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/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.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/ctype.h>
  38 #include <linux/log2.h>
  39 #include <linux/crc16.h>
  40 #include <linux/cleancache.h>
  41 #include <asm/uaccess.h>
  42
  43 #include <linux/kthread.h>
  44 #include <linux/freezer.h>
  45
  46 #include "ext4.h"
  47 #include "ext4_extents.h"       /* Needed for trace points definition */
  48 #include "ext4_jbd2.h"
  49 #include "xattr.h"
  50 #include "acl.h"
  51 #include "mballoc.h"
  52
  53 #define CREATE_TRACE_POINTS
  54 #include <trace/events/ext4.h>
  55
  56 static struct ext4_lazy_init *ext4_li_info;
  57 static struct mutex ext4_li_mtx;
  58 static int ext4_mballoc_ready;
  59 static struct ratelimit_state ext4_mount_msg_ratelimit;
  60
  61 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  62                              unsigned long journal_devnum);
  63 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  64 static int ext4_commit_super(struct super_block *sb, int sync);
  65 static void ext4_mark_recovery_complete(struct super_block *sb,
  66                                         struct ext4_super_block *es);
  67 static void ext4_clear_journal_err(struct super_block *sb,
  68                                    struct ext4_super_block *es);
  69 static int ext4_sync_fs(struct super_block *sb, int wait);
  70 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  71 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  72 static int ext4_unfreeze(struct super_block *sb);
  73 static int ext4_freeze(struct super_block *sb);
  74 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  75                        const char *dev_name, void *data);
  76 static inline int ext2_feature_set_ok(struct super_block *sb);
  77 static inline int ext3_feature_set_ok(struct super_block *sb);
  78 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  79 static void ext4_destroy_lazyinit_thread(void);
  80 static void ext4_unregister_li_request(struct super_block *sb);
  81 static void ext4_clear_request_list(void);
  82
  83 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
  84 static struct file_system_type ext2_fs_type = {
  85         .owner          = THIS_MODULE,
  86         .name           = "ext2",
  87         .mount          = ext4_mount,
  88         .kill_sb        = kill_block_super,
  89         .fs_flags       = FS_REQUIRES_DEV,
  90 };
  91 MODULE_ALIAS_FS("ext2");
  92 MODULE_ALIAS("ext2");
  93 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  94 #else
  95 #define IS_EXT2_SB(sb) (0)
  96 #endif
  97
  98
  99 static struct file_system_type ext3_fs_type = {
 100         .owner          = THIS_MODULE,
 101         .name           = "ext3",
 102         .mount          = ext4_mount,
 103         .kill_sb        = kill_block_super,
 104         .fs_flags       = FS_REQUIRES_DEV,
 105 };
 106 MODULE_ALIAS_FS("ext3");
 107 MODULE_ALIAS("ext3");
 108 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 109
 110 static int ext4_verify_csum_type(struct super_block *sb,
 111                                  struct ext4_super_block *es)
 112 {
 113         if (!ext4_has_feature_metadata_csum(sb))
 114                 return 1;
 115
 116         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 117 }
 118
 119 static __le32 ext4_superblock_csum(struct super_block *sb,
 120                                    struct ext4_super_block *es)
 121 {
 122         struct ext4_sb_info *sbi = EXT4_SB(sb);
 123         int offset = offsetof(struct ext4_super_block, s_checksum);
 124         __u32 csum;
 125
 126         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 127
 128         return cpu_to_le32(csum);
 129 }
 130
 131 static int ext4_superblock_csum_verify(struct super_block *sb,
 132                                        struct ext4_super_block *es)
 133 {
 134         if (!ext4_has_metadata_csum(sb))
 135                 return 1;
 136
 137         return es->s_checksum == ext4_superblock_csum(sb, es);
 138 }
 139
 140 void ext4_superblock_csum_set(struct super_block *sb)
 141 {
 142         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 143
 144         if (!ext4_has_metadata_csum(sb))
 145                 return;
 146
 147         es->s_checksum = ext4_superblock_csum(sb, es);
 148 }
 149
 150 void *ext4_kvmalloc(size_t size, gfp_t flags)
 151 {
 152         void *ret;
 153
 154         ret = kmalloc(size, flags | __GFP_NOWARN);
 155         if (!ret)
 156                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 157         return ret;
 158 }
 159
 160 void *ext4_kvzalloc(size_t size, gfp_t flags)
 161 {
 162         void *ret;
 163
 164         ret = kzalloc(size, flags | __GFP_NOWARN);
 165         if (!ret)
 166                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 167         return ret;
 168 }
 169
 170 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 171                                struct ext4_group_desc *bg)
 172 {
 173         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 174                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 175                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 176 }
 177
 178 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 179                                struct ext4_group_desc *bg)
 180 {
 181         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 182                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 183                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 184 }
 185
 186 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 187                               struct ext4_group_desc *bg)
 188 {
 189         return le32_to_cpu(bg->bg_inode_table_lo) |
 190                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 191                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 192 }
 193
 194 __u32 ext4_free_group_clusters(struct super_block *sb,
 195                                struct ext4_group_desc *bg)
 196 {
 197         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 198                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 199                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 200 }
 201
 202 __u32 ext4_free_inodes_count(struct super_block *sb,
 203                               struct ext4_group_desc *bg)
 204 {
 205         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 206                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 207                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 208 }
 209
 210 __u32 ext4_used_dirs_count(struct super_block *sb,
 211                               struct ext4_group_desc *bg)
 212 {
 213         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 214                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 215                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 216 }
 217
 218 __u32 ext4_itable_unused_count(struct super_block *sb,
 219                               struct ext4_group_desc *bg)
 220 {
 221         return le16_to_cpu(bg->bg_itable_unused_lo) |
 222                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 223                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 224 }
 225
 226 void ext4_block_bitmap_set(struct super_block *sb,
 227                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 228 {
 229         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 230         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 231                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 232 }
 233
 234 void ext4_inode_bitmap_set(struct super_block *sb,
 235                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 236 {
 237         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 238         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 239                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 240 }
 241
 242 void ext4_inode_table_set(struct super_block *sb,
 243                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 244 {
 245         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 246         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 247                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 248 }
 249
 250 void ext4_free_group_clusters_set(struct super_block *sb,
 251                                   struct ext4_group_desc *bg, __u32 count)
 252 {
 253         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 254         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 255                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 256 }
 257
 258 void ext4_free_inodes_set(struct super_block *sb,
 259                           struct ext4_group_desc *bg, __u32 count)
 260 {
 261         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 262         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 263                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 264 }
 265
 266 void ext4_used_dirs_set(struct super_block *sb,
 267                           struct ext4_group_desc *bg, __u32 count)
 268 {
 269         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 270         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 271                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 272 }
 273
 274 void ext4_itable_unused_set(struct super_block *sb,
 275                           struct ext4_group_desc *bg, __u32 count)
 276 {
 277         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 278         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 279                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 280 }
 281
 282
 283 static void __save_error_info(struct super_block *sb, const char *func,
 284                             unsigned int line)
 285 {
 286         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 287
 288         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 289         if (bdev_read_only(sb->s_bdev))
 290                 return;
 291         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 292         es->s_last_error_time = cpu_to_le32(get_seconds());
 293         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 294         es->s_last_error_line = cpu_to_le32(line);
 295         if (!es->s_first_error_time) {
 296                 es->s_first_error_time = es->s_last_error_time;
 297                 strncpy(es->s_first_error_func, func,
 298                         sizeof(es->s_first_error_func));
 299                 es->s_first_error_line = cpu_to_le32(line);
 300                 es->s_first_error_ino = es->s_last_error_ino;
 301                 es->s_first_error_block = es->s_last_error_block;
 302         }
 303         /*
 304          * Start the daily error reporting function if it hasn't been
 305          * started already
 306          */
 307         if (!es->s_error_count)
 308                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 309         le32_add_cpu(&es->s_error_count, 1);
 310 }
 311
 312 static void save_error_info(struct super_block *sb, const char *func,
 313                             unsigned int line)
 314 {
 315         __save_error_info(sb, func, line);
 316         ext4_commit_super(sb, 1);
 317 }
 318
 319 /*
 320  * The del_gendisk() function uninitializes the disk-specific data
 321  * structures, including the bdi structure, without telling anyone
 322  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 323  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 324  * This is a kludge to prevent these oops until we can put in a proper
 325  * hook in del_gendisk() to inform the VFS and file system layers.
 326  */
 327 static int block_device_ejected(struct super_block *sb)
 328 {
 329         struct inode *bd_inode = sb->s_bdev->bd_inode;
 330         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 331
 332         return bdi->dev == NULL;
 333 }
 334
 335 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 336 {
 337         struct super_block              *sb = journal->j_private;
 338         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 339         int                             error = is_journal_aborted(journal);
 340         struct ext4_journal_cb_entry    *jce;
 341
 342         BUG_ON(txn->t_state == T_FINISHED);
 343         spin_lock(&sbi->s_md_lock);
 344         while (!list_empty(&txn->t_private_list)) {
 345                 jce = list_entry(txn->t_private_list.next,
 346                                  struct ext4_journal_cb_entry, jce_list);
 347                 list_del_init(&jce->jce_list);
 348                 spin_unlock(&sbi->s_md_lock);
 349                 jce->jce_func(sb, jce, error);
 350                 spin_lock(&sbi->s_md_lock);
 351         }
 352         spin_unlock(&sbi->s_md_lock);
 353 }
 354
 355 /* Deal with the reporting of failure conditions on a filesystem such as
 356  * inconsistencies detected or read IO failures.
 357  *
 358  * On ext2, we can store the error state of the filesystem in the
 359  * superblock.  That is not possible on ext4, because we may have other
 360  * write ordering constraints on the superblock which prevent us from
 361  * writing it out straight away; and given that the journal is about to
 362  * be aborted, we can't rely on the current, or future, transactions to
 363  * write out the superblock safely.
 364  *
 365  * We'll just use the jbd2_journal_abort() error code to record an error in
 366  * the journal instead.  On recovery, the journal will complain about
 367  * that error until we've noted it down and cleared it.
 368  */
 369
 370 static void ext4_handle_error(struct super_block *sb)
 371 {
 372         if (sb->s_flags & MS_RDONLY)
 373                 return;
 374
 375         if (!test_opt(sb, ERRORS_CONT)) {
 376                 journal_t *journal = EXT4_SB(sb)->s_journal;
 377
 378                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 379                 if (journal)
 380                         jbd2_journal_abort(journal, -EIO);
 381         }
 382         if (test_opt(sb, ERRORS_RO)) {
 383                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 384                 /*
 385                  * Make sure updated value of ->s_mount_flags will be visible
 386                  * before ->s_flags update
 387                  */
 388                 smp_wmb();
 389                 sb->s_flags |= MS_RDONLY;
 390         }
 391         if (test_opt(sb, ERRORS_PANIC)) {
 392                 if (EXT4_SB(sb)->s_journal &&
 393                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 394                         return;
 395                 panic("EXT4-fs (device %s): panic forced after error\n",
 396                         sb->s_id);
 397         }
 398 }
 399
 400 #define ext4_error_ratelimit(sb)                                        \
 401                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 402                              "EXT4-fs error")
 403
 404 void __ext4_error(struct super_block *sb, const char *function,
 405                   unsigned int line, const char *fmt, ...)
 406 {
 407         struct va_format vaf;
 408         va_list args;
 409
 410         if (ext4_error_ratelimit(sb)) {
 411                 va_start(args, fmt);
 412                 vaf.fmt = fmt;
 413                 vaf.va = &args;
 414                 printk(KERN_CRIT
 415                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 416                        sb->s_id, function, line, current->comm, &vaf);
 417                 va_end(args);
 418         }
 419         save_error_info(sb, function, line);
 420         ext4_handle_error(sb);
 421 }
 422
 423 void __ext4_error_inode(struct inode *inode, const char *function,
 424                         unsigned int line, ext4_fsblk_t block,
 425                         const char *fmt, ...)
 426 {
 427         va_list args;
 428         struct va_format vaf;
 429         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 430
 431         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 432         es->s_last_error_block = cpu_to_le64(block);
 433         if (ext4_error_ratelimit(inode->i_sb)) {
 434                 va_start(args, fmt);
 435                 vaf.fmt = fmt;
 436                 vaf.va = &args;
 437                 if (block)
 438                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 439                                "inode #%lu: block %llu: comm %s: %pV\n",
 440                                inode->i_sb->s_id, function, line, inode->i_ino,
 441                                block, current->comm, &vaf);
 442                 else
 443                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 444                                "inode #%lu: comm %s: %pV\n",
 445                                inode->i_sb->s_id, function, line, inode->i_ino,
 446                                current->comm, &vaf);
 447                 va_end(args);
 448         }
 449         save_error_info(inode->i_sb, function, line);
 450         ext4_handle_error(inode->i_sb);
 451 }
 452
 453 void __ext4_error_file(struct file *file, const char *function,
 454                        unsigned int line, ext4_fsblk_t block,
 455                        const char *fmt, ...)
 456 {
 457         va_list args;
 458         struct va_format vaf;
 459         struct ext4_super_block *es;
 460         struct inode *inode = file_inode(file);
 461         char pathname[80], *path;
 462
 463         es = EXT4_SB(inode->i_sb)->s_es;
 464         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 465         if (ext4_error_ratelimit(inode->i_sb)) {
 466                 path = file_path(file, pathname, sizeof(pathname));
 467                 if (IS_ERR(path))
 468                         path = "(unknown)";
 469                 va_start(args, fmt);
 470                 vaf.fmt = fmt;
 471                 vaf.va = &args;
 472                 if (block)
 473                         printk(KERN_CRIT
 474                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 475                                "block %llu: comm %s: path %s: %pV\n",
 476                                inode->i_sb->s_id, function, line, inode->i_ino,
 477                                block, current->comm, path, &vaf);
 478                 else
 479                         printk(KERN_CRIT
 480                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 481                                "comm %s: path %s: %pV\n",
 482                                inode->i_sb->s_id, function, line, inode->i_ino,
 483                                current->comm, path, &vaf);
 484                 va_end(args);
 485         }
 486         save_error_info(inode->i_sb, function, line);
 487         ext4_handle_error(inode->i_sb);
 488 }
 489
 490 const char *ext4_decode_error(struct super_block *sb, int errno,
 491                               char nbuf[16])
 492 {
 493         char *errstr = NULL;
 494
 495         switch (errno) {
 496         case -EFSCORRUPTED:
 497                 errstr = "Corrupt filesystem";
 498                 break;
 499         case -EFSBADCRC:
 500                 errstr = "Filesystem failed CRC";
 501                 break;
 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         if (ext4_error_ratelimit(sb)) {
 547                 errstr = ext4_decode_error(sb, errno, nbuf);
 548                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 549                        sb->s_id, function, line, errstr);
 550         }
 551
 552         save_error_info(sb, function, line);
 553         ext4_handle_error(sb);
 554 }
 555
 556 /*
 557  * ext4_abort is a much stronger failure handler than ext4_error.  The
 558  * abort function may be used to deal with unrecoverable failures such
 559  * as journal IO errors or ENOMEM at a critical moment in log management.
 560  *
 561  * We unconditionally force the filesystem into an ABORT|READONLY state,
 562  * unless the error response on the fs has been set to panic in which
 563  * case we take the easy way out and panic immediately.
 564  */
 565
 566 void __ext4_abort(struct super_block *sb, const char *function,
 567                 unsigned int line, const char *fmt, ...)
 568 {
 569         va_list args;
 570
 571         save_error_info(sb, function, line);
 572         va_start(args, fmt);
 573         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 574                function, line);
 575         vprintk(fmt, args);
 576         printk("\n");
 577         va_end(args);
 578
 579         if ((sb->s_flags & MS_RDONLY) == 0) {
 580                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 581                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 582                 /*
 583                  * Make sure updated value of ->s_mount_flags will be visible
 584                  * before ->s_flags update
 585                  */
 586                 smp_wmb();
 587                 sb->s_flags |= MS_RDONLY;
 588                 if (EXT4_SB(sb)->s_journal)
 589                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 590                 save_error_info(sb, function, line);
 591         }
 592         if (test_opt(sb, ERRORS_PANIC)) {
 593                 if (EXT4_SB(sb)->s_journal &&
 594                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 595                         return;
 596                 panic("EXT4-fs panic from previous error\n");
 597         }
 598 }
 599
 600 void __ext4_msg(struct super_block *sb,
 601                 const char *prefix, const char *fmt, ...)
 602 {
 603         struct va_format vaf;
 604         va_list args;
 605
 606         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 607                 return;
 608
 609         va_start(args, fmt);
 610         vaf.fmt = fmt;
 611         vaf.va = &args;
 612         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 613         va_end(args);
 614 }
 615
 616 #define ext4_warning_ratelimit(sb)                                      \
 617                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 618                              "EXT4-fs warning")
 619
 620 void __ext4_warning(struct super_block *sb, const char *function,
 621                     unsigned int line, const char *fmt, ...)
 622 {
 623         struct va_format vaf;
 624         va_list args;
 625
 626         if (!ext4_warning_ratelimit(sb))
 627                 return;
 628
 629         va_start(args, fmt);
 630         vaf.fmt = fmt;
 631         vaf.va = &args;
 632         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 633                sb->s_id, function, line, &vaf);
 634         va_end(args);
 635 }
 636
 637 void __ext4_warning_inode(const struct inode *inode, const char *function,
 638                           unsigned int line, const char *fmt, ...)
 639 {
 640         struct va_format vaf;
 641         va_list args;
 642
 643         if (!ext4_warning_ratelimit(inode->i_sb))
 644                 return;
 645
 646         va_start(args, fmt);
 647         vaf.fmt = fmt;
 648         vaf.va = &args;
 649         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 650                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 651                function, line, inode->i_ino, current->comm, &vaf);
 652         va_end(args);
 653 }
 654
 655 void __ext4_grp_locked_error(const char *function, unsigned int line,
 656                              struct super_block *sb, ext4_group_t grp,
 657                              unsigned long ino, ext4_fsblk_t block,
 658                              const char *fmt, ...)
 659 __releases(bitlock)
 660 __acquires(bitlock)
 661 {
 662         struct va_format vaf;
 663         va_list args;
 664         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 665
 666         es->s_last_error_ino = cpu_to_le32(ino);
 667         es->s_last_error_block = cpu_to_le64(block);
 668         __save_error_info(sb, function, line);
 669
 670         if (ext4_error_ratelimit(sb)) {
 671                 va_start(args, fmt);
 672                 vaf.fmt = fmt;
 673                 vaf.va = &args;
 674                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 675                        sb->s_id, function, line, grp);
 676                 if (ino)
 677                         printk(KERN_CONT "inode %lu: ", ino);
 678                 if (block)
 679                         printk(KERN_CONT "block %llu:",
 680                                (unsigned long long) block);
 681                 printk(KERN_CONT "%pV\n", &vaf);
 682                 va_end(args);
 683         }
 684
 685         if (test_opt(sb, ERRORS_CONT)) {
 686                 ext4_commit_super(sb, 0);
 687                 return;
 688         }
 689
 690         ext4_unlock_group(sb, grp);
 691         ext4_handle_error(sb);
 692         /*
 693          * We only get here in the ERRORS_RO case; relocking the group
 694          * may be dangerous, but nothing bad will happen since the
 695          * filesystem will have already been marked read/only and the
 696          * journal has been aborted.  We return 1 as a hint to callers
 697          * who might what to use the return value from
 698          * ext4_grp_locked_error() to distinguish between the
 699          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 700          * aggressively from the ext4 function in question, with a
 701          * more appropriate error code.
 702          */
 703         ext4_lock_group(sb, grp);
 704         return;
 705 }
 706
 707 void ext4_update_dynamic_rev(struct super_block *sb)
 708 {
 709         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 710
 711         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 712                 return;
 713
 714         ext4_warning(sb,
 715                      "updating to rev %d because of new feature flag, "
 716                      "running e2fsck is recommended",
 717                      EXT4_DYNAMIC_REV);
 718
 719         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 720         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 721         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 722         /* leave es->s_feature_*compat flags alone */
 723         /* es->s_uuid will be set by e2fsck if empty */
 724
 725         /*
 726          * The rest of the superblock fields should be zero, and if not it
 727          * means they are likely already in use, so leave them alone.  We
 728          * can leave it up to e2fsck to clean up any inconsistencies there.
