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 aborted = 0;
 797         int i, err;
 798
 799         ext4_unregister_li_request(sb);
 800         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 801
 802         flush_workqueue(sbi->rsv_conversion_wq);
 803         destroy_workqueue(sbi->rsv_conversion_wq);
 804
 805         if (sbi->s_journal) {
 806                 aborted = is_journal_aborted(sbi->s_journal);
 807                 err = jbd2_journal_destroy(sbi->s_journal);
 808                 sbi->s_journal = NULL;
 809                 if ((err < 0) && !aborted)
 810                         ext4_abort(sb, "Couldn't clean up the journal");
 811         }
 812
 813         ext4_unregister_sysfs(sb);
 814         ext4_es_unregister_shrinker(sbi);
 815         del_timer_sync(&sbi->s_err_report);
 816         ext4_release_system_zone(sb);
 817         ext4_mb_release(sb);
 818         ext4_ext_release(sb);
 819         ext4_xattr_put_super(sb);
 820
 821         if (!(sb->s_flags & MS_RDONLY) && !aborted) {
 822                 ext4_clear_feature_journal_needs_recovery(sb);
 823                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 824         }
 825         if (!(sb->s_flags & MS_RDONLY))
 826                 ext4_commit_super(sb, 1);
 827
 828         for (i = 0; i < sbi->s_gdb_count; i++)
 829                 brelse(sbi->s_group_desc[i]);
 830         kvfree(sbi->s_group_desc);
 831         kvfree(sbi->s_flex_groups);
 832         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 833         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 834         percpu_counter_destroy(&sbi->s_dirs_counter);
 835         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 836         brelse(sbi->s_sbh);
 837 #ifdef CONFIG_QUOTA
 838         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 839                 kfree(sbi->s_qf_names[i]);
 840 #endif
 841
 842         /* Debugging code just in case the in-memory inode orphan list
 843          * isn't empty.  The on-disk one can be non-empty if we've
 844          * detected an error and taken the fs readonly, but the
 845          * in-memory list had better be clean by this point. */
 846         if (!list_empty(&sbi->s_orphan))
 847                 dump_orphan_list(sb, sbi);
 848         J_ASSERT(list_empty(&sbi->s_orphan));
 849
 850         sync_blockdev(sb->s_bdev);
 851         invalidate_bdev(sb->s_bdev);
 852         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 853                 /*
 854                  * Invalidate the journal device's buffers.  We don't want them
 855                  * floating about in memory - the physical journal device may
 856                  * hotswapped, and it breaks the `ro-after' testing code.
 857                  */
 858                 sync_blockdev(sbi->journal_bdev);
 859                 invalidate_bdev(sbi->journal_bdev);
 860                 ext4_blkdev_remove(sbi);
 861         }
 862         if (sbi->s_mb_cache) {
 863                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 864                 sbi->s_mb_cache = NULL;
 865         }
 866         if (sbi->s_mmp_tsk)
 867                 kthread_stop(sbi->s_mmp_tsk);
 868         sb->s_fs_info = NULL;
 869         /*
 870          * Now that we are completely done shutting down the
 871          * superblock, we need to actually destroy the kobject.
 872          */
 873         kobject_put(&sbi->s_kobj);
 874         wait_for_completion(&sbi->s_kobj_unregister);
 875         if (sbi->s_chksum_driver)
 876                 crypto_free_shash(sbi->s_chksum_driver);
 877         kfree(sbi->s_blockgroup_lock);
 878         kfree(sbi);
 879 }
 880
 881 static struct kmem_cache *ext4_inode_cachep;
 882
 883 /*
 884  * Called inside transaction, so use GFP_NOFS
 885  */
 886 static struct inode *ext4_alloc_inode(struct super_block *sb)
 887 {
 888         struct ext4_inode_info *ei;
 889
 890         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 891         if (!ei)
 892                 return NULL;
 893
 894         ei->vfs_inode.i_version = 1;
 895         spin_lock_init(&ei->i_raw_lock);
 896         INIT_LIST_HEAD(&ei->i_prealloc_list);
 897         spin_lock_init(&ei->i_prealloc_lock);
 898         ext4_es_init_tree(&ei->i_es_tree);
 899         rwlock_init(&ei->i_es_lock);
 900         INIT_LIST_HEAD(&ei->i_es_list);
 901         ei->i_es_all_nr = 0;
 902         ei->i_es_shk_nr = 0;
 903         ei->i_es_shrink_lblk = 0;
 904         ei->i_reserved_data_blocks = 0;
 905         ei->i_reserved_meta_blocks = 0;
 906         ei->i_allocated_meta_blocks = 0;
 907         ei->i_da_metadata_calc_len = 0;
 908         ei->i_da_metadata_calc_last_lblock = 0;
 909         spin_lock_init(&(ei->i_block_reservation_lock));
 910 #ifdef CONFIG_QUOTA
 911         ei->i_reserved_quota = 0;
 912         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 913 #endif
 914         ei->jinode = NULL;
 915         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 916         spin_lock_init(&ei->i_completed_io_lock);
 917         ei->i_sync_tid = 0;
 918         ei->i_datasync_tid = 0;
 919         atomic_set(&ei->i_ioend_count, 0);
