NULL)
 936                 return -ENOMEM;
 937         return 0;
 938 }
 939
 940 static void destroy_inodecache(void)
 941 {
 942         /*
 943          * Make sure all delayed rcu free inodes are flushed before we
 944          * destroy cache.
 945          */
 946         rcu_barrier();
 947         kmem_cache_destroy(ext4_inode_cachep);
 948 }
 949
 950 void ext4_clear_inode(struct inode *inode)
 951 {
 952         invalidate_inode_buffers(inode);
 953         clear_inode(inode);
 954         dquot_drop(inode);
 955         ext4_discard_preallocations(inode);
 956         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 957         ext4_es_lru_del(inode);
 958         if (EXT4_I(inode)->jinode) {
 959                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 960                                                EXT4_I(inode)->jinode);
 961                 jbd2_free_inode(EXT4_I(inode)->jinode);
 962                 EXT4_I(inode)->jinode = NULL;
 963         }
 964 }
 965
 966 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
 967                                         u64 ino, u32 generation)
 968 {
 969         struct inode *inode;
 970
 971         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
 972                 return ERR_PTR(-ESTALE);
 973         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
 974                 return ERR_PTR(-ESTALE);
 975
 976         /* iget isn't really right if the inode is currently unallocated!!
 977          *
 978          * ext4_read_inode will return a bad_inode if the inode had been
 979          * deleted, so we should be safe.
 980          *
 981          * Currently we don't know the generation for parent directory, so
 982          * a generation of 0 means "accept any"
 983          */
 984         inode = ext4_iget(sb, ino);
 985         if (IS_ERR(inode))
 986                 return ERR_CAST(inode);
 987         if (generation && inode->i_generation != generation) {
 988                 iput(inode);
 989                 return ERR_PTR(-ESTALE);
 990         }
 991
 992         return inode;
 993 }
 994
 995 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
 996                                         int fh_len, int fh_type)
 997 {
 998         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
 999                                     ext4_nfs_get_inode);
1000 }
1001
1002 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1003                                         int fh_len, int fh_type)
1004 {
1005         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1006                                     ext4_nfs_get_inode);
1007 }
1008
1009 /*
1010  * Try to release metadata pages (indirect blocks, directories) which are
1011  * mapped via the block device.  Since these pages could have journal heads
1012  * which would prevent try_to_free_buffers() from freeing them, we must use
1013  * jbd2 layer's try_to_free_buffers() function to release them.
1014  */
1015 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1016                                  gfp_t wait)
1017 {
1018         journal_t *journal = EXT4_SB(sb)->s_journal;
1019
1020         WARN_ON(PageChecked(page));
1021         if (!page_has_buffers(page))
1022                 return 0;
1023         if (journal)
1024                 return jbd2_journal_try_to_free_buffers(journal, page,
1025                                                         wait & ~__GFP_WAIT);
1026         return try_to_free_buffers(page);
1027 }
1028
1029 #ifdef CONFIG_QUOTA
1030 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1031 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1032
1033 static int ext4_write_dquot(struct dquot *dquot);
1034 static int ext4_acquire_dquot(struct dquot *dquot);
1035 static int ext4_release_dquot(struct dquot *dquot);
1036 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1037 static int ext4_write_info(struct super_block *sb, int type);
1038 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1039                          struct path *path);
1040 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
1041                                  int format_id);
1042 static int ext4_quota_off(struct super_block *sb, int type);
1043 static int ext4_quota_off_sysfile(struct super_block *sb, int type);
1044 static int ext4_quota_on_mount(struct super_block *sb, int type);
1045 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1046                                size_t len, loff_t off);
1047 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1048                                 const char *data, size_t len, loff_t off);
1049 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1050                              unsigned int flags);
1051 static int ext4_enable_quotas(struct super_block *sb);
1052
1053 static const struct dquot_operations ext4_quota_operations = {
1054         .get_reserved_space = ext4_get_reserved_space,
1055         .write_dquot    = ext4_write_dquot,
1056         .acquire_dquot  = ext4_acquire_dquot,
1057         .release_dquot  = ext4_release_dquot,
1058         .mark_dirty     = ext4_mark_dquot_dirty,
1059         .write_info     = ext4_write_info,
1060         .alloc_dquot    = dquot_alloc,
1061         .destroy_dquot  = dquot_destroy,
1062 };
1063
1064 static const struct quotactl_ops ext4_qctl_operations = {
1065         .quota_on       = ext4_quota_on,
1066         .quota_off      = ext4_quota_off,
1067         .quota_sync     = dquot_quota_sync,
1068         .get_info       = dquot_get_dqinfo,
1069         .set_info       = dquot_set_dqinfo,
1070         .get_dqblk      = dquot_get_dqblk,
1071         .set_dqblk      = dquot_set_dqblk
1072 };
1073
1074 static const struct quotactl_ops ext4_qctl_sysfile_operations = {
1075         .quota_on_meta  = ext4_quota_on_sysfile,
1076         .quota_off      = ext4_quota_off_sysfile,
1077         .quota_sync     = dquot_quota_sync,
1078         .get_info       = dquot_get_dqinfo,
1079         .set_info       = dquot_set_dqinfo,
1080         .get_dqblk      = dquot_get_dqblk,
1081         .set_dqblk      = dquot_set_dqblk
1082 };
1083 #endif
1084
1085 static const struct super_operations ext4_sops = {
1086         .alloc_inode    = ext4_alloc_inode,
1087         .destroy_inode  = ext4_destroy_inode,
1088         .write_inode    = ext4_write_inode,
1089         .dirty_inode    = ext4_dirty_inode,
1090         .drop_inode     = ext4_drop_inode,
1091         .evict_inode    = ext4_evict_inode,
1092         .put_super      = ext4_put_super,
1093         .sync_fs        = ext4_sync_fs,
1094         .freeze_fs      = ext4_freeze,
1095         .unfreeze_fs    = ext4_unfreeze,
1096         .statfs         = ext4_statfs,
1097         .remount_fs     = ext4_remount,
1098         .show_options   = ext4_show_options,
1099 #ifdef CONFIG_QUOTA
1100         .quota_read     = ext4_quota_read,
1101         .quota_write    = ext4_quota_write,
1102 #endif
1103         .bdev_try_to_free_page = bdev_try_to_free_page,
1104 };
1105
1106 static const struct super_operations ext4_nojournal_sops = {
1107         .alloc_inode    = ext4_alloc_inode,
1108         .destroy_inode  = ext4_destroy_inode,
1109         .write_inode    = ext4_write_inode,
1110         .dirty_inode    = ext4_dirty_inode,
1111         .drop_inode     = ext4_drop_inode,
1112         .evict_inode    = ext4_evict_inode,
1113         .sync_fs        = ext4_sync_fs_nojournal,
1114         .put_super      = ext4_put_super,
1115         .statfs         = ext4_statfs,
1116         .remount_fs     = ext4_remount,
1117         .show_options   = ext4_show_options,
1118 #ifdef CONFIG_QUOTA
1119         .quota_read     = ext4_quota_read,
1120         .quota_write    = ext4_quota_write,
1121 #endif
1122         .bdev_try_to_free_page = bdev_try_to_free_page,
1123 };
1124
1125 static const struct export_operations ext4_export_ops = {
1126         .fh_to_dentry = ext4_fh_to_dentry,
1127         .fh_to_parent = ext4_fh_to_parent,
1128         .get_parent = ext4_get_parent,
1129 };
1130
1131 enum {
1132         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1133         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1134         Opt_nouid32, Opt_debug, Opt_removed,
1135         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1136         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1137         Opt_commit, Opt_min_batch_time, Opt_max_batch_time,
1138         Opt_journal_dev, Opt_journal_checksum, Opt_journal_async_commit,
1139         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1140         Opt_data_err_abort, Opt_data_err_ignore,
1141         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1142         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1143         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1144         Opt_usrquota, Opt_grpquota, Opt_i_version,
1145         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1146         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1147         Opt_inode_readahead_blks, Opt_journal_ioprio,
1148         Opt_dioread_nolock, Opt_dioread_lock,
1149         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1150         Opt_max_dir_size_kb,
1151 };
1152
1153 static const match_table_t tokens = {
1154         {Opt_bsd_df, "bsddf"},
1155         {Opt_minix_df, "minixdf"},
1156         {Opt_grpid, "grpid"},
1157         {Opt_grpid, "bsdgroups"},
1158         {Opt_nogrpid, "nogrpid"},
1159         {Opt_nogrpid, "sysvgroups"},
1160         {Opt_resgid, "resgid=%u"},
1161         {Opt_resuid, "resuid=%u"},
1162         {Opt_sb, "sb=%u"},
1163         {Opt_err_cont, "errors=continue"},
1164         {Opt_err_panic, "errors=panic"},
1165         {Opt_err_ro, "errors=remount-ro"},
1166         {Opt_nouid32, "nouid32"},
1167         {Opt_debug, "debug"},
1168         {Opt_removed, "oldalloc"},
1169         {Opt_removed, "orlov"},
1170         {Opt_user_xattr, "user_xattr"},
1171         {Opt_nouser_xattr, "nouser_xattr"},
1172         {Opt_acl, "acl"},
1173         {Opt_noacl, "noacl"},
1174         {Opt_noload, "norecovery"},
1175         {Opt_noload, "noload"},
1176         {Opt_removed, "nobh"},
1177         {Opt_removed, "bh"},
1178         {Opt_commit, "commit=%u"},
1179         {Opt_min_batch_time, "min_batch_time=%u"},
1180         {Opt_max_batch_time, "max_batch_time=%u"},
1181         {Opt_journal_dev, "journal_dev=%u"},
1182         {Opt_journal_checksum, "journal_checksum"},
1183         {Opt_journal_async_commit, "journal_async_commit"},
1184         {Opt_abort, "abort"},
1185         {Opt_data_journal, "data=journal"},
1186         {Opt_data_ordered, "data=ordered"},
1187         {Opt_data_writeback, "data=writeback"},
1188         {Opt_data_err_abort, "data_err=abort"},
1189         {Opt_data_err_ignore, "data_err=ignore"},
1190         {Opt_offusrjquota, "usrjquota="},
1191         {Opt_usrjquota, "usrjquota=%s"},
1192         {Opt_offgrpjquota, "grpjquota="},
1193         {Opt_grpjquota, "grpjquota=%s"},
1194         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1195         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1196         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1197         {Opt_grpquota, "grpquota"},
1198         {Opt_noquota, "noquota"},
1199         {Opt_quota, "quota"},
1200         {Opt_usrquota, "usrquota"},
1201         {Opt_barrier, "barrier=%u"},
1202         {Opt_barrier, "barrier"},
1203         {Opt_nobarrier, "nobarrier"},
1204         {Opt_i_version, "i_version"},
1205         {Opt_stripe, "stripe=%u"},
1206         {Opt_delalloc, "delalloc"},
1207         {Opt_nodelalloc, "nodelalloc"},
1208         {Opt_removed, "mblk_io_submit"},
1209         {Opt_removed, "nomblk_io_submit"},
1210         {Opt_block_validity, "block_validity"},
1211         {Opt_noblock_validity, "noblock_validity"},
1212         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1213         {Opt_journal_ioprio, "journal_ioprio=%u"},
1214         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1215         {Opt_auto_da_alloc, "auto_da_alloc"},
1216         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1217         {Opt_dioread_nolock, "dioread_nolock"},
1218         {Opt_dioread_lock, "dioread_lock"},
1219         {Opt_discard, "discard"},
1220         {Opt_nodiscard, "nodiscard"},
1221         {Opt_init_itable, "init_itable=%u"},
1222         {Opt_init_itable, "init_itable"},
1223         {Opt_noinit_itable, "noinit_itable"},
1224         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1225         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1226         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1227         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1228         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1229         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1230         {Opt_err, NULL},
1231 };
1232
1233 static ext4_fsblk_t get_sb_block(void **data)
1234 {
1235         ext4_fsblk_t    sb_block;
1236         char            *options = (char *) *data;
1237
1238         if (!options || strncmp(options, "sb=", 3) != 0)
1239                 return 1;       /* Default location */
1240
1241         options += 3;
1242         /* TODO: use simple_strtoll with >32bit ext4 */
1243         sb_block = simple_strtoul(options, &options, 0);
1244         if (*options && *options != ',') {
1245                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1246                        (char *) *data);
1247                 return 1;
1248         }
1249         if (*options == ',')
1250                 options++;
1251         *data = (void *) options;
1252
1253         return sb_block;
1254 }
1255
1256 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1257 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1258         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1259
1260 #ifdef CONFIG_QUOTA
1261 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1262 {
1263         struct ext4_sb_info *sbi = EXT4_SB(sb);
1264         char *qname;
1265         int ret = -1;
1266
1267         if (sb_any_quota_loaded(sb) &&
1268                 !sbi->s_qf_names[qtype]) {
1269                 ext4_msg(sb, KERN_ERR,
1270                         "Cannot change journaled "
1271                         "quota options when quota turned on");
1272                 return -1;
1273         }
1274         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1275                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1276                          "when QUOTA feature is enabled");
1277                 return -1;
1278         }
1279         qname = match_strdup(args);
1280         if (!qname) {
1281                 ext4_msg(sb, KERN_ERR,
1282                         "Not enough memory for storing quotafile name");
1283                 return -1;
1284         }
1285         if (sbi->s_qf_names[qtype]) {
1286                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1287                         ret = 1;
1288                 else
1289                         ext4_msg(sb, KERN_ERR,
1290                                  "%s quota file already specified",
1291                                  QTYPE2NAME(qtype));
1292                 goto errout;
1293         }
1294         if (strchr(qname, '/')) {
1295                 ext4_msg(sb, KERN_ERR,
1296                         "quotafile must be on filesystem root");
1297                 goto errout;
1298         }
1299         sbi->s_qf_names[qtype] = qname;
1300         set_opt(sb, QUOTA);
1301         return 1;
1302 errout:
1303         kfree(qname);
1304         return ret;
1305 }
1306
1307 static int clear_qf_name(struct super_block *sb, int qtype)
1308 {
1309
1310         struct ext4_sb_info *sbi = EXT4_SB(sb);
1311
1312         if (sb_any_quota_loaded(sb) &&
1313                 sbi->s_qf_names[qtype]) {
1314                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1315                         " when quota turned on");
1316                 return -1;
1317         }
1318         kfree(sbi->s_qf_names[qtype]);
1319         sbi->s_qf_names[qtype] = NULL;
1320         return 1;
1321 }
1322 #endif
1323
1324 #define MOPT_SET        0x0001
1325 #define MOPT_CLEAR      0x0002
1326 #define MOPT_NOSUPPORT  0x0004
1327 #define MOPT_EXPLICIT   0x0008
1328 #define MOPT_CLEAR_ERR  0x0010
1329 #define MOPT_GTE0       0x0020
1330 #ifdef CONFIG_QUOTA
1331 #define MOPT_Q          0
1332 #define MOPT_QFMT       0x0040
1333 #else
1334 #define MOPT_Q          MOPT_NOSUPPORT
1335 #define MOPT_QFMT       MOPT_NOSUPPORT
1336 #endif
1337 #define MOPT_DATAJ      0x0080
1338 #define MOPT_NO_EXT2    0x0100
1339 #define MOPT_NO_EXT3    0x0200
1340 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1341
1342 static const struct mount_opts {
1343         int     token;
1344         int     mount_opt;
1345         int     flags;
1346 } ext4_mount_opts[] = {
1347         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1348         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1349         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1350         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1351         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1352         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1353         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1354          MOPT_EXT4_ONLY | MOPT_SET},
1355         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1356          MOPT_EXT4_ONLY | MOPT_CLEAR},
1357         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1358         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1359         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1360          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1361         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1362          MOPT_EXT4_ONLY | MOPT_CLEAR | MOPT_EXPLICIT},
1363         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1364          MOPT_EXT4_ONLY | MOPT_SET},
1365         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1366                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1367          MOPT_EXT4_ONLY | MOPT_SET},
1368         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1369         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1370         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1371         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1372         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1373          MOPT_NO_EXT2 | MOPT_SET},
1374         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1375          MOPT_NO_EXT2 | MOPT_CLEAR},
1376         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1377         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1378         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1379         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1380         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1381         {Opt_commit, 0, MOPT_GTE0},
1382         {Opt_max_batch_time, 0, MOPT_GTE0},
1383         {Opt_min_batch_time, 0, MOPT_GTE0},
1384         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1385         {Opt_init_itable, 0, MOPT_GTE0},
1386         {Opt_stripe, 0, MOPT_GTE0},
1387         {Opt_resuid, 0, MOPT_GTE0},
1388         {Opt_resgid, 0, MOPT_GTE0},
1389         {Opt_journal_dev, 0, MOPT_GTE0},
1390         {Opt_journal_ioprio, 0, MOPT_GTE0},
1391         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1392         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1393         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1394          MOPT_NO_EXT2 | MOPT_DATAJ},
1395         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1396         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1397 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1398         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1399         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1400 #else
1401         {Opt_acl, 0, MOPT_NOSUPPORT},
1402         {Opt_noacl, 0, MOPT_NOSUPPORT},
1403 #endif
1404         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1405         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1406         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1407         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1408                                                         MOPT_SET | MOPT_Q},
1409         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1410                                                         MOPT_SET | MOPT_Q},
1411         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1412                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1413         {Opt_usrjquota, 0, MOPT_Q},
1414         {Opt_grpjquota, 0, MOPT_Q},
1415         {Opt_offusrjquota, 0, MOPT_Q},
1416         {Opt_offgrpjquota, 0, MOPT_Q},
1417         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1418         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1419         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1420         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1421         {Opt_err, 0, 0}
1422 };
1423
1424 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1425                             substring_t *args, unsigned long *journal_devnum,
1426                             unsigned int *journal_ioprio, int is_remount)
1427 {
1428         struct ext4_sb_info *sbi = EXT4_SB(sb);
1429         const struct mount_opts *m;
1430         kuid_t uid;
1431         kgid_t gid;
1432         int arg = 0;
1433
1434 #ifdef CONFIG_QUOTA
1435         if (token == Opt_usrjquota)
1436                 return set_qf_name(sb, USRQUOTA, &args[0]);
1437         else if (token == Opt_grpjquota)
1438                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1439         else if (token == Opt_offusrjquota)
1440                 return clear_qf_name(sb, USRQUOTA);
1441         else if (token == Opt_offgrpjquota)
1442                 return clear_qf_name(sb, GRPQUOTA);
1443 #endif
1444         switch (token) {
1445         case Opt_noacl:
1446         case Opt_nouser_xattr:
1447                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1448                 break;
1449         case Opt_sb:
1450                 return 1;       /* handled by get_sb_block() */
1451         case Opt_removed:
1452                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1453                 return 1;
1454         case Opt_abort:
1455                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1456                 return 1;
1457         case Opt_i_version:
1458                 sb->s_flags |= MS_I_VERSION;
1459                 return 1;
1460         }
1461
1462         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1463                 if (token == m->token)
1464                         break;
1465
1466         if (m->token == Opt_err) {
1467                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1468                          "or missing value", opt);
1469                 return -1;
1470         }
1471
1472         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1473                 ext4_msg(sb, KERN_ERR,
1474                          "Mount option \"%s\" incompatible with ext2", opt);
1475                 return -1;
1476         }
1477         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1478                 ext4_msg(sb, KERN_ERR,
1479                          "Mount option \"%s\" incompatible with ext3", opt);
1480                 return -1;
1481         }
1482
1483         if (args->from && match_int(args, &arg))
1484                 return -1;
1485         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1486                 return -1;
1487         if (m->flags & MOPT_EXPLICIT)
1488                 set_opt2(sb, EXPLICIT_DELALLOC);
1489         if (m->flags & MOPT_CLEAR_ERR)
1490                 clear_opt(sb, ERRORS_MASK);
1491         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1492                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1493                          "options when quota turned on");
1494                 return -1;
1495         }
1496
1497         if (m->flags & MOPT_NOSUPPORT) {
1498                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1499         } else if (token == Opt_commit) {
1500                 if (arg == 0)
1501                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1502                 sbi->s_commit_interval = HZ * arg;
1503         } else if (token == Opt_max_batch_time) {
1504                 if (arg == 0)
1505                         arg = EXT4_DEF_MAX_BATCH_TIME;
1506                 sbi->s_max_batch_time = arg;
1507         } else if (token == Opt_min_batch_time) {
1508                 sbi->s_min_batch_time = arg;
1509         } else if (token == Opt_inode_readahead_blks) {
1510                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1511                         ext4_msg(sb, KERN_ERR,
1512                                  "EXT4-fs: inode_readahead_blks must be "
1513                                  "0 or a power of 2 smaller than 2^31");
1514                         return -1;
1515                 }
1516                 sbi->s_inode_readahead_blks = arg;
1517         } else if (token == Opt_init_itable) {
1518                 set_opt(sb, INIT_INODE_TABLE);
1519                 if (!args->from)
1520                         arg = EXT4_DEF_LI_WAIT_MULT;
1521                 sbi->s_li_wait_mult = arg;
1522         } else if (token == Opt_max_dir_size_kb) {
1523                 sbi->s_max_dir_size_kb = arg;
1524         } else if (token == Opt_stripe) {
1525                 sbi->s_stripe = arg;
1526         } else if (token == Opt_resuid) {
1527                 uid = make_kuid(current_user_ns(), arg);
1528                 if (!uid_valid(uid)) {
1529                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1530                         return -1;
1531                 }
1532                 sbi->s_resuid = uid;
1533         } else if (token == Opt_resgid) {
1534                 gid = make_kgid(current_user_ns(), arg);
1535                 if (!gid_valid(gid)) {
1536                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1537                         return -1;
1538                 }
1539                 sbi->s_resgid = gid;
1540         } else if (token == Opt_journal_dev) {
1541                 if (is_remount) {
1542                         ext4_msg(sb, KERN_ERR,
1543                                  "Cannot specify journal on remount");
1544                         return -1;
1545                 }
1546                 *journal_devnum = arg;
1547         } else if (token == Opt_journal_ioprio) {
1548                 if (arg > 7) {
1549                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1550                                  " (must be 0-7)");
1551                         return -1;
1552                 }
1553                 *journal_ioprio =
1554                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1555         } else if (m->flags & MOPT_DATAJ) {
1556                 if (is_remount) {
1557                         if (!sbi->s_journal)
1558                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1559                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1560                                 ext4_msg(sb, KERN_ERR,
1561                                          "Cannot change data mode on remount");
1562                                 return -1;
1563                         }
1564                 } else {
1565                         clear_opt(sb, DATA_FLAGS);
1566                         sbi->s_mount_opt |= m->mount_opt;
1567                 }
1568 #ifdef CONFIG_QUOTA
1569         } else if (m->flags & MOPT_QFMT) {
1570                 if (sb_any_quota_loaded(sb) &&
1571                     sbi->s_jquota_fmt != m->mount_opt) {
1572                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1573                                  "quota options when quota turned on");
1574                         return -1;
1575                 }
1576                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
1577                                                EXT4_FEATURE_RO_COMPAT_QUOTA)) {
1578                         ext4_msg(sb, KERN_ERR,
1579                                  "Cannot set journaled quota options "
1580                                  "when QUOTA feature is enabled");
1581                         return -1;
1582                 }
1583                 sbi->s_jquota_fmt = m->mount_opt;
1584 #endif
1585         } else {
1586                 if (!args->from)
1587                         arg = 1;
1588                 if (m->flags & MOPT_CLEAR)
1589                         arg = !arg;
1590                 else if (unlikely(!(m->flags & MOPT_SET))) {
1591                         ext4_msg(sb, KERN_WARNING,
1592                                  "buggy handling of option %s", opt);
1593                         WARN_ON(1);
1594                         return -1;
1595                 }
1596                 if (arg != 0)
1597                         sbi->s_mount_opt |= m->mount_opt;
1598                 else
1599                         sbi->s_mount_opt &= ~m->mount_opt;
1600         }
1601         return 1;
1602 }
1603
1604 static int parse_options(char *options, struct super_block *sb,
1605                          unsigned long *journal_devnum,
1606                          unsigned int *journal_ioprio,
1607                          int is_remount)
1608 {
1609         struct ext4_sb_info *sbi = EXT4_SB(sb);
1610         char *p;
1611         substring_t args[MAX_OPT_ARGS];
1612         int token;
1613
1614         if (!options)
1615                 return 1;
1616
1617         while ((p = strsep(&options, ",")) != NULL) {
1618                 if (!*p)
1619                         continue;
1620                 /*
1621                  * Initialize args struct so we know whether arg was
1622                  * found; some options take optional arguments.