 729          */
 730 }
 731
 732 /*
 733  * Open the external journal device
 734  */
 735 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 736 {
 737         struct block_device *bdev;
 738         char b[BDEVNAME_SIZE];
 739
 740         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 741         if (IS_ERR(bdev))
 742                 goto fail;
 743         return bdev;
 744
 745 fail:
 746         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 747                         __bdevname(dev, b), PTR_ERR(bdev));
 748         return NULL;
 749 }
 750
 751 /*
 752  * Release the journal device
 753  */
 754 static void ext4_blkdev_put(struct block_device *bdev)
 755 {
 756         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 757 }
 758
 759 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 760 {
 761         struct block_device *bdev;
 762         bdev = sbi->journal_bdev;
 763         if (bdev) {
 764                 ext4_blkdev_put(bdev);
 765                 sbi->journal_bdev = NULL;
 766         }
 767 }
 768
 769 static inline struct inode *orphan_list_entry(struct list_head *l)
 770 {
 771         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 772 }
 773
 774 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 775 {
 776         struct list_head *l;
 777
 778         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 779                  le32_to_cpu(sbi->s_es->s_last_orphan));
 780
 781         printk(KERN_ERR "sb_info orphan list:\n");
 782         list_for_each(l, &sbi->s_orphan) {
 783                 struct inode *inode = orphan_list_entry(l);
 784                 printk(KERN_ERR "  "
 785                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 786                        inode->i_sb->s_id, inode->i_ino, inode,
 787                        inode->i_mode, inode->i_nlink,
 788                        NEXT_ORPHAN(inode));
 789         }
 790 }
 791
 792 static void ext4_put_super(struct super_block *sb)
 793 {
 794         struct ext4_sb_info *sbi = EXT4_SB(sb);
 795         struct ext4_super_block *es = sbi->s_es;
 796         int i, err;
 797
 798         ext4_unregister_li_request(sb);
 799         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 800
 801         flush_workqueue(sbi->rsv_conversion_wq);
 802         destroy_workqueue(sbi->rsv_conversion_wq);
 803
 804         if (sbi->s_journal) {
 805                 err = jbd2_journal_destroy(sbi->s_journal);
 806                 sbi->s_journal = NULL;
 807                 if (err < 0)
 808                         ext4_abort(sb, "Couldn't clean up the journal");
 809         }
 810
 811         ext4_unregister_sysfs(sb);
 812         ext4_es_unregister_shrinker(sbi);
 813         del_timer_sync(&sbi->s_err_report);
 814         ext4_release_system_zone(sb);
 815         ext4_mb_release(sb);
 816         ext4_ext_release(sb);
 817         ext4_xattr_put_super(sb);
 818
 819         if (!(sb->s_flags & MS_RDONLY)) {
 820                 ext4_clear_feature_journal_needs_recovery(sb);
 821                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 822         }
 823         if (!(sb->s_flags & MS_RDONLY))
 824                 ext4_commit_super(sb, 1);
 825
 826         for (i = 0; i < sbi->s_gdb_count; i++)
 827                 brelse(sbi->s_group_desc[i]);
 828         kvfree(sbi->s_group_desc);
 829         kvfree(sbi->s_flex_groups);
 830         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 831         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 832         percpu_counter_destroy(&sbi->s_dirs_counter);
 833         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 834         brelse(sbi->s_sbh);
 835 #ifdef CONFIG_QUOTA
 836         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 837                 kfree(sbi->s_qf_names[i]);
 838 #endif
 839
 840         /* Debugging code just in case the in-memory inode orphan list
 841          * isn't empty.  The on-disk one can be non-empty if we've
 842          * detected an error and taken the fs readonly, but the
 843          * in-memory list had better be clean by this point. */
 844         if (!list_empty(&sbi->s_orphan))
 845                 dump_orphan_list(sb, sbi);
 846         J_ASSERT(list_empty(&sbi->s_orphan));
 847
 848         sync_blockdev(sb->s_bdev);
 849         invalidate_bdev(sb->s_bdev);
 850         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 851                 /*
 852                  * Invalidate the journal device's buffers.  We don't want them
 853                  * floating about in memory - the physical journal device may
 854                  * hotswapped, and it breaks the `ro-after' testing code.
 855                  */
 856                 sync_blockdev(sbi->journal_bdev);
 857                 invalidate_bdev(sbi->journal_bdev);
 858                 ext4_blkdev_remove(sbi);
 859         }
 860         if (sbi->s_mb_cache) {
 861                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 862                 sbi->s_mb_cache = NULL;
 863         }
 864         if (sbi->s_mmp_tsk)
 865                 kthread_stop(sbi->s_mmp_tsk);
 866         sb->s_fs_info = NULL;
 867         /*
 868          * Now that we are completely done shutting down the
 869          * superblock, we need to actually destroy the kobject.
 870          */
 871         kobject_put(&sbi->s_kobj);
 872         wait_for_completion(&sbi->s_kobj_unregister);
 873         if (sbi->s_chksum_driver)
 874                 crypto_free_shash(sbi->s_chksum_driver);
 875         kfree(sbi->s_blockgroup_lock);
 876         kfree(sbi);
 877 }
 878
 879 static struct kmem_cache *ext4_inode_cachep;
 880
 881 /*
 882  * Called inside transaction, so use GFP_NOFS
 883  */
 884 static struct inode *ext4_alloc_inode(struct super_block *sb)
 885 {
 886         struct ext4_inode_info *ei;
 887
 888         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 889         if (!ei)
 890                 return NULL;
 891
 892         ei->vfs_inode.i_version = 1;
 893         spin_lock_init(&ei->i_raw_lock);
 894         INIT_LIST_HEAD(&ei->i_prealloc_list);
 895         spin_lock_init(&ei->i_prealloc_lock);
 896         ext4_es_init_tree(&ei->i_es_tree);
 897         rwlock_init(&ei->i_es_lock);
 898         INIT_LIST_HEAD(&ei->i_es_list);
 899         ei->i_es_all_nr = 0;
 900         ei->i_es_shk_nr = 0;
 901         ei->i_es_shrink_lblk = 0;
 902         ei->i_reserved_data_blocks = 0;
 903         ei->i_reserved_meta_blocks = 0;
 904         ei->i_allocated_meta_blocks = 0;
 905         ei->i_da_metadata_calc_len = 0;
 906         ei->i_da_metadata_calc_last_lblock = 0;
 907         spin_lock_init(&(ei->i_block_reservation_lock));
 908 #ifdef CONFIG_QUOTA
 909         ei->i_reserved_quota = 0;
 910         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 911 #endif
 912         ei->jinode = NULL;
 913         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 914         spin_lock_init(&ei->i_completed_io_lock);
 915         ei->i_sync_tid = 0;
 916         ei->i_datasync_tid = 0;
 917         atomic_set(&ei->i_ioend_count, 0);
 918         atomic_set(&ei->i_unwritten, 0);
 919         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 920 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 921         ei->i_crypt_info = NULL;
 922 #endif
 923         return &ei->vfs_inode;
 924 }
 925
 926 static int ext4_drop_inode(struct inode *inode)
 927 {
 928         int drop = generic_drop_inode(inode);
 929
 930         trace_ext4_drop_inode(inode, drop);
 931         return drop;
 932 }
 933
 934 static void ext4_i_callback(struct rcu_head *head)
 935 {
 936         struct inode *inode = container_of(head, struct inode, i_rcu);
 937         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 938 }
 939
 940 static void ext4_destroy_inode(struct inode *inode)
 941 {
 942         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 943                 ext4_msg(inode->i_sb, KERN_ERR,
 944                          "Inode %lu (%p): orphan list check failed!",
 945                          inode->i_ino, EXT4_I(inode));
 946                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 947                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 948                                 true);
 949                 dump_stack();
 950         }
 951         call_rcu(&inode->i_rcu, ext4_i_callback);
 952 }
 953
 954 static void init_once(void *foo)
 955 {
 956         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 957
 958         INIT_LIST_HEAD(&ei->i_orphan);
 959         init_rwsem(&ei->xattr_sem);
 960         init_rwsem(&ei->i_data_sem);
 961         inode_init_once(&ei->vfs_inode);
 962 }
 963
 964 static int __init init_inodecache(void)
 965 {
 966         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 967                                              sizeof(struct ext4_inode_info),
 968                                              0, (SLAB_RECLAIM_ACCOUNT|
 969                                                 SLAB_MEM_SPREAD),
 970                                              init_once);
 971         if (ext4_inode_cachep == NULL)
 972                 return -ENOMEM;
 973         return 0;
 974 }
 975
 976 static void destroy_inodecache(void)
 977 {
 978         /*
 979          * Make sure all delayed rcu free inodes are flushed before we
 980          * destroy cache.
 981          */
 982         rcu_barrier();
 983         kmem_cache_destroy(ext4_inode_cachep);
 984 }
 985
 986 void ext4_clear_inode(struct inode *inode)
 987 {
 988         invalidate_inode_buffers(inode);
 989         clear_inode(inode);
 990         dquot_drop(inode);
 991         ext4_discard_preallocations(inode);
 992         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 993         if (EXT4_I(inode)->jinode) {
 994                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 995                                                EXT4_I(inode)->jinode);
 996                 jbd2_free_inode(EXT4_I(inode)->jinode);
 997                 EXT4_I(inode)->jinode = NULL;
 998         }
 999 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1000         if (EXT4_I(inode)->i_crypt_info)
1001                 ext4_free_encryption_info(inode, EXT4_I(inode)->i_crypt_info);
1002 #endif
1003 }
1004
1005 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1006                                         u64 ino, u32 generation)
1007 {
1008         struct inode *inode;
1009
1010         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1011                 return ERR_PTR(-ESTALE);
1012         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1013                 return ERR_PTR(-ESTALE);
1014
1015         /* iget isn't really right if the inode is currently unallocated!!
1016          *
1017          * ext4_read_inode will return a bad_inode if the inode had been
1018          * deleted, so we should be safe.
1019          *
1020          * Currently we don't know the generation for parent directory, so
1021          * a generation of 0 means "accept any"
1022          */
1023         inode = ext4_iget_normal(sb, ino);
1024         if (IS_ERR(inode))
1025                 return ERR_CAST(inode);
1026         if (generation && inode->i_generation != generation) {
1027                 iput(inode);
1028                 return ERR_PTR(-ESTALE);
1029         }
1030
1031         return inode;
1032 }
1033
1034 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1035                                         int fh_len, int fh_type)
1036 {
1037         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1038                                     ext4_nfs_get_inode);
1039 }
1040
1041 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1042                                         int fh_len, int fh_type)
1043 {
1044         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1045                                     ext4_nfs_get_inode);
1046 }
1047
1048 /*
1049  * Try to release metadata pages (indirect blocks, directories) which are
1050  * mapped via the block device.  Since these pages could have journal heads
1051  * which would prevent try_to_free_buffers() from freeing them, we must use
1052  * jbd2 layer's try_to_free_buffers() function to release them.
1053  */
1054 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1055                                  gfp_t wait)
1056 {
1057         journal_t *journal = EXT4_SB(sb)->s_journal;
1058
1059         WARN_ON(PageChecked(page));
1060         if (!page_has_buffers(page))
1061                 return 0;
1062         if (journal)
1063                 return jbd2_journal_try_to_free_buffers(journal, page,
1064                                                 wait & ~__GFP_DIRECT_RECLAIM);
1065         return try_to_free_buffers(page);
1066 }
1067
1068 #ifdef CONFIG_QUOTA
1069 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1070 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1071
1072 static int ext4_write_dquot(struct dquot *dquot);
1073 static int ext4_acquire_dquot(struct dquot *dquot);
1074 static int ext4_release_dquot(struct dquot *dquot);
1075 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1076 static int ext4_write_info(struct super_block *sb, int type);
1077 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1078                          struct path *path);
1079 static int ext4_quota_off(struct super_block *sb, int type);
1080 static int ext4_quota_on_mount(struct super_block *sb, int type);
1081 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1082                                size_t len, loff_t off);
1083 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1084                                 const char *data, size_t len, loff_t off);
1085 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1086                              unsigned int flags);
1087 static int ext4_enable_quotas(struct super_block *sb);
1088
1089 static struct dquot **ext4_get_dquots(struct inode *inode)
1090 {
1091         return EXT4_I(inode)->i_dquot;
1092 }
1093
1094 static const struct dquot_operations ext4_quota_operations = {
1095         .get_reserved_space = ext4_get_reserved_space,
1096         .write_dquot    = ext4_write_dquot,
1097         .acquire_dquot  = ext4_acquire_dquot,
1098         .release_dquot  = ext4_release_dquot,
1099         .mark_dirty     = ext4_mark_dquot_dirty,
1100         .write_info     = ext4_write_info,
1101         .alloc_dquot    = dquot_alloc,
1102         .destroy_dquot  = dquot_destroy,
1103 };
1104
1105 static const struct quotactl_ops ext4_qctl_operations = {
1106         .quota_on       = ext4_quota_on,
1107         .quota_off      = ext4_quota_off,
1108         .quota_sync     = dquot_quota_sync,
1109         .get_state      = dquot_get_state,
1110         .set_info       = dquot_set_dqinfo,
1111         .get_dqblk      = dquot_get_dqblk,
1112         .set_dqblk      = dquot_set_dqblk
1113 };
1114 #endif
1115
1116 static const struct super_operations ext4_sops = {
1117         .alloc_inode    = ext4_alloc_inode,
1118         .destroy_inode  = ext4_destroy_inode,
1119         .write_inode    = ext4_write_inode,
1120         .dirty_inode    = ext4_dirty_inode,
1121         .drop_inode     = ext4_drop_inode,
1122         .evict_inode    = ext4_evict_inode,
1123         .put_super      = ext4_put_super,
1124         .sync_fs        = ext4_sync_fs,
1125         .freeze_fs      = ext4_freeze,
1126         .unfreeze_fs    = ext4_unfreeze,
1127         .statfs         = ext4_statfs,
1128         .remount_fs     = ext4_remount,
1129         .show_options   = ext4_show_options,
1130 #ifdef CONFIG_QUOTA
1131         .quota_read     = ext4_quota_read,
1132         .quota_write    = ext4_quota_write,
1133         .get_dquots     = ext4_get_dquots,
1134 #endif
1135         .bdev_try_to_free_page = bdev_try_to_free_page,
1136 };
1137
1138 static const struct export_operations ext4_export_ops = {
1139         .fh_to_dentry = ext4_fh_to_dentry,
1140         .fh_to_parent = ext4_fh_to_parent,
1141         .get_parent = ext4_get_parent,
1142 };
1143
1144 enum {
1145         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1146         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1147         Opt_nouid32, Opt_debug, Opt_removed,
1148         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1149         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1150         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1151         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1152         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1153         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1154         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1155         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1156         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1157         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1158         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1159         Opt_lazytime, Opt_nolazytime,
1160         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1161         Opt_inode_readahead_blks, Opt_journal_ioprio,
1162         Opt_dioread_nolock, Opt_dioread_lock,
1163         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1164         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1165 };
1166
1167 static const match_table_t tokens = {
1168         {Opt_bsd_df, "bsddf"},
1169         {Opt_minix_df, "minixdf"},
1170         {Opt_grpid, "grpid"},
1171         {Opt_grpid, "bsdgroups"},
1172         {Opt_nogrpid, "nogrpid"},
1173         {Opt_nogrpid, "sysvgroups"},
1174         {Opt_resgid, "resgid=%u"},
1175         {Opt_resuid, "resuid=%u"},
1176         {Opt_sb, "sb=%u"},
1177         {Opt_err_cont, "errors=continue"},
1178         {Opt_err_panic, "errors=panic"},
1179         {Opt_err_ro, "errors=remount-ro"},
1180         {Opt_nouid32, "nouid32"},
1181         {Opt_debug, "debug"},
1182         {Opt_removed, "oldalloc"},
1183         {Opt_removed, "orlov"},
1184         {Opt_user_xattr, "user_xattr"},
1185         {Opt_nouser_xattr, "nouser_xattr"},
1186         {Opt_acl, "acl"},
1187         {Opt_noacl, "noacl"},
1188         {Opt_noload, "norecovery"},
1189         {Opt_noload, "noload"},
1190         {Opt_removed, "nobh"},
1191         {Opt_removed, "bh"},
1192         {Opt_commit, "commit=%u"},
1193         {Opt_min_batch_time, "min_batch_time=%u"},
1194         {Opt_max_batch_time, "max_batch_time=%u"},
1195         {Opt_journal_dev, "journal_dev=%u"},
1196         {Opt_journal_path, "journal_path=%s"},
1197         {Opt_journal_checksum, "journal_checksum"},
1198         {Opt_nojournal_checksum, "nojournal_checksum"},
1199         {Opt_journal_async_commit, "journal_async_commit"},
1200         {Opt_abort, "abort"},
1201         {Opt_data_journal, "data=journal"},
1202         {Opt_data_ordered, "data=ordered"},
1203         {Opt_data_writeback, "data=writeback"},
1204         {Opt_data_err_abort, "data_err=abort"},
1205         {Opt_data_err_ignore, "data_err=ignore"},
1206         {Opt_offusrjquota, "usrjquota="},
1207         {Opt_usrjquota, "usrjquota=%s"},
1208         {Opt_offgrpjquota, "grpjquota="},
1209         {Opt_grpjquota, "grpjquota=%s"},
1210         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1211         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1212         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1213         {Opt_grpquota, "grpquota"},
1214         {Opt_noquota, "noquota"},
1215         {Opt_quota, "quota"},
1216         {Opt_usrquota, "usrquota"},
1217         {Opt_barrier, "barrier=%u"},
1218         {Opt_barrier, "barrier"},
1219         {Opt_nobarrier, "nobarrier"},
1220         {Opt_i_version, "i_version"},
1221         {Opt_dax, "dax"},
1222         {Opt_stripe, "stripe=%u"},
1223         {Opt_delalloc, "delalloc"},
1224         {Opt_lazytime, "lazytime"},
1225         {Opt_nolazytime, "nolazytime"},
1226         {Opt_nodelalloc, "nodelalloc"},
1227         {Opt_removed, "mblk_io_submit"},
1228         {Opt_removed, "nomblk_io_submit"},
1229         {Opt_block_validity, "block_validity"},
1230         {Opt_noblock_validity, "noblock_validity"},
1231         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1232         {Opt_journal_ioprio, "journal_ioprio=%u"},
1233         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1234         {Opt_auto_da_alloc, "auto_da_alloc"},
1235         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1236         {Opt_dioread_nolock, "dioread_nolock"},
1237         {Opt_dioread_lock, "dioread_lock"},
1238         {Opt_discard, "discard"},
1239         {Opt_nodiscard, "nodiscard"},
1240         {Opt_init_itable, "init_itable=%u"},
1241         {Opt_init_itable, "init_itable"},
1242         {Opt_noinit_itable, "noinit_itable"},
1243         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1244         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1245         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1246         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1247         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1248         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1249         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1250         {Opt_err, NULL},
1251 };
1252
1253 static ext4_fsblk_t get_sb_block(void **data)
1254 {
1255         ext4_fsblk_t    sb_block;
1256         char            *options = (char *) *data;
1257
1258         if (!options || strncmp(options, "sb=", 3) != 0)
1259                 return 1;       /* Default location */
1260
1261         options += 3;
1262         /* TODO: use simple_strtoll with >32bit ext4 */
1263         sb_block = simple_strtoul(options, &options, 0);
1264         if (*options && *options != ',') {
1265                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1266                        (char *) *data);
1267                 return 1;
1268         }
1269         if (*options == ',')
1270                 options++;
1271         *data = (void *) options;
1272
1273         return sb_block;
1274 }
1275
1276 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1277 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1278         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1279
1280 #ifdef CONFIG_QUOTA
1281 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1282 {
1283         struct ext4_sb_info *sbi = EXT4_SB(sb);
1284         char *qname;
1285         int ret = -1;
1286
1287         if (sb_any_quota_loaded(sb) &&
1288                 !sbi->s_qf_names[qtype]) {
1289                 ext4_msg(sb, KERN_ERR,
1290                         "Cannot change journaled "
1291                         "quota options when quota turned on");
1292                 return -1;
1293         }
1294         if (ext4_has_feature_quota(sb)) {
1295                 ext4_msg(sb, KERN_ERR, "Cannot set journaled quota options "
1296                          "when QUOTA feature is enabled");
1297                 return -1;
1298         }
1299         qname = match_strdup(args);
1300         if (!qname) {
1301                 ext4_msg(sb, KERN_ERR,
1302                         "Not enough memory for storing quotafile name");
1303                 return -1;
1304         }
1305         if (sbi->s_qf_names[qtype]) {
1306                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1307                         ret = 1;
1308                 else
1309                         ext4_msg(sb, KERN_ERR,
1310                                  "%s quota file already specified",
1311                                  QTYPE2NAME(qtype));
1312                 goto errout;
1313         }
1314         if (strchr(qname, '/')) {
1315                 ext4_msg(sb, KERN_ERR,
1316                         "quotafile must be on filesystem root");
1317                 goto errout;
1318         }
1319         sbi->s_qf_names[qtype] = qname;
1320         set_opt(sb, QUOTA);
1321         return 1;
1322 errout:
1323         kfree(qname);
1324         return ret;
1325 }
1326
1327 static int clear_qf_name(struct super_block *sb, int qtype)
1328 {
1329
1330         struct ext4_sb_info *sbi = EXT4_SB(sb);
1331
1332         if (sb_any_quota_loaded(sb) &&
1333                 sbi->s_qf_names[qtype]) {
1334                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1335                         " when quota turned on");
1336                 return -1;
1337         }
1338         kfree(sbi->s_qf_names[qtype]);
1339         sbi->s_qf_names[qtype] = NULL;
1340         return 1;
1341 }
1342 #endif
1343
1344 #define MOPT_SET        0x0001
1345 #define MOPT_CLEAR      0x0002
1346 #define MOPT_NOSUPPORT  0x0004
1347 #define MOPT_EXPLICIT   0x0008
1348 #define MOPT_CLEAR_ERR  0x0010
1349 #define MOPT_GTE0       0x0020
1350 #ifdef CONFIG_QUOTA
1351 #define MOPT_Q          0
1352 #define MOPT_QFMT       0x0040
1353 #else
1354 #define MOPT_Q          MOPT_NOSUPPORT
1355 #define MOPT_QFMT       MOPT_NOSUPPORT
1356 #endif
1357 #define MOPT_DATAJ      0x0080
1358 #define MOPT_NO_EXT2    0x0100
1359 #define MOPT_NO_EXT3    0x0200
1360 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1361 #define MOPT_STRING     0x0400
1362
1363 static const struct mount_opts {
1364         int     token;
1365         int     mount_opt;
1366         int     flags;
1367 } ext4_mount_opts[] = {
1368         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1369         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1370         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1371         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1372         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1373         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1374         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1375          MOPT_EXT4_ONLY | MOPT_SET},
1376         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1377          MOPT_EXT4_ONLY | MOPT_CLEAR},
1378         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1379         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1380         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1381          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1382         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1383          MOPT_EXT4_ONLY | MOPT_CLEAR},
1384         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1385          MOPT_EXT4_ONLY | MOPT_CLEAR},
1386         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1387          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1388         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1389                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1390          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1391         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1392         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1393         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1394         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1395         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1396          MOPT_NO_EXT2 | MOPT_SET},
1397         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1398          MOPT_NO_EXT2 | MOPT_CLEAR},
1399         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1400         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1401         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1402         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1403         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1404         {Opt_commit, 0, MOPT_GTE0},
1405         {Opt_max_batch_time, 0, MOPT_GTE0},
1406         {Opt_min_batch_time, 0, MOPT_GTE0},
1407         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1408         {Opt_init_itable, 0, MOPT_GTE0},
1409         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1410         {Opt_stripe, 0, MOPT_GTE0},
1411         {Opt_resuid, 0, MOPT_GTE0},
1412         {Opt_resgid, 0, MOPT_GTE0},
1413         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1414         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1415         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1416         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1417         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1418         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1419          MOPT_NO_EXT2 | MOPT_DATAJ},
1420         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1421         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1422 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1423         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1424         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1425 #else
1426         {Opt_acl, 0, MOPT_NOSUPPORT},
1427         {Opt_noacl, 0, MOPT_NOSUPPORT},
1428 #endif
1429         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1430         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1431         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1432         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1433                                                         MOPT_SET | MOPT_Q},
1434         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1435                                                         MOPT_SET | MOPT_Q},
1436         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1437                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1438         {Opt_usrjquota, 0, MOPT_Q},
1439         {Opt_grpjquota, 0, MOPT_Q},
1440         {Opt_offusrjquota, 0, MOPT_Q},
1441         {Opt_offgrpjquota, 0, MOPT_Q},
1442         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1443         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1444         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1445         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1446         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1447         {Opt_err, 0, 0}
1448 };
1449
1450 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1451                             substring_t *args, unsigned long *journal_devnum,
1452                             unsigned int *journal_ioprio, int is_remount)
1453 {
1454         struct ext4_sb_info *sbi = EXT4_SB(sb);
1455         const struct mount_opts *m;
1456         kuid_t uid;
1457         kgid_t gid;
1458         int arg = 0;
1459
1460 #ifdef CONFIG_QUOTA
1461         if (token == Opt_usrjquota)
1462                 return set_qf_name(sb, USRQUOTA, &args[0]);
1463         else if (token == Opt_grpjquota)
1464                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1465         else if (token == Opt_offusrjquota)
1466                 return clear_qf_name(sb, USRQUOTA);
1467         else if (token == Opt_offgrpjquota)