 920         atomic_set(&ei->i_unwritten, 0);
 921         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 922 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 923         ei->i_crypt_info = NULL;
 924 #endif
 925         return &ei->vfs_inode;
 926 }
 927
 928 static int ext4_drop_inode(struct inode *inode)
 929 {
 930         int drop = generic_drop_inode(inode);
 931
 932         trace_ext4_drop_inode(inode, drop);
 933         return drop;
 934 }
 935
 936 static void ext4_i_callback(struct rcu_head *head)
 937 {
 938         struct inode *inode = container_of(head, struct inode, i_rcu);
 939         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 940 }
 941
 942 static void ext4_destroy_inode(struct inode *inode)
 943 {
 944         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 945                 ext4_msg(inode->i_sb, KERN_ERR,
 946                          "Inode %lu (%p): orphan list check failed!",
 947                          inode->i_ino, EXT4_I(inode));
 948                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 949                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 950                                 true);
 951                 dump_stack();
 952         }
 953         call_rcu(&inode->i_rcu, ext4_i_callback);
 954 }
 955
 956 static void init_once(void *foo)
 957 {
 958         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 959
 960         INIT_LIST_HEAD(&ei->i_orphan);
 961         init_rwsem(&ei->xattr_sem);
 962         init_rwsem(&ei->i_data_sem);
 963         init_rwsem(&ei->i_mmap_sem);
 964         inode_init_once(&ei->vfs_inode);
 965 }
 966
 967 static int __init init_inodecache(void)
 968 {
 969         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 970                                              sizeof(struct ext4_inode_info),
 971                                              0, (SLAB_RECLAIM_ACCOUNT|
 972                                                 SLAB_MEM_SPREAD),
 973                                              init_once);
 974         if (ext4_inode_cachep == NULL)
 975                 return -ENOMEM;
 976         return 0;
 977 }
 978
 979 static void destroy_inodecache(void)
 980 {
 981         /*
 982          * Make sure all delayed rcu free inodes are flushed before we
 983          * destroy cache.
 984          */
 985         rcu_barrier();
 986         kmem_cache_destroy(ext4_inode_cachep);
 987 }
 988
 989 void ext4_clear_inode(struct inode *inode)
 990 {
 991         invalidate_inode_buffers(inode);
 992         clear_inode(inode);
 993         dquot_drop(inode);
 994         ext4_discard_preallocations(inode);
 995         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 996         if (EXT4_I(inode)->jinode) {
 997                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 998                                                EXT4_I(inode)->jinode);
 999                 jbd2_free_inode(EXT4_I(inode)->jinode);
1000                 EXT4_I(inode)->jinode = NULL;
1001         }
1002 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1003         if (EXT4_I(inode)->i_crypt_info)
1004                 ext4_free_encryption_info(inode, EXT4_I(inode)->i_crypt_info);
1005 #endif
1006 }
1007
1008 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1009                                         u64 ino, u32 generation)
1010 {
1011         struct inode *inode;
1012
1013         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1014                 return ERR_PTR(-ESTALE);
1015         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1016                 return ERR_PTR(-ESTALE);
1017
1018         /* iget isn't really right if the inode is currently unallocated!!
1019          *
1020          * ext4_read_inode will return a bad_inode if the inode had been
1021          * deleted, so we should be safe.
1022          *
1023          * Currently we don't know the generation for parent directory, so
1024          * a generation of 0 means "accept any"
1025          */
1026         inode = ext4_iget_normal(sb, ino);
1027         if (IS_ERR(inode))
1028                 return ERR_CAST(inode);
1029         if (generation && inode->i_generation != generation) {
1030                 iput(inode);
1031                 return ERR_PTR(-ESTALE);
1032         }
1033
1034         return inode;
1035 }
1036
1037 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1038                                         int fh_len, int fh_type)
1039 {
1040         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1041                                     ext4_nfs_get_inode);
1042 }
1043
1044 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1045                                         int fh_len, int fh_type)
1046 {
1047         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1048                                     ext4_nfs_get_inode);
1049 }
1050
1051 /*
1052  * Try to release metadata pages (indirect blocks, directories) which are
1053  * mapped via the block device.  Since these pages could have journal heads
1054  * which would prevent try_to_free_buffers() from freeing them, we must use
1055  * jbd2 layer's try_to_free_buffers() function to release them.