1623                  */
1624                 args[0].to = args[0].from = NULL;
1625                 token = match_token(p, tokens, args);
1626                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1627                                      journal_ioprio, is_remount) < 0)
1628                         return 0;
1629         }
1630 #ifdef CONFIG_QUOTA
1631         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
1632             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1633                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1634                          "feature is enabled");
1635                 return 0;
1636         }
1637         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1638                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1639                         clear_opt(sb, USRQUOTA);
1640
1641                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1642                         clear_opt(sb, GRPQUOTA);
1643
1644                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1645                         ext4_msg(sb, KERN_ERR, "old and new quota "
1646                                         "format mixing");
1647                         return 0;
1648                 }
1649
1650                 if (!sbi->s_jquota_fmt) {
1651                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1652                                         "not specified");
1653                         return 0;
1654                 }
1655         } else {
1656                 if (sbi->s_jquota_fmt) {
1657                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1658                                         "specified with no journaling "
1659                                         "enabled");
1660                         return 0;
1661                 }
1662         }
1663 #endif
1664         if (test_opt(sb, DIOREAD_NOLOCK)) {
1665                 int blocksize =
1666                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1667
1668                 if (blocksize < PAGE_CACHE_SIZE) {
1669                         ext4_msg(sb, KERN_ERR, "can't mount with "
1670                                  "dioread_nolock if block size != PAGE_SIZE");
1671                         return 0;
1672                 }
1673         }
1674         return 1;
1675 }
1676
1677 static inline void ext4_show_quota_options(struct seq_file *seq,
1678                                            struct super_block *sb)
1679 {
1680 #if defined(CONFIG_QUOTA)
1681         struct ext4_sb_info *sbi = EXT4_SB(sb);
1682
1683         if (sbi->s_jquota_fmt) {
1684                 char *fmtname = "";
1685
1686                 switch (sbi->s_jquota_fmt) {
1687                 case QFMT_VFS_OLD:
1688                         fmtname = "vfsold";
1689                         break;
1690                 case QFMT_VFS_V0:
1691                         fmtname = "vfsv0";
1692                         break;
1693                 case QFMT_VFS_V1:
1694                         fmtname = "vfsv1";
1695                         break;
1696                 }
1697                 seq_printf(seq, ",jqfmt=%s", fmtname);
1698         }
1699
1700         if (sbi->s_qf_names[USRQUOTA])
1701                 seq_printf(seq, ",usrjquota=%s", sbi->s_qf_names[USRQUOTA]);
1702
1703         if (sbi->s_qf_names[GRPQUOTA])
1704                 seq_printf(seq, ",grpjquota=%s", sbi->s_qf_names[GRPQUOTA]);
1705
1706         if (test_opt(sb, USRQUOTA))
1707                 seq_puts(seq, ",usrquota");
1708
1709         if (test_opt(sb, GRPQUOTA))
1710                 seq_puts(seq, ",grpquota");
1711 #endif
1712 }
1713
1714 static const char *token2str(int token)
1715 {
1716         const struct match_token *t;
1717
1718         for (t = tokens; t->token != Opt_err; t++)
1719                 if (t->token == token && !strchr(t->pattern, '='))
1720                         break;
1721         return t->pattern;
1722 }
1723
1724 /*
1725  * Show an option if
1726  *  - it's set to a non-default value OR
1727  *  - if the per-sb default is different from the global default
1728  */
1729 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1730                               int nodefs)
1731 {
1732         struct ext4_sb_info *sbi = EXT4_SB(sb);
1733         struct ext4_super_block *es = sbi->s_es;
1734         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1735         const struct mount_opts *m;
1736         char sep = nodefs ? '\n' : ',';
1737
1738 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1739 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1740
1741         if (sbi->s_sb_block != 1)
1742                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1743
1744         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1745                 int want_set = m->flags & MOPT_SET;
1746                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1747                     (m->flags & MOPT_CLEAR_ERR))
1748                         continue;
1749                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1750                         continue; /* skip if same as the default */
1751                 if ((want_set &&
1752                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1753                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1754                         continue; /* select Opt_noFoo vs Opt_Foo */
1755                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1756         }
1757
1758         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1759             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1760                 SEQ_OPTS_PRINT("resuid=%u",
1761                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1762         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1763             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1764                 SEQ_OPTS_PRINT("resgid=%u",
1765                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1766         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1767         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1768                 SEQ_OPTS_PUTS("errors=remount-ro");
1769         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1770                 SEQ_OPTS_PUTS("errors=continue");
1771         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1772                 SEQ_OPTS_PUTS("errors=panic");
1773         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1774                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1775         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1776                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1777         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1778                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1779         if (sb->s_flags & MS_I_VERSION)
1780                 SEQ_OPTS_PUTS("i_version");
1781         if (nodefs || sbi->s_stripe)
1782                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1783         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1784                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1785                         SEQ_OPTS_PUTS("data=journal");
1786                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1787                         SEQ_OPTS_PUTS("data=ordered");
1788                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1789                         SEQ_OPTS_PUTS("data=writeback");
1790         }
1791         if (nodefs ||
1792             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1793                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1794                                sbi->s_inode_readahead_blks);
1795
1796         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1797                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1798                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1799         if (nodefs || sbi->s_max_dir_size_kb)
1800                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1801
1802         ext4_show_quota_options(seq, sb);
1803         return 0;
1804 }
1805
1806 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1807 {
1808         return _ext4_show_options(seq, root->d_sb, 0);
1809 }
1810
1811 static int options_seq_show(struct seq_file *seq, void *offset)
1812 {
1813         struct super_block *sb = seq->private;
1814         int rc;
1815
1816         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1817         rc = _ext4_show_options(seq, sb, 1);
1818         seq_puts(seq, "\n");
1819         return rc;
1820 }
1821
1822 static int options_open_fs(struct inode *inode, struct file *file)
1823 {
1824         return single_open(file, options_seq_show, PDE_DATA(inode));
1825 }
1826
1827 static const struct file_operations ext4_seq_options_fops = {
1828         .owner = THIS_MODULE,
1829         .open = options_open_fs,
1830         .read = seq_read,
1831         .llseek = seq_lseek,
1832         .release = single_release,
1833 };
1834
1835 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1836                             int read_only)
1837 {
1838         struct ext4_sb_info *sbi = EXT4_SB(sb);
1839         int res = 0;
1840
1841         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1842                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1843                          "forcing read-only mode");
1844                 res = MS_RDONLY;
1845         }
1846         if (read_only)
1847                 goto done;
1848         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1849                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1850                          "running e2fsck is recommended");
1851         else if ((sbi->s_mount_state & EXT4_ERROR_FS))
1852                 ext4_msg(sb, KERN_WARNING,
1853                          "warning: mounting fs with errors, "
1854                          "running e2fsck is recommended");
1855         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1856                  le16_to_cpu(es->s_mnt_count) >=
1857                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1858                 ext4_msg(sb, KERN_WARNING,
1859                          "warning: maximal mount count reached, "
1860                          "running e2fsck is recommended");
1861         else if (le32_to_cpu(es->s_checkinterval) &&
1862                 (le32_to_cpu(es->s_lastcheck) +
1863                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1864                 ext4_msg(sb, KERN_WARNING,
1865                          "warning: checktime reached, "
1866                          "running e2fsck is recommended");
1867         if (!sbi->s_journal)
1868                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1869         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1870                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1871         le16_add_cpu(&es->s_mnt_count, 1);
1872         es->s_mtime = cpu_to_le32(get_seconds());
1873         ext4_update_dynamic_rev(sb);
1874         if (sbi->s_journal)
1875                 EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
1876
1877         ext4_commit_super(sb, 1);
1878 done:
1879         if (test_opt(sb, DEBUG))
1880                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1881                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1882                         sb->s_blocksize,
1883                         sbi->s_groups_count,
1884                         EXT4_BLOCKS_PER_GROUP(sb),
1885                         EXT4_INODES_PER_GROUP(sb),
1886                         sbi->s_mount_opt, sbi->s_mount_opt2);
1887
1888         cleancache_init_fs(sb);
1889         return res;
1890 }
1891
1892 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1893 {
1894         struct ext4_sb_info *sbi = EXT4_SB(sb);
1895         struct flex_groups *new_groups;
1896         int size;
1897
1898         if (!sbi->s_log_groups_per_flex)
1899                 return 0;
1900
1901         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1902         if (size <= sbi->s_flex_groups_allocated)
1903                 return 0;
1904
1905         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1906         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1907         if (!new_groups) {
1908                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1909                          size / (int) sizeof(struct flex_groups));
1910                 return -ENOMEM;
1911         }
1912
1913         if (sbi->s_flex_groups) {
1914                 memcpy(new_groups, sbi->s_flex_groups,
1915                        (sbi->s_flex_groups_allocated *
1916                         sizeof(struct flex_groups)));
1917                 ext4_kvfree(sbi->s_flex_groups);
1918         }
1919         sbi->s_flex_groups = new_groups;
1920         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1921         return 0;
1922 }
1923
1924 static int ext4_fill_flex_info(struct super_block *sb)
1925 {
1926         struct ext4_sb_info *sbi = EXT4_SB(sb);
1927         struct ext4_group_desc *gdp = NULL;
1928         ext4_group_t flex_group;
1929         int i, err;
1930
1931         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1932         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
1933                 sbi->s_log_groups_per_flex = 0;
1934                 return 1;
1935         }
1936
1937         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
1938         if (err)
1939                 goto failed;
1940
1941         for (i = 0; i < sbi->s_groups_count; i++) {
1942                 gdp = ext4_get_group_desc(sb, i, NULL);
1943
1944                 flex_group = ext4_flex_group(sbi, i);
1945                 atomic_add(ext4_free_inodes_count(sb, gdp),
1946                            &sbi->s_flex_groups[flex_group].free_inodes);
1947                 atomic64_add(ext4_free_group_clusters(sb, gdp),
1948                              &sbi->s_flex_groups[flex_group].free_clusters);
1949                 atomic_add(ext4_used_dirs_count(sb, gdp),
1950                            &sbi->s_flex_groups[flex_group].used_dirs);
1951         }
1952
1953         return 1;
1954 failed:
1955         return 0;
1956 }
1957
1958 static __le16 ext4_group_desc_csum(struct ext4_sb_info *sbi, __u32 block_group,
1959                                    struct ext4_group_desc *gdp)
1960 {
1961         int offset;
1962         __u16 crc = 0;
1963         __le32 le_group = cpu_to_le32(block_group);
1964
1965         if ((sbi->s_es->s_feature_ro_compat &
1966              cpu_to_le32(EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))) {
1967                 /* Use new metadata_csum algorithm */
1968                 __le16 save_csum;
1969                 __u32 csum32;
1970
1971                 save_csum = gdp->bg_checksum;
1972                 gdp->bg_checksum = 0;
1973                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
1974                                      sizeof(le_group));
1975                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
1976                                      sbi->s_desc_size);
1977                 gdp->bg_checksum = save_csum;
1978
1979                 crc = csum32 & 0xFFFF;
1980                 goto out;
1981         }
1982
1983         /* old crc16 code */
1984         offset = offsetof(struct ext4_group_desc, bg_checksum);
1985
1986         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
1987         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
1988         crc = crc16(crc, (__u8 *)gdp, offset);
1989         offset += sizeof(gdp->bg_checksum); /* skip checksum */
1990         /* for checksum of struct ext4_group_desc do the rest...*/
1991         if ((sbi->s_es->s_feature_incompat &
1992              cpu_to_le32(EXT4_FEATURE_INCOMPAT_64BIT)) &&
1993             offset < le16_to_cpu(sbi->s_es->s_desc_size))
1994                 crc = crc16(crc, (__u8 *)gdp + offset,
1995                             le16_to_cpu(sbi->s_es->s_desc_size) -
1996                                 offset);
1997
1998 out:
1999         return cpu_to_le16(crc);
2000 }
2001
2002 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2003                                 struct ext4_group_desc *gdp)
2004 {
2005         if (ext4_has_group_desc_csum(sb) &&
2006             (gdp->bg_checksum != ext4_group_desc_csum(EXT4_SB(sb),
2007                                                       block_group, gdp)))
2008                 return 0;
2009
2010         return 1;
2011 }
2012
2013 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2014                               struct ext4_group_desc *gdp)
2015 {
2016         if (!ext4_has_group_desc_csum(sb))
2017                 return;
2018         gdp->bg_checksum = ext4_group_desc_csum(EXT4_SB(sb), block_group, gdp);
2019 }
2020
2021 /* Called at mount-time, super-block is locked */
2022 static int ext4_check_descriptors(struct super_block *sb,
2023                                   ext4_group_t *first_not_zeroed)
2024 {
2025         struct ext4_sb_info *sbi = EXT4_SB(sb);
2026         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2027         ext4_fsblk_t last_block;
2028         ext4_fsblk_t block_bitmap;
2029         ext4_fsblk_t inode_bitmap;
2030         ext4_fsblk_t inode_table;
2031         int flexbg_flag = 0;
2032         ext4_group_t i, grp = sbi->s_groups_count;
2033
2034         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
2035                 flexbg_flag = 1;
2036
2037         ext4_debug("Checking group descriptors");
2038
2039         for (i = 0; i < sbi->s_groups_count; i++) {
2040                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2041
2042                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2043                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2044                 else
2045                         last_block = first_block +
2046                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2047
2048                 if ((grp == sbi->s_groups_count) &&
2049                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2050                         grp = i;
2051
2052                 block_bitmap = ext4_block_bitmap(sb, gdp);
2053                 if (block_bitmap < first_block || block_bitmap > last_block) {
2054                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2055                                "Block bitmap for group %u not in group "
2056                                "(block %llu)!", i, block_bitmap);
2057                         return 0;
2058                 }
2059                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2060                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2061                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2062                                "Inode bitmap for group %u not in group "
2063                                "(block %llu)!", i, inode_bitmap);
2064                         return 0;
2065                 }
2066                 inode_table = ext4_inode_table(sb, gdp);
2067                 if (inode_table < first_block ||
2068                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2069                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2070                                "Inode table for group %u not in group "
2071                                "(block %llu)!", i, inode_table);
2072                         return 0;
2073                 }
2074                 ext4_lock_group(sb, i);
2075                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2076                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2077                                  "Checksum for group %u failed (%u!=%u)",
2078                                  i, le16_to_cpu(ext4_group_desc_csum(sbi, i,
2079                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2080                         if (!(sb->s_flags & MS_RDONLY)) {
2081                                 ext4_unlock_group(sb, i);
2082                                 return 0;
2083                         }
2084                 }
2085                 ext4_unlock_group(sb, i);
2086                 if (!flexbg_flag)
2087                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2088         }
2089         if (NULL != first_not_zeroed)
2090                 *first_not_zeroed = grp;
2091
2092         ext4_free_blocks_count_set(sbi->s_es,
2093                                    EXT4_C2B(sbi, ext4_count_free_clusters(sb)));
2094         sbi->s_es->s_free_inodes_count =cpu_to_le32(ext4_count_free_inodes(sb));
2095         return 1;
2096 }
2097
2098 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2099  * the superblock) which were deleted from all directories, but held open by
2100  * a process at the time of a crash.  We walk the list and try to delete these
2101  * inodes at recovery time (only with a read-write filesystem).