1468                 return clear_qf_name(sb, GRPQUOTA);
1469 #endif
1470         switch (token) {
1471         case Opt_noacl:
1472         case Opt_nouser_xattr:
1473                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1474                 break;
1475         case Opt_sb:
1476                 return 1;       /* handled by get_sb_block() */
1477         case Opt_removed:
1478                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1479                 return 1;
1480         case Opt_abort:
1481                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1482                 return 1;
1483         case Opt_i_version:
1484                 sb->s_flags |= MS_I_VERSION;
1485                 return 1;
1486         case Opt_lazytime:
1487                 sb->s_flags |= MS_LAZYTIME;
1488                 return 1;
1489         case Opt_nolazytime:
1490                 sb->s_flags &= ~MS_LAZYTIME;
1491                 return 1;
1492         }
1493
1494         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1495                 if (token == m->token)
1496                         break;
1497
1498         if (m->token == Opt_err) {
1499                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1500                          "or missing value", opt);
1501                 return -1;
1502         }
1503
1504         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1505                 ext4_msg(sb, KERN_ERR,
1506                          "Mount option \"%s\" incompatible with ext2", opt);
1507                 return -1;
1508         }
1509         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1510                 ext4_msg(sb, KERN_ERR,
1511                          "Mount option \"%s\" incompatible with ext3", opt);
1512                 return -1;
1513         }
1514
1515         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1516                 return -1;
1517         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1518                 return -1;
1519         if (m->flags & MOPT_EXPLICIT) {
1520                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1521                         set_opt2(sb, EXPLICIT_DELALLOC);
1522                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1523                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1524                 } else
1525                         return -1;
1526         }
1527         if (m->flags & MOPT_CLEAR_ERR)
1528                 clear_opt(sb, ERRORS_MASK);
1529         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1530                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1531                          "options when quota turned on");
1532                 return -1;
1533         }
1534
1535         if (m->flags & MOPT_NOSUPPORT) {
1536                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1537         } else if (token == Opt_commit) {
1538                 if (arg == 0)
1539                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1540                 sbi->s_commit_interval = HZ * arg;
1541         } else if (token == Opt_max_batch_time) {
1542                 sbi->s_max_batch_time = arg;
1543         } else if (token == Opt_min_batch_time) {
1544                 sbi->s_min_batch_time = arg;
1545         } else if (token == Opt_inode_readahead_blks) {
1546                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1547                         ext4_msg(sb, KERN_ERR,
1548                                  "EXT4-fs: inode_readahead_blks must be "
1549                                  "0 or a power of 2 smaller than 2^31");
1550                         return -1;
1551                 }
1552                 sbi->s_inode_readahead_blks = arg;
1553         } else if (token == Opt_init_itable) {
1554                 set_opt(sb, INIT_INODE_TABLE);
1555                 if (!args->from)
1556                         arg = EXT4_DEF_LI_WAIT_MULT;
1557                 sbi->s_li_wait_mult = arg;
1558         } else if (token == Opt_max_dir_size_kb) {
1559                 sbi->s_max_dir_size_kb = arg;
1560         } else if (token == Opt_stripe) {
1561                 sbi->s_stripe = arg;
1562         } else if (token == Opt_resuid) {
1563                 uid = make_kuid(current_user_ns(), arg);
1564                 if (!uid_valid(uid)) {
1565                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1566                         return -1;
1567                 }
1568                 sbi->s_resuid = uid;
1569         } else if (token == Opt_resgid) {
1570                 gid = make_kgid(current_user_ns(), arg);
1571                 if (!gid_valid(gid)) {
1572                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1573                         return -1;
1574                 }
1575                 sbi->s_resgid = gid;
1576         } else if (token == Opt_journal_dev) {
1577                 if (is_remount) {
1578                         ext4_msg(sb, KERN_ERR,
1579                                  "Cannot specify journal on remount");
1580                         return -1;
1581                 }
1582                 *journal_devnum = arg;
1583         } else if (token == Opt_journal_path) {
1584                 char *journal_path;
1585                 struct inode *journal_inode;
1586                 struct path path;
1587                 int error;
1588
1589                 if (is_remount) {
1590                         ext4_msg(sb, KERN_ERR,
1591                                  "Cannot specify journal on remount");
1592                         return -1;
1593                 }
1594                 journal_path = match_strdup(&args[0]);
1595                 if (!journal_path) {
1596                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1597                                 "journal device string");
1598                         return -1;
1599                 }
1600
1601                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1602                 if (error) {
1603                         ext4_msg(sb, KERN_ERR, "error: could not find "
1604                                 "journal device path: error %d", error);
1605                         kfree(journal_path);
1606                         return -1;
1607                 }
1608
1609                 journal_inode = d_inode(path.dentry);
1610                 if (!S_ISBLK(journal_inode->i_mode)) {
1611                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1612                                 "is not a block device", journal_path);
1613                         path_put(&path);
1614                         kfree(journal_path);
1615                         return -1;
1616                 }
1617
1618                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1619                 path_put(&path);
1620                 kfree(journal_path);
1621         } else if (token == Opt_journal_ioprio) {
1622                 if (arg > 7) {
1623                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1624                                  " (must be 0-7)");
1625                         return -1;
1626                 }
1627                 *journal_ioprio =
1628                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1629         } else if (token == Opt_test_dummy_encryption) {
1630 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1631                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1632                 ext4_msg(sb, KERN_WARNING,
1633                          "Test dummy encryption mode enabled");
1634 #else
1635                 ext4_msg(sb, KERN_WARNING,
1636                          "Test dummy encryption mount option ignored");
1637 #endif
1638         } else if (m->flags & MOPT_DATAJ) {
1639                 if (is_remount) {
1640                         if (!sbi->s_journal)
1641                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1642                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1643                                 ext4_msg(sb, KERN_ERR,
1644                                          "Cannot change data mode on remount");
1645                                 return -1;
1646                         }
1647                 } else {
1648                         clear_opt(sb, DATA_FLAGS);
1649                         sbi->s_mount_opt |= m->mount_opt;
1650                 }
1651 #ifdef CONFIG_QUOTA
1652         } else if (m->flags & MOPT_QFMT) {
1653                 if (sb_any_quota_loaded(sb) &&
1654                     sbi->s_jquota_fmt != m->mount_opt) {
1655                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1656                                  "quota options when quota turned on");
1657                         return -1;
1658                 }
1659                 if (ext4_has_feature_quota(sb)) {
1660                         ext4_msg(sb, KERN_ERR,
1661                                  "Cannot set journaled quota options "
1662                                  "when QUOTA feature is enabled");
1663                         return -1;
1664                 }
1665                 sbi->s_jquota_fmt = m->mount_opt;
1666 #endif
1667 #ifndef CONFIG_FS_DAX
1668         } else if (token == Opt_dax) {
1669                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1670                 return -1;
1671 #endif
1672         } else {
1673                 if (!args->from)
1674                         arg = 1;
1675                 if (m->flags & MOPT_CLEAR)
1676                         arg = !arg;
1677                 else if (unlikely(!(m->flags & MOPT_SET))) {
1678                         ext4_msg(sb, KERN_WARNING,
1679                                  "buggy handling of option %s", opt);
1680                         WARN_ON(1);
1681                         return -1;
1682                 }
1683                 if (arg != 0)
1684                         sbi->s_mount_opt |= m->mount_opt;
1685                 else
1686                         sbi->s_mount_opt &= ~m->mount_opt;
1687         }
1688         return 1;
1689 }
1690
1691 static int parse_options(char *options, struct super_block *sb,
1692                          unsigned long *journal_devnum,
1693                          unsigned int *journal_ioprio,
1694                          int is_remount)
1695 {
1696         struct ext4_sb_info *sbi = EXT4_SB(sb);
1697         char *p;
1698         substring_t args[MAX_OPT_ARGS];
1699         int token;
1700
1701         if (!options)
1702                 return 1;
1703
1704         while ((p = strsep(&options, ",")) != NULL) {
1705                 if (!*p)
1706                         continue;
1707                 /*
1708                  * Initialize args struct so we know whether arg was
1709                  * found; some options take optional arguments.
1710                  */
1711                 args[0].to = args[0].from = NULL;
1712                 token = match_token(p, tokens, args);
1713                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1714                                      journal_ioprio, is_remount) < 0)
1715                         return 0;
1716         }
1717 #ifdef CONFIG_QUOTA
1718         if (ext4_has_feature_quota(sb) &&
1719             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1720                 ext4_msg(sb, KERN_ERR, "Cannot set quota options when QUOTA "
1721                          "feature is enabled");
1722                 return 0;
1723         }
1724         if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1725                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1726                         clear_opt(sb, USRQUOTA);
1727
1728                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1729                         clear_opt(sb, GRPQUOTA);
1730
1731                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1732                         ext4_msg(sb, KERN_ERR, "old and new quota "
1733                                         "format mixing");
1734                         return 0;
1735                 }
1736
1737                 if (!sbi->s_jquota_fmt) {
1738                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1739                                         "not specified");
1740                         return 0;
1741                 }
1742         }
1743 #endif
1744         if (test_opt(sb, DIOREAD_NOLOCK)) {
1745                 int blocksize =
1746                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1747
1748                 if (blocksize < PAGE_CACHE_SIZE) {
1749                         ext4_msg(sb, KERN_ERR, "can't mount with "
1750                                  "dioread_nolock if block size != PAGE_SIZE");
1751                         return 0;
1752                 }
1753         }
1754         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1755             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1756                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1757                          "in data=ordered mode");
1758                 return 0;
1759         }
1760         return 1;
1761 }
1762
1763 static inline void ext4_show_quota_options(struct seq_file *seq,
1764                                            struct super_block *sb)
1765 {
1766 #if defined(CONFIG_QUOTA)
1767         struct ext4_sb_info *sbi = EXT4_SB(sb);
1768
1769         if (sbi->s_jquota_fmt) {
1770                 char *fmtname = "";
1771
1772                 switch (sbi->s_jquota_fmt) {
1773                 case QFMT_VFS_OLD:
1774                         fmtname = "vfsold";
1775                         break;
1776                 case QFMT_VFS_V0:
1777                         fmtname = "vfsv0";
1778                         break;
1779                 case QFMT_VFS_V1:
1780                         fmtname = "vfsv1";
1781                         break;
1782                 }
1783                 seq_printf(seq, ",jqfmt=%s", fmtname);
1784         }
1785
1786         if (sbi->s_qf_names[USRQUOTA])
1787                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1788
1789         if (sbi->s_qf_names[GRPQUOTA])
1790                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1791 #endif
1792 }
1793
1794 static const char *token2str(int token)
1795 {
1796         const struct match_token *t;
1797
1798         for (t = tokens; t->token != Opt_err; t++)
1799                 if (t->token == token && !strchr(t->pattern, '='))
1800                         break;
1801         return t->pattern;
1802 }
1803
1804 /*
1805  * Show an option if
1806  *  - it's set to a non-default value OR
1807  *  - if the per-sb default is different from the global default
1808  */
1809 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1810                               int nodefs)
1811 {
1812         struct ext4_sb_info *sbi = EXT4_SB(sb);
1813         struct ext4_super_block *es = sbi->s_es;
1814         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1815         const struct mount_opts *m;
1816         char sep = nodefs ? '\n' : ',';
1817
1818 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1819 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1820
1821         if (sbi->s_sb_block != 1)
1822                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1823
1824         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1825                 int want_set = m->flags & MOPT_SET;
1826                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1827                     (m->flags & MOPT_CLEAR_ERR))
1828                         continue;
1829                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1830                         continue; /* skip if same as the default */
1831                 if ((want_set &&
1832                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1833                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1834                         continue; /* select Opt_noFoo vs Opt_Foo */
1835                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1836         }
1837
1838         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1839             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1840                 SEQ_OPTS_PRINT("resuid=%u",
1841                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1842         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1843             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1844                 SEQ_OPTS_PRINT("resgid=%u",
1845                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1846         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1847         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1848                 SEQ_OPTS_PUTS("errors=remount-ro");
1849         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1850                 SEQ_OPTS_PUTS("errors=continue");
1851         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1852                 SEQ_OPTS_PUTS("errors=panic");
1853         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1854                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1855         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1856                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1857         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1858                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1859         if (sb->s_flags & MS_I_VERSION)
1860                 SEQ_OPTS_PUTS("i_version");
1861         if (nodefs || sbi->s_stripe)
1862                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1863         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1864                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1865                         SEQ_OPTS_PUTS("data=journal");
1866                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1867                         SEQ_OPTS_PUTS("data=ordered");
1868                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1869                         SEQ_OPTS_PUTS("data=writeback");
1870         }
1871         if (nodefs ||
1872             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1873                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1874                                sbi->s_inode_readahead_blks);
1875
1876         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1877                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1878                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1879         if (nodefs || sbi->s_max_dir_size_kb)
1880                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1881
1882         ext4_show_quota_options(seq, sb);
1883         return 0;
1884 }
1885
1886 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1887 {
1888         return _ext4_show_options(seq, root->d_sb, 0);
1889 }
1890
1891 int ext4_seq_options_show(struct seq_file *seq, void *offset)
1892 {
1893         struct super_block *sb = seq->private;
1894         int rc;
1895
1896         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1897         rc = _ext4_show_options(seq, sb, 1);
1898         seq_puts(seq, "\n");
1899         return rc;
1900 }
1901
1902 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1903                             int read_only)
1904 {
1905         struct ext4_sb_info *sbi = EXT4_SB(sb);
1906         int res = 0;
1907
1908         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1909                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1910                          "forcing read-only mode");
1911                 res = MS_RDONLY;
1912         }
1913         if (read_only)
1914                 goto done;
1915         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1916                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1917                          "running e2fsck is recommended");
1918         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1919                 ext4_msg(sb, KERN_WARNING,
1920                          "warning: mounting fs with errors, "
1921                          "running e2fsck is recommended");
1922         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1923                  le16_to_cpu(es->s_mnt_count) >=
1924                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1925                 ext4_msg(sb, KERN_WARNING,
1926                          "warning: maximal mount count reached, "
1927                          "running e2fsck is recommended");
1928         else if (le32_to_cpu(es->s_checkinterval) &&
1929                 (le32_to_cpu(es->s_lastcheck) +
1930                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1931                 ext4_msg(sb, KERN_WARNING,
1932                          "warning: checktime reached, "
1933                          "running e2fsck is recommended");
1934         if (!sbi->s_journal)
1935                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1936         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1937                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1938         le16_add_cpu(&es->s_mnt_count, 1);
1939         es->s_mtime = cpu_to_le32(get_seconds());
1940         ext4_update_dynamic_rev(sb);
1941         if (sbi->s_journal)
1942                 ext4_set_feature_journal_needs_recovery(sb);
1943
1944         ext4_commit_super(sb, 1);
1945 done:
1946         if (test_opt(sb, DEBUG))
1947                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1948                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1949                         sb->s_blocksize,
1950                         sbi->s_groups_count,
1951                         EXT4_BLOCKS_PER_GROUP(sb),
1952                         EXT4_INODES_PER_GROUP(sb),
1953                         sbi->s_mount_opt, sbi->s_mount_opt2);
1954
1955         cleancache_init_fs(sb);
1956         return res;
1957 }
1958
1959 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1960 {
1961         struct ext4_sb_info *sbi = EXT4_SB(sb);
1962         struct flex_groups *new_groups;
1963         int size;
1964
1965         if (!sbi->s_log_groups_per_flex)
1966                 return 0;
1967
1968         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1969         if (size <= sbi->s_flex_groups_allocated)
1970                 return 0;
1971
1972         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1973         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1974         if (!new_groups) {
1975                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1976                          size / (int) sizeof(struct flex_groups));
1977                 return -ENOMEM;
1978         }
1979
1980         if (sbi->s_flex_groups) {
1981                 memcpy(new_groups, sbi->s_flex_groups,
1982                        (sbi->s_flex_groups_allocated *
1983                         sizeof(struct flex_groups)));
1984                 kvfree(sbi->s_flex_groups);
1985         }
1986         sbi->s_flex_groups = new_groups;
1987         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1988         return 0;
1989 }
1990
1991 static int ext4_fill_flex_info(struct super_block *sb)
1992 {
1993         struct ext4_sb_info *sbi = EXT4_SB(sb);
1994         struct ext4_group_desc *gdp = NULL;
1995         ext4_group_t flex_group;
1996         int i, err;
1997
1998         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
1999         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2000                 sbi->s_log_groups_per_flex = 0;
2001                 return 1;
2002         }
2003
2004         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2005         if (err)
2006                 goto failed;
2007
2008         for (i = 0; i < sbi->s_groups_count; i++) {
2009                 gdp = ext4_get_group_desc(sb, i, NULL);
2010
2011                 flex_group = ext4_flex_group(sbi, i);
2012                 atomic_add(ext4_free_inodes_count(sb, gdp),
2013                            &sbi->s_flex_groups[flex_group].free_inodes);
2014                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2015                              &sbi->s_flex_groups[flex_group].free_clusters);
2016                 atomic_add(ext4_used_dirs_count(sb, gdp),
2017                            &sbi->s_flex_groups[flex_group].used_dirs);
2018         }
2019
2020         return 1;
2021 failed:
2022         return 0;
2023 }
2024
2025 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2026                                    struct ext4_group_desc *gdp)
2027 {
2028         int offset;
2029         __u16 crc = 0;
2030         __le32 le_group = cpu_to_le32(block_group);
2031         struct ext4_sb_info *sbi = EXT4_SB(sb);
2032
2033         if (ext4_has_metadata_csum(sbi->s_sb)) {
2034                 /* Use new metadata_csum algorithm */
2035                 __le16 save_csum;
2036                 __u32 csum32;
2037
2038                 save_csum = gdp->bg_checksum;
2039                 gdp->bg_checksum = 0;
2040                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2041                                      sizeof(le_group));
2042                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp,
2043                                      sbi->s_desc_size);
2044                 gdp->bg_checksum = save_csum;
2045
2046                 crc = csum32 & 0xFFFF;
2047                 goto out;
2048         }
2049
2050         /* old crc16 code */
2051         if (!ext4_has_feature_gdt_csum(sb))
2052                 return 0;
2053
2054         offset = offsetof(struct ext4_group_desc, bg_checksum);
2055
2056         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2057         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2058         crc = crc16(crc, (__u8 *)gdp, offset);
2059         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2060         /* for checksum of struct ext4_group_desc do the rest...*/
2061         if (ext4_has_feature_64bit(sb) &&
2062             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2063                 crc = crc16(crc, (__u8 *)gdp + offset,
2064                             le16_to_cpu(sbi->s_es->s_desc_size) -
2065                                 offset);
2066
2067 out:
2068         return cpu_to_le16(crc);
2069 }
2070
2071 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2072                                 struct ext4_group_desc *gdp)
2073 {
2074         if (ext4_has_group_desc_csum(sb) &&
2075             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2076                 return 0;
2077
2078         return 1;
2079 }
2080
2081 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2082                               struct ext4_group_desc *gdp)
2083 {
2084         if (!ext4_has_group_desc_csum(sb))
2085                 return;
2086         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2087 }
2088
2089 /* Called at mount-time, super-block is locked */
2090 static int ext4_check_descriptors(struct super_block *sb,
2091                                   ext4_group_t *first_not_zeroed)
2092 {
2093         struct ext4_sb_info *sbi = EXT4_SB(sb);
2094         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2095         ext4_fsblk_t last_block;
2096         ext4_fsblk_t block_bitmap;
2097         ext4_fsblk_t inode_bitmap;
2098         ext4_fsblk_t inode_table;
2099         int flexbg_flag = 0;
2100         ext4_group_t i, grp = sbi->s_groups_count;
2101
2102         if (ext4_has_feature_flex_bg(sb))
2103                 flexbg_flag = 1;
2104
2105         ext4_debug("Checking group descriptors");
2106
2107         for (i = 0; i < sbi->s_groups_count; i++) {
2108                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2109
2110                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2111                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2112                 else
2113                         last_block = first_block +
2114                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2115
2116                 if ((grp == sbi->s_groups_count) &&
2117                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2118                         grp = i;
2119
2120                 block_bitmap = ext4_block_bitmap(sb, gdp);
2121                 if (block_bitmap < first_block || block_bitmap > last_block) {
2122                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2123                                "Block bitmap for group %u not in group "
2124                                "(block %llu)!", i, block_bitmap);
2125                         return 0;
2126                 }
2127                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2128                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2129                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2130                                "Inode bitmap for group %u not in group "
2131                                "(block %llu)!", i, inode_bitmap);
2132                         return 0;
2133                 }
2134                 inode_table = ext4_inode_table(sb, gdp);
2135                 if (inode_table < first_block ||
2136                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2137                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2138                                "Inode table for group %u not in group "
2139                                "(block %llu)!", i, inode_table);
2140                         return 0;
2141                 }
2142                 ext4_lock_group(sb, i);
2143                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2144                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2145                                  "Checksum for group %u failed (%u!=%u)",
2146                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2147                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2148                         if (!(sb->s_flags & MS_RDONLY)) {
2149                                 ext4_unlock_group(sb, i);
2150                                 return 0;
2151                         }
2152                 }
2153                 ext4_unlock_group(sb, i);
2154                 if (!flexbg_flag)
2155                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2156         }
2157         if (NULL != first_not_zeroed)
2158                 *first_not_zeroed = grp;
2159         return 1;
2160 }
2161
2162 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2163  * the superblock) which were deleted from all directories, but held open by
2164  * a process at the time of a crash.  We walk the list and try to delete these
2165  * inodes at recovery time (only with a read-write filesystem).