1056  */
1057 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1058                                  gfp_t wait)
1059 {
1060         journal_t *journal = EXT4_SB(sb)->s_journal;
1061
1062         WARN_ON(PageChecked(page));
1063         if (!page_has_buffers(page))
1064                 return 0;
1065         if (journal)
1066                 return jbd2_journal_try_to_free_buffers(journal, page,
1067                                                 wait & ~__GFP_DIRECT_RECLAIM);
1068         return try_to_free_buffers(page);
1069 }
1070
1071 #ifdef CONFIG_QUOTA
1072 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1073 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1074
1075 static int ext4_write_dquot(struct dquot *dquot);
1076 static int ext4_acquire_dquot(struct dquot *dquot);
1077 static int ext4_release_dquot(struct dquot *dquot);
1078 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1079 static int ext4_write_info(struct super_block *sb, int type);
1080 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1081                          struct path *path);
1082 static int ext4_quota_off(struct super_block *sb, int type);
1083 static int ext4_quota_on_mount(struct super_block *sb, int type);
1084 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1085                                size_t len, loff_t off);
1086 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1087                                 const char *data, size_t len, loff_t off);
1088 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1089                              unsigned int flags);
1090 static int ext4_enable_quotas(struct super_block *sb);
1091
1092 static struct dquot **ext4_get_dquots(struct inode *inode)
1093 {
1094         return EXT4_I(inode)->i_dquot;
1095 }
1096
1097 static const struct dquot_operations ext4_quota_operations = {
1098         .get_reserved_space = ext4_get_reserved_space,
1099         .write_dquot    = ext4_write_dquot,
1100         .acquire_dquot  = ext4_acquire_dquot,
1101         .release_dquot  = ext4_release_dquot,
1102         .mark_dirty     = ext4_mark_dquot_dirty,
1103         .write_info     = ext4_write_info,
1104         .alloc_dquot    = dquot_alloc,
1105         .destroy_dquot  = dquot_destroy,
1106 };
1107
1108 static const struct quotactl_ops ext4_qctl_operations = {
1109         .quota_on       = ext4_quota_on,
1110         .quota_off      = ext4_quota_off,
1111         .quota_sync     = dquot_quota_sync,
1112         .get_state      = dquot_get_state,
1113         .set_info       = dquot_set_dqinfo,
1114         .get_dqblk      = dquot_get_dqblk,
1115         .set_dqblk      = dquot_set_dqblk
1116 };
1117 #endif
1118
1119 static const struct super_operations ext4_sops = {
1120         .alloc_inode    = ext4_alloc_inode,
1121         .destroy_inode  = ext4_destroy_inode,
1122         .write_inode    = ext4_write_inode,
1123         .dirty_inode    = ext4_dirty_inode,
1124         .drop_inode     = ext4_drop_inode,
1125         .evict_inode    = ext4_evict_inode,
1126         .put_super      = ext4_put_super,
1127         .sync_fs        = ext4_sync_fs,
1128         .freeze_fs      = ext4_freeze,
1129         .unfreeze_fs    = ext4_unfreeze,
1130         .statfs         = ext4_statfs,
1131         .remount_fs     = ext4_remount,
1132         .show_options   = ext4_show_options,
1133 #ifdef CONFIG_QUOTA
1134         .quota_read     = ext4_quota_read,
1135         .quota_write    = ext4_quota_write,
1136         .get_dquots     = ext4_get_dquots,
1137 #endif
1138         .bdev_try_to_free_page = bdev_try_to_free_page,
1139 };
1140
1141 static const struct export_operations ext4_export_ops = {
1142         .fh_to_dentry = ext4_fh_to_dentry,
1143         .fh_to_parent = ext4_fh_to_parent,
1144         .get_parent = ext4_get_parent,
1145 };
1146
1147 enum {
1148         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1149         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1150         Opt_nouid32, Opt_debug, Opt_removed,
1151         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1152         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1153         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1154         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1155         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1156         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1157         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1158         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1159         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1160         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1161         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1162         Opt_lazytime, Opt_nolazytime,
1163         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1164         Opt_inode_readahead_blks, Opt_journal_ioprio,
1165         Opt_dioread_nolock, Opt_dioread_lock,
1166         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1167         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1168 };
1169
1170 static const match_table_t tokens = {
1171         {Opt_bsd_df, "bsddf"},
1172         {Opt_minix_df, "minixdf"},
1173         {Opt_grpid, "grpid"},
1174         {Opt_grpid, "bsdgroups"},
1175         {Opt_nogrpid, "nogrpid"},
1176         {Opt_nogrpid, "sysvgroups"},
1177         {Opt_resgid, "resgid=%u"},
1178         {Opt_resuid, "resuid=%u"},
1179         {Opt_sb, "sb=%u"},
1180         {Opt_err_cont, "errors=continue"},
1181         {Opt_err_panic, "errors=panic"},
1182         {Opt_err_ro, "errors=remount-ro"},
1183         {Opt_nouid32, "nouid32"},
1184         {Opt_debug, "debug"},
1185         {Opt_removed, "oldalloc"},
1186         {Opt_removed, "orlov"},
1187         {Opt_user_xattr, "user_xattr"},
1188         {Opt_nouser_xattr, "nouser_xattr"},
1189         {Opt_acl, "acl"},
1190         {Opt_noacl, "noacl"},
1191         {Opt_noload, "norecovery"},
1192         {Opt_noload, "noload"},
1193         {Opt_removed, "nobh"},
1194         {Opt_removed, "bh"},
1195         {Opt_commit, "commit=%u"},
1196         {Opt_min_batch_time, "min_batch_time=%u"},
1197         {Opt_max_batch_time, "max_batch_time=%u"},
1198         {Opt_journal_dev, "journal_dev=%u"},
1199         {Opt_journal_path, "journal_path=%s"},
1200         {Opt_journal_checksum, "journal_checksum"},
1201         {Opt_nojournal_checksum, "nojournal_checksum"},
1202         {Opt_journal_async_commit, "journal_async_commit"},
1203         {Opt_abort, "abort"},
1204         {Opt_data_journal, "data=journal"},
1205         {Opt_data_ordered, "data=ordered"},
1206         {Opt_data_writeback, "data=writeback"},
1207         {Opt_data_err_abort, "data_err=abort"},
1208         {Opt_data_err_ignore, "data_err=ignore"},
1209         {Opt_offusrjquota, "usrjquota="},
1210         {Opt_usrjquota, "usrjquota=%s"},
1211         {Opt_offgrpjquota, "grpjquota="},
1212         {Opt_grpjquota, "grpjquota=%s"},
1213         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1214         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1215         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1216         {Opt_grpquota, "grpquota"},
1217         {Opt_noquota, "noquota"},
1218         {Opt_quota, "quota"},
1219         {Opt_usrquota, "usrquota"},
1220         {Opt_barrier, "barrier=%u"},
1221         {Opt_barrier, "barrier"},