2102  *
2103  * In order to keep the orphan inode chain consistent during traversal (in
2104  * case of crash during recovery), we link each inode into the superblock
2105  * orphan list_head and handle it the same way as an inode deletion during
2106  * normal operation (which journals the operations for us).
2107  *
2108  * We only do an iget() and an iput() on each inode, which is very safe if we
2109  * accidentally point at an in-use or already deleted inode.  The worst that
2110  * can happen in this case is that we get a "bit already cleared" message from
2111  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2112  * e2fsck was run on this filesystem, and it must have already done the orphan
2113  * inode cleanup for us, so we can safely abort without any further action.
2114  */
2115 static void ext4_orphan_cleanup(struct super_block *sb,
2116                                 struct ext4_super_block *es)
2117 {
2118         unsigned int s_flags = sb->s_flags;
2119         int nr_orphans = 0, nr_truncates = 0;
2120 #ifdef CONFIG_QUOTA
2121         int i;
2122 #endif
2123         if (!es->s_last_orphan) {
2124                 jbd_debug(4, "no orphan inodes to clean up\n");
2125                 return;
2126         }
2127
2128         if (bdev_read_only(sb->s_bdev)) {
2129                 ext4_msg(sb, KERN_ERR, "write access "
2130                         "unavailable, skipping orphan cleanup");
2131                 return;
2132         }
2133
2134         /* Check if feature set would not allow a r/w mount */
2135         if (!ext4_feature_set_ok(sb, 0)) {
2136                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2137                          "unknown ROCOMPAT features");
2138                 return;
2139         }
2140
2141         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2142                 /* don't clear list on RO mount w/ errors */
2143                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2144                         jbd_debug(1, "Errors on filesystem, "
2145                                   "clearing orphan list.\n");
2146                         es->s_last_orphan = 0;
2147                 }
2148                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2149                 return;
2150         }
2151
2152         if (s_flags & MS_RDONLY) {
2153                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2154                 sb->s_flags &= ~MS_RDONLY;
2155         }
2156 #ifdef CONFIG_QUOTA
2157         /* Needed for iput() to work correctly and not trash data */
2158         sb->s_flags |= MS_ACTIVE;
2159         /* Turn on quotas so that they are updated correctly */
2160         for (i = 0; i < MAXQUOTAS; i++) {
2161                 if (EXT4_SB(sb)->s_qf_names[i]) {
2162                         int ret = ext4_quota_on_mount(sb, i);
2163                         if (ret < 0)
2164                                 ext4_msg(sb, KERN_ERR,
2165                                         "Cannot turn on journaled "
2166                                         "quota: error %d", ret);
2167                 }
2168         }
2169 #endif
2170
2171         while (es->s_last_orphan) {
2172                 struct inode *inode;
2173
2174                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2175                 if (IS_ERR(inode)) {
2176                         es->s_last_orphan = 0;
2177                         break;
2178                 }
2179
2180                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2181                 dquot_initialize(inode);
2182                 if (inode->i_nlink) {
2183                         if (test_opt(sb, DEBUG))
2184                                 ext4_msg(sb, KERN_DEBUG,
2185                                         "%s: truncating inode %lu to %lld bytes",
2186                                         __func__, inode->i_ino, inode->i_size);
2187                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2188                                   inode->i_ino, inode->i_size);
2189                         mutex_lock(&inode->i_mutex);
2190                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2191                         ext4_truncate(inode);
2192                         mutex_unlock(&inode->i_mutex);
2193                         nr_truncates++;
2194                 } else {
2195                         if (test_opt(sb, DEBUG))
2196                                 ext4_msg(sb, KERN_DEBUG,
2197                                         "%s: deleting unreferenced inode %lu",
2198                                         __func__, inode->i_ino);
2199                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2200                                   inode->i_ino);
2201                         nr_orphans++;
2202                 }
2203                 iput(inode);  /* The delete magic happens here! */
2204         }
2205
2206 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2207
2208         if (nr_orphans)
2209                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2210                        PLURAL(nr_orphans));
2211         if (nr_truncates)
2212                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2213                        PLURAL(nr_truncates));
2214 #ifdef CONFIG_QUOTA
2215         /* Turn quotas off */
2216         for (i = 0; i < MAXQUOTAS; i++) {
2217                 if (sb_dqopt(sb)->files[i])
2218                         dquot_quota_off(sb, i);
2219         }
2220 #endif
2221         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2222 }
2223
2224 /*
2225  * Maximal extent format file size.
2226  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2227  * extent format containers, within a sector_t, and within i_blocks
2228  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2229  * so that won't be a limiting factor.
2230  *
2231  * However there is other limiting factor. We do store extents in the form
2232  * of starting block and length, hence the resulting length of the extent
2233  * covering maximum file size must fit into on-disk format containers as
2234  * well. Given that length is always by 1 unit bigger than max unit (because
2235  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2236  *
2237  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2238  */
2239 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2240 {
2241         loff_t res;
2242         loff_t upper_limit = MAX_LFS_FILESIZE;
2243
2244         /* small i_blocks in vfs inode? */
2245         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2246                 /*
2247                  * CONFIG_LBDAF is not enabled implies the inode
2248                  * i_block represent total blocks in 512 bytes
2249                  * 32 == size of vfs inode i_blocks * 8
2250                  */
2251                 upper_limit = (1LL << 32) - 1;
2252
2253                 /* total blocks in file system block size */
2254                 upper_limit >>= (blkbits - 9);
2255                 upper_limit <<= blkbits;
2256         }
2257
2258         /*
2259          * 32-bit extent-start container, ee_block. We lower the maxbytes
2260          * by one fs block, so ee_len can cover the extent of maximum file
2261          * size
2262          */
2263         res = (1LL << 32) - 1;
2264         res <<= blkbits;
2265
2266         /* Sanity check against vm- & vfs- imposed limits */
2267         if (res > upper_limit)
2268                 res = upper_limit;
2269
2270         return res;
2271 }
2272
2273 /*
2274  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2275  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2276  * We need to be 1 filesystem block less than the 2^48 sector limit.
2277  */
2278 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2279 {
2280         loff_t res = EXT4_NDIR_BLOCKS;
2281         int meta_blocks;
2282         loff_t upper_limit;
2283         /* This is calculated to be the largest file size for a dense, block
2284          * mapped file such that the file's total number of 512-byte sectors,
2285          * including data and all indirect blocks, does not exceed (2^48 - 1).
2286          *
2287          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2288          * number of 512-byte sectors of the file.
2289          */
2290
2291         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2292                 /*
2293                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2294                  * the inode i_block field represents total file blocks in
2295                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2296                  */
2297                 upper_limit = (1LL << 32) - 1;
2298
2299                 /* total blocks in file system block size */
2300                 upper_limit >>= (bits - 9);
2301
2302         } else {
2303                 /*
2304                  * We use 48 bit ext4_inode i_blocks
2305                  * With EXT4_HUGE_FILE_FL set the i_blocks
2306                  * represent total number of blocks in
2307                  * file system block size
2308                  */
2309                 upper_limit = (1LL << 48) - 1;
2310
2311         }
2312
2313         /* indirect blocks */
2314         meta_blocks = 1;
2315         /* double indirect blocks */
2316         meta_blocks += 1 + (1LL << (bits-2));
2317         /* tripple indirect blocks */
2318         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2319
2320         upper_limit -= meta_blocks;
2321         upper_limit <<= bits;
2322
2323         res += 1LL << (bits-2);
2324         res += 1LL << (2*(bits-2));
2325         res += 1LL << (3*(bits-2));
2326         res <<= bits;
2327         if (res > upper_limit)
2328                 res = upper_limit;
2329
2330         if (res > MAX_LFS_FILESIZE)
2331                 res = MAX_LFS_FILESIZE;
2332
2333         return res;
2334 }
2335
2336 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2337                                    ext4_fsblk_t logical_sb_block, int nr)
2338 {
2339         struct ext4_sb_info *sbi = EXT4_SB(sb);
2340         ext4_group_t bg, first_meta_bg;
2341         int has_super = 0;
2342
2343         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2344
2345         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_META_BG) ||
2346             nr < first_meta_bg)
2347                 return logical_sb_block + nr + 1;
2348         bg = sbi->s_desc_per_block * nr;
2349         if (ext4_bg_has_super(sb, bg))
2350                 has_super = 1;
2351
2352         return (has_super + ext4_group_first_block_no(sb, bg));
2353 }
2354
2355 /**
2356  * ext4_get_stripe_size: Get the stripe size.
2357  * @sbi: In memory super block info
2358  *
2359  * If we have specified it via mount option, then
2360  * use the mount option value. If the value specified at mount time is
2361  * greater than the blocks per group use the super block value.
2362  * If the super block value is greater than blocks per group return 0.
2363  * Allocator needs it be less than blocks per group.
2364  *
2365  */
2366 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2367 {
2368         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2369         unsigned long stripe_width =
2370                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2371         int ret;
2372
2373         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2374                 ret = sbi->s_stripe;
2375         else if (stripe_width <= sbi->s_blocks_per_group)
2376                 ret = stripe_width;
2377         else if (stride <= sbi->s_blocks_per_group)
2378                 ret = stride;
2379         else
2380                 ret = 0;
2381
2382         /*
2383          * If the stripe width is 1, this makes no sense and
2384          * we set it to 0 to turn off stripe handling code.
2385          */
2386         if (ret <= 1)
2387                 ret = 0;
2388
2389         return ret;
2390 }
2391
2392 /* sysfs supprt */
2393
2394 struct ext4_attr {
2395         struct attribute attr;
2396         ssize_t (*show)(struct ext4_attr *, struct ext4_sb_info *, char *);
2397         ssize_t (*store)(struct ext4_attr *, struct ext4_sb_info *,
2398                          const char *, size_t);
2399         union {
2400                 int offset;
2401                 int deprecated_val;
2402         } u;
2403 };
2404
2405 static int parse_strtoull(const char *buf,
2406                 unsigned long long max, unsigned long long *value)
2407 {
2408         int ret;
2409
2410         ret = kstrtoull(skip_spaces(buf), 0, value);
2411         if (!ret && *value > max)
2412                 ret = -EINVAL;
2413         return ret;
2414 }
2415
2416 static ssize_t delayed_allocation_blocks_show(struct ext4_attr *a,
2417                                               struct ext4_sb_info *sbi,
2418                                               char *buf)
2419 {
2420         return snprintf(buf, PAGE_SIZE, "%llu\n",
2421                 (s64) EXT4_C2B(sbi,
2422                         percpu_counter_sum(&sbi->s_dirtyclusters_counter)));
2423 }
2424
2425 static ssize_t session_write_kbytes_show(struct ext4_attr *a,
2426                                          struct ext4_sb_info *sbi, char *buf)
2427 {
2428         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2429
2430         if (!sb->s_bdev->bd_part)
2431                 return snprintf(buf, PAGE_SIZE, "0\n");
2432         return snprintf(buf, PAGE_SIZE, "%lu\n",
2433                         (part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2434                          sbi->s_sectors_written_start) >> 1);
2435 }
2436
2437 static ssize_t lifetime_write_kbytes_show(struct ext4_attr *a,
2438                                           struct ext4_sb_info *sbi, char *buf)
2439 {
2440         struct super_block *sb = sbi->s_buddy_cache->i_sb;
2441
2442         if (!sb->s_bdev->bd_part)
2443                 return snprintf(buf, PAGE_SIZE, "0\n");
2444         return snprintf(buf, PAGE_SIZE, "%llu\n",
2445                         (unsigned long long)(sbi->s_kbytes_written +
2446                         ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
2447                           EXT4_SB(sb)->s_sectors_written_start) >> 1)));
2448 }
2449
2450 static ssize_t inode_readahead_blks_store(struct ext4_attr *a,
2451                                           struct ext4_sb_info *sbi,
2452                                           const char *buf, size_t count)
2453 {
2454         unsigned long t;
2455         int ret;
2456
2457         ret = kstrtoul(skip_spaces(buf), 0, &t);
2458         if (ret)
2459                 return ret;
2460
2461         if (t && (!is_power_of_2(t) || t > 0x40000000))
2462                 return -EINVAL;
2463
2464         sbi->s_inode_readahead_blks = t;
2465         return count;
2466 }
2467
2468 static ssize_t sbi_ui_show(struct ext4_attr *a,
2469                            struct ext4_sb_info *sbi, char *buf)
2470 {
2471         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2472
2473         return snprintf(buf, PAGE_SIZE, "%u\n", *ui);
2474 }
2475
2476 static ssize_t sbi_ui_store(struct ext4_attr *a,
2477                             struct ext4_sb_info *sbi,
2478                             const char *buf, size_t count)
2479 {
2480         unsigned int *ui = (unsigned int *) (((char *) sbi) + a->u.offset);
2481         unsigned long t;
2482         int ret;
2483
2484         ret = kstrtoul(skip_spaces(buf), 0, &t);
2485         if (ret)
2486                 return ret;
2487         *ui = t;
2488         return count;
2489 }
2490
2491 static ssize_t reserved_clusters_show(struct ext4_attr *a,
2492                                   struct ext4_sb_info *sbi, char *buf)
2493 {
2494         return snprintf(buf, PAGE_SIZE, "%llu\n",
2495                 (unsigned long long) atomic64_read(&sbi->s_resv_clusters));
2496 }
2497
2498 static ssize_t reserved_clusters_store(struct ext4_attr *a,
2499                                    struct ext4_sb_info *sbi,
2500                                    const char *buf, size_t count)
2501 {
2502         unsigned long long val;
2503         int ret;
2504
2505         if (parse_strtoull(buf, -1ULL, &val))
2506                 return -EINVAL;
2507         ret = ext4_reserve_clusters(sbi, val);
2508
2509         return ret ? ret : count;
2510 }
2511
2512 static ssize_t trigger_test_error(struct ext4_attr *a,
2513                                   struct ext4_sb_info *sbi,
2514                                   const char *buf, size_t count)
2515 {
2516         int len = count;
2517
2518         if (!capable(CAP_SYS_ADMIN))
2519                 return -EPERM;
2520
2521         if (len && buf[len-1] == '\n')
2522                 len--;
2523
2524         if (len)
2525                 ext4_error(sbi->s_sb, "%.*s", len, buf);
2526         return count;
2527 }
2528
2529 static ssize_t sbi_deprecated_show(struct ext4_attr *a,
2530                                    struct ext4_sb_info *sbi, char *buf)
2531 {
2532         return snprintf(buf, PAGE_SIZE, "%d\n", a->u.deprecated_val);
2533 }
2534
2535 #define EXT4_ATTR_OFFSET(_name,_mode,_show,_store,_elname) \
2536 static struct ext4_attr ext4_attr_##_name = {                   \
2537         .attr = {.name = __stringify(_name), .mode = _mode },   \
2538         .show   = _show,                                        \
2539         .store  = _store,                                       \
2540         .u = {                                                  \
2541                 .offset = offsetof(struct ext4_sb_info, _elname),\
2542         },                                                      \
2543 }
2544 #define EXT4_ATTR(name, mode, show, store) \
2545 static struct ext4_attr ext4_attr_##name = __ATTR(name, mode, show, store)
2546
2547 #define EXT4_INFO_ATTR(name) EXT4_ATTR(name, 0444, NULL, NULL)
2548 #define EXT4_RO_ATTR(name) EXT4_ATTR(name, 0444, name##_show, NULL)
2549 #define EXT4_RW_ATTR(name) EXT4_ATTR(name, 0644, name##_show, name##_store)
2550 #define EXT4_RW_ATTR_SBI_UI(name, elname)       \
2551         EXT4_ATTR_OFFSET(name, 0644, sbi_ui_show, sbi_ui_store, elname)
2552 #define ATTR_LIST(name) &ext4_attr_##name.attr
2553 #define EXT4_DEPRECATED_ATTR(_name, _val)       \
2554 static struct ext4_attr ext4_attr_##_name = {                   \
2555         .attr = {.name = __stringify(_name), .mode = 0444 },    \
2556         .show   = sbi_deprecated_show,                          \
2557         .u = {                                                  \
2558                 .deprecated_val = _val,                         \
2559         },                                                      \
2560 }
2561
2562 EXT4_RO_ATTR(delayed_allocation_blocks);
2563 EXT4_RO_ATTR(session_write_kbytes);
2564 EXT4_RO_ATTR(lifetime_write_kbytes);
2565 EXT4_RW_ATTR(reserved_clusters);
2566 EXT4_ATTR_OFFSET(inode_readahead_blks, 0644, sbi_ui_show,
2567                  inode_readahead_blks_store, s_inode_readahead_blks);
2568 EXT4_RW_ATTR_SBI_UI(inode_goal, s_inode_goal);
2569 EXT4_RW_ATTR_SBI_UI(mb_stats, s_mb_stats);
2570 EXT4_RW_ATTR_SBI_UI(mb_max_to_scan, s_mb_max_to_scan);
2571 EXT4_RW_ATTR_SBI_UI(mb_min_to_scan, s_mb_min_to_scan);
2572 EXT4_RW_ATTR_SBI_UI(mb_order2_req, s_mb_order2_reqs);
2573 EXT4_RW_ATTR_SBI_UI(mb_stream_req, s_mb_stream_request);
2574 EXT4_RW_ATTR_SBI_UI(mb_group_prealloc, s_mb_group_prealloc);
2575 EXT4_DEPRECATED_ATTR(max_writeback_mb_bump, 128);
2576 EXT4_RW_ATTR_SBI_UI(extent_max_zeroout_kb, s_extent_max_zeroout_kb);
2577 EXT4_ATTR(trigger_fs_error, 0200, NULL, trigger_test_error);
2578
2579 static struct attribute *ext4_attrs[] = {
2580         ATTR_LIST(delayed_allocation_blocks),
2581         ATTR_LIST(session_write_kbytes),
2582         ATTR_LIST(lifetime_write_kbytes),
2583         ATTR_LIST(reserved_clusters),
2584         ATTR_LIST(inode_readahead_blks),
2585         ATTR_LIST(inode_goal),
2586         ATTR_LIST(mb_stats),
2587         ATTR_LIST(mb_max_to_scan),
2588         ATTR_LIST(mb_min_to_scan),
2589         ATTR_LIST(mb_order2_req),
2590         ATTR_LIST(mb_stream_req),
2591         ATTR_LIST(mb_group_prealloc),
2592         ATTR_LIST(max_writeback_mb_bump),
2593         ATTR_LIST(extent_max_zeroout_kb),
2594         ATTR_LIST(trigger_fs_error),
2595         NULL,
2596 };
2597
2598 /* Features this copy of ext4 supports */
2599 EXT4_INFO_ATTR(lazy_itable_init);
2600 EXT4_INFO_ATTR(batched_discard);
2601 EXT4_INFO_ATTR(meta_bg_resize);
2602
2603 static struct attribute *ext4_feat_attrs[] = {
2604         ATTR_LIST(lazy_itable_init),
2605         ATTR_LIST(batched_discard),
2606         ATTR_LIST(meta_bg_resize),
2607         NULL,
2608 };
2609
2610 static ssize_t ext4_attr_show(struct kobject *kobj,
2611                               struct attribute *attr, char *buf)
2612 {
2613         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2614                                                 s_kobj);
2615         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2616
2617         return a->show ? a->show(a, sbi, buf) : 0;
2618 }
2619
2620 static ssize_t ext4_attr_store(struct kobject *kobj,
2621                                struct attribute *attr,
2622                                const char *buf, size_t len)
2623 {
2624         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2625                                                 s_kobj);
2626         struct ext4_attr *a = container_of(attr, struct ext4_attr, attr);
2627
2628         return a->store ? a->store(a, sbi, buf, len) : 0;
2629 }
2630
2631 static void ext4_sb_release(struct kobject *kobj)
2632 {
2633         struct ext4_sb_info *sbi = container_of(kobj, struct ext4_sb_info,
2634                                                 s_kobj);
2635         complete(&sbi->s_kobj_unregister);
2636 }
2637
2638 static const struct sysfs_ops ext4_attr_ops = {
2639         .show   = ext4_attr_show,
2640         .store  = ext4_attr_store,
2641 };
2642
2643 static struct kobj_type ext4_ktype = {
2644         .default_attrs  = ext4_attrs,
2645         .sysfs_ops      = &ext4_attr_ops,
2646         .release        = ext4_sb_release,
2647 };
2648
2649 static void ext4_feat_release(struct kobject *kobj)
2650 {
2651         complete(&ext4_feat->f_kobj_unregister);
2652 }
2653
2654 static struct kobj_type ext4_feat_ktype = {
2655         .default_attrs  = ext4_feat_attrs,
2656         .sysfs_ops      = &ext4_attr_ops,
2657         .release        = ext4_feat_release,
2658 };
2659
2660 /*
2661  * Check whether this filesystem can be mounted based on
2662  * the features present and the RDONLY/RDWR mount requested.