2166  *
2167  * In order to keep the orphan inode chain consistent during traversal (in
2168  * case of crash during recovery), we link each inode into the superblock
2169  * orphan list_head and handle it the same way as an inode deletion during
2170  * normal operation (which journals the operations for us).
2171  *
2172  * We only do an iget() and an iput() on each inode, which is very safe if we
2173  * accidentally point at an in-use or already deleted inode.  The worst that
2174  * can happen in this case is that we get a "bit already cleared" message from
2175  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2176  * e2fsck was run on this filesystem, and it must have already done the orphan
2177  * inode cleanup for us, so we can safely abort without any further action.
2178  */
2179 static void ext4_orphan_cleanup(struct super_block *sb,
2180                                 struct ext4_super_block *es)
2181 {
2182         unsigned int s_flags = sb->s_flags;
2183         int nr_orphans = 0, nr_truncates = 0;
2184 #ifdef CONFIG_QUOTA
2185         int i;
2186 #endif
2187         if (!es->s_last_orphan) {
2188                 jbd_debug(4, "no orphan inodes to clean up\n");
2189                 return;
2190         }
2191
2192         if (bdev_read_only(sb->s_bdev)) {
2193                 ext4_msg(sb, KERN_ERR, "write access "
2194                         "unavailable, skipping orphan cleanup");
2195                 return;
2196         }
2197
2198         /* Check if feature set would not allow a r/w mount */
2199         if (!ext4_feature_set_ok(sb, 0)) {
2200                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2201                          "unknown ROCOMPAT features");
2202                 return;
2203         }
2204
2205         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2206                 /* don't clear list on RO mount w/ errors */
2207                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2208                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2209                                   "clearing orphan list.\n");
2210                         es->s_last_orphan = 0;
2211                 }
2212                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2213                 return;
2214         }
2215
2216         if (s_flags & MS_RDONLY) {
2217                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2218                 sb->s_flags &= ~MS_RDONLY;
2219         }
2220 #ifdef CONFIG_QUOTA
2221         /* Needed for iput() to work correctly and not trash data */
2222         sb->s_flags |= MS_ACTIVE;
2223         /* Turn on quotas so that they are updated correctly */
2224         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2225                 if (EXT4_SB(sb)->s_qf_names[i]) {
2226                         int ret = ext4_quota_on_mount(sb, i);
2227                         if (ret < 0)
2228                                 ext4_msg(sb, KERN_ERR,
2229                                         "Cannot turn on journaled "
2230                                         "quota: error %d", ret);
2231                 }
2232         }
2233 #endif
2234
2235         while (es->s_last_orphan) {
2236                 struct inode *inode;
2237
2238                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2239                 if (IS_ERR(inode)) {
2240                         es->s_last_orphan = 0;
2241                         break;
2242                 }
2243
2244                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2245                 dquot_initialize(inode);
2246                 if (inode->i_nlink) {
2247                         if (test_opt(sb, DEBUG))
2248                                 ext4_msg(sb, KERN_DEBUG,
2249                                         "%s: truncating inode %lu to %lld bytes",
2250                                         __func__, inode->i_ino, inode->i_size);
2251                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2252                                   inode->i_ino, inode->i_size);
2253                         mutex_lock(&inode->i_mutex);
2254                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2255                         ext4_truncate(inode);
2256                         mutex_unlock(&inode->i_mutex);
2257                         nr_truncates++;
2258                 } else {
2259                         if (test_opt(sb, DEBUG))
2260                                 ext4_msg(sb, KERN_DEBUG,
2261                                         "%s: deleting unreferenced inode %lu",
2262                                         __func__, inode->i_ino);
2263                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2264                                   inode->i_ino);
2265                         nr_orphans++;
2266                 }
2267                 iput(inode);  /* The delete magic happens here! */
2268         }
2269
2270 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2271
2272         if (nr_orphans)
2273                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2274                        PLURAL(nr_orphans));
2275         if (nr_truncates)
2276                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2277                        PLURAL(nr_truncates));
2278 #ifdef CONFIG_QUOTA
2279         /* Turn quotas off */
2280         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2281                 if (sb_dqopt(sb)->files[i])
2282                         dquot_quota_off(sb, i);
2283         }
2284 #endif
2285         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2286 }
2287
2288 /*
2289  * Maximal extent format file size.
2290  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2291  * extent format containers, within a sector_t, and within i_blocks
2292  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2293  * so that won't be a limiting factor.
2294  *
2295  * However there is other limiting factor. We do store extents in the form
2296  * of starting block and length, hence the resulting length of the extent
2297  * covering maximum file size must fit into on-disk format containers as
2298  * well. Given that length is always by 1 unit bigger than max unit (because
2299  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2300  *
2301  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2302  */
2303 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2304 {
2305         loff_t res;
2306         loff_t upper_limit = MAX_LFS_FILESIZE;
2307
2308         /* small i_blocks in vfs inode? */
2309         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2310                 /*
2311                  * CONFIG_LBDAF is not enabled implies the inode
2312                  * i_block represent total blocks in 512 bytes
2313                  * 32 == size of vfs inode i_blocks * 8
2314                  */
2315                 upper_limit = (1LL << 32) - 1;
2316
2317                 /* total blocks in file system block size */
2318                 upper_limit >>= (blkbits - 9);
2319                 upper_limit <<= blkbits;
2320         }
2321
2322         /*
2323          * 32-bit extent-start container, ee_block. We lower the maxbytes
2324          * by one fs block, so ee_len can cover the extent of maximum file
2325          * size
2326          */
2327         res = (1LL << 32) - 1;
2328         res <<= blkbits;
2329
2330         /* Sanity check against vm- & vfs- imposed limits */
2331         if (res > upper_limit)
2332                 res = upper_limit;
2333
2334         return res;
2335 }
2336
2337 /*
2338  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2339  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2340  * We need to be 1 filesystem block less than the 2^48 sector limit.
2341  */
2342 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2343 {
2344         loff_t res = EXT4_NDIR_BLOCKS;
2345         int meta_blocks;
2346         loff_t upper_limit;
2347         /* This is calculated to be the largest file size for a dense, block
2348          * mapped file such that the file's total number of 512-byte sectors,
2349          * including data and all indirect blocks, does not exceed (2^48 - 1).
2350          *
2351          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2352          * number of 512-byte sectors of the file.
2353          */
2354
2355         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2356                 /*
2357                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2358                  * the inode i_block field represents total file blocks in
2359                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2360                  */
2361                 upper_limit = (1LL << 32) - 1;
2362
2363                 /* total blocks in file system block size */
2364                 upper_limit >>= (bits - 9);
2365
2366         } else {
2367                 /*
2368                  * We use 48 bit ext4_inode i_blocks
2369                  * With EXT4_HUGE_FILE_FL set the i_blocks
2370                  * represent total number of blocks in
2371                  * file system block size
2372                  */
2373                 upper_limit = (1LL << 48) - 1;
2374
2375         }
2376
2377         /* indirect blocks */
2378         meta_blocks = 1;
2379         /* double indirect blocks */
2380         meta_blocks += 1 + (1LL << (bits-2));
2381         /* tripple indirect blocks */
2382         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2383
2384         upper_limit -= meta_blocks;
2385         upper_limit <<= bits;
2386
2387         res += 1LL << (bits-2);
2388         res += 1LL << (2*(bits-2));
2389         res += 1LL << (3*(bits-2));
2390         res <<= bits;
2391         if (res > upper_limit)
2392                 res = upper_limit;
2393
2394         if (res > MAX_LFS_FILESIZE)
2395                 res = MAX_LFS_FILESIZE;
2396
2397         return res;
2398 }
2399
2400 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2401                                    ext4_fsblk_t logical_sb_block, int nr)
2402 {
2403         struct ext4_sb_info *sbi = EXT4_SB(sb);
2404         ext4_group_t bg, first_meta_bg;
2405         int has_super = 0;
2406
2407         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2408
2409         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2410                 return logical_sb_block + nr + 1;
2411         bg = sbi->s_desc_per_block * nr;
2412         if (ext4_bg_has_super(sb, bg))
2413                 has_super = 1;
2414
2415         /*
2416          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2417          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2418          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2419          * compensate.
2420          */
2421         if (sb->s_blocksize == 1024 && nr == 0 &&
2422             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2423                 has_super++;
2424
2425         return (has_super + ext4_group_first_block_no(sb, bg));
2426 }
2427
2428 /**
2429  * ext4_get_stripe_size: Get the stripe size.
2430  * @sbi: In memory super block info
2431  *
2432  * If we have specified it via mount option, then
2433  * use the mount option value. If the value specified at mount time is
2434  * greater than the blocks per group use the super block value.
2435  * If the super block value is greater than blocks per group return 0.
2436  * Allocator needs it be less than blocks per group.
2437  *
2438  */
2439 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2440 {
2441         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2442         unsigned long stripe_width =
2443                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2444         int ret;
2445
2446         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2447                 ret = sbi->s_stripe;
2448         else if (stripe_width <= sbi->s_blocks_per_group)
2449                 ret = stripe_width;
2450         else if (stride <= sbi->s_blocks_per_group)
2451                 ret = stride;
2452         else
2453                 ret = 0;
2454
2455         /*
2456          * If the stripe width is 1, this makes no sense and
2457          * we set it to 0 to turn off stripe handling code.
2458          */
2459         if (ret <= 1)
2460                 ret = 0;
2461
2462         return ret;
2463 }
2464
2465 /*
2466  * Check whether this filesystem can be mounted based on
2467  * the features present and the RDONLY/RDWR mount requested.
2468  * Returns 1 if this filesystem can be mounted as requested,
2469  * 0 if it cannot be.
2470  */
2471 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2472 {
2473         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2474                 ext4_msg(sb, KERN_ERR,
2475                         "Couldn't mount because of "
2476                         "unsupported optional features (%x)",
2477                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2478                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2479                 return 0;
2480         }
2481
2482         if (readonly)
2483                 return 1;
2484
2485         if (ext4_has_feature_readonly(sb)) {
2486                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2487                 sb->s_flags |= MS_RDONLY;
2488                 return 1;
2489         }
2490
2491         /* Check that feature set is OK for a read-write mount */
2492         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2493                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2494                          "unsupported optional features (%x)",
2495                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2496                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2497                 return 0;
2498         }
2499         /*
2500          * Large file size enabled file system can only be mounted
2501          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2502          */
2503         if (ext4_has_feature_huge_file(sb)) {
2504                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2505                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2506                                  "cannot be mounted RDWR without "
2507                                  "CONFIG_LBDAF");
2508                         return 0;
2509                 }
2510         }
2511         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2512                 ext4_msg(sb, KERN_ERR,
2513                          "Can't support bigalloc feature without "
2514                          "extents feature\n");
2515                 return 0;
2516         }
2517
2518 #ifndef CONFIG_QUOTA
2519         if (ext4_has_feature_quota(sb) && !readonly) {
2520                 ext4_msg(sb, KERN_ERR,
2521                          "Filesystem with quota feature cannot be mounted RDWR "
2522                          "without CONFIG_QUOTA");
2523                 return 0;
2524         }
2525 #endif  /* CONFIG_QUOTA */
2526         return 1;
2527 }
2528
2529 /*
2530  * This function is called once a day if we have errors logged
2531  * on the file system
2532  */
2533 static void print_daily_error_info(unsigned long arg)
2534 {
2535         struct super_block *sb = (struct super_block *) arg;
2536         struct ext4_sb_info *sbi;
2537         struct ext4_super_block *es;
2538
2539         sbi = EXT4_SB(sb);
2540         es = sbi->s_es;
2541
2542         if (es->s_error_count)
2543                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2544                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2545                          le32_to_cpu(es->s_error_count));
2546         if (es->s_first_error_time) {
2547                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2548                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2549                        (int) sizeof(es->s_first_error_func),
2550                        es->s_first_error_func,
2551                        le32_to_cpu(es->s_first_error_line));
2552                 if (es->s_first_error_ino)
2553                         printk(": inode %u",
2554                                le32_to_cpu(es->s_first_error_ino));
2555                 if (es->s_first_error_block)
2556                         printk(": block %llu", (unsigned long long)
2557                                le64_to_cpu(es->s_first_error_block));
2558                 printk("\n");
2559         }
2560         if (es->s_last_error_time) {
2561                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2562                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2563                        (int) sizeof(es->s_last_error_func),
2564                        es->s_last_error_func,
2565                        le32_to_cpu(es->s_last_error_line));
2566                 if (es->s_last_error_ino)
2567                         printk(": inode %u",
2568                                le32_to_cpu(es->s_last_error_ino));
2569                 if (es->s_last_error_block)
2570                         printk(": block %llu", (unsigned long long)
2571                                le64_to_cpu(es->s_last_error_block));
2572                 printk("\n");
2573         }
2574         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2575 }
2576
2577 /* Find next suitable group and run ext4_init_inode_table */
2578 static int ext4_run_li_request(struct ext4_li_request *elr)
2579 {
2580         struct ext4_group_desc *gdp = NULL;
2581         ext4_group_t group, ngroups;
2582         struct super_block *sb;
2583         unsigned long timeout = 0;
2584         int ret = 0;
2585
2586         sb = elr->lr_super;
2587         ngroups = EXT4_SB(sb)->s_groups_count;
2588
2589         sb_start_write(sb);
2590         for (group = elr->lr_next_group; group < ngroups; group++) {
2591                 gdp = ext4_get_group_desc(sb, group, NULL);
2592                 if (!gdp) {
2593                         ret = 1;
2594                         break;
2595                 }
2596
2597                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2598                         break;
2599         }
2600
2601         if (group >= ngroups)
2602                 ret = 1;
2603
2604         if (!ret) {
2605                 timeout = jiffies;
2606                 ret = ext4_init_inode_table(sb, group,
2607                                             elr->lr_timeout ? 0 : 1);
2608                 if (elr->lr_timeout == 0) {
2609                         timeout = (jiffies - timeout) *
2610                                   elr->lr_sbi->s_li_wait_mult;
2611                         elr->lr_timeout = timeout;
2612                 }
2613                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2614                 elr->lr_next_group = group + 1;
2615         }
2616         sb_end_write(sb);
2617
2618         return ret;
2619 }
2620
2621 /*
2622  * Remove lr_request from the list_request and free the
2623  * request structure. Should be called with li_list_mtx held
2624  */
2625 static void ext4_remove_li_request(struct ext4_li_request *elr)
2626 {
2627         struct ext4_sb_info *sbi;
2628
2629         if (!elr)
2630                 return;
2631
2632         sbi = elr->lr_sbi;
2633
2634         list_del(&elr->lr_request);
2635         sbi->s_li_request = NULL;
2636         kfree(elr);
2637 }
2638
2639 static void ext4_unregister_li_request(struct super_block *sb)
2640 {
2641         mutex_lock(&ext4_li_mtx);
2642         if (!ext4_li_info) {
2643                 mutex_unlock(&ext4_li_mtx);
2644                 return;
2645         }
2646
2647         mutex_lock(&ext4_li_info->li_list_mtx);
2648         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2649         mutex_unlock(&ext4_li_info->li_list_mtx);
2650         mutex_unlock(&ext4_li_mtx);
2651 }
2652
2653 static struct task_struct *ext4_lazyinit_task;
2654
2655 /*
2656  * This is the function where ext4lazyinit thread lives. It walks
2657  * through the request list searching for next scheduled filesystem.
2658  * When such a fs is found, run the lazy initialization request
2659  * (ext4_rn_li_request) and keep track of the time spend in this
2660  * function. Based on that time we compute next schedule time of
2661  * the request. When walking through the list is complete, compute
2662  * next waking time and put itself into sleep.