1222         {Opt_nobarrier, "nobarrier"},
1223         {Opt_i_version, "i_version"},
1224         {Opt_dax, "dax"},
1225         {Opt_stripe, "stripe=%u"},
1226         {Opt_delalloc, "delalloc"},
1227         {Opt_lazytime, "lazytime"},
1228         {Opt_nolazytime, "nolazytime"},
1229         {Opt_nodelalloc, "nodelalloc"},
1230         {Opt_removed, "mblk_io_submit"},
1231         {Opt_removed, "nomblk_io_submit"},
1232         {Opt_block_validity, "block_validity"},
1233         {Opt_noblock_validity, "noblock_validity"},
1234         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1235         {Opt_journal_ioprio, "journal_ioprio=%u"},
1236         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1237         {Opt_auto_da_alloc, "auto_da_alloc"},
1238         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1239         {Opt_dioread_nolock, "dioread_nolock"},
1240         {Opt_dioread_lock, "dioread_lock"},
1241         {Opt_discard, "discard"},
1242         {Opt_nodiscard, "nodiscard"},
1243         {Opt_init_itable, "init_itable=%u"},
1244         {Opt_init_itable, "init_itable"},
1245         {Opt_noinit_itable, "noinit_itable"},
1246         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1247         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1248         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1249         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1250         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1251         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1252         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1253         {Opt_err, NULL},
1254 };
1255
1256 static ext4_fsblk_t get_sb_block(void **data)
1257 {
1258         ext4_fsblk_t    sb_block;
1259         char            *options = (char *) *data;
1260
1261         if (!options || strncmp(options, "sb=", 3) != 0)
1262                 return 1;       /* Default location */
1263
1264         options += 3;
1265         /* TODO: use simple_strtoll with >32bit ext4 */
1266         sb_block = simple_strtoul(options, &options, 0);
1267         if (*options && *options != ',') {
1268                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1269                        (char *) *data);
1270                 return 1;
1271         }
1272         if (*options == ',')
1273                 options++;
1274         *data = (void *) options;
1275
1276         return sb_block;
1277 }
1278
1279 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1280 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1281         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1282
1283 #ifdef CONFIG_QUOTA
1284 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1285 {
1286         struct ext4_sb_info *sbi = EXT4_SB(sb);
1287         char *qname;
1288         int ret = -1;
1289
1290         if (sb_any_quota_loaded(sb) &&
1291                 !sbi->s_qf_names[qtype]) {
1292                 ext4_msg(sb, KERN_ERR,
1293                         "Cannot change journaled "
1294                         "quota options when quota turned on");
1295                 return -1;
1296         }
1297         if (ext4_has_feature_quota(sb)) {
1298                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1299                          "ignored when QUOTA feature is enabled");
1300                 return 1;
1301         }
1302         qname = match_strdup(args);
1303         if (!qname) {
1304                 ext4_msg(sb, KERN_ERR,
1305                         "Not enough memory for storing quotafile name");
1306                 return -1;
1307         }
1308         if (sbi->s_qf_names[qtype]) {
1309                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1310                         ret = 1;
1311                 else
1312                         ext4_msg(sb, KERN_ERR,
1313                                  "%s quota file already specified",
1314                                  QTYPE2NAME(qtype));
1315                 goto errout;
1316         }
1317         if (strchr(qname, '/')) {
1318                 ext4_msg(sb, KERN_ERR,
1319                         "quotafile must be on filesystem root");
1320                 goto errout;
1321         }
1322         sbi->s_qf_names[qtype] = qname;
1323         set_opt(sb, QUOTA);
1324         return 1;
1325 errout:
1326         kfree(qname);
1327         return ret;
1328 }
1329
1330 static int clear_qf_name(struct super_block *sb, int qtype)
1331 {
1332
1333         struct ext4_sb_info *sbi = EXT4_SB(sb);
1334
1335         if (sb_any_quota_loaded(sb) &&
1336                 sbi->s_qf_names[qtype]) {
1337                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1338                         " when quota turned on");
1339                 return -1;
1340         }
1341         kfree(sbi->s_qf_names[qtype]);
1342         sbi->s_qf_names[qtype] = NULL;
1343         return 1;
1344 }
1345 #endif
1346
1347 #define MOPT_SET        0x0001
1348 #define MOPT_CLEAR      0x0002
1349 #define MOPT_NOSUPPORT  0x0004
1350 #define MOPT_EXPLICIT   0x0008
1351 #define MOPT_CLEAR_ERR  0x0010
1352 #define MOPT_GTE0       0x0020
1353 #ifdef CONFIG_QUOTA
1354 #define MOPT_Q          0
1355 #define MOPT_QFMT       0x0040
1356 #else
1357 #define MOPT_Q          MOPT_NOSUPPORT
1358 #define MOPT_QFMT       MOPT_NOSUPPORT
1359 #endif
1360 #define MOPT_DATAJ      0x0080
1361 #define MOPT_NO_EXT2    0x0100
1362 #define MOPT_NO_EXT3    0x0200
1363 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1364 #define MOPT_STRING     0x0400
1365
1366 static const struct mount_opts {
1367         int     token;
1368         int     mount_opt;
1369         int     flags;
1370 } ext4_mount_opts[] = {
1371         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1372         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1373         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1374         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1375         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1376         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1377         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1378          MOPT_EXT4_ONLY | MOPT_SET},
1379         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1380          MOPT_EXT4_ONLY | MOPT_CLEAR},
1381         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1382         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1383         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1384          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1385         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1386          MOPT_EXT4_ONLY | MOPT_CLEAR},
1387         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1388          MOPT_EXT4_ONLY | MOPT_CLEAR},
1389         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1390          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1391         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1392                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1393          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1394         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1395         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1396         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1397         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1398         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1399          MOPT_NO_EXT2 | MOPT_SET},