2663  * Returns 1 if this filesystem can be mounted as requested,
2664  * 0 if it cannot be.
2665  */
2666 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2667 {
2668         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT4_FEATURE_INCOMPAT_SUPP)) {
2669                 ext4_msg(sb, KERN_ERR,
2670                         "Couldn't mount because of "
2671                         "unsupported optional features (%x)",
2672                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2673                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2674                 return 0;
2675         }
2676
2677         if (readonly)
2678                 return 1;
2679
2680         /* Check that feature set is OK for a read-write mount */
2681         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT4_FEATURE_RO_COMPAT_SUPP)) {
2682                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2683                          "unsupported optional features (%x)",
2684                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2685                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2686                 return 0;
2687         }
2688         /*
2689          * Large file size enabled file system can only be mounted
2690          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2691          */
2692         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_HUGE_FILE)) {
2693                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2694                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2695                                  "cannot be mounted RDWR without "
2696                                  "CONFIG_LBDAF");
2697                         return 0;
2698                 }
2699         }
2700         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC) &&
2701             !EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_EXTENTS)) {
2702                 ext4_msg(sb, KERN_ERR,
2703                          "Can't support bigalloc feature without "
2704                          "extents feature\n");
2705                 return 0;
2706         }
2707
2708 #ifndef CONFIG_QUOTA
2709         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
2710             !readonly) {
2711                 ext4_msg(sb, KERN_ERR,
2712                          "Filesystem with quota feature cannot be mounted RDWR "
2713                          "without CONFIG_QUOTA");
2714                 return 0;
2715         }
2716 #endif  /* CONFIG_QUOTA */
2717         return 1;
2718 }
2719
2720 /*
2721  * This function is called once a day if we have errors logged
2722  * on the file system
2723  */
2724 static void print_daily_error_info(unsigned long arg)
2725 {
2726         struct super_block *sb = (struct super_block *) arg;
2727         struct ext4_sb_info *sbi;
2728         struct ext4_super_block *es;
2729
2730         sbi = EXT4_SB(sb);
2731         es = sbi->s_es;
2732
2733         if (es->s_error_count)
2734                 ext4_msg(sb, KERN_NOTICE, "error count: %u",
2735                          le32_to_cpu(es->s_error_count));
2736         if (es->s_first_error_time) {
2737                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at %u: %.*s:%d",
2738                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2739                        (int) sizeof(es->s_first_error_func),
2740                        es->s_first_error_func,
2741                        le32_to_cpu(es->s_first_error_line));
2742                 if (es->s_first_error_ino)
2743                         printk(": inode %u",
2744                                le32_to_cpu(es->s_first_error_ino));
2745                 if (es->s_first_error_block)
2746                         printk(": block %llu", (unsigned long long)
2747                                le64_to_cpu(es->s_first_error_block));
2748                 printk("\n");
2749         }
2750         if (es->s_last_error_time) {
2751                 printk(KERN_NOTICE "EXT4-fs (%s): last error at %u: %.*s:%d",
2752                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2753                        (int) sizeof(es->s_last_error_func),
2754                        es->s_last_error_func,
2755                        le32_to_cpu(es->s_last_error_line));
2756                 if (es->s_last_error_ino)
2757                         printk(": inode %u",
2758                                le32_to_cpu(es->s_last_error_ino));
2759                 if (es->s_last_error_block)
2760                         printk(": block %llu", (unsigned long long)
2761                                le64_to_cpu(es->s_last_error_block));
2762                 printk("\n");
2763         }
2764         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2765 }
2766
2767 /* Find next suitable group and run ext4_init_inode_table */
2768 static int ext4_run_li_request(struct ext4_li_request *elr)
2769 {
2770         struct ext4_group_desc *gdp = NULL;
2771         ext4_group_t group, ngroups;
2772         struct super_block *sb;
2773         unsigned long timeout = 0;
2774         int ret = 0;
2775
2776         sb = elr->lr_super;
2777         ngroups = EXT4_SB(sb)->s_groups_count;
2778
2779         sb_start_write(sb);
2780         for (group = elr->lr_next_group; group < ngroups; group++) {
2781                 gdp = ext4_get_group_desc(sb, group, NULL);
2782                 if (!gdp) {
2783                         ret = 1;
2784                         break;
2785                 }
2786
2787                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2788                         break;
2789         }
2790
2791         if (group >= ngroups)
2792                 ret = 1;
2793
2794         if (!ret) {
2795                 timeout = jiffies;
2796                 ret = ext4_init_inode_table(sb, group,
2797                                             elr->lr_timeout ? 0 : 1);
2798                 if (elr->lr_timeout == 0) {
2799                         timeout = (jiffies - timeout) *
2800                                   elr->lr_sbi->s_li_wait_mult;
2801                         elr->lr_timeout = timeout;
2802                 }
2803                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2804                 elr->lr_next_group = group + 1;
2805         }
2806         sb_end_write(sb);
2807
2808         return ret;
2809 }
2810
2811 /*
2812  * Remove lr_request from the list_request and free the
2813  * request structure. Should be called with li_list_mtx held
2814  */
2815 static void ext4_remove_li_request(struct ext4_li_request *elr)
2816 {
2817         struct ext4_sb_info *sbi;
2818
2819         if (!elr)
2820                 return;
2821
2822         sbi = elr->lr_sbi;
2823
2824         list_del(&elr->lr_request);
2825         sbi->s_li_request = NULL;
2826         kfree(elr);
2827 }
2828
2829 static void ext4_unregister_li_request(struct super_block *sb)
2830 {
2831         mutex_lock(&ext4_li_mtx);
2832         if (!ext4_li_info) {
2833                 mutex_unlock(&ext4_li_mtx);
2834                 return;
2835         }
2836
2837         mutex_lock(&ext4_li_info->li_list_mtx);
2838         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2839         mutex_unlock(&ext4_li_info->li_list_mtx);
2840         mutex_unlock(&ext4_li_mtx);
2841 }
2842
2843 static struct task_struct *ext4_lazyinit_task;
2844
2845 /*
2846  * This is the function where ext4lazyinit thread lives. It walks
2847  * through the request list searching for next scheduled filesystem.
2848  * When such a fs is found, run the lazy initialization request
2849  * (ext4_rn_li_request) and keep track of the time spend in this
2850  * function. Based on that time we compute next schedule time of
2851  * the request. When walking through the list is complete, compute
2852  * next waking time and put itself into sleep.
2853  */
2854 static int ext4_lazyinit_thread(void *arg)
2855 {
2856         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2857         struct list_head *pos, *n;
2858         struct ext4_li_request *elr;
2859         unsigned long next_wakeup, cur;
2860
2861         BUG_ON(NULL == eli);
2862
2863 cont_thread:
2864         while (true) {
2865                 next_wakeup = MAX_JIFFY_OFFSET;
2866
2867                 mutex_lock(&eli->li_list_mtx);
2868                 if (list_empty(&eli->li_request_list)) {
2869                         mutex_unlock(&eli->li_list_mtx);
2870                         goto exit_thread;
2871                 }
2872
2873                 list_for_each_safe(pos, n, &eli->li_request_list) {
2874                         elr = list_entry(pos, struct ext4_li_request,
2875                                          lr_request);
2876
2877                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2878                                 if (ext4_run_li_request(elr) != 0) {
2879                                         /* error, remove the lazy_init job */
2880                                         ext4_remove_li_request(elr);
2881                                         continue;
2882                                 }
2883                         }
2884
2885                         if (time_before(elr->lr_next_sched, next_wakeup))
2886                                 next_wakeup = elr->lr_next_sched;
2887                 }
2888                 mutex_unlock(&eli->li_list_mtx);
2889
2890                 try_to_freeze();
2891
2892                 cur = jiffies;
2893                 if ((time_after_eq(cur, next_wakeup)) ||
2894                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2895                         cond_resched();
2896                         continue;
2897                 }
2898
2899                 schedule_timeout_interruptible(next_wakeup - cur);
2900
2901                 if (kthread_should_stop()) {
2902                         ext4_clear_request_list();
2903                         goto exit_thread;
2904                 }
2905         }
2906
2907 exit_thread:
2908         /*
2909          * It looks like the request list is empty, but we need
2910          * to check it under the li_list_mtx lock, to prevent any
2911          * additions into it, and of course we should lock ext4_li_mtx
2912          * to atomically free the list and ext4_li_info, because at
2913          * this point another ext4 filesystem could be registering
2914          * new one.
2915          */
2916         mutex_lock(&ext4_li_mtx);
2917         mutex_lock(&eli->li_list_mtx);
2918         if (!list_empty(&eli->li_request_list)) {
2919                 mutex_unlock(&eli->li_list_mtx);
2920                 mutex_unlock(&ext4_li_mtx);
2921                 goto cont_thread;
2922         }
2923         mutex_unlock(&eli->li_list_mtx);
2924         kfree(ext4_li_info);
2925         ext4_li_info = NULL;
2926         mutex_unlock(&ext4_li_mtx);
2927
2928         return 0;
2929 }
2930
2931 static void ext4_clear_request_list(void)
2932 {
2933         struct list_head *pos, *n;
2934         struct ext4_li_request *elr;
2935
2936         mutex_lock(&ext4_li_info->li_list_mtx);
2937         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2938                 elr = list_entry(pos, struct ext4_li_request,
2939                                  lr_request);
2940                 ext4_remove_li_request(elr);
2941         }
2942         mutex_unlock(&ext4_li_info->li_list_mtx);
2943 }
2944
2945 static int ext4_run_lazyinit_thread(void)
2946 {
2947         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2948                                          ext4_li_info, "ext4lazyinit");
2949         if (IS_ERR(ext4_lazyinit_task)) {
2950                 int err = PTR_ERR(ext4_lazyinit_task);
2951                 ext4_clear_request_list();
2952                 kfree(ext4_li_info);
2953                 ext4_li_info = NULL;
2954                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2955                                  "initialization thread\n",
2956                                  err);
2957                 return err;
2958         }
2959         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2960         return 0;
2961 }
2962
2963 /*
2964  * Check whether it make sense to run itable init. thread or not.
2965  * If there is at least one uninitialized inode table, return
2966  * corresponding group number, else the loop goes through all
2967  * groups and return total number of groups.
2968  */
2969 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2970 {
2971         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2972         struct ext4_group_desc *gdp = NULL;
2973
2974         for (group = 0; group < ngroups; group++) {
2975                 gdp = ext4_get_group_desc(sb, group, NULL);
2976                 if (!gdp)
2977                         continue;
2978
2979                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2980                         break;
2981         }
2982
2983         return group;
2984 }
2985
2986 static int ext4_li_info_new(void)
2987 {
2988         struct ext4_lazy_init *eli = NULL;
2989
2990         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2991         if (!eli)
2992                 return -ENOMEM;
2993
2994         INIT_LIST_HEAD(&eli->li_request_list);
2995         mutex_init(&eli->li_list_mtx);
2996
2997         eli->li_state |= EXT4_LAZYINIT_QUIT;
2998
2999         ext4_li_info = eli;
3000
3001         return 0;
3002 }
3003
3004 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
3005                                             ext4_group_t start)
3006 {
3007         struct ext4_sb_info *sbi = EXT4_SB(sb);
3008         struct ext4_li_request *elr;
3009         unsigned long rnd;
3010
3011         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
3012         if (!elr)
3013                 return NULL;
3014
3015         elr->lr_super = sb;
3016         elr->lr_sbi = sbi;
3017         elr->lr_next_group = start;
3018
3019         /*
3020          * Randomize first schedule time of the request to
3021          * spread the inode table initialization requests
3022          * better.
3023          */
3024         get_random_bytes(&rnd, sizeof(rnd));
3025         elr->lr_next_sched = jiffies + (unsigned long)rnd %
3026                              (EXT4_DEF_LI_MAX_START_DELAY * HZ);
3027
3028         return elr;
3029 }
3030
3031 int ext4_register_li_request(struct super_block *sb,
3032                              ext4_group_t first_not_zeroed)
3033 {
3034         struct ext4_sb_info *sbi = EXT4_SB(sb);
3035         struct ext4_li_request *elr = NULL;
3036         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
3037         int ret = 0;
3038
3039         mutex_lock(&ext4_li_mtx);
3040         if (sbi->s_li_request != NULL) {
3041                 /*
3042                  * Reset timeout so it can be computed again, because
3043                  * s_li_wait_mult might have changed.
3044                  */
3045                 sbi->s_li_request->lr_timeout = 0;
3046                 goto out;
3047         }
3048
3049         if (first_not_zeroed == ngroups ||
3050             (sb->s_flags & MS_RDONLY) ||
3051             !test_opt(sb, INIT_INODE_TABLE))
3052                 goto out;
3053
3054         elr = ext4_li_request_new(sb, first_not_zeroed);
3055         if (!elr) {
3056                 ret = -ENOMEM;
3057                 goto out;
3058         }
3059
3060         if (NULL == ext4_li_info) {
3061                 ret = ext4_li_info_new();
3062                 if (ret)
3063                         goto out;
3064         }
3065
3066         mutex_lock(&ext4_li_info->li_list_mtx);
3067         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
3068         mutex_unlock(&ext4_li_info->li_list_mtx);
3069
3070         sbi->s_li_request = elr;
3071         /*
3072          * set elr to NULL here since it has been inserted to
3073          * the request_list and the removal and free of it is
3074          * handled by ext4_clear_request_list from now on.
3075          */
3076         elr = NULL;
3077
3078         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
3079                 ret = ext4_run_lazyinit_thread();
3080                 if (ret)
3081                         goto out;
3082         }
3083 out:
3084         mutex_unlock(&ext4_li_mtx);
3085         if (ret)
3086                 kfree(elr);
3087         return ret;
3088 }
3089
3090 /*
3091  * We do not need to lock anything since this is called on
3092  * module unload.
3093  */
3094 static void ext4_destroy_lazyinit_thread(void)
3095 {
3096         /*
3097          * If thread exited earlier
3098          * there's nothing to be done.
3099          */
3100         if (!ext4_li_info || !ext4_lazyinit_task)
3101                 return;
3102
3103         kthread_stop(ext4_lazyinit_task);
3104 }
3105
3106 static int set_journal_csum_feature_set(struct super_block *sb)
3107 {
3108         int ret = 1;
3109         int compat, incompat;
3110         struct ext4_sb_info *sbi = EXT4_SB(sb);
3111
3112         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3113                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3114                 /* journal checksum v2 */
3115                 compat = 0;
3116                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V2;
3117         } else {
3118                 /* journal checksum v1 */
3119                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
3120                 incompat = 0;
3121         }
3122
3123         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3124                 ret = jbd2_journal_set_features(sbi->s_journal,
3125                                 compat, 0,
3126                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3127                                 incompat);
3128         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
3129                 ret = jbd2_journal_set_features(sbi->s_journal,
3130                                 compat, 0,
3131                                 incompat);
3132                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3133                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3134         } else {
3135                 jbd2_journal_clear_features(sbi->s_journal,
3136                                 JBD2_FEATURE_COMPAT_CHECKSUM, 0,
3137                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
3138                                 JBD2_FEATURE_INCOMPAT_CSUM_V2);
3139         }
3140
3141         return ret;
3142 }
3143
3144 /*
3145  * Note: calculating the overhead so we can be compatible with
3146  * historical BSD practice is quite difficult in the face of
3147  * clusters/bigalloc.  This is because multiple metadata blocks from
3148  * different block group can end up in the same allocation cluster.