2663  */
2664 static int ext4_lazyinit_thread(void *arg)
2665 {
2666         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2667         struct list_head *pos, *n;
2668         struct ext4_li_request *elr;
2669         unsigned long next_wakeup, cur;
2670
2671         BUG_ON(NULL == eli);
2672
2673 cont_thread:
2674         while (true) {
2675                 next_wakeup = MAX_JIFFY_OFFSET;
2676
2677                 mutex_lock(&eli->li_list_mtx);
2678                 if (list_empty(&eli->li_request_list)) {
2679                         mutex_unlock(&eli->li_list_mtx);
2680                         goto exit_thread;
2681                 }
2682
2683                 list_for_each_safe(pos, n, &eli->li_request_list) {
2684                         elr = list_entry(pos, struct ext4_li_request,
2685                                          lr_request);
2686
2687                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2688                                 if (ext4_run_li_request(elr) != 0) {
2689                                         /* error, remove the lazy_init job */
2690                                         ext4_remove_li_request(elr);
2691                                         continue;
2692                                 }
2693                         }
2694
2695                         if (time_before(elr->lr_next_sched, next_wakeup))
2696                                 next_wakeup = elr->lr_next_sched;
2697                 }
2698                 mutex_unlock(&eli->li_list_mtx);
2699
2700                 try_to_freeze();
2701
2702                 cur = jiffies;
2703                 if ((time_after_eq(cur, next_wakeup)) ||
2704                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2705                         cond_resched();
2706                         continue;
2707                 }
2708
2709                 schedule_timeout_interruptible(next_wakeup - cur);
2710
2711                 if (kthread_should_stop()) {
2712                         ext4_clear_request_list();
2713                         goto exit_thread;
2714                 }
2715         }
2716
2717 exit_thread:
2718         /*
2719          * It looks like the request list is empty, but we need
2720          * to check it under the li_list_mtx lock, to prevent any
2721          * additions into it, and of course we should lock ext4_li_mtx
2722          * to atomically free the list and ext4_li_info, because at
2723          * this point another ext4 filesystem could be registering
2724          * new one.
2725          */
2726         mutex_lock(&ext4_li_mtx);
2727         mutex_lock(&eli->li_list_mtx);
2728         if (!list_empty(&eli->li_request_list)) {
2729                 mutex_unlock(&eli->li_list_mtx);
2730                 mutex_unlock(&ext4_li_mtx);
2731                 goto cont_thread;
2732         }
2733         mutex_unlock(&eli->li_list_mtx);
2734         kfree(ext4_li_info);
2735         ext4_li_info = NULL;
2736         mutex_unlock(&ext4_li_mtx);
2737
2738         return 0;
2739 }
2740
2741 static void ext4_clear_request_list(void)
2742 {
2743         struct list_head *pos, *n;
2744         struct ext4_li_request *elr;
2745
2746         mutex_lock(&ext4_li_info->li_list_mtx);
2747         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2748                 elr = list_entry(pos, struct ext4_li_request,
2749                                  lr_request);
2750                 ext4_remove_li_request(elr);
2751         }
2752         mutex_unlock(&ext4_li_info->li_list_mtx);
2753 }
2754
2755 static int ext4_run_lazyinit_thread(void)
2756 {
2757         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2758                                          ext4_li_info, "ext4lazyinit");
2759         if (IS_ERR(ext4_lazyinit_task)) {
2760                 int err = PTR_ERR(ext4_lazyinit_task);
2761                 ext4_clear_request_list();
2762                 kfree(ext4_li_info);
2763                 ext4_li_info = NULL;
2764                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2765                                  "initialization thread\n",
2766                                  err);
2767                 return err;
2768         }
2769         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2770         return 0;
2771 }
2772
2773 /*
2774  * Check whether it make sense to run itable init. thread or not.
2775  * If there is at least one uninitialized inode table, return
2776  * corresponding group number, else the loop goes through all
2777  * groups and return total number of groups.
2778  */
2779 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2780 {
2781         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2782         struct ext4_group_desc *gdp = NULL;
2783
2784         for (group = 0; group < ngroups; group++) {
2785                 gdp = ext4_get_group_desc(sb, group, NULL);
2786                 if (!gdp)
2787                         continue;
2788
2789                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2790                         break;
2791         }
2792
2793         return group;
2794 }
2795
2796 static int ext4_li_info_new(void)
2797 {
2798         struct ext4_lazy_init *eli = NULL;
2799
2800         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2801         if (!eli)
2802                 return -ENOMEM;
2803
2804         INIT_LIST_HEAD(&eli->li_request_list);
2805         mutex_init(&eli->li_list_mtx);
2806
2807         eli->li_state |= EXT4_LAZYINIT_QUIT;
2808
2809         ext4_li_info = eli;
2810
2811         return 0;
2812 }
2813
2814 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2815                                             ext4_group_t start)
2816 {
2817         struct ext4_sb_info *sbi = EXT4_SB(sb);
2818         struct ext4_li_request *elr;
2819
2820         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2821         if (!elr)
2822                 return NULL;
2823
2824         elr->lr_super = sb;
2825         elr->lr_sbi = sbi;
2826         elr->lr_next_group = start;
2827
2828         /*
2829          * Randomize first schedule time of the request to
2830          * spread the inode table initialization requests
2831          * better.
2832          */
2833         elr->lr_next_sched = jiffies + (prandom_u32() %
2834                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2835         return elr;
2836 }
2837
2838 int ext4_register_li_request(struct super_block *sb,
2839                              ext4_group_t first_not_zeroed)
2840 {
2841         struct ext4_sb_info *sbi = EXT4_SB(sb);
2842         struct ext4_li_request *elr = NULL;
2843         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2844         int ret = 0;
2845
2846         mutex_lock(&ext4_li_mtx);
2847         if (sbi->s_li_request != NULL) {
2848                 /*
2849                  * Reset timeout so it can be computed again, because
2850                  * s_li_wait_mult might have changed.
2851                  */
2852                 sbi->s_li_request->lr_timeout = 0;
2853                 goto out;
2854         }
2855
2856         if (first_not_zeroed == ngroups ||
2857             (sb->s_flags & MS_RDONLY) ||
2858             !test_opt(sb, INIT_INODE_TABLE))
2859                 goto out;
2860
2861         elr = ext4_li_request_new(sb, first_not_zeroed);
2862         if (!elr) {
2863                 ret = -ENOMEM;
2864                 goto out;
2865         }
2866
2867         if (NULL == ext4_li_info) {
2868                 ret = ext4_li_info_new();
2869                 if (ret)
2870                         goto out;
2871         }
2872
2873         mutex_lock(&ext4_li_info->li_list_mtx);
2874         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2875         mutex_unlock(&ext4_li_info->li_list_mtx);
2876
2877         sbi->s_li_request = elr;
2878         /*
2879          * set elr to NULL here since it has been inserted to
2880          * the request_list and the removal and free of it is
2881          * handled by ext4_clear_request_list from now on.
2882          */
2883         elr = NULL;
2884
2885         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2886                 ret = ext4_run_lazyinit_thread();
2887                 if (ret)
2888                         goto out;
2889         }
2890 out:
2891         mutex_unlock(&ext4_li_mtx);
2892         if (ret)
2893                 kfree(elr);
2894         return ret;
2895 }
2896
2897 /*
2898  * We do not need to lock anything since this is called on
2899  * module unload.
2900  */
2901 static void ext4_destroy_lazyinit_thread(void)
2902 {
2903         /*
2904          * If thread exited earlier
2905          * there's nothing to be done.
2906          */
2907         if (!ext4_li_info || !ext4_lazyinit_task)
2908                 return;
2909
2910         kthread_stop(ext4_lazyinit_task);
2911 }
2912
2913 static int set_journal_csum_feature_set(struct super_block *sb)
2914 {
2915         int ret = 1;
2916         int compat, incompat;
2917         struct ext4_sb_info *sbi = EXT4_SB(sb);
2918
2919         if (ext4_has_metadata_csum(sb)) {
2920                 /* journal checksum v3 */
2921                 compat = 0;
2922                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
2923         } else {
2924                 /* journal checksum v1 */
2925                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
2926                 incompat = 0;
2927         }
2928
2929         jbd2_journal_clear_features(sbi->s_journal,
2930                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2931                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
2932                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
2933         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2934                 ret = jbd2_journal_set_features(sbi->s_journal,
2935                                 compat, 0,
2936                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
2937                                 incompat);
2938         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2939                 ret = jbd2_journal_set_features(sbi->s_journal,
2940                                 compat, 0,
2941                                 incompat);
2942                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2943                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2944         } else {
2945                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2946                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2947         }
2948
2949         return ret;
2950 }
2951
2952 /*
2953  * Note: calculating the overhead so we can be compatible with
2954  * historical BSD practice is quite difficult in the face of
2955  * clusters/bigalloc.  This is because multiple metadata blocks from
2956  * different block group can end up in the same allocation cluster.
2957  * Calculating the exact overhead in the face of clustered allocation
2958  * requires either O(all block bitmaps) in memory or O(number of block
2959  * groups**2) in time.  We will still calculate the superblock for
2960  * older file systems --- and if we come across with a bigalloc file
2961  * system with zero in s_overhead_clusters the estimate will be close to
2962  * correct especially for very large cluster sizes --- but for newer
2963  * file systems, it's better to calculate this figure once at mkfs
2964  * time, and store it in the superblock.  If the superblock value is
2965  * present (even for non-bigalloc file systems), we will use it.
2966  */
2967 static int count_overhead(struct super_block *sb, ext4_group_t grp,
2968                           char *buf)
2969 {
2970         struct ext4_sb_info     *sbi = EXT4_SB(sb);
2971         struct ext4_group_desc  *gdp;
2972         ext4_fsblk_t            first_block, last_block, b;
2973         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
2974         int                     s, j, count = 0;
2975
2976         if (!ext4_has_feature_bigalloc(sb))
2977                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
2978                         sbi->s_itb_per_group + 2);
2979
2980         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
2981                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
2982         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
2983         for (i = 0; i < ngroups; i++) {
2984                 gdp = ext4_get_group_desc(sb, i, NULL);
2985                 b = ext4_block_bitmap(sb, gdp);
2986                 if (b >= first_block && b <= last_block) {
2987                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
2988                         count++;
2989                 }
2990                 b = ext4_inode_bitmap(sb, gdp);
2991                 if (b >= first_block && b <= last_block) {
2992                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
2993                         count++;
2994                 }
2995                 b = ext4_inode_table(sb, gdp);
2996                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
2997                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
2998                                 int c = EXT4_B2C(sbi, b - first_block);
2999                                 ext4_set_bit(c, buf);
3000                                 count++;
3001                         }
3002                 if (i != grp)
3003                         continue;
3004                 s = 0;
3005                 if (ext4_bg_has_super(sb, grp)) {
3006                         ext4_set_bit(s++, buf);
3007                         count++;
3008                 }
3009                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3010                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3011                         count++;
3012                 }
3013         }
3014         if (!count)
3015                 return 0;
3016         return EXT4_CLUSTERS_PER_GROUP(sb) -
3017                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3018 }
3019
3020 /*
3021  * Compute the overhead and stash it in sbi->s_overhead
3022  */
3023 int ext4_calculate_overhead(struct super_block *sb)
3024 {
3025         struct ext4_sb_info *sbi = EXT4_SB(sb);
3026         struct ext4_super_block *es = sbi->s_es;
3027         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3028         ext4_fsblk_t overhead = 0;
3029         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3030
3031         if (!buf)
3032                 return -ENOMEM;
3033
3034         /*
3035          * Compute the overhead (FS structures).  This is constant
3036          * for a given filesystem unless the number of block groups
3037          * changes so we cache the previous value until it does.
3038          */
3039
3040         /*
3041          * All of the blocks before first_data_block are overhead
3042          */
3043         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3044
3045         /*
3046          * Add the overhead found in each block group
3047          */
3048         for (i = 0; i < ngroups; i++) {
3049                 int blks;
3050
3051                 blks = count_overhead(sb, i, buf);
3052                 overhead += blks;
3053                 if (blks)
3054                         memset(buf, 0, PAGE_SIZE);
3055                 cond_resched();
3056         }
3057         /* Add the internal journal blocks as well */
3058         if (sbi->s_journal && !sbi->journal_bdev)
3059                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3060
3061         sbi->s_overhead = overhead;
3062         smp_wmb();
3063         free_page((unsigned long) buf);
3064         return 0;
3065 }
3066
3067 static void ext4_set_resv_clusters(struct super_block *sb)
3068 {
3069         ext4_fsblk_t resv_clusters;
3070         struct ext4_sb_info *sbi = EXT4_SB(sb);
3071
3072         /*
3073          * There's no need to reserve anything when we aren't using extents.
3074          * The space estimates are exact, there are no unwritten extents,
3075          * hole punching doesn't need new metadata... This is needed especially
3076          * to keep ext2/3 backward compatibility.
3077          */
3078         if (!ext4_has_feature_extents(sb))
3079                 return;
3080         /*
3081          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3082          * This should cover the situations where we can not afford to run
3083          * out of space like for example punch hole, or converting
3084          * unwritten extents in delalloc path. In most cases such
3085          * allocation would require 1, or 2 blocks, higher numbers are
3086          * very rare.
3087          */
3088         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3089                          sbi->s_cluster_bits);
3090
3091         do_div(resv_clusters, 50);
3092         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3093
3094         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3095 }
3096
3097 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3098 {
3099         char *orig_data = kstrdup(data, GFP_KERNEL);
3100         struct buffer_head *bh;
3101         struct ext4_super_block *es = NULL;
3102         struct ext4_sb_info *sbi;
3103         ext4_fsblk_t block;
3104         ext4_fsblk_t sb_block = get_sb_block(&data);
3105         ext4_fsblk_t logical_sb_block;
3106         unsigned long offset = 0;
3107         unsigned long journal_devnum = 0;
3108         unsigned long def_mount_opts;
3109         struct inode *root;
3110         const char *descr;
3111         int ret = -ENOMEM;
3112         int blocksize, clustersize;
3113         unsigned int db_count;
3114         unsigned int i;
3115         int needs_recovery, has_huge_files, has_bigalloc;
3116         __u64 blocks_count;
3117         int err = 0;
3118         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3119         ext4_group_t first_not_zeroed;
3120
3121         sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3122         if (!sbi)
3123                 goto out_free_orig;
3124
3125         sbi->s_blockgroup_lock =
3126                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3127         if (!sbi->s_blockgroup_lock) {
3128                 kfree(sbi);
3129                 goto out_free_orig;
3130         }
3131         sb->s_fs_info = sbi;
3132         sbi->s_sb = sb;
3133         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3134         sbi->s_sb_block = sb_block;
3135         if (sb->s_bdev->bd_part)
3136                 sbi->s_sectors_written_start =
3137                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3138
3139         /* Cleanup superblock name */
3140         strreplace(sb->s_id, '/', '!');
3141
3142         /* -EINVAL is default */
3143         ret = -EINVAL;
3144         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3145         if (!blocksize) {
3146                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3147                 goto out_fail;
3148         }
3149
3150         /*
3151          * The ext4 superblock will not be buffer aligned for other than 1kB
3152          * block sizes.  We need to calculate the offset from buffer start.
3153          */
3154         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3155                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3156                 offset = do_div(logical_sb_block, blocksize);
3157         } else {
3158                 logical_sb_block = sb_block;
3159         }
3160
3161         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3162                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3163                 goto out_fail;
3164         }
3165         /*
3166          * Note: s_es must be initialized as soon as possible because
3167          *       some ext4 macro-instructions depend on its value
3168          */
3169         es = (struct ext4_super_block *) (bh->b_data + offset);
3170         sbi->s_es = es;
3171         sb->s_magic = le16_to_cpu(es->s_magic);
3172         if (sb->s_magic != EXT4_SUPER_MAGIC)
3173                 goto cantfind_ext4;
3174         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3175
3176         /* Warn if metadata_csum and gdt_csum are both set. */
3177         if (ext4_has_feature_metadata_csum(sb) &&
3178             ext4_has_feature_gdt_csum(sb))
3179                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3180                              "redundant flags; please run fsck.");
3181
3182         /* Check for a known checksum algorithm */
3183         if (!ext4_verify_csum_type(sb, es)) {
3184                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3185                          "unknown checksum algorithm.");
3186                 silent = 1;
3187                 goto cantfind_ext4;
3188         }
3189
3190         /* Load the checksum driver */
3191         if (ext4_has_feature_metadata_csum(sb)) {
3192                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3193                 if (IS_ERR(sbi->s_chksum_driver)) {
3194                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3195                         ret = PTR_ERR(sbi->s_chksum_driver);
3196                         sbi->s_chksum_driver = NULL;
3197                         goto failed_mount;
3198                 }
3199         }
3200
3201         /* Check superblock checksum */
3202         if (!ext4_superblock_csum_verify(sb, es)) {
3203                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3204                          "invalid superblock checksum.  Run e2fsck?");
3205                 silent = 1;
3206                 ret = -EFSBADCRC;
3207                 goto cantfind_ext4;
3208         }
3209
3210         /* Precompute checksum seed for all metadata */
3211         if (ext4_has_feature_csum_seed(sb))
3212                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3213         else if (ext4_has_metadata_csum(sb))
3214                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3215                                                sizeof(es->s_uuid));
3216
3217         /* Set defaults before we parse the mount options */
3218         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3219         set_opt(sb, INIT_INODE_TABLE);
3220         if (def_mount_opts & EXT4_DEFM_DEBUG)
3221                 set_opt(sb, DEBUG);
3222         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3223                 set_opt(sb, GRPID);
3224         if (def_mount_opts & EXT4_DEFM_UID16)
3225                 set_opt(sb, NO_UID32);
3226         /* xattr user namespace & acls are now defaulted on */
3227         set_opt(sb, XATTR_USER);
3228 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3229         set_opt(sb, POSIX_ACL);
3230 #endif
3231         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3232         if (ext4_has_metadata_csum(sb))
3233                 set_opt(sb, JOURNAL_CHECKSUM);
3234
3235         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3236                 set_opt(sb, JOURNAL_DATA);
3237         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3238                 set_opt(sb, ORDERED_DATA);
3239         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3240                 set_opt(sb, WRITEBACK_DATA);
3241
3242         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3243                 set_opt(sb, ERRORS_PANIC);
3244         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3245                 set_opt(sb, ERRORS_CONT);
3246         else
3247                 set_opt(sb, ERRORS_RO);
3248         /* block_validity enabled by default; disable with noblock_validity */
3249         set_opt(sb, BLOCK_VALIDITY);
3250         if (def_mount_opts & EXT4_DEFM_DISCARD)
3251                 set_opt(sb, DISCARD);
3252
3253         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3254         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3255         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3256         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3257         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3258
3259         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3260                 set_opt(sb, BARRIER);
3261
3262         /*
3263          * enable delayed allocation by default
3264          * Use -o nodelalloc to turn it off
3265          */
3266         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3267             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3268                 set_opt(sb, DELALLOC);
3269
3270         /*
3271          * set default s_li_wait_mult for lazyinit, for the case there is
3272          * no mount option specified.
3273          */
3274         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3275
3276         if (!parse_options((char *) sbi->s_es->s_mount_opts, sb,
3277                            &journal_devnum, &journal_ioprio, 0)) {
3278                 ext4_msg(sb, KERN_WARNING,
3279                          "failed to parse options in superblock: %s",
3280                          sbi->s_es->s_mount_opts);
3281         }
3282         sbi->s_def_mount_opt = sbi->s_mount_opt;
3283         if (!parse_options((char *) data, sb, &journal_devnum,
3284                            &journal_ioprio, 0))
3285                 goto failed_mount;
3286
3287         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3288                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3289                             "with data=journal disables delayed "
3290                             "allocation and O_DIRECT support!\n");
3291                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3292                         ext4_msg(sb, KERN_ERR, "can't mount with "
3293                                  "both data=journal and delalloc");
3294                         goto failed_mount;
3295                 }
3296                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3297                         ext4_msg(sb, KERN_ERR, "can't mount with "
3298                                  "both data=journal and dioread_nolock");
3299                         goto failed_mount;
3300                 }
3301                 if (test_opt(sb, DAX)) {
3302                         ext4_msg(sb, KERN_ERR, "can't mount with "
3303                                  "both data=journal and dax");
3304                         goto failed_mount;
3305                 }
3306                 if (test_opt(sb, DELALLOC))
3307                         clear_opt(sb, DELALLOC);
3308         } else {
3309                 sb->s_iflags |= SB_I_CGROUPWB;
3310         }
3311
3312         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3313                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3314
3315         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3316             (ext4_has_compat_features(sb) ||
3317              ext4_has_ro_compat_features(sb) ||
3318              ext4_has_incompat_features(sb)))
3319                 ext4_msg(sb, KERN_WARNING,
3320                        "feature flags set on rev 0 fs, "
3321                        "running e2fsck is recommended");
3322
3323         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3324                 set_opt2(sb, HURD_COMPAT);
3325                 if (ext4_has_feature_64bit(sb)) {
3326                         ext4_msg(sb, KERN_ERR,
3327                                  "The Hurd can't support 64-bit file systems");
3328                         goto failed_mount;
3329                 }
3330         }
3331
3332         if (IS_EXT2_SB(sb)) {
3333                 if (ext2_feature_set_ok(sb))
3334                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3335                                  "using the ext4 subsystem");
3336                 else {
3337                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3338                                  "to feature incompatibilities");
3339                         goto failed_mount;
3340                 }
3341         }
3342
3343         if (IS_EXT3_SB(sb)) {
3344                 if (ext3_feature_set_ok(sb))
3345                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3346                                  "using the ext4 subsystem");
3347                 else {
3348                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3349                                  "to feature incompatibilities");
3350                         goto failed_mount;
3351                 }
3352         }
3353
3354         /*
3355          * Check feature flags regardless of the revision level, since we
3356          * previously didn't change the revision level when setting the flags,
3357          * so there is a chance incompat flags are set on a rev 0 filesystem.