1400         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1401          MOPT_NO_EXT2 | MOPT_CLEAR},
1402         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1403         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1404         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1405         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1406         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1407         {Opt_commit, 0, MOPT_GTE0},
1408         {Opt_max_batch_time, 0, MOPT_GTE0},
1409         {Opt_min_batch_time, 0, MOPT_GTE0},
1410         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1411         {Opt_init_itable, 0, MOPT_GTE0},
1412         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1413         {Opt_stripe, 0, MOPT_GTE0},
1414         {Opt_resuid, 0, MOPT_GTE0},
1415         {Opt_resgid, 0, MOPT_GTE0},
1416         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1417         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1418         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1419         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1420         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1421         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1422          MOPT_NO_EXT2 | MOPT_DATAJ},
1423         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1424         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1425 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1426         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1427         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1428 #else
1429         {Opt_acl, 0, MOPT_NOSUPPORT},
1430         {Opt_noacl, 0, MOPT_NOSUPPORT},
1431 #endif
1432         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1433         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1434         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1435         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1436                                                         MOPT_SET | MOPT_Q},
1437         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1438                                                         MOPT_SET | MOPT_Q},
1439         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1440                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1441         {Opt_usrjquota, 0, MOPT_Q},
1442         {Opt_grpjquota, 0, MOPT_Q},
1443         {Opt_offusrjquota, 0, MOPT_Q},
1444         {Opt_offgrpjquota, 0, MOPT_Q},
1445         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1446         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1447         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1448         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1449         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1450         {Opt_err, 0, 0}
1451 };
1452
1453 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1454                             substring_t *args, unsigned long *journal_devnum,
1455                             unsigned int *journal_ioprio, int is_remount)
1456 {
1457         struct ext4_sb_info *sbi = EXT4_SB(sb);
1458         const struct mount_opts *m;
1459         kuid_t uid;
1460         kgid_t gid;
1461         int arg = 0;
1462
1463 #ifdef CONFIG_QUOTA
1464         if (token == Opt_usrjquota)
1465                 return set_qf_name(sb, USRQUOTA, &args[0]);
1466         else if (token == Opt_grpjquota)
1467                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1468         else if (token == Opt_offusrjquota)
1469                 return clear_qf_name(sb, USRQUOTA);
1470         else if (token == Opt_offgrpjquota)
1471                 return clear_qf_name(sb, GRPQUOTA);
1472 #endif
1473         switch (token) {
1474         case Opt_noacl:
1475         case Opt_nouser_xattr:
1476                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1477                 break;
1478         case Opt_sb:
1479                 return 1;       /* handled by get_sb_block() */
1480         case Opt_removed:
1481                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1482                 return 1;
1483         case Opt_abort:
1484                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1485                 return 1;
1486         case Opt_i_version:
1487                 sb->s_flags |= MS_I_VERSION;
1488                 return 1;
1489         case Opt_lazytime:
1490                 sb->s_flags |= MS_LAZYTIME;
1491                 return 1;
1492         case Opt_nolazytime:
1493                 sb->s_flags &= ~MS_LAZYTIME;
1494                 return 1;
1495         }
1496
1497         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1498                 if (token == m->token)
1499                         break;
1500
1501         if (m->token == Opt_err) {
1502                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1503                          "or missing value", opt);
1504                 return -1;
1505         }
1506
1507         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1508                 ext4_msg(sb, KERN_ERR,
1509                          "Mount option \"%s\" incompatible with ext2", opt);
1510                 return -1;
1511         }
1512         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1513                 ext4_msg(sb, KERN_ERR,
1514                          "Mount option \"%s\" incompatible with ext3", opt);
1515                 return -1;
1516         }
1517
1518         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1519                 return -1;
1520         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1521                 return -1;
1522         if (m->flags & MOPT_EXPLICIT) {
1523                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1524                         set_opt2(sb, EXPLICIT_DELALLOC);
1525                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1526                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1527                 } else
1528                         return -1;
1529         }
1530         if (m->flags & MOPT_CLEAR_ERR)
1531                 clear_opt(sb, ERRORS_MASK);
1532         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1533                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1534                          "options when quota turned on");
1535                 return -1;
1536         }
1537
1538         if (m->flags & MOPT_NOSUPPORT) {
1539                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1540         } else if (token == Opt_commit) {
1541                 if (arg == 0)
1542                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1543                 sbi->s_commit_interval = HZ * arg;
1544         } else if (token == Opt_max_batch_time) {
1545                 sbi->s_max_batch_time = arg;
1546         } else if (token == Opt_min_batch_time) {
1547                 sbi->s_min_batch_time = arg;
1548         } else if (token == Opt_inode_readahead_blks) {
1549                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1550                         ext4_msg(sb, KERN_ERR,
1551                                  "EXT4-fs: inode_readahead_blks must be "
1552                                  "0 or a power of 2 smaller than 2^31");