3149  * Calculating the exact overhead in the face of clustered allocation
3150  * requires either O(all block bitmaps) in memory or O(number of block
3151  * groups**2) in time.  We will still calculate the superblock for
3152  * older file systems --- and if we come across with a bigalloc file
3153  * system with zero in s_overhead_clusters the estimate will be close to
3154  * correct especially for very large cluster sizes --- but for newer
3155  * file systems, it's better to calculate this figure once at mkfs
3156  * time, and store it in the superblock.  If the superblock value is
3157  * present (even for non-bigalloc file systems), we will use it.
3158  */
3159 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3160                           char *buf)
3161 {
3162         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3163         struct ext4_group_desc  *gdp;
3164         ext4_fsblk_t            first_block, last_block, b;
3165         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3166         int                     s, j, count = 0;
3167
3168         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_BIGALLOC))
3169                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3170                         sbi->s_itb_per_group + 2);
3171
3172         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3173                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3174         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3175         for (i = 0; i < ngroups; i++) {
3176                 gdp = ext4_get_group_desc(sb, i, NULL);
3177                 b = ext4_block_bitmap(sb, gdp);
3178                 if (b >= first_block && b <= last_block) {
3179                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3180                         count++;
3181                 }
3182                 b = ext4_inode_bitmap(sb, gdp);
3183                 if (b >= first_block && b <= last_block) {
3184                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3185                         count++;
3186                 }
3187                 b = ext4_inode_table(sb, gdp);
3188                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3189                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3190                                 int c = EXT4_B2C(sbi, b - first_block);
3191                                 ext4_set_bit(c, buf);
3192                                 count++;
3193                         }
3194                 if (i != grp)
3195                         continue;
3196                 s = 0;
3197                 if (ext4_bg_has_super(sb, grp)) {
3198                         ext4_set_bit(s++, buf);
3199                         count++;
3200                 }
3201                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3202                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3203                         count++;
3204                 }
3205         }
3206         if (!count)
3207                 return 0;
3208         return EXT4_CLUSTERS_PER_GROUP(sb) -
3209                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3210 }
3211
3212 /*
3213  * Compute the overhead and stash it in sbi->s_overhead
3214  */
3215 int ext4_calculate_overhead(struct super_block *sb)
3216 {
3217         struct ext4_sb_info *sbi = EXT4_SB(sb);
3218         struct ext4_super_block *es = sbi->s_es;
3219         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3220         ext4_fsblk_t overhead = 0;
3221         char *buf = (char *) get_zeroed_page(GFP_KERNEL);
3222
3223         if (!buf)
3224                 return -ENOMEM;
3225
3226         /*
3227          * Compute the overhead (FS structures).  This is constant
3228          * for a given filesystem unless the number of block groups
3229          * changes so we cache the previous value until it does.
3230          */
3231
3232         /*
3233          * All of the blocks before first_data_block are overhead
3234          */
3235         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3236
3237         /*
3238          * Add the overhead found in each block group
3239          */
3240         for (i = 0; i < ngroups; i++) {
3241                 int blks;
3242
3243                 blks = count_overhead(sb, i, buf);
3244                 overhead += blks;
3245                 if (blks)
3246                         memset(buf, 0, PAGE_SIZE);
3247                 cond_resched();
3248         }
3249         /* Add the journal blocks as well */
3250         if (sbi->s_journal)
3251                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3252
3253         sbi->s_overhead = overhead;
3254         smp_wmb();
3255         free_page((unsigned long) buf);
3256         return 0;
3257 }
3258
3259
3260 static ext4_fsblk_t ext4_calculate_resv_clusters(struct ext4_sb_info *sbi)
3261 {
3262         ext4_fsblk_t resv_clusters;
3263
3264         /*
3265          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3266          * This should cover the situations where we can not afford to run
3267          * out of space like for example punch hole, or converting
3268          * uninitialized extents in delalloc path. In most cases such
3269          * allocation would require 1, or 2 blocks, higher numbers are
3270          * very rare.
3271          */
3272         resv_clusters = ext4_blocks_count(sbi->s_es) >> sbi->s_cluster_bits;
3273
3274         do_div(resv_clusters, 50);
3275         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3276
3277         return resv_clusters;
3278 }
3279
3280
3281 static int ext4_reserve_clusters(struct ext4_sb_info *sbi, ext4_fsblk_t count)
3282 {
3283         ext4_fsblk_t clusters = ext4_blocks_count(sbi->s_es) >>
3284                                 sbi->s_cluster_bits;
3285
3286         if (count >= clusters)
3287                 return -EINVAL;
3288
3289         atomic64_set(&sbi->s_resv_clusters, count);
3290         return 0;
3291 }
3292
3293 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3294 {
3295         char *orig_data = kstrdup(data, GFP_KERNEL);
3296         struct buffer_head *bh;
3297         struct ext4_super_block *es = NULL;
3298         struct ext4_sb_info *sbi;
3299         ext4_fsblk_t block;
3300         ext4_fsblk_t sb_block = get_sb_block(&data);
3301         ext4_fsblk_t logical_sb_block;
3302         unsigned long offset = 0;
3303         unsigned long journal_devnum = 0;
3304         unsigned long def_mount_opts;
3305         struct inode *root;
3306         char *cp;
3307         const char *descr;
3308         int ret = -ENOMEM;
3309         int blocksize, clustersize;
3310         unsigned int db_count;
3311         unsigned int i;
3312         int needs_recovery, has_huge_files, has_bigalloc;
3313         __u64 blocks_count;
3314         int err = 0;
3315         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3316         ext4_group_t first_not_zeroed;
3317
3318         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3319         if (!sbi)
3320                 goto out_free_orig;
3321
3322         sbi->s_blockgroup_lock =
3323                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3324         if (!sbi->s_blockgroup_lock) {
3325                 kfree(sbi);
3326                 goto out_free_orig;
3327         }
3328         sb->s_fs_info = sbi;
3329         sbi->s_sb = sb;
3330         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3331         sbi->s_sb_block = sb_block;
3332         if (sb->s_bdev->bd_part)
3333                 sbi->s_sectors_written_start =
3334                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3335
3336         /* Cleanup superblock name */
3337         for (cp = sb->s_id; (cp = strchr(cp, '/'));)
3338                 *cp = '!';
3339
3340         /* -EINVAL is default */
3341         ret = -EINVAL;
3342         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3343         if (!blocksize) {
3344                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3345                 goto out_fail;
3346         }
3347
3348         /*
3349          * The ext4 superblock will not be buffer aligned for other than 1kB
3350          * block sizes.  We need to calculate the offset from buffer start.
3351          */
3352         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3353                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3354                 offset = do_div(logical_sb_block, blocksize);
3355         } else {
3356                 logical_sb_block = sb_block;
3357         }
3358
3359         if (!(bh = sb_bread(sb, logical_sb_block))) {
3360                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3361                 goto out_fail;
3362         }
3363         /*
3364          * Note: s_es must be initialized as soon as possible because
3365          *       some ext4 macro-instructions depend on its value
3366          */
3367         es = (struct ext4_super_block *) (bh->b_data + offset);
3368         sbi->s_es = es;
3369         sb->s_magic = le16_to_cpu(es->s_magic);
3370         if (sb->s_magic != EXT4_SUPER_MAGIC)
3371                 goto cantfind_ext4;
3372         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3373
3374         /* Warn if metadata_csum and gdt_csum are both set. */
3375         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3376                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
3377             EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_GDT_CSUM))
3378                 ext4_warning(sb, KERN_INFO "metadata_csum and uninit_bg are "
3379                              "redundant flags; please run fsck.");
3380
3381         /* Check for a known checksum algorithm */
3382         if (!ext4_verify_csum_type(sb, es)) {
3383                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3384                          "unknown checksum algorithm.");
3385                 silent = 1;
3386                 goto cantfind_ext4;
3387         }
3388
3389         /* Load the checksum driver */
3390         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3391                                        EXT4_FEATURE_RO_COMPAT_METADATA_CSUM)) {
3392                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3393                 if (IS_ERR(sbi->s_chksum_driver)) {
3394                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3395                         ret = PTR_ERR(sbi->s_chksum_driver);
3396                         sbi->s_chksum_driver = NULL;
3397                         goto failed_mount;
3398                 }
3399         }
3400
3401         /* Check superblock checksum */
3402         if (!ext4_superblock_csum_verify(sb, es)) {
3403                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3404                          "invalid superblock checksum.  Run e2fsck?");
3405                 silent = 1;
3406                 goto cantfind_ext4;
3407         }
3408
3409         /* Precompute checksum seed for all metadata */
3410         if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
3411                         EXT4_FEATURE_RO_COMPAT_METADATA_CSUM))
3412                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3413                                                sizeof(es->s_uuid));
3414
3415         /* Set defaults before we parse the mount options */
3416         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3417         set_opt(sb, INIT_INODE_TABLE);
3418         if (def_mount_opts & EXT4_DEFM_DEBUG)
3419                 set_opt(sb, DEBUG);
3420         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3421                 set_opt(sb, GRPID);
3422         if (def_mount_opts & EXT4_DEFM_UID16)
3423                 set_opt(sb, NO_UID32);
3424         /* xattr user namespace & acls are now defaulted on */
3425         set_opt(sb, XATTR_USER);
3426 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3427         set_opt(sb, POSIX_ACL);
3428 #endif
3429         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3430                 set_opt(sb, JOURNAL_DATA);
3431         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3432                 set_opt(sb, ORDERED_DATA);
3433         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3434                 set_opt(sb, WRITEBACK_DATA);
3435
3436         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3437                 set_opt(sb, ERRORS_PANIC);
3438         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3439                 set_opt(sb, ERRORS_CONT);
3440         else
3441                 set_opt(sb, ERRORS_RO);
3442         if (def_mount_opts & EXT4_DEFM_BLOCK_VALIDITY)
3443                 set_opt(sb, BLOCK_VALIDITY);
3444         if (def_mount_opts & EXT4_DEFM_DISCARD)
3445                 set_opt(sb, DISCARD);
3446
3447         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3448         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3449         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3450         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3451         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3452
3453         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3454                 set_opt(sb, BARRIER);
3455
3456         /*
3457          * enable delayed allocation by default
3458          * Use -o nodelalloc to turn it off
3459          */
3460         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3461             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3462                 set_opt(sb, DELALLOC);
3463
3464         /*
3465          * set default s_li_wait_mult for lazyinit, for the case there is
3466          * no mount option specified.
3467          */
3468         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3469
3470         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3471                            &journal_devnum, &journal_ioprio, 0)) {
3472                 ext4_msg(sb, KERN_WARNING,
3473                          "failed to parse options in superblock: %s",
3474                          sbi->s_es->s_mount_opts);
3475         }
3476         sbi->s_def_mount_opt = sbi->s_mount_opt;
3477         if (!parse_options((char *) data, sb, &journal_devnum,
3478                            &journal_ioprio, 0))
3479                 goto failed_mount;
3480
3481         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3482                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3483                             "with data=journal disables delayed "
3484                             "allocation and O_DIRECT support!\n");
3485                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3486                         ext4_msg(sb, KERN_ERR, "can't mount with "
3487                                  "both data=journal and delalloc");
3488                         goto failed_mount;
3489                 }
3490                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3491                         ext4_msg(sb, KERN_ERR, "can't mount with "
3492                                  "both data=journal and delalloc");
3493                         goto failed_mount;
3494                 }
3495                 if (test_opt(sb, DELALLOC))
3496                         clear_opt(sb, DELALLOC);
3497         }
3498
3499         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3500                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3501
3502         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3503             (EXT4_HAS_COMPAT_FEATURE(sb, ~0U) ||
3504              EXT4_HAS_RO_COMPAT_FEATURE(sb, ~0U) ||
3505              EXT4_HAS_INCOMPAT_FEATURE(sb, ~0U)))
3506                 ext4_msg(sb, KERN_WARNING,
3507                        "feature flags set on rev 0 fs, "
3508                        "running e2fsck is recommended");
3509
3510         if (IS_EXT2_SB(sb)) {
3511                 if (ext2_feature_set_ok(sb))
3512                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3513                                  "using the ext4 subsystem");
3514                 else {
3515                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3516                                  "to feature incompatibilities");
3517                         goto failed_mount;
3518                 }
3519         }
3520
3521         if (IS_EXT3_SB(sb)) {
3522                 if (ext3_feature_set_ok(sb))
3523                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3524                                  "using the ext4 subsystem");
3525                 else {
3526                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3527                                  "to feature incompatibilities");
3528                         goto failed_mount;
3529                 }
3530         }
3531
3532         /*
3533          * Check feature flags regardless of the revision level, since we
3534          * previously didn't change the revision level when setting the flags,
3535          * so there is a chance incompat flags are set on a rev 0 filesystem.
3536          */
3537         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3538                 goto failed_mount;
3539
3540         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3541         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3542             blocksize > EXT4_MAX_BLOCK_SIZE) {
3543                 ext4_msg(sb, KERN_ERR,
3544                        "Unsupported filesystem blocksize %d", blocksize);
3545                 goto failed_mount;
3546         }
3547
3548         if (sb->s_blocksize != blocksize) {
3549                 /* Validate the filesystem blocksize */
3550                 if (!sb_set_blocksize(sb, blocksize)) {
3551                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3552                                         blocksize);
3553                         goto failed_mount;
3554                 }
3555
3556                 brelse(bh);
3557                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3558                 offset = do_div(logical_sb_block, blocksize);
3559                 bh = sb_bread(sb, logical_sb_block);
3560                 if (!bh) {
3561                         ext4_msg(sb, KERN_ERR,
3562                                "Can't read superblock on 2nd try");
3563                         goto failed_mount;
3564                 }
3565                 es = (struct ext4_super_block *)(bh->b_data + offset);
3566                 sbi->s_es = es;
3567                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3568                         ext4_msg(sb, KERN_ERR,
3569                                "Magic mismatch, very weird!");
3570                         goto failed_mount;
3571                 }
3572         }
3573
3574         has_huge_files = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3575                                 EXT4_FEATURE_RO_COMPAT_HUGE_FILE);
3576         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3577                                                       has_huge_files);
3578         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3579
3580         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3581                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3582                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3583         } else {
3584                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3585                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3586                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3587                     (!is_power_of_2(sbi->s_inode_size)) ||
3588                     (sbi->s_inode_size > blocksize)) {
3589                         ext4_msg(sb, KERN_ERR,
3590                                "unsupported inode size: %d",
3591                                sbi->s_inode_size);
3592                         goto failed_mount;
3593                 }
3594                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3595                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3596         }
3597
3598         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3599         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT)) {
3600                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3601                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3602                     !is_power_of_2(sbi->s_desc_size)) {
3603                         ext4_msg(sb, KERN_ERR,
3604                                "unsupported descriptor size %lu",
3605                                sbi->s_desc_size);
3606                         goto failed_mount;
3607                 }
3608         } else
3609                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3610
3611         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3612         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3613         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3614                 goto cantfind_ext4;
3615
3616         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3617         if (sbi->s_inodes_per_block == 0)
3618                 goto cantfind_ext4;
3619         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3620                                         sbi->s_inodes_per_block;
3621         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3622         sbi->s_sbh = bh;
3623         sbi->s_mount_state = le16_to_cpu(es->s_state);
3624         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3625         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3626
3627         /* Do we have standard group size of blocksize * 8 blocks ? */
3628         if (sbi->s_blocks_per_group == blocksize << 3)
3629                 set_opt2(sb, STD_GROUP_SIZE);
3630
3631         for (i = 0; i < 4; i++)
3632                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3633         sbi->s_def_hash_version = es->s_def_hash_version;
3634         i = le32_to_cpu(es->s_flags);
3635         if (i & EXT2_FLAGS_UNSIGNED_HASH)
3636                 sbi->s_hash_unsigned = 3;
3637         else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3638 #ifdef __CHAR_UNSIGNED__
3639                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3640                 sbi->s_hash_unsigned = 3;
3641 #else
3642                 es->s_flags |= cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3643 #endif
3644         }
3645
3646         /* Handle clustersize */
3647         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3648         has_bigalloc = EXT4_HAS_RO_COMPAT_FEATURE(sb,
3649                                 EXT4_FEATURE_RO_COMPAT_BIGALLOC);
3650         if (has_bigalloc) {
3651                 if (clustersize < blocksize) {
3652                         ext4_msg(sb, KERN_ERR,
3653                                  "cluster size (%d) smaller than "
3654                                  "block size (%d)", clustersize, blocksize);
3655                         goto failed_mount;
3656                 }
3657                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3658                         le32_to_cpu(es->s_log_block_size);
3659                 sbi->s_clusters_per_group =
3660                         le32_to_cpu(es->s_clusters_per_group);
3661                 if (sbi->s_clusters_per_group > blocksize * 8) {
3662                         ext4_msg(sb, KERN_ERR,
3663                                  "#clusters per group too big: %lu",
3664                                  sbi->s_clusters_per_group);
3665                         goto failed_mount;
3666                 }
3667                 if (sbi->s_blocks_per_group !=
3668                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3669                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3670                                  "clusters per group (%lu) inconsistent",
3671                                  sbi->s_blocks_per_group,
3672                                  sbi->s_clusters_per_group);
3673                         goto failed_mount;
3674                 }
3675         } else {
3676                 if (clustersize != blocksize) {
3677                         ext4_warning(sb, "fragment/cluster size (%d) != "
3678                                      "block size (%d)", clustersize,
3679                                      blocksize);
3680                         clustersize = blocksize;
3681                 }
3682                 if (sbi->s_blocks_per_group > blocksize * 8) {
3683                         ext4_msg(sb, KERN_ERR,
3684                                  "#blocks per group too big: %lu",
3685                                  sbi->s_blocks_per_group);
3686                         goto failed_mount;
3687                 }
3688                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3689                 sbi->s_cluster_bits = 0;
3690         }
3691         sbi->s_cluster_ratio = clustersize / blocksize;
3692
3693         if (sbi->s_inodes_per_group > blocksize * 8) {
3694                 ext4_msg(sb, KERN_ERR,
3695                        "#inodes per group too big: %lu",
3696                        sbi->s_inodes_per_group);
3697                 goto failed_mount;
3698         }
3699
3700         /*
3701          * Test whether we have more sectors than will fit in sector_t,
3702          * and whether the max offset is addressable by the page cache.
3703          */
3704         err = generic_check_addressable(sb->s_blocksize_bits,
3705                                         ext4_blocks_count(es));
3706         if (err) {
3707                 ext4_msg(sb, KERN_ERR, "filesystem"
3708                          " too large to mount safely on this system");
3709                 if (sizeof(sector_t) < 8)
3710                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3711                 goto failed_mount;
3712         }
3713
3714         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3715                 goto cantfind_ext4;
3716
3717         /* check blocks count against device size */
3718         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3719         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3720                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3721                        "exceeds size of device (%llu blocks)",
3722                        ext4_blocks_count(es), blocks_count);
3723                 goto failed_mount;
3724         }
3725
3726         /*
3727          * It makes no sense for the first data block to be beyond the end
3728          * of the filesystem.