3358          */
3359         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3360                 goto failed_mount;
3361
3362         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3363         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3364             blocksize > EXT4_MAX_BLOCK_SIZE) {
3365                 ext4_msg(sb, KERN_ERR,
3366                        "Unsupported filesystem blocksize %d", blocksize);
3367                 goto failed_mount;
3368         }
3369
3370         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3371                 if (blocksize != PAGE_SIZE) {
3372                         ext4_msg(sb, KERN_ERR,
3373                                         "error: unsupported blocksize for dax");
3374                         goto failed_mount;
3375                 }
3376                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3377                         ext4_msg(sb, KERN_ERR,
3378                                         "error: device does not support dax");
3379                         goto failed_mount;
3380                 }
3381         }
3382
3383         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3384                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3385                          es->s_encryption_level);
3386                 goto failed_mount;
3387         }
3388
3389         if (sb->s_blocksize != blocksize) {
3390                 /* Validate the filesystem blocksize */
3391                 if (!sb_set_blocksize(sb, blocksize)) {
3392                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3393                                         blocksize);
3394                         goto failed_mount;
3395                 }
3396
3397                 brelse(bh);
3398                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3399                 offset = do_div(logical_sb_block, blocksize);
3400                 bh = sb_bread_unmovable(sb, logical_sb_block);
3401                 if (!bh) {
3402                         ext4_msg(sb, KERN_ERR,
3403                                "Can't read superblock on 2nd try");
3404                         goto failed_mount;
3405                 }
3406                 es = (struct ext4_super_block *)(bh->b_data + offset);
3407                 sbi->s_es = es;
3408                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3409                         ext4_msg(sb, KERN_ERR,
3410                                "Magic mismatch, very weird!");
3411                         goto failed_mount;
3412                 }
3413         }
3414
3415         has_huge_files = ext4_has_feature_huge_file(sb);
3416         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3417                                                       has_huge_files);
3418         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3419
3420         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3421                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3422                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3423         } else {
3424                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3425                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3426                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3427                     (!is_power_of_2(sbi->s_inode_size)) ||
3428                     (sbi->s_inode_size > blocksize)) {
3429                         ext4_msg(sb, KERN_ERR,
3430                                "unsupported inode size: %d",
3431                                sbi->s_inode_size);
3432                         goto failed_mount;
3433                 }
3434                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3435                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3436         }
3437
3438         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3439         if (ext4_has_feature_64bit(sb)) {
3440                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3441                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3442                     !is_power_of_2(sbi->s_desc_size)) {
3443                         ext4_msg(sb, KERN_ERR,
3444                                "unsupported descriptor size %lu",
3445                                sbi->s_desc_size);
3446                         goto failed_mount;
3447                 }
3448         } else
3449                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3450
3451         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3452         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3453         if (EXT4_INODE_SIZE(sb) == 0 || EXT4_INODES_PER_GROUP(sb) == 0)
3454                 goto cantfind_ext4;
3455
3456         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3457         if (sbi->s_inodes_per_block == 0)
3458                 goto cantfind_ext4;
3459         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3460                                         sbi->s_inodes_per_block;
3461         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3462         sbi->s_sbh = bh;
3463         sbi->s_mount_state = le16_to_cpu(es->s_state);
3464         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3465         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3466
3467         for (i = 0; i < 4; i++)
3468                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3469         sbi->s_def_hash_version = es->s_def_hash_version;
3470         if (ext4_has_feature_dir_index(sb)) {
3471                 i = le32_to_cpu(es->s_flags);
3472                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3473                         sbi->s_hash_unsigned = 3;
3474                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3475 #ifdef __CHAR_UNSIGNED__
3476                         if (!(sb->s_flags & MS_RDONLY))
3477                                 es->s_flags |=
3478                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3479                         sbi->s_hash_unsigned = 3;
3480 #else
3481                         if (!(sb->s_flags & MS_RDONLY))
3482                                 es->s_flags |=
3483                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3484 #endif
3485                 }
3486         }
3487
3488         /* Handle clustersize */
3489         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3490         has_bigalloc = ext4_has_feature_bigalloc(sb);
3491         if (has_bigalloc) {
3492                 if (clustersize < blocksize) {
3493                         ext4_msg(sb, KERN_ERR,
3494                                  "cluster size (%d) smaller than "
3495                                  "block size (%d)", clustersize, blocksize);
3496                         goto failed_mount;
3497                 }
3498                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3499                         le32_to_cpu(es->s_log_block_size);
3500                 sbi->s_clusters_per_group =
3501                         le32_to_cpu(es->s_clusters_per_group);
3502                 if (sbi->s_clusters_per_group > blocksize * 8) {
3503                         ext4_msg(sb, KERN_ERR,
3504                                  "#clusters per group too big: %lu",
3505                                  sbi->s_clusters_per_group);
3506                         goto failed_mount;
3507                 }
3508                 if (sbi->s_blocks_per_group !=
3509                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3510                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3511                                  "clusters per group (%lu) inconsistent",
3512                                  sbi->s_blocks_per_group,
3513                                  sbi->s_clusters_per_group);
3514                         goto failed_mount;
3515                 }
3516         } else {
3517                 if (clustersize != blocksize) {
3518                         ext4_warning(sb, "fragment/cluster size (%d) != "
3519                                      "block size (%d)", clustersize,
3520                                      blocksize);
3521                         clustersize = blocksize;
3522                 }
3523                 if (sbi->s_blocks_per_group > blocksize * 8) {
3524                         ext4_msg(sb, KERN_ERR,
3525                                  "#blocks per group too big: %lu",
3526                                  sbi->s_blocks_per_group);
3527                         goto failed_mount;
3528                 }
3529                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3530                 sbi->s_cluster_bits = 0;
3531         }
3532         sbi->s_cluster_ratio = clustersize / blocksize;
3533
3534         if (sbi->s_inodes_per_group > blocksize * 8) {
3535                 ext4_msg(sb, KERN_ERR,
3536                        "#inodes per group too big: %lu",
3537                        sbi->s_inodes_per_group);
3538                 goto failed_mount;
3539         }
3540
3541         /* Do we have standard group size of clustersize * 8 blocks ? */
3542         if (sbi->s_blocks_per_group == clustersize << 3)
3543                 set_opt2(sb, STD_GROUP_SIZE);
3544
3545         /*
3546          * Test whether we have more sectors than will fit in sector_t,
3547          * and whether the max offset is addressable by the page cache.
3548          */
3549         err = generic_check_addressable(sb->s_blocksize_bits,
3550                                         ext4_blocks_count(es));
3551         if (err) {
3552                 ext4_msg(sb, KERN_ERR, "filesystem"
3553                          " too large to mount safely on this system");
3554                 if (sizeof(sector_t) < 8)
3555                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3556                 goto failed_mount;
3557         }
3558
3559         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3560                 goto cantfind_ext4;
3561
3562         /* check blocks count against device size */
3563         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3564         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3565                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3566                        "exceeds size of device (%llu blocks)",
3567                        ext4_blocks_count(es), blocks_count);
3568                 goto failed_mount;
3569         }
3570
3571         /*
3572          * It makes no sense for the first data block to be beyond the end
3573          * of the filesystem.
3574          */
3575         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3576                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3577                          "block %u is beyond end of filesystem (%llu)",
3578                          le32_to_cpu(es->s_first_data_block),
3579                          ext4_blocks_count(es));
3580                 goto failed_mount;
3581         }
3582         blocks_count = (ext4_blocks_count(es) -
3583                         le32_to_cpu(es->s_first_data_block) +
3584                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3585         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3586         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3587                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3588                        "(block count %llu, first data block %u, "
3589                        "blocks per group %lu)", sbi->s_groups_count,
3590                        ext4_blocks_count(es),
3591                        le32_to_cpu(es->s_first_data_block),
3592                        EXT4_BLOCKS_PER_GROUP(sb));
3593                 goto failed_mount;
3594         }
3595         sbi->s_groups_count = blocks_count;
3596         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3597                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3598         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3599                    EXT4_DESC_PER_BLOCK(sb);
3600         sbi->s_group_desc = ext4_kvmalloc(db_count *
3601                                           sizeof(struct buffer_head *),
3602                                           GFP_KERNEL);
3603         if (sbi->s_group_desc == NULL) {
3604                 ext4_msg(sb, KERN_ERR, "not enough memory");
3605                 ret = -ENOMEM;
3606                 goto failed_mount;
3607         }
3608
3609         bgl_lock_init(sbi->s_blockgroup_lock);
3610
3611         for (i = 0; i < db_count; i++) {
3612                 block = descriptor_loc(sb, logical_sb_block, i);
3613                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3614                 if (!sbi->s_group_desc[i]) {
3615                         ext4_msg(sb, KERN_ERR,
3616                                "can't read group descriptor %d", i);
3617                         db_count = i;
3618                         goto failed_mount2;
3619                 }
3620         }
3621         if (!ext4_check_descriptors(sb, &first_not_zeroed)) {
3622                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3623                 ret = -EFSCORRUPTED;
3624                 goto failed_mount2;
3625         }
3626
3627         sbi->s_gdb_count = db_count;
3628         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3629         spin_lock_init(&sbi->s_next_gen_lock);
3630
3631         setup_timer(&sbi->s_err_report, print_daily_error_info,
3632                 (unsigned long) sb);
3633
3634         /* Register extent status tree shrinker */
3635         if (ext4_es_register_shrinker(sbi))
3636                 goto failed_mount3;
3637
3638         sbi->s_stripe = ext4_get_stripe_size(sbi);
3639         sbi->s_extent_max_zeroout_kb = 32;
3640
3641         /*
3642          * set up enough so that it can read an inode
3643          */
3644         sb->s_op = &ext4_sops;
3645         sb->s_export_op = &ext4_export_ops;
3646         sb->s_xattr = ext4_xattr_handlers;
3647 #ifdef CONFIG_QUOTA
3648         sb->dq_op = &ext4_quota_operations;
3649         if (ext4_has_feature_quota(sb))
3650                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3651         else
3652                 sb->s_qcop = &ext4_qctl_operations;
3653         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
3654 #endif
3655         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3656
3657         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3658         mutex_init(&sbi->s_orphan_lock);
3659
3660         sb->s_root = NULL;
3661
3662         needs_recovery = (es->s_last_orphan != 0 ||
3663                           ext4_has_feature_journal_needs_recovery(sb));
3664
3665         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3666                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3667                         goto failed_mount3a;
3668
3669         /*
3670          * The first inode we look at is the journal inode.  Don't try
3671          * root first: it may be modified in the journal!
3672          */
3673         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3674                 if (ext4_load_journal(sb, es, journal_devnum))
3675                         goto failed_mount3a;
3676         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3677                    ext4_has_feature_journal_needs_recovery(sb)) {
3678                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3679                        "suppressed and not mounted read-only");
3680                 goto failed_mount_wq;
3681         } else {
3682                 /* Nojournal mode, all journal mount options are illegal */
3683                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3684                         ext4_msg(sb, KERN_ERR, "can't mount with "
3685                                  "journal_checksum, fs mounted w/o journal");
3686                         goto failed_mount_wq;
3687                 }
3688                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3689                         ext4_msg(sb, KERN_ERR, "can't mount with "
3690                                  "journal_async_commit, fs mounted w/o journal");
3691                         goto failed_mount_wq;
3692                 }
3693                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3694                         ext4_msg(sb, KERN_ERR, "can't mount with "
3695                                  "commit=%lu, fs mounted w/o journal",
3696                                  sbi->s_commit_interval / HZ);
3697                         goto failed_mount_wq;
3698                 }
3699                 if (EXT4_MOUNT_DATA_FLAGS &
3700                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3701                         ext4_msg(sb, KERN_ERR, "can't mount with "
3702                                  "data=, fs mounted w/o journal");
3703                         goto failed_mount_wq;
3704                 }
3705                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
3706                 clear_opt(sb, JOURNAL_CHECKSUM);
3707                 clear_opt(sb, DATA_FLAGS);
3708                 sbi->s_journal = NULL;
3709                 needs_recovery = 0;
3710                 goto no_journal;
3711         }
3712
3713         if (ext4_has_feature_64bit(sb) &&
3714             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3715                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3716                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3717                 goto failed_mount_wq;
3718         }
3719
3720         if (!set_journal_csum_feature_set(sb)) {
3721                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3722                          "feature set");
3723                 goto failed_mount_wq;
3724         }
3725
3726         /* We have now updated the journal if required, so we can
3727          * validate the data journaling mode. */
3728         switch (test_opt(sb, DATA_FLAGS)) {
3729         case 0:
3730                 /* No mode set, assume a default based on the journal
3731                  * capabilities: ORDERED_DATA if the journal can
3732                  * cope, else JOURNAL_DATA
3733                  */
3734                 if (jbd2_journal_check_available_features
3735                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3736                         set_opt(sb, ORDERED_DATA);
3737                 else
3738                         set_opt(sb, JOURNAL_DATA);
3739                 break;
3740
3741         case EXT4_MOUNT_ORDERED_DATA:
3742         case EXT4_MOUNT_WRITEBACK_DATA:
3743                 if (!jbd2_journal_check_available_features
3744                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3745                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3746                                "requested data journaling mode");
3747                         goto failed_mount_wq;
3748                 }
3749         default:
3750                 break;
3751         }
3752         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3753
3754         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3755
3756 no_journal:
3757         if (ext4_mballoc_ready) {
3758                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
3759                 if (!sbi->s_mb_cache) {
3760                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
3761                         goto failed_mount_wq;
3762                 }
3763         }
3764
3765         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
3766             (blocksize != PAGE_CACHE_SIZE)) {
3767                 ext4_msg(sb, KERN_ERR,
3768                          "Unsupported blocksize for fs encryption");
3769                 goto failed_mount_wq;
3770         }
3771
3772         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
3773             !ext4_has_feature_encrypt(sb)) {
3774                 ext4_set_feature_encrypt(sb);
3775                 ext4_commit_super(sb, 1);
3776         }
3777
3778         /*
3779          * Get the # of file system overhead blocks from the
3780          * superblock if present.
3781          */
3782         if (es->s_overhead_clusters)
3783                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3784         else {
3785                 err = ext4_calculate_overhead(sb);
3786                 if (err)
3787                         goto failed_mount_wq;
3788         }
3789
3790         /*
3791          * The maximum number of concurrent works can be high and
3792          * concurrency isn't really necessary.  Limit it to 1.
3793          */
3794         EXT4_SB(sb)->rsv_conversion_wq =
3795                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3796         if (!EXT4_SB(sb)->rsv_conversion_wq) {
3797                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3798                 ret = -ENOMEM;
3799                 goto failed_mount4;
3800         }
3801
3802         /*
3803          * The jbd2_journal_load will have done any necessary log recovery,
3804          * so we can safely mount the rest of the filesystem now.
3805          */
3806
3807         root = ext4_iget(sb, EXT4_ROOT_INO);
3808         if (IS_ERR(root)) {
3809                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3810                 ret = PTR_ERR(root);
3811                 root = NULL;
3812                 goto failed_mount4;
3813         }
3814         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3815                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3816                 iput(root);
3817                 goto failed_mount4;
3818         }
3819         sb->s_root = d_make_root(root);
3820         if (!sb->s_root) {
3821                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3822                 ret = -ENOMEM;
3823                 goto failed_mount4;
3824         }
3825
3826         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3827                 sb->s_flags |= MS_RDONLY;
3828
3829         /* determine the minimum size of new large inodes, if present */
3830         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3831                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3832                                                      EXT4_GOOD_OLD_INODE_SIZE;
3833                 if (ext4_has_feature_extra_isize(sb)) {
3834                         if (sbi->s_want_extra_isize <
3835                             le16_to_cpu(es->s_want_extra_isize))
3836                                 sbi->s_want_extra_isize =
3837                                         le16_to_cpu(es->s_want_extra_isize);
3838                         if (sbi->s_want_extra_isize <
3839                             le16_to_cpu(es->s_min_extra_isize))
3840                                 sbi->s_want_extra_isize =
3841                                         le16_to_cpu(es->s_min_extra_isize);
3842                 }
3843         }
3844         /* Check if enough inode space is available */
3845         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3846                                                         sbi->s_inode_size) {
3847                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3848                                                        EXT4_GOOD_OLD_INODE_SIZE;
3849                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3850                          "available");
3851         }
3852
3853         ext4_set_resv_clusters(sb);
3854
3855         err = ext4_setup_system_zone(sb);
3856         if (err) {
3857                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3858                          "zone (%d)", err);
3859                 goto failed_mount4a;
3860         }
3861
3862         ext4_ext_init(sb);
3863         err = ext4_mb_init(sb);
3864         if (err) {
3865                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3866                          err);
3867                 goto failed_mount5;
3868         }
3869
3870         block = ext4_count_free_clusters(sb);
3871         ext4_free_blocks_count_set(sbi->s_es,
3872                                    EXT4_C2B(sbi, block));
3873         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
3874                                   GFP_KERNEL);
3875         if (!err) {
3876                 unsigned long freei = ext4_count_free_inodes(sb);
3877                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
3878                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
3879                                           GFP_KERNEL);
3880         }
3881         if (!err)
3882                 err = percpu_counter_init(&sbi->s_dirs_counter,
3883                                           ext4_count_dirs(sb), GFP_KERNEL);
3884         if (!err)
3885                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
3886                                           GFP_KERNEL);
3887         if (err) {
3888                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3889                 goto failed_mount6;
3890         }
3891
3892         if (ext4_has_feature_flex_bg(sb))
3893                 if (!ext4_fill_flex_info(sb)) {
3894                         ext4_msg(sb, KERN_ERR,
3895                                "unable to initialize "
3896                                "flex_bg meta info!");
3897                         goto failed_mount6;
3898                 }
3899
3900         err = ext4_register_li_request(sb, first_not_zeroed);
3901         if (err)
3902                 goto failed_mount6;
3903
3904         err = ext4_register_sysfs(sb);
3905         if (err)
3906                 goto failed_mount7;
3907
3908 #ifdef CONFIG_QUOTA
3909         /* Enable quota usage during mount. */
3910         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
3911                 err = ext4_enable_quotas(sb);
3912                 if (err)
3913                         goto failed_mount8;
3914         }
3915 #endif  /* CONFIG_QUOTA */
3916
3917         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3918         ext4_orphan_cleanup(sb, es);
3919         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3920         if (needs_recovery) {
3921                 ext4_msg(sb, KERN_INFO, "recovery complete");
3922                 ext4_mark_recovery_complete(sb, es);
3923         }
3924         if (EXT4_SB(sb)->s_journal) {
3925                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3926                         descr = " journalled data mode";
3927                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3928                         descr = " ordered data mode";
3929                 else
3930                         descr = " writeback data mode";
3931         } else
3932                 descr = "out journal";
3933
3934         if (test_opt(sb, DISCARD)) {
3935                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
3936                 if (!blk_queue_discard(q))
3937                         ext4_msg(sb, KERN_WARNING,
3938                                  "mounting with \"discard\" option, but "
3939                                  "the device does not support discard");
3940         }
3941
3942         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
3943                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
3944                          "Opts: %s%s%s", descr, sbi->s_es->s_mount_opts,
3945                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
3946
3947         if (es->s_error_count)
3948                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
3949
3950         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
3951         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
3952         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
3953         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
3954
3955         kfree(orig_data);
3956         return 0;
3957
3958 cantfind_ext4:
3959         if (!silent)
3960                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
3961         goto failed_mount;
3962
3963 #ifdef CONFIG_QUOTA
3964 failed_mount8:
3965         ext4_unregister_sysfs(sb);
3966 #endif
3967 failed_mount7:
3968         ext4_unregister_li_request(sb);
3969 failed_mount6:
3970         ext4_mb_release(sb);
3971         if (sbi->s_flex_groups)
3972                 kvfree(sbi->s_flex_groups);
3973         percpu_counter_destroy(&sbi->s_freeclusters_counter);
3974         percpu_counter_destroy(&sbi->s_freeinodes_counter);
3975         percpu_counter_destroy(&sbi->s_dirs_counter);
3976         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
3977 failed_mount5:
3978         ext4_ext_release(sb);
3979         ext4_release_system_zone(sb);
3980 failed_mount4a:
3981         dput(sb->s_root);
3982         sb->s_root = NULL;
3983 failed_mount4:
3984         ext4_msg(sb, KERN_ERR, "mount failed");
3985         if (EXT4_SB(sb)->rsv_conversion_wq)
3986                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
3987 failed_mount_wq:
3988         if (sbi->s_journal) {
3989                 jbd2_journal_destroy(sbi->s_journal);
3990                 sbi->s_journal = NULL;
3991         }
3992 failed_mount3a:
3993         ext4_es_unregister_shrinker(sbi);
3994 failed_mount3:
3995         del_timer_sync(&sbi->s_err_report);
3996         if (sbi->s_mmp_tsk)
3997                 kthread_stop(sbi->s_mmp_tsk);
3998 failed_mount2:
3999         for (i = 0; i < db_count; i++)
4000                 brelse(sbi->s_group_desc[i]);
4001         kvfree(sbi->s_group_desc);
4002 failed_mount:
4003         if (sbi->s_chksum_driver)
4004                 crypto_free_shash(sbi->s_chksum_driver);
4005 #ifdef CONFIG_QUOTA
4006         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4007                 kfree(sbi->s_qf_names[i]);
4008 #endif
4009         ext4_blkdev_remove(sbi);
4010         brelse(bh);
4011 out_fail:
4012         sb->s_fs_info = NULL;
4013         kfree(sbi->s_blockgroup_lock);
4014         kfree(sbi);
4015 out_free_orig:
4016         kfree(orig_data);
4017         return err ? err : ret;
4018 }
4019
4020 /*
4021  * Setup any per-fs journal parameters now.  We'll do this both on
4022  * initial mount, once the journal has been initialised but before we've
4023  * done any recovery; and again on any subsequent remount.