1553                         return -1;
1554                 }
1555                 sbi->s_inode_readahead_blks = arg;
1556         } else if (token == Opt_init_itable) {
1557                 set_opt(sb, INIT_INODE_TABLE);
1558                 if (!args->from)
1559                         arg = EXT4_DEF_LI_WAIT_MULT;
1560                 sbi->s_li_wait_mult = arg;
1561         } else if (token == Opt_max_dir_size_kb) {
1562                 sbi->s_max_dir_size_kb = arg;
1563         } else if (token == Opt_stripe) {
1564                 sbi->s_stripe = arg;
1565         } else if (token == Opt_resuid) {
1566                 uid = make_kuid(current_user_ns(), arg);
1567                 if (!uid_valid(uid)) {
1568                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1569                         return -1;
1570                 }
1571                 sbi->s_resuid = uid;
1572         } else if (token == Opt_resgid) {
1573                 gid = make_kgid(current_user_ns(), arg);
1574                 if (!gid_valid(gid)) {
1575                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1576                         return -1;
1577                 }
1578                 sbi->s_resgid = gid;
1579         } else if (token == Opt_journal_dev) {
1580                 if (is_remount) {
1581                         ext4_msg(sb, KERN_ERR,
1582                                  "Cannot specify journal on remount");
1583                         return -1;
1584                 }
1585                 *journal_devnum = arg;
1586         } else if (token == Opt_journal_path) {
1587                 char *journal_path;
1588                 struct inode *journal_inode;
1589                 struct path path;
1590                 int error;
1591
1592                 if (is_remount) {
1593                         ext4_msg(sb, KERN_ERR,
1594                                  "Cannot specify journal on remount");
1595                         return -1;
1596                 }
1597                 journal_path = match_strdup(&args[0]);
1598                 if (!journal_path) {
1599                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1600                                 "journal device string");
1601                         return -1;
1602                 }
1603
1604                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1605                 if (error) {
1606                         ext4_msg(sb, KERN_ERR, "error: could not find "
1607                                 "journal device path: error %d", error);
1608                         kfree(journal_path);
1609                         return -1;
1610                 }
1611
1612                 journal_inode = d_inode(path.dentry);
1613                 if (!S_ISBLK(journal_inode->i_mode)) {
1614                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1615                                 "is not a block device", journal_path);
1616                         path_put(&path);
1617                         kfree(journal_path);
1618                         return -1;
1619                 }
1620
1621                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1622                 path_put(&path);
1623                 kfree(journal_path);
1624         } else if (token == Opt_journal_ioprio) {
1625                 if (arg > 7) {
1626                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1627                                  " (must be 0-7)");
1628                         return -1;
1629                 }
1630                 *journal_ioprio =
1631                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1632         } else if (token == Opt_test_dummy_encryption) {
1633 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1634                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1635                 ext4_msg(sb, KERN_WARNING,
1636                          "Test dummy encryption mode enabled");
1637 #else
1638                 ext4_msg(sb, KERN_WARNING,
1639                          "Test dummy encryption mount option ignored");
1640 #endif
1641         } else if (m->flags & MOPT_DATAJ) {
1642                 if (is_remount) {
1643                         if (!sbi->s_journal)
1644                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1645                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1646                                 ext4_msg(sb, KERN_ERR,
1647                                          "Cannot change data mode on remount");
1648                                 return -1;
1649                         }
1650                 } else {
1651                         clear_opt(sb, DATA_FLAGS);
1652                         sbi->s_mount_opt |= m->mount_opt;
1653                 }
1654 #ifdef CONFIG_QUOTA
1655         } else if (m->flags & MOPT_QFMT) {
1656                 if (sb_any_quota_loaded(sb) &&
1657                     sbi->s_jquota_fmt != m->mount_opt) {
1658                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1659                                  "quota options when quota turned on");
1660                         return -1;
1661                 }
1662                 if (ext4_has_feature_quota(sb)) {
1663                         ext4_msg(sb, KERN_INFO,
1664                                  "Quota format mount options ignored "
1665                                  "when QUOTA feature is enabled");
1666                         return 1;
1667                 }
1668                 sbi->s_jquota_fmt = m->mount_opt;
1669 #endif
1670         } else if (token == Opt_dax) {
1671 #ifdef CONFIG_FS_DAX
1672                 ext4_msg(sb, KERN_WARNING,
1673                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1674                         sbi->s_mount_opt |= m->mount_opt;
1675 #else
1676                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1677                 return -1;
1678 #endif
1679         } else {
1680                 if (!args->from)
1681                         arg = 1;
1682                 if (m->flags & MOPT_CLEAR)
1683                         arg = !arg;
1684                 else if (unlikely(!(m->flags & MOPT_SET))) {
1685                         ext4_msg(sb, KERN_WARNING,
1686                                  "buggy handling of option %s", opt);
1687                         WARN_ON(1);
1688                         return -1;
1689                 }
1690                 if (arg != 0)
1691                         sbi->s_mount_opt |= m->mount_opt;
1692                 else
1693                         sbi->s_mount_opt &= ~m->mount_opt;
1694         }
1695         return 1;
1696 }
1697
1698 static int parse_options(char *options, struct super_block *sb,
1699                          unsigned long *journal_devnum,
1700                          unsigned int *journal_ioprio,
1701                          int is_remount)
1702 {
1703         struct ext4_sb_info *sbi = EXT4_SB(sb);
1704         char *p;
1705         substring_t args[MAX_OPT_ARGS];
1706         int token;
1707
1708         if (!options)
1709                 return 1;
1710
1711         while ((p = strsep(&options, ",")) != NULL) {
1712                 if (!*p)
1713                         continue;
1714                 /*
1715                  * Initialize args struct so we know whether arg was
1716                  * found; some options take optional arguments.