3729          */
3730         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3731                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3732                          "block %u is beyond end of filesystem (%llu)",
3733                          le32_to_cpu(es->s_first_data_block),
3734                          ext4_blocks_count(es));
3735                 goto failed_mount;
3736         }
3737         blocks_count = (ext4_blocks_count(es) -
3738                         le32_to_cpu(es->s_first_data_block) +
3739                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3740         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3741         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3742                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3743                        "(block count %llu, first data block %u, "
3744                        "blocks per group %lu)", sbi->s_groups_count,
3745                        ext4_blocks_count(es),
3746                        le32_to_cpu(es->s_first_data_block),
3747                        EXT4_BLOCKS_PER_GROUP(sb));
3748                 goto failed_mount;
3749         }
3750         sbi->s_groups_count = blocks_count;
3751         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3752                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3753         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3754                    EXT4_DESC_PER_BLOCK(sb);
3755         sbi->s_group_desc = ext4_kvmalloc(db_count *
3756                                           sizeof(struct buffer_head *),
3757                                           GFP_KERNEL);
3758         if (sbi->s_group_desc == NULL) {
3759                 ext4_msg(sb, KERN_ERR, "not enough memory");
3760                 ret = -ENOMEM;
3761                 goto failed_mount;
3762         }
3763
3764         if (ext4_proc_root)
3765                 sbi->s_proc = proc_mkdir(sb->s_id, ext4_proc_root);
3766
3767         if (sbi->s_proc)
3768                 proc_create_data("options", S_IRUGO, sbi->s_proc,
3769                                  &ext4_seq_options_fops, sb);
3770
3771         bgl_lock_init(sbi->s_blockgroup_lock);
3772
3773         for (i = 0; i < db_count; i++) {
3774                 block = descriptor_loc(sb, logical_sb_block, i);
3775                 sbi->s_group_desc[i] = sb_bread(sb, block);
3776                 if (!sbi->s_group_desc[i]) {
3777                         ext4_msg(sb, KERN_ERR,
3778                                "can't read group descriptor %d", i);
3779                         db_count = i;
3780                         goto failed_mount2;
3781                 }
3782         }
3783         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3784                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3785                 goto failed_mount2;
3786         }
3787         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_FLEX_BG))
3788                 if (!ext4_fill_flex_info(sb)) {
3789                         ext4_msg(sb, KERN_ERR,
3790                                "unable to initialize "
3791                                "flex_bg meta info!");
3792                         goto failed_mount2;
3793                 }
3794
3795         sbi->s_gdb_count = db_count;
3796         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3797         spin_lock_init(&sbi->s_next_gen_lock);
3798
3799         init_timer(&sbi->s_err_report);
3800         sbi->s_err_report.function = print_daily_error_info;
3801         sbi->s_err_report.data = (unsigned long) sb;
3802
3803         /* Register extent status tree shrinker */
3804         ext4_es_register_shrinker(sbi);
3805
3806         err = percpu_counter_init(&sbi->s_freeclusters_counter,
3807                         ext4_count_free_clusters(sb));
3808         if (!err) {
3809                 err = percpu_counter_init(&sbi->s_freeinodes_counter,
3810                                 ext4_count_free_inodes(sb));
3811         }
3812         if (!err) {
3813                 err = percpu_counter_init(&sbi->s_dirs_counter,
3814                                 ext4_count_dirs(sb));
3815         }
3816         if (!err) {
3817                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0);
3818         }
3819         if (!err) {
3820                 err = percpu_counter_init(&sbi->s_extent_cache_cnt, 0);
3821         }
3822         if (err) {
3823                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3824                 goto failed_mount3;
3825         }
3826
3827         sbi->s_stripe = ext4_get_stripe_size(sbi);
3828         sbi->s_extent_max_zeroout_kb = 32;
3829
3830         /*
3831          * set up enough so that it can read an inode
3832          */
3833         if (!test_opt(sb, NOLOAD) &&
3834             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
3835                 sb->s_op = &ext4_sops;
3836         else
3837                 sb->s_op = &ext4_nojournal_sops;
3838         sb->s_export_op = &ext4_export_ops;
3839         sb->s_xattr = ext4_xattr_handlers;
3840 #ifdef CONFIG_QUOTA
3841         sb->dq_op = &ext4_quota_operations;
3842         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
3843                 sb->s_qcop = &ext4_qctl_sysfile_operations;
3844         else
3845                 sb->s_qcop = &ext4_qctl_operations;
3846 #endif
3847         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3848
3849         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3850         mutex_init(&sbi->s_orphan_lock);
3851
3852         sb->s_root = NULL;
3853
3854         needs_recovery = (es->s_last_orphan != 0 ||
3855                           EXT4_HAS_INCOMPAT_FEATURE(sb,
3856                                     EXT4_FEATURE_INCOMPAT_RECOVER));
3857
3858         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_MMP) &&
3859             !(sb->s_flags & MS_RDONLY))
3860                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3861                         goto failed_mount3;
3862
3863         /*
3864          * The first inode we look at is the journal inode.  Don't try
3865          * root first: it may be modified in the journal!
3866          */
3867         if (!test_opt(sb, NOLOAD) &&
3868             EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
3869                 if (ext4_load_journal(sb, es, journal_devnum))
3870                         goto failed_mount3;
3871         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3872               EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
3873                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3874                        "suppressed and not mounted read-only");
3875                 goto failed_mount_wq;
3876         } else {
3877                 clear_opt(sb, DATA_FLAGS);
3878                 sbi->s_journal = NULL;
3879                 needs_recovery = 0;
3880                 goto no_journal;
3881         }
3882
3883         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_64BIT) &&
3884             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3885                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3886                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3887                 goto failed_mount_wq;
3888         }
3889
3890         if (!set_journal_csum_feature_set(sb)) {
3891                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3892                          "feature set");
3893                 goto failed_mount_wq;
3894         }
3895
3896         /* We have now updated the journal if required, so we can
3897          * validate the data journaling mode. */
3898         switch (test_opt(sb, DATA_FLAGS)) {
3899         case 0:
3900                 /* No mode set, assume a default based on the journal
3901                  * capabilities: ORDERED_DATA if the journal can
3902                  * cope, else JOURNAL_DATA
3903                  */
3904                 if (jbd2_journal_check_available_features
3905                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3906                         set_opt(sb, ORDERED_DATA);
3907                 else
3908                         set_opt(sb, JOURNAL_DATA);
3909                 break;
3910
3911         case EXT4_MOUNT_ORDERED_DATA:
3912         case EXT4_MOUNT_WRITEBACK_DATA:
3913                 if (!jbd2_journal_check_available_features
3914                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3915                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3916                                "requested data journaling mode");
3917                         goto failed_mount_wq;
3918                 }
3919         default:
3920                 break;
3921         }
3922         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3923
3924         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3925
3926         /*
3927          * The journal may have updated the bg summary counts, so we
3928          * need to update the global counters.
3929          */
3930         percpu_counter_set(&sbi->s_freeclusters_counter,
3931                            ext4_count_free_clusters(sb));
3932         percpu_counter_set(&sbi->s_freeinodes_counter,
3933                            ext4_count_free_inodes(sb));
3934         percpu_counter_set(&sbi->s_dirs_counter,
3935                            ext4_count_dirs(sb));
3936         percpu_counter_set(&sbi->s_dirtyclusters_counter, 0);
3937
3938 no_journal:
3939         /*
3940          * Get the # of file system overhead blocks from the
3941          * superblock if present.
3942          */
3943         if (es->s_overhead_clusters)
3944                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3945         else {
3946                 err = ext4_calculate_overhead(sb);
3947                 if (err)
3948                         goto failed_mount_wq;
3949         }
3950
3951         /*
3952          * The maximum number of concurrent works can be high and
3953          * concurrency isn't really necessary.  Limit it to 1.
3954          */
3955         EXT4_SB(sb)->rsv_conversion_wq =
3956                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3957         if (!EXT4_SB(sb)->rsv_conversion_wq) {
3958                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3959                 ret = -ENOMEM;
3960                 goto failed_mount4;
3961         }
3962
3963         EXT4_SB(sb)->unrsv_conversion_wq =
3964                 alloc_workqueue("ext4-unrsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3965         if (!EXT4_SB(sb)->unrsv_conversion_wq) {
3966                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3967                 ret = -ENOMEM;
3968                 goto failed_mount4;
3969         }
3970
3971         /*
3972          * The jbd2_journal_load will have done any necessary log recovery,
3973          * so we can safely mount the rest of the filesystem now.
3974          */
3975
3976         root = ext4_iget(sb, EXT4_ROOT_INO);
3977         if (IS_ERR(root)) {
3978                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3979                 ret = PTR_ERR(root);
3980                 root = NULL;
3981                 goto failed_mount4;
3982         }
3983         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3984                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3985                 iput(root);
3986                 goto failed_mount4;
3987         }
3988         sb->s_root = d_make_root(root);
3989         if (!sb->s_root) {
3990                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3991                 ret = -ENOMEM;
3992                 goto failed_mount4;
3993         }
3994
3995         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3996                 sb->s_flags |= MS_RDONLY;
3997
3998         /* determine the minimum size of new large inodes, if present */
3999         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
4000                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4001                                                      EXT4_GOOD_OLD_INODE_SIZE;
4002                 if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4003                                        EXT4_FEATURE_RO_COMPAT_EXTRA_ISIZE)) {
4004                         if (sbi->s_want_extra_isize <
4005                             le16_to_cpu(es->s_want_extra_isize))
4006                                 sbi->s_want_extra_isize =
4007                                         le16_to_cpu(es->s_want_extra_isize);
4008                         if (sbi->s_want_extra_isize <
4009                             le16_to_cpu(es->s_min_extra_isize))
4010                                 sbi->s_want_extra_isize =
4011                                         le16_to_cpu(es->s_min_extra_isize);
4012                 }
4013         }
4014         /* Check if enough inode space is available */
4015         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
4016                                                         sbi->s_inode_size) {
4017                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
4018                                                        EXT4_GOOD_OLD_INODE_SIZE;
4019                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
4020                          "available");
4021         }
4022
4023         err = ext4_reserve_clusters(sbi, ext4_calculate_resv_clusters(sbi));
4024         if (err) {
4025                 ext4_msg(sb, KERN_ERR, "failed to reserve %llu clusters for "
4026                          "reserved pool", ext4_calculate_resv_clusters(sbi));
4027                 goto failed_mount4a;
4028         }
4029
4030         err = ext4_setup_system_zone(sb);
4031         if (err) {
4032                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
4033                          "zone (%d)", err);
4034                 goto failed_mount4a;
4035         }
4036
4037         ext4_ext_init(sb);
4038         err = ext4_mb_init(sb);
4039         if (err) {
4040                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
4041                          err);
4042                 goto failed_mount5;
4043         }
4044
4045         err = ext4_register_li_request(sb, first_not_zeroed);
4046         if (err)
4047                 goto failed_mount6;
4048
4049         sbi->s_kobj.kset = ext4_kset;
4050         init_completion(&sbi->s_kobj_unregister);
4051         err = kobject_init_and_add(&sbi->s_kobj, &ext4_ktype, NULL,
4052                                    "%s", sb->s_id);
4053         if (err)
4054                 goto failed_mount7;
4055
4056 #ifdef CONFIG_QUOTA
4057         /* Enable quota usage during mount. */
4058         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) &&
4059             !(sb->s_flags & MS_RDONLY)) {
4060                 err = ext4_enable_quotas(sb);
4061                 if (err)
4062                         goto failed_mount8;
4063         }
4064 #endif  /* CONFIG_QUOTA */
4065
4066         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4067         ext4_orphan_cleanup(sb, es);
4068         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4069         if (needs_recovery) {
4070                 ext4_msg(sb, KERN_INFO, "recovery complete");
4071                 ext4_mark_recovery_complete(sb, es);
4072         }
4073         if (EXT4_SB(sb)->s_journal) {
4074                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4075                         descr = " journalled data mode";
4076                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4077                         descr = " ordered data mode";
4078                 else
4079                         descr = " writeback data mode";
4080         } else
4081                 descr = "out journal";
4082
4083         if (test_opt(sb, DISCARD)) {
4084                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4085                 if (!blk_queue_discard(q))
4086                         ext4_msg(sb, KERN_WARNING,
4087                                  "mounting with \"discard\" option, but "
4088                                  "the device does not support discard");
4089         }
4090
4091         ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4092                  "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
4093                  *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4094
4095         if (es->s_error_count)
4096                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4097
4098         kfree(orig_data);
4099         return 0;
4100
4101 cantfind_ext4:
4102         if (!silent)
4103                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4104         goto failed_mount;
4105
4106 #ifdef CONFIG_QUOTA
4107 failed_mount8:
4108         kobject_del(&sbi->s_kobj);
4109 #endif
4110 failed_mount7:
4111         ext4_unregister_li_request(sb);
4112 failed_mount6:
4113         ext4_mb_release(sb);
4114 failed_mount5:
4115         ext4_ext_release(sb);
4116         ext4_release_system_zone(sb);
4117 failed_mount4a:
4118         dput(sb->s_root);
4119         sb->s_root = NULL;
4120 failed_mount4:
4121         ext4_msg(sb, KERN_ERR, "mount failed");
4122         if (EXT4_SB(sb)->rsv_conversion_wq)
4123                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4124         if (EXT4_SB(sb)->unrsv_conversion_wq)
4125                 destroy_workqueue(EXT4_SB(sb)->unrsv_conversion_wq);
4126 failed_mount_wq:
4127         if (sbi->s_journal) {
4128                 jbd2_journal_destroy(sbi->s_journal);
4129                 sbi->s_journal = NULL;
4130         }
4131 failed_mount3:
4132         ext4_es_unregister_shrinker(sbi);
4133         del_timer(&sbi->s_err_report);
4134         if (sbi->s_flex_groups)
4135                 ext4_kvfree(sbi->s_flex_groups);
4136         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4137         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4138         percpu_counter_destroy(&sbi->s_dirs_counter);
4139         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4140         percpu_counter_destroy(&sbi->s_extent_cache_cnt);
4141         if (sbi->s_mmp_tsk)
4142                 kthread_stop(sbi->s_mmp_tsk);
4143 failed_mount2:
4144         for (i = 0; i < db_count; i++)
4145                 brelse(sbi->s_group_desc[i]);
4146         ext4_kvfree(sbi->s_group_desc);
4147 failed_mount:
4148         if (sbi->s_chksum_driver)
4149                 crypto_free_shash(sbi->s_chksum_driver);
4150         if (sbi->s_proc) {
4151                 remove_proc_entry("options", sbi->s_proc);
4152                 remove_proc_entry(sb->s_id, ext4_proc_root);
4153         }
4154 #ifdef CONFIG_QUOTA
4155         for (i = 0; i < MAXQUOTAS; i++)
4156                 kfree(sbi->s_qf_names[i]);
4157 #endif
4158         ext4_blkdev_remove(sbi);
4159         brelse(bh);
4160 out_fail:
4161         sb->s_fs_info = NULL;
4162         kfree(sbi->s_blockgroup_lock);
4163         kfree(sbi);
4164 out_free_orig:
4165         kfree(orig_data);
4166         return err ? err : ret;
4167 }
4168
4169 /*
4170  * Setup any per-fs journal parameters now.  We'll do this both on
4171  * initial mount, once the journal has been initialised but before we've
4172  * done any recovery; and again on any subsequent remount.
4173  */
4174 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4175 {
4176         struct ext4_sb_info *sbi = EXT4_SB(sb);
4177
4178         journal->j_commit_interval = sbi->s_commit_interval;
4179         journal->j_min_batch_time = sbi->s_min_batch_time;
4180         journal->j_max_batch_time = sbi->s_max_batch_time;
4181
4182         write_lock(&journal->j_state_lock);
4183         if (test_opt(sb, BARRIER))
4184                 journal->j_flags |= JBD2_BARRIER;
4185         else
4186                 journal->j_flags &= ~JBD2_BARRIER;
4187         if (test_opt(sb, DATA_ERR_ABORT))
4188                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4189         else
4190                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4191         write_unlock(&journal->j_state_lock);
4192 }
4193
4194 static journal_t *ext4_get_journal(struct super_block *sb,
4195                                    unsigned int journal_inum)
4196 {
4197         struct inode *journal_inode;
4198         journal_t *journal;
4199
4200         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4201
4202         /* First, test for the existence of a valid inode on disk.  Bad
4203          * things happen if we iget() an unused inode, as the subsequent
4204          * iput() will try to delete it. */
4205
4206         journal_inode = ext4_iget(sb, journal_inum);
4207         if (IS_ERR(journal_inode)) {
4208                 ext4_msg(sb, KERN_ERR, "no journal found");
4209                 return NULL;
4210         }
4211         if (!journal_inode->i_nlink) {
4212                 make_bad_inode(journal_inode);
4213                 iput(journal_inode);
4214                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4215                 return NULL;
4216         }
4217
4218         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4219                   journal_inode, journal_inode->i_size);
4220         if (!S_ISREG(journal_inode->i_mode)) {
4221                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4222                 iput(journal_inode);
4223                 return NULL;
4224         }
4225
4226         journal = jbd2_journal_init_inode(journal_inode);
4227         if (!journal) {
4228                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4229                 iput(journal_inode);
4230                 return NULL;
4231         }
4232         journal->j_private = sb;
4233         ext4_init_journal_params(sb, journal);
4234         return journal;
4235 }
4236
4237 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4238                                        dev_t j_dev)
4239 {
4240         struct buffer_head *bh;
4241         journal_t *journal;
4242         ext4_fsblk_t start;
4243         ext4_fsblk_t len;
4244         int hblock, blocksize;
4245         ext4_fsblk_t sb_block;
4246         unsigned long offset;
4247         struct ext4_super_block *es;
4248         struct block_device *bdev;
4249
4250         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4251
4252         bdev = ext4_blkdev_get(j_dev, sb);
4253         if (bdev == NULL)
4254                 return NULL;
4255
4256         blocksize = sb->s_blocksize;
4257         hblock = bdev_logical_block_size(bdev);
4258         if (blocksize < hblock) {
4259                 ext4_msg(sb, KERN_ERR,
4260                         "blocksize too small for journal device");
4261                 goto out_bdev;
4262         }
4263
4264         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4265         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4266         set_blocksize(bdev, blocksize);
4267         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4268                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4269                        "external journal");
4270                 goto out_bdev;
4271         }
4272
4273         es = (struct ext4_super_block *) (bh->b_data + offset);
4274         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4275             !(le32_to_cpu(es->s_feature_incompat) &
4276               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4277                 ext4_msg(sb, KERN_ERR, "external journal has "
4278                                         "bad superblock");
4279                 brelse(bh);
4280                 goto out_bdev;
4281         }
4282
4283         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4284                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4285                 brelse(bh);
4286                 goto out_bdev;
4287         }
4288
4289         len = ext4_blocks_count(es);
4290         start = sb_block + 1;
4291         brelse(bh);     /* we're done with the superblock */
4292
4293         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4294                                         start, len, blocksize);
4295         if (!journal) {
4296                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4297                 goto out_bdev;
4298         }
4299         journal->j_private = sb;
4300         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4301         wait_on_buffer(journal->j_sb_buffer);
4302         if (!buffer_uptodate(journal->j_sb_buffer)) {
4303                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4304                 goto out_journal;
4305         }
4306         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4307                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4308                                         "user (unsupported) - %d",
4309                         be32_to_cpu(journal->j_superblock->s_nr_users));
4310                 goto out_journal;
4311         }
4312         EXT4_SB(sb)->journal_bdev = bdev;
4313         ext4_init_journal_params(sb, journal);
4314         return journal;
4315
4316 out_journal:
4317         jbd2_journal_destroy(journal);
4318 out_bdev:
4319         ext4_blkdev_put(bdev);
4320         return NULL;
4321 }
4322
4323 static int ext4_load_journal(struct super_block *sb,
4324                              struct ext4_super_block *es,
4325                              unsigned long journal_devnum)
4326 {
4327         journal_t *journal;
4328         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4329         dev_t journal_dev;
4330         int err = 0;
4331         int really_read_only;
4332
4333         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4334
4335         if (journal_devnum &&
4336             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4337                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4338                         "numbers have changed");
4339                 journal_dev = new_decode_dev(journal_devnum);
4340         } else
4341                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4342
4343         really_read_only = bdev_read_only(sb->s_bdev);
4344
4345         /*
4346          * Are we loading a blank journal or performing recovery after a
4347          * crash?  For recovery, we need to check in advance whether we
4348          * can get read-write access to the device.