4024  */
4025 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4026 {
4027         struct ext4_sb_info *sbi = EXT4_SB(sb);
4028
4029         journal->j_commit_interval = sbi->s_commit_interval;
4030         journal->j_min_batch_time = sbi->s_min_batch_time;
4031         journal->j_max_batch_time = sbi->s_max_batch_time;
4032
4033         write_lock(&journal->j_state_lock);
4034         if (test_opt(sb, BARRIER))
4035                 journal->j_flags |= JBD2_BARRIER;
4036         else
4037                 journal->j_flags &= ~JBD2_BARRIER;
4038         if (test_opt(sb, DATA_ERR_ABORT))
4039                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4040         else
4041                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4042         write_unlock(&journal->j_state_lock);
4043 }
4044
4045 static journal_t *ext4_get_journal(struct super_block *sb,
4046                                    unsigned int journal_inum)
4047 {
4048         struct inode *journal_inode;
4049         journal_t *journal;
4050
4051         BUG_ON(!ext4_has_feature_journal(sb));
4052
4053         /* First, test for the existence of a valid inode on disk.  Bad
4054          * things happen if we iget() an unused inode, as the subsequent
4055          * iput() will try to delete it. */
4056
4057         journal_inode = ext4_iget(sb, journal_inum);
4058         if (IS_ERR(journal_inode)) {
4059                 ext4_msg(sb, KERN_ERR, "no journal found");
4060                 return NULL;
4061         }
4062         if (!journal_inode->i_nlink) {
4063                 make_bad_inode(journal_inode);
4064                 iput(journal_inode);
4065                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4066                 return NULL;
4067         }
4068
4069         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4070                   journal_inode, journal_inode->i_size);
4071         if (!S_ISREG(journal_inode->i_mode)) {
4072                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4073                 iput(journal_inode);
4074                 return NULL;
4075         }
4076
4077         journal = jbd2_journal_init_inode(journal_inode);
4078         if (!journal) {
4079                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4080                 iput(journal_inode);
4081                 return NULL;
4082         }
4083         journal->j_private = sb;
4084         ext4_init_journal_params(sb, journal);
4085         return journal;
4086 }
4087
4088 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4089                                        dev_t j_dev)
4090 {
4091         struct buffer_head *bh;
4092         journal_t *journal;
4093         ext4_fsblk_t start;
4094         ext4_fsblk_t len;
4095         int hblock, blocksize;
4096         ext4_fsblk_t sb_block;
4097         unsigned long offset;
4098         struct ext4_super_block *es;
4099         struct block_device *bdev;
4100
4101         BUG_ON(!ext4_has_feature_journal(sb));
4102
4103         bdev = ext4_blkdev_get(j_dev, sb);
4104         if (bdev == NULL)
4105                 return NULL;
4106
4107         blocksize = sb->s_blocksize;
4108         hblock = bdev_logical_block_size(bdev);
4109         if (blocksize < hblock) {
4110                 ext4_msg(sb, KERN_ERR,
4111                         "blocksize too small for journal device");
4112                 goto out_bdev;
4113         }
4114
4115         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4116         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4117         set_blocksize(bdev, blocksize);
4118         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4119                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4120                        "external journal");
4121                 goto out_bdev;
4122         }
4123
4124         es = (struct ext4_super_block *) (bh->b_data + offset);
4125         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4126             !(le32_to_cpu(es->s_feature_incompat) &
4127               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4128                 ext4_msg(sb, KERN_ERR, "external journal has "
4129                                         "bad superblock");
4130                 brelse(bh);
4131                 goto out_bdev;
4132         }
4133
4134         if ((le32_to_cpu(es->s_feature_ro_compat) &
4135              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4136             es->s_checksum != ext4_superblock_csum(sb, es)) {
4137                 ext4_msg(sb, KERN_ERR, "external journal has "
4138                                        "corrupt superblock");
4139                 brelse(bh);
4140                 goto out_bdev;
4141         }
4142
4143         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4144                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4145                 brelse(bh);
4146                 goto out_bdev;
4147         }
4148
4149         len = ext4_blocks_count(es);
4150         start = sb_block + 1;
4151         brelse(bh);     /* we're done with the superblock */
4152
4153         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4154                                         start, len, blocksize);
4155         if (!journal) {
4156                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4157                 goto out_bdev;
4158         }
4159         journal->j_private = sb;
4160         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4161         wait_on_buffer(journal->j_sb_buffer);
4162         if (!buffer_uptodate(journal->j_sb_buffer)) {
4163                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4164                 goto out_journal;
4165         }
4166         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4167                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4168                                         "user (unsupported) - %d",
4169                         be32_to_cpu(journal->j_superblock->s_nr_users));
4170                 goto out_journal;
4171         }
4172         EXT4_SB(sb)->journal_bdev = bdev;
4173         ext4_init_journal_params(sb, journal);
4174         return journal;
4175
4176 out_journal:
4177         jbd2_journal_destroy(journal);
4178 out_bdev:
4179         ext4_blkdev_put(bdev);
4180         return NULL;
4181 }
4182
4183 static int ext4_load_journal(struct super_block *sb,
4184                              struct ext4_super_block *es,
4185                              unsigned long journal_devnum)
4186 {
4187         journal_t *journal;
4188         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4189         dev_t journal_dev;
4190         int err = 0;
4191         int really_read_only;
4192
4193         BUG_ON(!ext4_has_feature_journal(sb));
4194
4195         if (journal_devnum &&
4196             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4197                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4198                         "numbers have changed");
4199                 journal_dev = new_decode_dev(journal_devnum);
4200         } else
4201                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4202
4203         really_read_only = bdev_read_only(sb->s_bdev);
4204
4205         /*
4206          * Are we loading a blank journal or performing recovery after a
4207          * crash?  For recovery, we need to check in advance whether we
4208          * can get read-write access to the device.
4209          */
4210         if (ext4_has_feature_journal_needs_recovery(sb)) {
4211                 if (sb->s_flags & MS_RDONLY) {
4212                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4213                                         "required on readonly filesystem");
4214                         if (really_read_only) {
4215                                 ext4_msg(sb, KERN_ERR, "write access "
4216                                         "unavailable, cannot proceed");
4217                                 return -EROFS;
4218                         }
4219                         ext4_msg(sb, KERN_INFO, "write access will "
4220                                "be enabled during recovery");
4221                 }
4222         }
4223
4224         if (journal_inum && journal_dev) {
4225                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4226                        "and inode journals!");
4227                 return -EINVAL;
4228         }
4229
4230         if (journal_inum) {
4231                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4232                         return -EINVAL;
4233         } else {
4234                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4235                         return -EINVAL;
4236         }
4237
4238         if (!(journal->j_flags & JBD2_BARRIER))
4239                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4240
4241         if (!ext4_has_feature_journal_needs_recovery(sb))
4242                 err = jbd2_journal_wipe(journal, !really_read_only);
4243         if (!err) {
4244                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4245                 if (save)
4246                         memcpy(save, ((char *) es) +
4247                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4248                 err = jbd2_journal_load(journal);
4249                 if (save)
4250                         memcpy(((char *) es) + EXT4_S_ERR_START,
4251                                save, EXT4_S_ERR_LEN);
4252                 kfree(save);
4253         }
4254
4255         if (err) {
4256                 ext4_msg(sb, KERN_ERR, "error loading journal");
4257                 jbd2_journal_destroy(journal);
4258                 return err;
4259         }
4260
4261         EXT4_SB(sb)->s_journal = journal;
4262         ext4_clear_journal_err(sb, es);
4263
4264         if (!really_read_only && journal_devnum &&
4265             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4266                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4267
4268                 /* Make sure we flush the recovery flag to disk. */
4269                 ext4_commit_super(sb, 1);
4270         }
4271
4272         return 0;
4273 }
4274
4275 static int ext4_commit_super(struct super_block *sb, int sync)
4276 {
4277         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4278         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4279         int error = 0;
4280
4281         if (!sbh || block_device_ejected(sb))
4282                 return error;
4283         if (buffer_write_io_error(sbh)) {
4284                 /*
4285                  * Oh, dear.  A previous attempt to write the
4286                  * superblock failed.  This could happen because the
4287                  * USB device was yanked out.  Or it could happen to
4288                  * be a transient write error and maybe the block will
4289                  * be remapped.  Nothing we can do but to retry the
4290                  * write and hope for the best.
4291                  */
4292                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4293                        "superblock detected");
4294                 clear_buffer_write_io_error(sbh);
4295                 set_buffer_uptodate(sbh);
4296         }
4297         /*
4298          * If the file system is mounted read-only, don't update the
4299          * superblock write time.  This avoids updating the superblock
4300          * write time when we are mounting the root file system
4301          * read/only but we need to replay the journal; at that point,
4302          * for people who are east of GMT and who make their clock
4303          * tick in localtime for Windows bug-for-bug compatibility,
4304          * the clock is set in the future, and this will cause e2fsck
4305          * to complain and force a full file system check.
4306          */
4307         if (!(sb->s_flags & MS_RDONLY))
4308                 es->s_wtime = cpu_to_le32(get_seconds());
4309         if (sb->s_bdev->bd_part)
4310                 es->s_kbytes_written =
4311                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4312                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4313                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4314         else
4315                 es->s_kbytes_written =
4316                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4317         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4318                 ext4_free_blocks_count_set(es,
4319                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4320                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4321         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4322                 es->s_free_inodes_count =
4323                         cpu_to_le32(percpu_counter_sum_positive(
4324                                 &EXT4_SB(sb)->s_freeinodes_counter));
4325         BUFFER_TRACE(sbh, "marking dirty");
4326         ext4_superblock_csum_set(sb);
4327         mark_buffer_dirty(sbh);
4328         if (sync) {
4329                 error = __sync_dirty_buffer(sbh,
4330                         test_opt(sb, BARRIER) ? WRITE_FUA : WRITE_SYNC);
4331                 if (error)
4332                         return error;
4333
4334                 error = buffer_write_io_error(sbh);
4335                 if (error) {
4336                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4337                                "superblock");
4338                         clear_buffer_write_io_error(sbh);
4339                         set_buffer_uptodate(sbh);
4340                 }
4341         }
4342         return error;
4343 }
4344
4345 /*
4346  * Have we just finished recovery?  If so, and if we are mounting (or
4347  * remounting) the filesystem readonly, then we will end up with a
4348  * consistent fs on disk.  Record that fact.
4349  */
4350 static void ext4_mark_recovery_complete(struct super_block *sb,
4351                                         struct ext4_super_block *es)
4352 {
4353         journal_t *journal = EXT4_SB(sb)->s_journal;
4354
4355         if (!ext4_has_feature_journal(sb)) {
4356                 BUG_ON(journal != NULL);
4357                 return;
4358         }
4359         jbd2_journal_lock_updates(journal);
4360         if (jbd2_journal_flush(journal) < 0)
4361                 goto out;
4362
4363         if (ext4_has_feature_journal_needs_recovery(sb) &&
4364             sb->s_flags & MS_RDONLY) {
4365                 ext4_clear_feature_journal_needs_recovery(sb);
4366                 ext4_commit_super(sb, 1);
4367         }
4368
4369 out:
4370         jbd2_journal_unlock_updates(journal);
4371 }
4372
4373 /*
4374  * If we are mounting (or read-write remounting) a filesystem whose journal
4375  * has recorded an error from a previous lifetime, move that error to the
4376  * main filesystem now.
4377  */
4378 static void ext4_clear_journal_err(struct super_block *sb,
4379                                    struct ext4_super_block *es)
4380 {
4381         journal_t *journal;
4382         int j_errno;
4383         const char *errstr;
4384
4385         BUG_ON(!ext4_has_feature_journal(sb));
4386
4387         journal = EXT4_SB(sb)->s_journal;
4388
4389         /*
4390          * Now check for any error status which may have been recorded in the
4391          * journal by a prior ext4_error() or ext4_abort()
4392          */
4393
4394         j_errno = jbd2_journal_errno(journal);
4395         if (j_errno) {
4396                 char nbuf[16];
4397
4398                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4399                 ext4_warning(sb, "Filesystem error recorded "
4400                              "from previous mount: %s", errstr);
4401                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4402
4403                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4404                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4405                 ext4_commit_super(sb, 1);
4406
4407                 jbd2_journal_clear_err(journal);
4408                 jbd2_journal_update_sb_errno(journal);
4409         }
4410 }
4411
4412 /*
4413  * Force the running and committing transactions to commit,
4414  * and wait on the commit.
4415  */
4416 int ext4_force_commit(struct super_block *sb)
4417 {
4418         journal_t *journal;
4419
4420         if (sb->s_flags & MS_RDONLY)
4421                 return 0;
4422
4423         journal = EXT4_SB(sb)->s_journal;
4424         return ext4_journal_force_commit(journal);
4425 }
4426
4427 static int ext4_sync_fs(struct super_block *sb, int wait)
4428 {
4429         int ret = 0;
4430         tid_t target;
4431         bool needs_barrier = false;
4432         struct ext4_sb_info *sbi = EXT4_SB(sb);
4433
4434         trace_ext4_sync_fs(sb, wait);
4435         flush_workqueue(sbi->rsv_conversion_wq);
4436         /*
4437          * Writeback quota in non-journalled quota case - journalled quota has
4438          * no dirty dquots
4439          */
4440         dquot_writeback_dquots(sb, -1);
4441         /*
4442          * Data writeback is possible w/o journal transaction, so barrier must
4443          * being sent at the end of the function. But we can skip it if
4444          * transaction_commit will do it for us.
4445          */
4446         if (sbi->s_journal) {
4447                 target = jbd2_get_latest_transaction(sbi->s_journal);
4448                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4449                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4450                         needs_barrier = true;
4451
4452                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4453                         if (wait)
4454                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4455                                                            target);
4456                 }
4457         } else if (wait && test_opt(sb, BARRIER))
4458                 needs_barrier = true;
4459         if (needs_barrier) {
4460                 int err;
4461                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4462                 if (!ret)
4463                         ret = err;
4464         }
4465
4466         return ret;
4467 }
4468
4469 /*
4470  * LVM calls this function before a (read-only) snapshot is created.  This
4471  * gives us a chance to flush the journal completely and mark the fs clean.
4472  *
4473  * Note that only this function cannot bring a filesystem to be in a clean
4474  * state independently. It relies on upper layer to stop all data & metadata
4475  * modifications.
4476  */
4477 static int ext4_freeze(struct super_block *sb)
4478 {
4479         int error = 0;
4480         journal_t *journal;
4481
4482         if (sb->s_flags & MS_RDONLY)
4483                 return 0;
4484
4485         journal = EXT4_SB(sb)->s_journal;
4486
4487         if (journal) {
4488                 /* Now we set up the journal barrier. */
4489                 jbd2_journal_lock_updates(journal);
4490
4491                 /*
4492                  * Don't clear the needs_recovery flag if we failed to
4493                  * flush the journal.
4494                  */
4495                 error = jbd2_journal_flush(journal);
4496                 if (error < 0)
4497                         goto out;
4498
4499                 /* Journal blocked and flushed, clear needs_recovery flag. */
4500                 ext4_clear_feature_journal_needs_recovery(sb);
4501         }
4502
4503         error = ext4_commit_super(sb, 1);
4504 out:
4505         if (journal)
4506                 /* we rely on upper layer to stop further updates */
4507                 jbd2_journal_unlock_updates(journal);
4508         return error;
4509 }
4510
4511 /*
4512  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4513  * flag here, even though the filesystem is not technically dirty yet.
4514  */
4515 static int ext4_unfreeze(struct super_block *sb)
4516 {
4517         if (sb->s_flags & MS_RDONLY)
4518                 return 0;
4519
4520         if (EXT4_SB(sb)->s_journal) {
4521                 /* Reset the needs_recovery flag before the fs is unlocked. */
4522                 ext4_set_feature_journal_needs_recovery(sb);
4523         }
4524
4525         ext4_commit_super(sb, 1);
4526         return 0;
4527 }
4528
4529 /*
4530  * Structure to save mount options for ext4_remount's benefit
4531  */
4532 struct ext4_mount_options {
4533         unsigned long s_mount_opt;
4534         unsigned long s_mount_opt2;
4535         kuid_t s_resuid;
4536         kgid_t s_resgid;
4537         unsigned long s_commit_interval;
4538         u32 s_min_batch_time, s_max_batch_time;
4539 #ifdef CONFIG_QUOTA
4540         int s_jquota_fmt;
4541         char *s_qf_names[EXT4_MAXQUOTAS];
4542 #endif
4543 };
4544
4545 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4546 {
4547         struct ext4_super_block *es;
4548         struct ext4_sb_info *sbi = EXT4_SB(sb);
4549         unsigned long old_sb_flags;
4550         struct ext4_mount_options old_opts;
4551         int enable_quota = 0;
4552         ext4_group_t g;
4553         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4554         int err = 0;
4555 #ifdef CONFIG_QUOTA
4556         int i, j;
4557 #endif
4558         char *orig_data = kstrdup(data, GFP_KERNEL);
4559
4560         /* Store the original options */
4561         old_sb_flags = sb->s_flags;
4562         old_opts.s_mount_opt = sbi->s_mount_opt;
4563         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4564         old_opts.s_resuid = sbi->s_resuid;
4565         old_opts.s_resgid = sbi->s_resgid;
4566         old_opts.s_commit_interval = sbi->s_commit_interval;
4567         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4568         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4569 #ifdef CONFIG_QUOTA
4570         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4571         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4572                 if (sbi->s_qf_names[i]) {
4573                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4574                                                          GFP_KERNEL);
4575                         if (!old_opts.s_qf_names[i]) {
4576                                 for (j = 0; j < i; j++)
4577                                         kfree(old_opts.s_qf_names[j]);
4578                                 kfree(orig_data);
4579                                 return -ENOMEM;
4580                         }
4581                 } else
4582                         old_opts.s_qf_names[i] = NULL;
4583 #endif
4584         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4585                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4586
4587         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4588                 err = -EINVAL;
4589                 goto restore_opts;
4590         }
4591
4592         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4593             test_opt(sb, JOURNAL_CHECKSUM)) {
4594                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4595                          "during remount not supported; ignoring");
4596                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4597         }
4598
4599         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4600                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4601                         ext4_msg(sb, KERN_ERR, "can't mount with "
4602                                  "both data=journal and delalloc");
4603                         err = -EINVAL;
4604                         goto restore_opts;
4605                 }
4606                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4607                         ext4_msg(sb, KERN_ERR, "can't mount with "
4608                                  "both data=journal and dioread_nolock");
4609                         err = -EINVAL;
4610                         goto restore_opts;
4611                 }
4612                 if (test_opt(sb, DAX)) {
4613                         ext4_msg(sb, KERN_ERR, "can't mount with "
4614                                  "both data=journal and dax");
4615                         err = -EINVAL;
4616                         goto restore_opts;
4617                 }
4618         }
4619
4620         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4621                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4622                         "dax flag with busy inodes while remounting");
4623                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4624         }
4625
4626         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4627                 ext4_abort(sb, "Abort forced by user");
4628
4629         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4630                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4631
4632         es = sbi->s_es;
4633
4634         if (sbi->s_journal) {
4635                 ext4_init_journal_params(sb, sbi->s_journal);
4636                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4637         }
4638
4639         if (*flags & MS_LAZYTIME)
4640                 sb->s_flags |= MS_LAZYTIME;
4641
4642         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4643                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4644                         err = -EROFS;
4645                         goto restore_opts;
4646                 }
4647
4648                 if (*flags & MS_RDONLY) {
4649                         err = sync_filesystem(sb);
4650                         if (err < 0)
4651                                 goto restore_opts;
4652                         err = dquot_suspend(sb, -1);
4653                         if (err < 0)
4654                                 goto restore_opts;
4655
4656                         /*
4657                          * First of all, the unconditional stuff we have to do
4658                          * to disable replay of the journal when we next remount
4659                          */
4660                         sb->s_flags |= MS_RDONLY;
4661
4662                         /*
4663                          * OK, test if we are remounting a valid rw partition
4664                          * readonly, and if so set the rdonly flag and then
4665                          * mark the partition as valid again.