1717                  */
1718                 args[0].to = args[0].from = NULL;
1719                 token = match_token(p, tokens, args);
1720                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1721                                      journal_ioprio, is_remount) < 0)
1722                         return 0;
1723         }
1724 #ifdef CONFIG_QUOTA
1725         if (ext4_has_feature_quota(sb) &&
1726             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1727                 ext4_msg(sb, KERN_INFO, "Quota feature enabled, usrquota and grpquota "
1728                          "mount options ignored.");
1729                 clear_opt(sb, USRQUOTA);
1730                 clear_opt(sb, GRPQUOTA);
1731         } else if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1732                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1733                         clear_opt(sb, USRQUOTA);
1734
1735                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1736                         clear_opt(sb, GRPQUOTA);
1737
1738                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1739                         ext4_msg(sb, KERN_ERR, "old and new quota "
1740                                         "format mixing");
1741                         return 0;
1742                 }
1743
1744                 if (!sbi->s_jquota_fmt) {
1745                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1746                                         "not specified");
1747                         return 0;
1748                 }
1749         }
1750 #endif
1751         if (test_opt(sb, DIOREAD_NOLOCK)) {
1752                 int blocksize =
1753                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1754
1755                 if (blocksize < PAGE_CACHE_SIZE) {
1756                         ext4_msg(sb, KERN_ERR, "can't mount with "
1757                                  "dioread_nolock if block size != PAGE_SIZE");
1758                         return 0;
1759                 }
1760         }
1761         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1762             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1763                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1764                          "in data=ordered mode");
1765                 return 0;
1766         }
1767         return 1;
1768 }
1769
1770 static inline void ext4_show_quota_options(struct seq_file *seq,
1771                                            struct super_block *sb)
1772 {
1773 #if defined(CONFIG_QUOTA)
1774         struct ext4_sb_info *sbi = EXT4_SB(sb);
1775
1776         if (sbi->s_jquota_fmt) {
1777                 char *fmtname = "";
1778
1779                 switch (sbi->s_jquota_fmt) {
1780                 case QFMT_VFS_OLD:
1781                         fmtname = "vfsold";
1782                         break;
1783                 case QFMT_VFS_V0:
1784                         fmtname = "vfsv0";
1785                         break;
1786                 case QFMT_VFS_V1:
1787                         fmtname = "vfsv1";
1788                         break;
1789                 }
1790                 seq_printf(seq, ",jqfmt=%s", fmtname);
1791         }
1792
1793         if (sbi->s_qf_names[USRQUOTA])
1794                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1795
1796         if (sbi->s_qf_names[GRPQUOTA])
1797                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1798 #endif
1799 }
1800
1801 static const char *token2str(int token)
1802 {
1803         const struct match_token *t;
1804
1805         for (t = tokens; t->token != Opt_err; t++)
1806                 if (t->token == token && !strchr(t->pattern, '='))
1807                         break;
1808         return t->pattern;
1809 }
1810
1811 /*
1812  * Show an option if
1813  *  - it's set to a non-default value OR
1814  *  - if the per-sb default is different from the global default
1815  */
1816 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1817                               int nodefs)
1818 {
1819         struct ext4_sb_info *sbi = EXT4_SB(sb);
1820         struct ext4_super_block *es = sbi->s_es;
1821         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1822         const struct mount_opts *m;
1823         char sep = nodefs ? '\n' : ',';
1824
1825 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1826 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1827
1828         if (sbi->s_sb_block != 1)
1829                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1830
1831         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1832                 int want_set = m->flags & MOPT_SET;
1833                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1834                     (m->flags & MOPT_CLEAR_ERR))
1835                         continue;
1836                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1837                         continue; /* skip if same as the default */
1838                 if ((want_set &&
1839                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1840                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1841                         continue; /* select Opt_noFoo vs Opt_Foo */
1842                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1843         }
1844
1845         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1846             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1847                 SEQ_OPTS_PRINT("resuid=%u",
1848                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1849         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1850             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1851                 SEQ_OPTS_PRINT("resgid=%u",
1852                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1853         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1854         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1855                 SEQ_OPTS_PUTS("errors=remount-ro");
1856         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1857                 SEQ_OPTS_PUTS("errors=continue");
1858         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1859                 SEQ_OPTS_PUTS("errors=panic");
1860         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1861                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1862         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1863                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1864         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1865                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1866         if (sb->s_flags & MS_I_VERSION)
1867                 SEQ_OPTS_PUTS("i_version");
1868         if (nodefs || sbi->s_stripe)