4349          */
4350         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER)) {
4351                 if (sb->s_flags & MS_RDONLY) {
4352                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4353                                         "required on readonly filesystem");
4354                         if (really_read_only) {
4355                                 ext4_msg(sb, KERN_ERR, "write access "
4356                                         "unavailable, cannot proceed");
4357                                 return -EROFS;
4358                         }
4359                         ext4_msg(sb, KERN_INFO, "write access will "
4360                                "be enabled during recovery");
4361                 }
4362         }
4363
4364         if (journal_inum && journal_dev) {
4365                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4366                        "and inode journals!");
4367                 return -EINVAL;
4368         }
4369
4370         if (journal_inum) {
4371                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4372                         return -EINVAL;
4373         } else {
4374                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4375                         return -EINVAL;
4376         }
4377
4378         if (!(journal->j_flags & JBD2_BARRIER))
4379                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4380
4381         if (!EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER))
4382                 err = jbd2_journal_wipe(journal, !really_read_only);
4383         if (!err) {
4384                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4385                 if (save)
4386                         memcpy(save, ((char *) es) +
4387                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4388                 err = jbd2_journal_load(journal);
4389                 if (save)
4390                         memcpy(((char *) es) + EXT4_S_ERR_START,
4391                                save, EXT4_S_ERR_LEN);
4392                 kfree(save);
4393         }
4394
4395         if (err) {
4396                 ext4_msg(sb, KERN_ERR, "error loading journal");
4397                 jbd2_journal_destroy(journal);
4398                 return err;
4399         }
4400
4401         EXT4_SB(sb)->s_journal = journal;
4402         ext4_clear_journal_err(sb, es);
4403
4404         if (!really_read_only && journal_devnum &&
4405             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4406                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4407
4408                 /* Make sure we flush the recovery flag to disk. */
4409                 ext4_commit_super(sb, 1);
4410         }
4411
4412         return 0;
4413 }
4414
4415 static int ext4_commit_super(struct super_block *sb, int sync)
4416 {
4417         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4418         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4419         int error = 0;
4420
4421         if (!sbh || block_device_ejected(sb))
4422                 return error;
4423         if (buffer_write_io_error(sbh)) {
4424                 /*
4425                  * Oh, dear.  A previous attempt to write the
4426                  * superblock failed.  This could happen because the
4427                  * USB device was yanked out.  Or it could happen to
4428                  * be a transient write error and maybe the block will
4429                  * be remapped.  Nothing we can do but to retry the
4430                  * write and hope for the best.
4431                  */
4432                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4433                        "superblock detected");
4434                 clear_buffer_write_io_error(sbh);
4435                 set_buffer_uptodate(sbh);
4436         }
4437         /*
4438          * If the file system is mounted read-only, don't update the
4439          * superblock write time.  This avoids updating the superblock
4440          * write time when we are mounting the root file system
4441          * read/only but we need to replay the journal; at that point,
4442          * for people who are east of GMT and who make their clock
4443          * tick in localtime for Windows bug-for-bug compatibility,
4444          * the clock is set in the future, and this will cause e2fsck
4445          * to complain and force a full file system check.
4446          */
4447         if (!(sb->s_flags & MS_RDONLY))
4448                 es->s_wtime = cpu_to_le32(get_seconds());
4449         if (sb->s_bdev->bd_part)
4450                 es->s_kbytes_written =
4451                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4452                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4453                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4454         else
4455                 es->s_kbytes_written =
4456                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4457         ext4_free_blocks_count_set(es,
4458                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4459                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4460         es->s_free_inodes_count =
4461                 cpu_to_le32(percpu_counter_sum_positive(
4462                                 &EXT4_SB(sb)->s_freeinodes_counter));
4463         BUFFER_TRACE(sbh, "marking dirty");
4464         ext4_superblock_csum_set(sb);
4465         mark_buffer_dirty(sbh);
4466         if (sync) {
4467                 error = sync_dirty_buffer(sbh);
4468                 if (error)
4469                         return error;
4470
4471                 error = buffer_write_io_error(sbh);
4472                 if (error) {
4473                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4474                                "superblock");
4475                         clear_buffer_write_io_error(sbh);
4476                         set_buffer_uptodate(sbh);
4477                 }
4478         }
4479         return error;
4480 }
4481
4482 /*
4483  * Have we just finished recovery?  If so, and if we are mounting (or
4484  * remounting) the filesystem readonly, then we will end up with a
4485  * consistent fs on disk.  Record that fact.
4486  */
4487 static void ext4_mark_recovery_complete(struct super_block *sb,
4488                                         struct ext4_super_block *es)
4489 {
4490         journal_t *journal = EXT4_SB(sb)->s_journal;
4491
4492         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL)) {
4493                 BUG_ON(journal != NULL);
4494                 return;
4495         }
4496         jbd2_journal_lock_updates(journal);
4497         if (jbd2_journal_flush(journal) < 0)
4498                 goto out;
4499
4500         if (EXT4_HAS_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER) &&
4501             sb->s_flags & MS_RDONLY) {
4502                 EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4503                 ext4_commit_super(sb, 1);
4504         }
4505
4506 out:
4507         jbd2_journal_unlock_updates(journal);
4508 }
4509
4510 /*
4511  * If we are mounting (or read-write remounting) a filesystem whose journal
4512  * has recorded an error from a previous lifetime, move that error to the
4513  * main filesystem now.
4514  */
4515 static void ext4_clear_journal_err(struct super_block *sb,
4516                                    struct ext4_super_block *es)
4517 {
4518         journal_t *journal;
4519         int j_errno;
4520         const char *errstr;
4521
4522         BUG_ON(!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL));
4523
4524         journal = EXT4_SB(sb)->s_journal;
4525
4526         /*
4527          * Now check for any error status which may have been recorded in the
4528          * journal by a prior ext4_error() or ext4_abort()
4529          */
4530
4531         j_errno = jbd2_journal_errno(journal);
4532         if (j_errno) {
4533                 char nbuf[16];
4534
4535                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4536                 ext4_warning(sb, "Filesystem error recorded "
4537                              "from previous mount: %s", errstr);
4538                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4539
4540                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4541                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4542                 ext4_commit_super(sb, 1);
4543
4544                 jbd2_journal_clear_err(journal);
4545                 jbd2_journal_update_sb_errno(journal);
4546         }
4547 }
4548
4549 /*
4550  * Force the running and committing transactions to commit,
4551  * and wait on the commit.
4552  */
4553 int ext4_force_commit(struct super_block *sb)
4554 {
4555         journal_t *journal;
4556
4557         if (sb->s_flags & MS_RDONLY)
4558                 return 0;
4559
4560         journal = EXT4_SB(sb)->s_journal;
4561         return ext4_journal_force_commit(journal);
4562 }
4563
4564 static int ext4_sync_fs(struct super_block *sb, int wait)
4565 {
4566         int ret = 0;
4567         tid_t target;
4568         bool needs_barrier = false;
4569         struct ext4_sb_info *sbi = EXT4_SB(sb);
4570
4571         trace_ext4_sync_fs(sb, wait);
4572         flush_workqueue(sbi->rsv_conversion_wq);
4573         flush_workqueue(sbi->unrsv_conversion_wq);
4574         /*
4575          * Writeback quota in non-journalled quota case - journalled quota has
4576          * no dirty dquots
4577          */
4578         dquot_writeback_dquots(sb, -1);
4579         /*
4580          * Data writeback is possible w/o journal transaction, so barrier must
4581          * being sent at the end of the function. But we can skip it if
4582          * transaction_commit will do it for us.
4583          */
4584         target = jbd2_get_latest_transaction(sbi->s_journal);
4585         if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4586             !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4587                 needs_barrier = true;
4588
4589         if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4590                 if (wait)
4591                         ret = jbd2_log_wait_commit(sbi->s_journal, target);
4592         }
4593         if (needs_barrier) {
4594                 int err;
4595                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4596                 if (!ret)
4597                         ret = err;
4598         }
4599
4600         return ret;
4601 }
4602
4603 static int ext4_sync_fs_nojournal(struct super_block *sb, int wait)
4604 {
4605         int ret = 0;
4606
4607         trace_ext4_sync_fs(sb, wait);
4608         flush_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4609         flush_workqueue(EXT4_SB(sb)->unrsv_conversion_wq);
4610         dquot_writeback_dquots(sb, -1);
4611         if (wait && test_opt(sb, BARRIER))
4612                 ret = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4613
4614         return ret;
4615 }
4616
4617 /*
4618  * LVM calls this function before a (read-only) snapshot is created.  This
4619  * gives us a chance to flush the journal completely and mark the fs clean.
4620  *
4621  * Note that only this function cannot bring a filesystem to be in a clean
4622  * state independently. It relies on upper layer to stop all data & metadata
4623  * modifications.
4624  */
4625 static int ext4_freeze(struct super_block *sb)
4626 {
4627         int error = 0;
4628         journal_t *journal;
4629
4630         if (sb->s_flags & MS_RDONLY)
4631                 return 0;
4632
4633         journal = EXT4_SB(sb)->s_journal;
4634
4635         /* Now we set up the journal barrier. */
4636         jbd2_journal_lock_updates(journal);
4637
4638         /*
4639          * Don't clear the needs_recovery flag if we failed to flush
4640          * the journal.
4641          */
4642         error = jbd2_journal_flush(journal);
4643         if (error < 0)
4644                 goto out;
4645
4646         /* Journal blocked and flushed, clear needs_recovery flag. */
4647         EXT4_CLEAR_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4648         error = ext4_commit_super(sb, 1);
4649 out:
4650         /* we rely on upper layer to stop further updates */
4651         jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4652         return error;
4653 }
4654
4655 /*
4656  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4657  * flag here, even though the filesystem is not technically dirty yet.
4658  */
4659 static int ext4_unfreeze(struct super_block *sb)
4660 {
4661         if (sb->s_flags & MS_RDONLY)
4662                 return 0;
4663
4664         /* Reset the needs_recovery flag before the fs is unlocked. */
4665         EXT4_SET_INCOMPAT_FEATURE(sb, EXT4_FEATURE_INCOMPAT_RECOVER);
4666         ext4_commit_super(sb, 1);
4667         return 0;
4668 }
4669
4670 /*
4671  * Structure to save mount options for ext4_remount's benefit
4672  */
4673 struct ext4_mount_options {
4674         unsigned long s_mount_opt;
4675         unsigned long s_mount_opt2;
4676         kuid_t s_resuid;
4677         kgid_t s_resgid;
4678         unsigned long s_commit_interval;
4679         u32 s_min_batch_time, s_max_batch_time;
4680 #ifdef CONFIG_QUOTA
4681         int s_jquota_fmt;
4682         char *s_qf_names[MAXQUOTAS];
4683 #endif
4684 };
4685
4686 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4687 {
4688         struct ext4_super_block *es;
4689         struct ext4_sb_info *sbi = EXT4_SB(sb);
4690         unsigned long old_sb_flags;
4691         struct ext4_mount_options old_opts;
4692         int enable_quota = 0;
4693         ext4_group_t g;
4694         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4695         int err = 0;
4696 #ifdef CONFIG_QUOTA
4697         int i, j;
4698 #endif
4699         char *orig_data = kstrdup(data, GFP_KERNEL);
4700
4701         /* Store the original options */
4702         old_sb_flags = sb->s_flags;
4703         old_opts.s_mount_opt = sbi->s_mount_opt;
4704         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4705         old_opts.s_resuid = sbi->s_resuid;
4706         old_opts.s_resgid = sbi->s_resgid;
4707         old_opts.s_commit_interval = sbi->s_commit_interval;
4708         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4709         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4710 #ifdef CONFIG_QUOTA
4711         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4712         for (i = 0; i < MAXQUOTAS; i++)
4713                 if (sbi->s_qf_names[i]) {
4714                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4715                                                          GFP_KERNEL);
4716                         if (!old_opts.s_qf_names[i]) {
4717                                 for (j = 0; j < i; j++)
4718                                         kfree(old_opts.s_qf_names[j]);
4719                                 kfree(orig_data);
4720                                 return -ENOMEM;
4721                         }
4722                 } else
4723                         old_opts.s_qf_names[i] = NULL;
4724 #endif
4725         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4726                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4727
4728         /*
4729          * Allow the "check" option to be passed as a remount option.
4730          */
4731         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4732                 err = -EINVAL;
4733                 goto restore_opts;
4734         }
4735
4736         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4737                 ext4_abort(sb, "Abort forced by user");
4738
4739         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4740                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4741
4742         es = sbi->s_es;
4743
4744         if (sbi->s_journal) {
4745                 ext4_init_journal_params(sb, sbi->s_journal);
4746                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4747         }
4748
4749         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4750                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4751                         err = -EROFS;
4752                         goto restore_opts;
4753                 }
4754
4755                 if (*flags & MS_RDONLY) {
4756                         err = dquot_suspend(sb, -1);
4757                         if (err < 0)
4758                                 goto restore_opts;
4759
4760                         /*
4761                          * First of all, the unconditional stuff we have to do
4762                          * to disable replay of the journal when we next remount
4763                          */
4764                         sb->s_flags |= MS_RDONLY;
4765
4766                         /*
4767                          * OK, test if we are remounting a valid rw partition
4768                          * readonly, and if so set the rdonly flag and then
4769                          * mark the partition as valid again.
4770                          */
4771                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4772                             (sbi->s_mount_state & EXT4_VALID_FS))
4773                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4774
4775                         if (sbi->s_journal)
4776                                 ext4_mark_recovery_complete(sb, es);
4777                 } else {
4778                         /* Make sure we can mount this feature set readwrite */
4779                         if (!ext4_feature_set_ok(sb, 0)) {
4780                                 err = -EROFS;
4781                                 goto restore_opts;
4782                         }
4783                         /*
4784                          * Make sure the group descriptor checksums
4785                          * are sane.  If they aren't, refuse to remount r/w.
4786                          */
4787                         for (g = 0; g < sbi->s_groups_count; g++) {
4788                                 struct ext4_group_desc *gdp =
4789                                         ext4_get_group_desc(sb, g, NULL);
4790
4791                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4792                                         ext4_msg(sb, KERN_ERR,
4793                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4794                 g, le16_to_cpu(ext4_group_desc_csum(sbi, g, gdp)),
4795                                                le16_to_cpu(gdp->bg_checksum));
4796                                         err = -EINVAL;
4797                                         goto restore_opts;
4798                                 }
4799                         }
4800
4801                         /*
4802                          * If we have an unprocessed orphan list hanging
4803                          * around from a previously readonly bdev mount,
4804                          * require a full umount/remount for now.
4805                          */
4806                         if (es->s_last_orphan) {
4807                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4808                                        "remount RDWR because of unprocessed "
4809                                        "orphan inode list.  Please "
4810                                        "umount/remount instead");
4811                                 err = -EINVAL;
4812                                 goto restore_opts;
4813                         }
4814
4815                         /*
4816                          * Mounting a RDONLY partition read-write, so reread
4817                          * and store the current valid flag.  (It may have
4818                          * been changed by e2fsck since we originally mounted
4819                          * the partition.)