4666                          */
4667                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4668                             (sbi->s_mount_state & EXT4_VALID_FS))
4669                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4670
4671                         if (sbi->s_journal)
4672                                 ext4_mark_recovery_complete(sb, es);
4673                 } else {
4674                         /* Make sure we can mount this feature set readwrite */
4675                         if (ext4_has_feature_readonly(sb) ||
4676                             !ext4_feature_set_ok(sb, 0)) {
4677                                 err = -EROFS;
4678                                 goto restore_opts;
4679                         }
4680                         /*
4681                          * Make sure the group descriptor checksums
4682                          * are sane.  If they aren't, refuse to remount r/w.
4683                          */
4684                         for (g = 0; g < sbi->s_groups_count; g++) {
4685                                 struct ext4_group_desc *gdp =
4686                                         ext4_get_group_desc(sb, g, NULL);
4687
4688                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4689                                         ext4_msg(sb, KERN_ERR,
4690                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4691                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
4692                                                le16_to_cpu(gdp->bg_checksum));
4693                                         err = -EFSBADCRC;
4694                                         goto restore_opts;
4695                                 }
4696                         }
4697
4698                         /*
4699                          * If we have an unprocessed orphan list hanging
4700                          * around from a previously readonly bdev mount,
4701                          * require a full umount/remount for now.
4702                          */
4703                         if (es->s_last_orphan) {
4704                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4705                                        "remount RDWR because of unprocessed "
4706                                        "orphan inode list.  Please "
4707                                        "umount/remount instead");
4708                                 err = -EINVAL;
4709                                 goto restore_opts;
4710                         }
4711
4712                         /*
4713                          * Mounting a RDONLY partition read-write, so reread
4714                          * and store the current valid flag.  (It may have
4715                          * been changed by e2fsck since we originally mounted
4716                          * the partition.)
4717                          */
4718                         if (sbi->s_journal)
4719                                 ext4_clear_journal_err(sb, es);
4720                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4721                         if (!ext4_setup_super(sb, es, 0))
4722                                 sb->s_flags &= ~MS_RDONLY;
4723                         if (ext4_has_feature_mmp(sb))
4724                                 if (ext4_multi_mount_protect(sb,
4725                                                 le64_to_cpu(es->s_mmp_block))) {
4726                                         err = -EROFS;
4727                                         goto restore_opts;
4728                                 }
4729                         enable_quota = 1;
4730                 }
4731         }
4732
4733         /*
4734          * Reinitialize lazy itable initialization thread based on
4735          * current settings
4736          */
4737         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4738                 ext4_unregister_li_request(sb);
4739         else {
4740                 ext4_group_t first_not_zeroed;
4741                 first_not_zeroed = ext4_has_uninit_itable(sb);
4742                 ext4_register_li_request(sb, first_not_zeroed);
4743         }
4744
4745         ext4_setup_system_zone(sb);
4746         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4747                 ext4_commit_super(sb, 1);
4748
4749 #ifdef CONFIG_QUOTA
4750         /* Release old quota file names */
4751         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4752                 kfree(old_opts.s_qf_names[i]);
4753         if (enable_quota) {
4754                 if (sb_any_quota_suspended(sb))
4755                         dquot_resume(sb, -1);
4756                 else if (ext4_has_feature_quota(sb)) {
4757                         err = ext4_enable_quotas(sb);
4758                         if (err)
4759                                 goto restore_opts;
4760                 }
4761         }
4762 #endif
4763
4764         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
4765         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4766         kfree(orig_data);
4767         return 0;
4768
4769 restore_opts:
4770         sb->s_flags = old_sb_flags;
4771         sbi->s_mount_opt = old_opts.s_mount_opt;
4772         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4773         sbi->s_resuid = old_opts.s_resuid;
4774         sbi->s_resgid = old_opts.s_resgid;
4775         sbi->s_commit_interval = old_opts.s_commit_interval;
4776         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4777         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4778 #ifdef CONFIG_QUOTA
4779         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4780         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
4781                 kfree(sbi->s_qf_names[i]);
4782                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4783         }
4784 #endif
4785         kfree(orig_data);
4786         return err;
4787 }
4788
4789 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4790 {
4791         struct super_block *sb = dentry->d_sb;
4792         struct ext4_sb_info *sbi = EXT4_SB(sb);
4793         struct ext4_super_block *es = sbi->s_es;
4794         ext4_fsblk_t overhead = 0, resv_blocks;
4795         u64 fsid;
4796         s64 bfree;
4797         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4798
4799         if (!test_opt(sb, MINIX_DF))
4800                 overhead = sbi->s_overhead;
4801
4802         buf->f_type = EXT4_SUPER_MAGIC;
4803         buf->f_bsize = sb->s_blocksize;
4804         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4805         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4806                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4807         /* prevent underflow in case that few free space is available */
4808         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4809         buf->f_bavail = buf->f_bfree -
4810                         (ext4_r_blocks_count(es) + resv_blocks);
4811         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4812                 buf->f_bavail = 0;
4813         buf->f_files = le32_to_cpu(es->s_inodes_count);
4814         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4815         buf->f_namelen = EXT4_NAME_LEN;
4816         fsid = le64_to_cpup((void *)es->s_uuid) ^
4817                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4818         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4819         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4820
4821         return 0;
4822 }
4823
4824 /* Helper function for writing quotas on sync - we need to start transaction
4825  * before quota file is locked for write. Otherwise the are possible deadlocks:
4826  * Process 1                         Process 2
4827  * ext4_create()                     quota_sync()
4828  *   jbd2_journal_start()                  write_dquot()
4829  *   dquot_initialize()                         down(dqio_mutex)
4830  *     down(dqio_mutex)                    jbd2_journal_start()
4831  *
4832  */
4833
4834 #ifdef CONFIG_QUOTA
4835
4836 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4837 {
4838         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4839 }
4840
4841 static int ext4_write_dquot(struct dquot *dquot)
4842 {
4843         int ret, err;
4844         handle_t *handle;
4845         struct inode *inode;
4846
4847         inode = dquot_to_inode(dquot);
4848         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4849                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4850         if (IS_ERR(handle))
4851                 return PTR_ERR(handle);
4852         ret = dquot_commit(dquot);
4853         err = ext4_journal_stop(handle);
4854         if (!ret)
4855                 ret = err;
4856         return ret;
4857 }
4858
4859 static int ext4_acquire_dquot(struct dquot *dquot)
4860 {
4861         int ret, err;
4862         handle_t *handle;
4863
4864         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4865                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4866         if (IS_ERR(handle))
4867                 return PTR_ERR(handle);
4868         ret = dquot_acquire(dquot);
4869         err = ext4_journal_stop(handle);
4870         if (!ret)
4871                 ret = err;
4872         return ret;
4873 }
4874
4875 static int ext4_release_dquot(struct dquot *dquot)
4876 {
4877         int ret, err;
4878         handle_t *handle;
4879
4880         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4881                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4882         if (IS_ERR(handle)) {
4883                 /* Release dquot anyway to avoid endless cycle in dqput() */
4884                 dquot_release(dquot);
4885                 return PTR_ERR(handle);
4886         }
4887         ret = dquot_release(dquot);
4888         err = ext4_journal_stop(handle);
4889         if (!ret)
4890                 ret = err;
4891         return ret;
4892 }
4893
4894 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4895 {
4896         struct super_block *sb = dquot->dq_sb;
4897         struct ext4_sb_info *sbi = EXT4_SB(sb);
4898
4899         /* Are we journaling quotas? */
4900         if (ext4_has_feature_quota(sb) ||
4901             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4902                 dquot_mark_dquot_dirty(dquot);
4903                 return ext4_write_dquot(dquot);
4904         } else {
4905                 return dquot_mark_dquot_dirty(dquot);
4906         }
4907 }
4908
4909 static int ext4_write_info(struct super_block *sb, int type)
4910 {
4911         int ret, err;
4912         handle_t *handle;
4913
4914         /* Data block + inode block */
4915         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
4916         if (IS_ERR(handle))
4917                 return PTR_ERR(handle);
4918         ret = dquot_commit_info(sb, type);
4919         err = ext4_journal_stop(handle);
4920         if (!ret)
4921                 ret = err;
4922         return ret;
4923 }
4924
4925 /*
4926  * Turn on quotas during mount time - we need to find
4927  * the quota file and such...
4928  */
4929 static int ext4_quota_on_mount(struct super_block *sb, int type)
4930 {
4931         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4932                                         EXT4_SB(sb)->s_jquota_fmt, type);
4933 }
4934
4935 /*
4936  * Standard function to be called on quota_on
4937  */
4938 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
4939                          struct path *path)
4940 {
4941         int err;
4942
4943         if (!test_opt(sb, QUOTA))
4944                 return -EINVAL;
4945
4946         /* Quotafile not on the same filesystem? */
4947         if (path->dentry->d_sb != sb)
4948                 return -EXDEV;
4949         /* Journaling quota? */
4950         if (EXT4_SB(sb)->s_qf_names[type]) {
4951                 /* Quotafile not in fs root? */
4952                 if (path->dentry->d_parent != sb->s_root)
4953                         ext4_msg(sb, KERN_WARNING,
4954                                 "Quota file not on filesystem root. "
4955                                 "Journaled quota will not work");
4956         }
4957
4958         /*
4959          * When we journal data on quota file, we have to flush journal to see
4960          * all updates to the file when we bypass pagecache...
4961          */
4962         if (EXT4_SB(sb)->s_journal &&
4963             ext4_should_journal_data(d_inode(path->dentry))) {
4964                 /*
4965                  * We don't need to lock updates but journal_flush() could
4966                  * otherwise be livelocked...
4967                  */
4968                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
4969                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
4970                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
4971                 if (err)
4972                         return err;
4973         }
4974
4975         return dquot_quota_on(sb, type, format_id, path);
4976 }
4977
4978 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
4979                              unsigned int flags)
4980 {
4981         int err;
4982         struct inode *qf_inode;
4983         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
4984                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
4985                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
4986         };
4987
4988         BUG_ON(!ext4_has_feature_quota(sb));
4989
4990         if (!qf_inums[type])
4991                 return -EPERM;
4992
4993         qf_inode = ext4_iget(sb, qf_inums[type]);
4994         if (IS_ERR(qf_inode)) {
4995                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
4996                 return PTR_ERR(qf_inode);
4997         }
4998
4999         /* Don't account quota for quota files to avoid recursion */
5000         qf_inode->i_flags |= S_NOQUOTA;
5001         err = dquot_enable(qf_inode, type, format_id, flags);
5002         iput(qf_inode);
5003
5004         return err;
5005 }
5006
5007 /* Enable usage tracking for all quota types. */
5008 static int ext4_enable_quotas(struct super_block *sb)
5009 {
5010         int type, err = 0;
5011         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5012                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5013                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5014         };
5015
5016         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5017         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5018                 if (qf_inums[type]) {
5019                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5020                                                 DQUOT_USAGE_ENABLED);
5021                         if (err) {
5022                                 ext4_warning(sb,
5023                                         "Failed to enable quota tracking "
5024                                         "(type=%d, err=%d). Please run "
5025                                         "e2fsck to fix.", type, err);
5026                                 return err;
5027                         }
5028                 }
5029         }
5030         return 0;
5031 }
5032
5033 static int ext4_quota_off(struct super_block *sb, int type)
5034 {
5035         struct inode *inode = sb_dqopt(sb)->files[type];
5036         handle_t *handle;
5037
5038         /* Force all delayed allocation blocks to be allocated.
5039          * Caller already holds s_umount sem */
5040         if (test_opt(sb, DELALLOC))
5041                 sync_filesystem(sb);
5042
5043         if (!inode)
5044                 goto out;
5045
5046         /* Update modification times of quota files when userspace can
5047          * start looking at them */
5048         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5049         if (IS_ERR(handle))
5050                 goto out;
5051         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5052         ext4_mark_inode_dirty(handle, inode);
5053         ext4_journal_stop(handle);
5054
5055 out:
5056         return dquot_quota_off(sb, type);
5057 }
5058
5059 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5060  * acquiring the locks... As quota files are never truncated and quota code
5061  * itself serializes the operations (and no one else should touch the files)
5062  * we don't have to be afraid of races */
5063 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5064                                size_t len, loff_t off)
5065 {
5066         struct inode *inode = sb_dqopt(sb)->files[type];
5067         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5068         int offset = off & (sb->s_blocksize - 1);
5069         int tocopy;
5070         size_t toread;
5071         struct buffer_head *bh;
5072         loff_t i_size = i_size_read(inode);
5073
5074         if (off > i_size)
5075                 return 0;
5076         if (off+len > i_size)
5077                 len = i_size-off;
5078         toread = len;
5079         while (toread > 0) {
5080                 tocopy = sb->s_blocksize - offset < toread ?
5081                                 sb->s_blocksize - offset : toread;
5082                 bh = ext4_bread(NULL, inode, blk, 0);
5083                 if (IS_ERR(bh))
5084                         return PTR_ERR(bh);
5085                 if (!bh)        /* A hole? */
5086                         memset(data, 0, tocopy);
5087                 else
5088                         memcpy(data, bh->b_data+offset, tocopy);
5089                 brelse(bh);
5090                 offset = 0;
5091                 toread -= tocopy;
5092                 data += tocopy;
5093                 blk++;
5094         }
5095         return len;
5096 }
5097
5098 /* Write to quotafile (we know the transaction is already started and has
5099  * enough credits) */
5100 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5101                                 const char *data, size_t len, loff_t off)
5102 {
5103         struct inode *inode = sb_dqopt(sb)->files[type];
5104         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5105         int err, offset = off & (sb->s_blocksize - 1);
5106         int retries = 0;
5107         struct buffer_head *bh;
5108         handle_t *handle = journal_current_handle();
5109
5110         if (EXT4_SB(sb)->s_journal && !handle) {
5111                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5112                         " cancelled because transaction is not started",
5113                         (unsigned long long)off, (unsigned long long)len);
5114                 return -EIO;
5115         }
5116         /*
5117          * Since we account only one data block in transaction credits,
5118          * then it is impossible to cross a block boundary.
5119          */
5120         if (sb->s_blocksize - offset < len) {
5121                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5122                         " cancelled because not block aligned",
5123                         (unsigned long long)off, (unsigned long long)len);
5124                 return -EIO;
5125         }
5126
5127         do {
5128                 bh = ext4_bread(handle, inode, blk,
5129                                 EXT4_GET_BLOCKS_CREATE |
5130                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5131         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5132                  ext4_should_retry_alloc(inode->i_sb, &retries));
5133         if (IS_ERR(bh))
5134                 return PTR_ERR(bh);
5135         if (!bh)
5136                 goto out;
5137         BUFFER_TRACE(bh, "get write access");
5138         err = ext4_journal_get_write_access(handle, bh);
5139         if (err) {
5140                 brelse(bh);
5141                 return err;
5142         }
5143         lock_buffer(bh);
5144         memcpy(bh->b_data+offset, data, len);
5145         flush_dcache_page(bh->b_page);
5146         unlock_buffer(bh);
5147         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5148         brelse(bh);
5149 out:
5150         if (inode->i_size < off + len) {
5151                 i_size_write(inode, off + len);
5152                 EXT4_I(inode)->i_disksize = inode->i_size;
5153                 ext4_mark_inode_dirty(handle, inode);
5154         }
5155         return len;
5156 }
5157
5158 #endif
5159
5160 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5161                        const char *dev_name, void *data)
5162 {
5163         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5164 }
5165
5166 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5167 static inline void register_as_ext2(void)
5168 {
5169         int err = register_filesystem(&ext2_fs_type);
5170         if (err)
5171                 printk(KERN_WARNING
5172                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5173 }
5174
5175 static inline void unregister_as_ext2(void)
5176 {
5177         unregister_filesystem(&ext2_fs_type);
5178 }
5179
5180 static inline int ext2_feature_set_ok(struct super_block *sb)
5181 {
5182         if (ext4_has_unknown_ext2_incompat_features(sb))
5183                 return 0;
5184         if (sb->s_flags & MS_RDONLY)
5185                 return 1;
5186         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5187                 return 0;
5188         return 1;
5189 }
5190 #else
5191 static inline void register_as_ext2(void) { }
5192 static inline void unregister_as_ext2(void) { }
5193 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5194 #endif
5195
5196 static inline void register_as_ext3(void)
5197 {
5198         int err = register_filesystem(&ext3_fs_type);
5199         if (err)
5200                 printk(KERN_WARNING
5201                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5202 }
5203
5204 static inline void unregister_as_ext3(void)
5205 {
5206         unregister_filesystem(&ext3_fs_type);
5207 }
5208
5209 static inline int ext3_feature_set_ok(struct super_block *sb)
5210 {
5211         if (ext4_has_unknown_ext3_incompat_features(sb))
5212                 return 0;
5213         if (!ext4_has_feature_journal(sb))
5214                 return 0;
5215         if (sb->s_flags & MS_RDONLY)
5216                 return 1;
5217         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5218                 return 0;
5219         return 1;
5220 }
5221
5222 static struct file_system_type ext4_fs_type = {
5223         .owner          = THIS_MODULE,
5224         .name           = "ext4",
5225         .mount          = ext4_mount,
5226         .kill_sb        = kill_block_super,
5227         .fs_flags       = FS_REQUIRES_DEV,
5228 };
5229 MODULE_ALIAS_FS("ext4");
5230
5231 /* Shared across all ext4 file systems */
5232 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5233 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5234
5235 static int __init ext4_init_fs(void)
5236 {
5237         int i, err;
5238
5239         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5240         ext4_li_info = NULL;
5241         mutex_init(&ext4_li_mtx);
5242
5243         /* Build-time check for flags consistency */
5244         ext4_check_flag_values();
5245
5246         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5247                 mutex_init(&ext4__aio_mutex[i]);
5248                 init_waitqueue_head(&ext4__ioend_wq[i]);
5249         }
5250
5251         err = ext4_init_es();
5252         if (err)
5253                 return err;
5254
5255         err = ext4_init_pageio();
5256         if (err)
5257                 goto out5;
5258
5259         err = ext4_init_system_zone();
5260         if (err)
5261                 goto out4;
5262
5263         err = ext4_init_sysfs();
5264         if (err)
5265                 goto out3;
5266
5267         err = ext4_init_mballoc();
5268         if (err)
5269                 goto out2;
5270         else
5271                 ext4_mballoc_ready = 1;
5272         err = init_inodecache();
5273         if (err)
5274                 goto out1;
5275         register_as_ext3();
5276         register_as_ext2();
5277         err = register_filesystem(&ext4_fs_type);
5278         if (err)
5279                 goto out;
5280
5281         return 0;
5282 out:
5283         unregister_as_ext2();
5284         unregister_as_ext3();
5285         destroy_inodecache();
5286 out1:
5287         ext4_mballoc_ready = 0;
5288         ext4_exit_mballoc();
5289 out2:
5290         ext4_exit_sysfs();
5291 out3:
5292         ext4_exit_system_zone();
5293 out4:
5294         ext4_exit_pageio();
5295 out5:
5296         ext4_exit_es();
5297
5298         return err;
5299 }
5300
5301 static void __exit ext4_exit_fs(void)
5302 {
5303         ext4_exit_crypto();
5304         ext4_destroy_lazyinit_thread();
5305         unregister_as_ext2();
5306         unregister_as_ext3();
5307         unregister_filesystem(&ext4_fs_type);
5308         destroy_inodecache();
5309         ext4_exit_mballoc();
5310         ext4_exit_sysfs();
5311         ext4_exit_system_zone();
5312         ext4_exit_pageio();
5313         ext4_exit_es();
5314 }
5315
5316 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5317 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5318 MODULE_LICENSE("GPL");
5319 module_init(ext4_init_fs)
5320 module_exit(ext4_exit_fs)