4820                          */
4821                         if (sbi->s_journal)
4822                                 ext4_clear_journal_err(sb, es);
4823                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4824                         if (!ext4_setup_super(sb, es, 0))
4825                                 sb->s_flags &= ~MS_RDONLY;
4826                         if (EXT4_HAS_INCOMPAT_FEATURE(sb,
4827                                                      EXT4_FEATURE_INCOMPAT_MMP))
4828                                 if (ext4_multi_mount_protect(sb,
4829                                                 le64_to_cpu(es->s_mmp_block))) {
4830                                         err = -EROFS;
4831                                         goto restore_opts;
4832                                 }
4833                         enable_quota = 1;
4834                 }
4835         }
4836
4837         /*
4838          * Reinitialize lazy itable initialization thread based on
4839          * current settings
4840          */
4841         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4842                 ext4_unregister_li_request(sb);
4843         else {
4844                 ext4_group_t first_not_zeroed;
4845                 first_not_zeroed = ext4_has_uninit_itable(sb);
4846                 ext4_register_li_request(sb, first_not_zeroed);
4847         }
4848
4849         ext4_setup_system_zone(sb);
4850         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4851                 ext4_commit_super(sb, 1);
4852
4853 #ifdef CONFIG_QUOTA
4854         /* Release old quota file names */
4855         for (i = 0; i < MAXQUOTAS; i++)
4856                 kfree(old_opts.s_qf_names[i]);
4857         if (enable_quota) {
4858                 if (sb_any_quota_suspended(sb))
4859                         dquot_resume(sb, -1);
4860                 else if (EXT4_HAS_RO_COMPAT_FEATURE(sb,
4861                                         EXT4_FEATURE_RO_COMPAT_QUOTA)) {
4862                         err = ext4_enable_quotas(sb);
4863                         if (err)
4864                                 goto restore_opts;
4865                 }
4866         }
4867 #endif
4868
4869         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4870         kfree(orig_data);
4871         return 0;
4872
4873 restore_opts:
4874         sb->s_flags = old_sb_flags;
4875         sbi->s_mount_opt = old_opts.s_mount_opt;
4876         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4877         sbi->s_resuid = old_opts.s_resuid;
4878         sbi->s_resgid = old_opts.s_resgid;
4879         sbi->s_commit_interval = old_opts.s_commit_interval;
4880         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4881         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4882 #ifdef CONFIG_QUOTA
4883         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4884         for (i = 0; i < MAXQUOTAS; i++) {
4885                 kfree(sbi->s_qf_names[i]);
4886                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4887         }
4888 #endif
4889         kfree(orig_data);
4890         return err;
4891 }
4892
4893 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4894 {
4895         struct super_block *sb = dentry->d_sb;
4896         struct ext4_sb_info *sbi = EXT4_SB(sb);
4897         struct ext4_super_block *es = sbi->s_es;
4898         ext4_fsblk_t overhead = 0, resv_blocks;
4899         u64 fsid;
4900         s64 bfree;
4901         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4902
4903         if (!test_opt(sb, MINIX_DF))
4904                 overhead = sbi->s_overhead;
4905
4906         buf->f_type = EXT4_SUPER_MAGIC;
4907         buf->f_bsize = sb->s_blocksize;
4908         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4909         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4910                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4911         /* prevent underflow in case that few free space is available */
4912         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4913         buf->f_bavail = buf->f_bfree -
4914                         (ext4_r_blocks_count(es) + resv_blocks);
4915         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4916                 buf->f_bavail = 0;
4917         buf->f_files = le32_to_cpu(es->s_inodes_count);
4918         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4919         buf->f_namelen = EXT4_NAME_LEN;
4920         fsid = le64_to_cpup((void *)es->s_uuid) ^
4921                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4922         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4923         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4924
4925         return 0;
4926 }
4927
4928 /* Helper function for writing quotas on sync - we need to start transaction
4929  * before quota file is locked for write. Otherwise the are possible deadlocks:
4930  * Process 1                         Process 2
4931  * ext4_create()                     quota_sync()
4932  *   jbd2_journal_start()                  write_dquot()
4933  *   dquot_initialize()                         down(dqio_mutex)
4934  *     down(dqio_mutex)                    jbd2_journal_start()
4935  *
4936  */
4937
4938 #ifdef CONFIG_QUOTA
4939
4940 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4941 {
4942         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4943 }
4944
4945 static int ext4_write_dquot(struct dquot *dquot)
4946 {
4947         int ret, err;
4948         handle_t *handle;
4949         struct inode *inode;
4950
4951         inode = dquot_to_inode(dquot);
4952         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4953                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4954         if (IS_ERR(handle))
4955                 return PTR_ERR(handle);
4956         ret = dquot_commit(dquot);
4957         err = ext4_journal_stop(handle);
4958         if (!ret)
4959                 ret = err;
4960         return ret;
4961 }
4962
4963 static int ext4_acquire_dquot(struct dquot *dquot)
4964 {
4965         int ret, err;
4966         handle_t *handle;
4967
4968         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4969                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4970         if (IS_ERR(handle))
4971                 return PTR_ERR(handle);
4972         ret = dquot_acquire(dquot);
4973         err = ext4_journal_stop(handle);
4974         if (!ret)
4975                 ret = err;
4976         return ret;
4977 }
4978
4979 static int ext4_release_dquot(struct dquot *dquot)
4980 {
4981         int ret, err;
4982         handle_t *handle;
4983
4984         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4985                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4986         if (IS_ERR(handle)) {
4987                 /* Release dquot anyway to avoid endless cycle in dqput() */
4988                 dquot_release(dquot);
4989                 return PTR_ERR(handle);
4990         }
4991         ret = dquot_release(dquot);
4992         err = ext4_journal_stop(handle);
4993         if (!ret)
4994                 ret = err;
4995         return ret;
4996 }
4997
4998 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4999 {
5000         struct super_block *sb = dquot->dq_sb;
5001         struct ext4_sb_info *sbi = EXT4_SB(sb);
5002
5003         /* Are we journaling quotas? */
5004         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA) ||
5005             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5006                 dquot_mark_dquot_dirty(dquot);
5007                 return ext4_write_dquot(dquot);
5008         } else {
5009                 return dquot_mark_dquot_dirty(dquot);
5010         }
5011 }
5012
5013 static int ext4_write_info(struct super_block *sb, int type)
5014 {
5015         int ret, err;
5016         handle_t *handle;
5017
5018         /* Data block + inode block */
5019         handle = ext4_journal_start(sb->s_root->d_inode, EXT4_HT_QUOTA, 2);
5020         if (IS_ERR(handle))
5021                 return PTR_ERR(handle);
5022         ret = dquot_commit_info(sb, type);
5023         err = ext4_journal_stop(handle);
5024         if (!ret)
5025                 ret = err;
5026         return ret;
5027 }
5028
5029 /*
5030  * Turn on quotas during mount time - we need to find
5031  * the quota file and such...
5032  */
5033 static int ext4_quota_on_mount(struct super_block *sb, int type)
5034 {
5035         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5036                                         EXT4_SB(sb)->s_jquota_fmt, type);
5037 }
5038
5039 /*
5040  * Standard function to be called on quota_on
5041  */
5042 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5043                          struct path *path)
5044 {
5045         int err;
5046
5047         if (!test_opt(sb, QUOTA))
5048                 return -EINVAL;
5049
5050         /* Quotafile not on the same filesystem? */
5051         if (path->dentry->d_sb != sb)
5052                 return -EXDEV;
5053         /* Journaling quota? */
5054         if (EXT4_SB(sb)->s_qf_names[type]) {
5055                 /* Quotafile not in fs root? */
5056                 if (path->dentry->d_parent != sb->s_root)
5057                         ext4_msg(sb, KERN_WARNING,
5058                                 "Quota file not on filesystem root. "
5059                                 "Journaled quota will not work");
5060         }
5061
5062         /*
5063          * When we journal data on quota file, we have to flush journal to see
5064          * all updates to the file when we bypass pagecache...
5065          */
5066         if (EXT4_SB(sb)->s_journal &&
5067             ext4_should_journal_data(path->dentry->d_inode)) {
5068                 /*
5069                  * We don't need to lock updates but journal_flush() could
5070                  * otherwise be livelocked...
5071                  */
5072                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5073                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5074                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5075                 if (err)
5076                         return err;
5077         }
5078
5079         return dquot_quota_on(sb, type, format_id, path);
5080 }
5081
5082 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5083                              unsigned int flags)
5084 {
5085         int err;
5086         struct inode *qf_inode;
5087         unsigned long qf_inums[MAXQUOTAS] = {
5088                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5089                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5090         };
5091
5092         BUG_ON(!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA));
5093
5094         if (!qf_inums[type])
5095                 return -EPERM;
5096
5097         qf_inode = ext4_iget(sb, qf_inums[type]);
5098         if (IS_ERR(qf_inode)) {
5099                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5100                 return PTR_ERR(qf_inode);
5101         }
5102
5103         /* Don't account quota for quota files to avoid recursion */
5104         qf_inode->i_flags |= S_NOQUOTA;
5105         err = dquot_enable(qf_inode, type, format_id, flags);
5106         iput(qf_inode);
5107
5108         return err;
5109 }
5110
5111 /* Enable usage tracking for all quota types. */
5112 static int ext4_enable_quotas(struct super_block *sb)
5113 {
5114         int type, err = 0;
5115         unsigned long qf_inums[MAXQUOTAS] = {
5116                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5117                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5118         };
5119
5120         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5121         for (type = 0; type < MAXQUOTAS; type++) {
5122                 if (qf_inums[type]) {
5123                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5124                                                 DQUOT_USAGE_ENABLED);
5125                         if (err) {
5126                                 ext4_warning(sb,
5127                                         "Failed to enable quota tracking "
5128                                         "(type=%d, err=%d). Please run "
5129                                         "e2fsck to fix.", type, err);
5130                                 return err;
5131                         }
5132                 }
5133         }
5134         return 0;
5135 }
5136
5137 /*
5138  * quota_on function that is used when QUOTA feature is set.
5139  */
5140 static int ext4_quota_on_sysfile(struct super_block *sb, int type,
5141                                  int format_id)
5142 {
5143         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5144                 return -EINVAL;
5145
5146         /*
5147          * USAGE was enabled at mount time. Only need to enable LIMITS now.
5148          */
5149         return ext4_quota_enable(sb, type, format_id, DQUOT_LIMITS_ENABLED);
5150 }
5151
5152 static int ext4_quota_off(struct super_block *sb, int type)
5153 {
5154         struct inode *inode = sb_dqopt(sb)->files[type];
5155         handle_t *handle;
5156
5157         /* Force all delayed allocation blocks to be allocated.
5158          * Caller already holds s_umount sem */
5159         if (test_opt(sb, DELALLOC))
5160                 sync_filesystem(sb);
5161
5162         if (!inode)
5163                 goto out;
5164
5165         /* Update modification times of quota files when userspace can
5166          * start looking at them */
5167         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5168         if (IS_ERR(handle))
5169                 goto out;
5170         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5171         ext4_mark_inode_dirty(handle, inode);
5172         ext4_journal_stop(handle);
5173
5174 out:
5175         return dquot_quota_off(sb, type);
5176 }
5177
5178 /*
5179  * quota_off function that is used when QUOTA feature is set.
5180  */
5181 static int ext4_quota_off_sysfile(struct super_block *sb, int type)
5182 {
5183         if (!EXT4_HAS_RO_COMPAT_FEATURE(sb, EXT4_FEATURE_RO_COMPAT_QUOTA))
5184                 return -EINVAL;
5185
5186         /* Disable only the limits. */
5187         return dquot_disable(sb, type, DQUOT_LIMITS_ENABLED);
5188 }
5189
5190 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5191  * acquiring the locks... As quota files are never truncated and quota code
5192  * itself serializes the operations (and no one else should touch the files)
5193  * we don't have to be afraid of races */
5194 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5195                                size_t len, loff_t off)
5196 {
5197         struct inode *inode = sb_dqopt(sb)->files[type];
5198         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5199         int err = 0;
5200         int offset = off & (sb->s_blocksize - 1);
5201         int tocopy;
5202         size_t toread;
5203         struct buffer_head *bh;
5204         loff_t i_size = i_size_read(inode);
5205
5206         if (off > i_size)
5207                 return 0;
5208         if (off+len > i_size)
5209                 len = i_size-off;
5210         toread = len;
5211         while (toread > 0) {
5212                 tocopy = sb->s_blocksize - offset < toread ?
5213                                 sb->s_blocksize - offset : toread;
5214                 bh = ext4_bread(NULL, inode, blk, 0, &err);
5215                 if (err)
5216                         return err;
5217                 if (!bh)        /* A hole? */
5218                         memset(data, 0, tocopy);
5219                 else
5220                         memcpy(data, bh->b_data+offset, tocopy);
5221                 brelse(bh);
5222                 offset = 0;
5223                 toread -= tocopy;
5224                 data += tocopy;
5225                 blk++;
5226         }
5227         return len;
5228 }
5229
5230 /* Write to quotafile (we know the transaction is already started and has
5231  * enough credits) */
5232 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5233                                 const char *data, size_t len, loff_t off)
5234 {
5235         struct inode *inode = sb_dqopt(sb)->files[type];
5236         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5237         int err = 0;
5238         int offset = off & (sb->s_blocksize - 1);
5239         struct buffer_head *bh;
5240         handle_t *handle = journal_current_handle();
5241
5242         if (EXT4_SB(sb)->s_journal && !handle) {
5243                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5244                         " cancelled because transaction is not started",
5245                         (unsigned long long)off, (unsigned long long)len);
5246                 return -EIO;
5247         }
5248         /*
5249          * Since we account only one data block in transaction credits,
5250          * then it is impossible to cross a block boundary.
5251          */
5252         if (sb->s_blocksize - offset < len) {
5253                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5254                         " cancelled because not block aligned",
5255                         (unsigned long long)off, (unsigned long long)len);
5256                 return -EIO;
5257         }
5258
5259         bh = ext4_bread(handle, inode, blk, 1, &err);
5260         if (!bh)
5261                 goto out;
5262         err = ext4_journal_get_write_access(handle, bh);
5263         if (err) {
5264                 brelse(bh);
5265                 goto out;
5266         }
5267         lock_buffer(bh);
5268         memcpy(bh->b_data+offset, data, len);
5269         flush_dcache_page(bh->b_page);
5270         unlock_buffer(bh);
5271         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5272         brelse(bh);
5273 out:
5274         if (err)
5275                 return err;
5276         if (inode->i_size < off + len) {
5277                 i_size_write(inode, off + len);
5278                 EXT4_I(inode)->i_disksize = inode->i_size;
5279                 ext4_mark_inode_dirty(handle, inode);
5280         }
5281         return len;
5282 }
5283
5284 #endif
5285
5286 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5287                        const char *dev_name, void *data)
5288 {
5289         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5290 }
5291
5292 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5293 static inline void register_as_ext2(void)
5294 {
5295         int err = register_filesystem(&ext2_fs_type);
5296         if (err)
5297                 printk(KERN_WARNING
5298                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5299 }
5300
5301 static inline void unregister_as_ext2(void)
5302 {
5303         unregister_filesystem(&ext2_fs_type);
5304 }
5305
5306 static inline int ext2_feature_set_ok(struct super_block *sb)
5307 {
5308         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT2_FEATURE_INCOMPAT_SUPP))
5309                 return 0;
5310         if (sb->s_flags & MS_RDONLY)
5311                 return 1;
5312         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT2_FEATURE_RO_COMPAT_SUPP))
5313                 return 0;
5314         return 1;
5315 }
5316 #else
5317 static inline void register_as_ext2(void) { }
5318 static inline void unregister_as_ext2(void) { }
5319 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5320 #endif
5321
5322 #if !defined(CONFIG_EXT3_FS) && !defined(CONFIG_EXT3_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT23)
5323 static inline void register_as_ext3(void)
5324 {
5325         int err = register_filesystem(&ext3_fs_type);
5326         if (err)
5327                 printk(KERN_WARNING
5328                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5329 }
5330
5331 static inline void unregister_as_ext3(void)
5332 {
5333         unregister_filesystem(&ext3_fs_type);
5334 }
5335
5336 static inline int ext3_feature_set_ok(struct super_block *sb)
5337 {
5338         if (EXT4_HAS_INCOMPAT_FEATURE(sb, ~EXT3_FEATURE_INCOMPAT_SUPP))
5339                 return 0;
5340         if (!EXT4_HAS_COMPAT_FEATURE(sb, EXT4_FEATURE_COMPAT_HAS_JOURNAL))
5341                 return 0;
5342         if (sb->s_flags & MS_RDONLY)
5343                 return 1;
5344         if (EXT4_HAS_RO_COMPAT_FEATURE(sb, ~EXT3_FEATURE_RO_COMPAT_SUPP))
5345                 return 0;
5346         return 1;
5347 }
5348 #else
5349 static inline void register_as_ext3(void) { }
5350 static inline void unregister_as_ext3(void) { }
5351 static inline int ext3_feature_set_ok(struct super_block *sb) { return 0; }
5352 #endif
5353
5354 static struct file_system_type ext4_fs_type = {
5355         .owner          = THIS_MODULE,
5356         .name           = "ext4",
5357         .mount          = ext4_mount,
5358         .kill_sb        = kill_block_super,
5359         .fs_flags       = FS_REQUIRES_DEV,
5360 };
5361 MODULE_ALIAS_FS("ext4");
5362
5363 static int __init ext4_init_feat_adverts(void)
5364 {
5365         struct ext4_features *ef;
5366         int ret = -ENOMEM;
5367
5368         ef = kzalloc(sizeof(struct ext4_features), GFP_KERNEL);
5369         if (!ef)
5370                 goto out;
5371
5372         ef->f_kobj.kset = ext4_kset;
5373         init_completion(&ef->f_kobj_unregister);
5374         ret = kobject_init_and_add(&ef->f_kobj, &ext4_feat_ktype, NULL,
5375                                    "features");
5376         if (ret) {
5377                 kfree(ef);
5378                 goto out;
5379         }
5380
5381         ext4_feat = ef;
5382         ret = 0;
5383 out:
5384         return ret;
5385 }
5386
5387 static void ext4_exit_feat_adverts(void)
5388 {
5389         kobject_put(&ext4_feat->f_kobj);
5390         wait_for_completion(&ext4_feat->f_kobj_unregister);
5391         kfree(ext4_feat);
5392 }
5393
5394 /* Shared across all ext4 file systems */
5395 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5396 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5397
5398 static int __init ext4_init_fs(void)
5399 {
5400         int i, err;
5401
5402         ext4_li_info = NULL;
5403         mutex_init(&ext4_li_mtx);
5404
5405         /* Build-time check for flags consistency */
5406         ext4_check_flag_values();
5407
5408         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5409                 mutex_init(&ext4__aio_mutex[i]);
5410                 init_waitqueue_head(&ext4__ioend_wq[i]);
5411         }
5412
5413         err = ext4_init_es();
5414         if (err)
5415                 return err;
5416
5417         err = ext4_init_pageio();
5418         if (err)
5419                 goto out7;
5420
5421         err = ext4_init_system_zone();
5422         if (err)
5423                 goto out6;
5424         ext4_kset = kset_create_and_add("ext4", NULL, fs_kobj);
5425         if (!ext4_kset) {
5426                 err = -ENOMEM;
5427                 goto out5;
5428         }
5429         ext4_proc_root = proc_mkdir("fs/ext4", NULL);
5430
5431         err = ext4_init_feat_adverts();
5432         if (err)
5433                 goto out4;
5434
5435         err = ext4_init_mballoc();
5436         if (err)
5437                 goto out3;
5438
5439         err = ext4_init_xattr();
5440         if (err)
5441                 goto out2;
5442         err = init_inodecache();
5443         if (err)
5444                 goto out1;
5445         register_as_ext3();
5446         register_as_ext2();
5447         err = register_filesystem(&ext4_fs_type);
5448         if (err)
5449                 goto out;
5450
5451         return 0;
5452 out:
5453         unregister_as_ext2();
5454         unregister_as_ext3();
5455         destroy_inodecache();
5456 out1:
5457         ext4_exit_xattr();
5458 out2:
5459         ext4_exit_mballoc();
5460 out3:
5461         ext4_exit_feat_adverts();
5462 out4:
5463         if (ext4_proc_root)
5464                 remove_proc_entry("fs/ext4", NULL);
5465         kset_unregister(ext4_kset);
5466 out5:
5467         ext4_exit_system_zone();
5468 out6:
5469         ext4_exit_pageio();
5470 out7:
5471         ext4_exit_es();
5472
5473         return err;
5474 }
5475
5476 static void __exit ext4_exit_fs(void)
5477 {
5478         ext4_destroy_lazyinit_thread();
5479         unregister_as_ext2();
5480         unregister_as_ext3();
5481         unregister_filesystem(&ext4_fs_type);
5482         destroy_inodecache();
5483         ext4_exit_xattr();
5484         ext4_exit_mballoc();
5485         ext4_exit_feat_adverts();
5486         remove_proc_entry("fs/ext4", NULL);
5487         kset_unregister(ext4_kset);
5488         ext4_exit_system_zone();
5489         ext4_exit_pageio();
5490 }
5491
5492 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5493 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5494 MODULE_LICENSE("GPL");
5495 module_init(ext4_init_fs)
5496 module_exit(ext4_exit_fs)