1869                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1870         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1871                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1872                         SEQ_OPTS_PUTS("data=journal");
1873                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1874                         SEQ_OPTS_PUTS("data=ordered");
1875                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1876                         SEQ_OPTS_PUTS("data=writeback");
1877         }
1878         if (nodefs ||
1879             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1880                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1881                                sbi->s_inode_readahead_blks);
1882
1883         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1884                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1885                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1886         if (nodefs || sbi->s_max_dir_size_kb)
1887                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1888
1889         ext4_show_quota_options(seq, sb);
1890         return 0;
1891 }
1892
1893 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1894 {
1895         return _ext4_show_options(seq, root->d_sb, 0);
1896 }
1897
1898 int ext4_seq_options_show(struct seq_file *seq, void *offset)
1899 {
1900         struct super_block *sb = seq->private;
1901         int rc;
1902
1903         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1904         rc = _ext4_show_options(seq, sb, 1);
1905         seq_puts(seq, "\n");
1906         return rc;
1907 }
1908
1909 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1910                             int read_only)
1911 {
1912         struct ext4_sb_info *sbi = EXT4_SB(sb);
1913         int res = 0;
1914
1915         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1916                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1917                          "forcing read-only mode");
1918                 res = MS_RDONLY;
1919         }
1920         if (read_only)
1921                 goto done;
1922         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1923                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1924                          "running e2fsck is recommended");
1925         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1926                 ext4_msg(sb, KERN_WARNING,
1927                          "warning: mounting fs with errors, "
1928                          "running e2fsck is recommended");
1929         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1930                  le16_to_cpu(es->s_mnt_count) >=
1931                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1932                 ext4_msg(sb, KERN_WARNING,
1933                          "warning: maximal mount count reached, "
1934                          "running e2fsck is recommended");
1935         else if (le32_to_cpu(es->s_checkinterval) &&
1936                 (le32_to_cpu(es->s_lastcheck) +
1937                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1938                 ext4_msg(sb, KERN_WARNING,
1939                          "warning: checktime reached, "
1940                          "running e2fsck is recommended");
1941         if (!sbi->s_journal)
1942                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1943         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1944                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1945         le16_add_cpu(&es->s_mnt_count, 1);
1946         es->s_mtime = cpu_to_le32(get_seconds());
1947         ext4_update_dynamic_rev(sb);
1948         if (sbi->s_journal)
1949                 ext4_set_feature_journal_needs_recovery(sb);
1950
1951         ext4_commit_super(sb, 1);
1952 done:
1953         if (test_opt(sb, DEBUG))
1954                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1955                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1956                         sb->s_blocksize,
1957                         sbi->s_groups_count,
1958                         EXT4_BLOCKS_PER_GROUP(sb),
1959                         EXT4_INODES_PER_GROUP(sb),
1960                         sbi->s_mount_opt, sbi->s_mount_opt2);
1961
1962         cleancache_init_fs(sb);
1963         return res;
1964 }
1965
1966 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1967 {
1968         struct ext4_sb_info *sbi = EXT4_SB(sb);
1969         struct flex_groups *new_groups;
1970         int size;
1971
1972         if (!sbi->s_log_groups_per_flex)
1973                 return 0;
1974
1975         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1976         if (size <= sbi->s_flex_groups_allocated)
1977                 return 0;
1978
1979         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1980         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1981         if (!new_groups) {
1982                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1983                          size / (int) sizeof(struct flex_groups));
1984                 return -ENOMEM;
1985         }
1986
1987         if (sbi->s_flex_groups) {
1988                 memcpy(new_groups, sbi->s_flex_groups,
1989                        (sbi->s_flex_groups_allocated *
1990                         sizeof(struct flex_groups)));
1991                 kvfree(sbi->s_flex_groups);
1992         }
1993         sbi->s_flex_groups = new_groups;
1994         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1995         return 0;
1996 }
1997
1998 static int ext4_fill_flex_info(struct super_block *sb)
1999 {
2000         struct ext4_sb_info *sbi = EXT4_SB(sb);
2001         struct ext4_group_desc *gdp = NULL;
2002         ext4_group_t flex_group;
2003         int i, err;
2004
2005         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2006         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2007                 sbi->s_log_groups_per_flex = 0;
2008                 return 1;
2009         }
2010
2011         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2012         if (err)
2013                 goto failed;
2014
2015         for (i = 0; i < sbi->s_groups_count; i++) {
2016                 gdp = ext4_get_group_desc(sb, i, NULL);
2017
2018                 flex_group = ext4_flex_group(sbi, i);
2019                 atomic_add(ext4_free_inodes_count(sb, gdp),
2020                            &sbi->s_flex_groups[flex_group].free_inodes);
2021                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2022                              &sbi->s_flex_groups[flex_group].free_clusters);
2023                 atomic_add(ext4_used_dirs_count(sb, gdp),
2024                            &sbi->s_flex_groups[flex_group].used_dirs);
2025         }
2026
2027         return 1;
2028 failed:
2029         return 0;
2030 }
2031
2032 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2033                                    struct ext4_group_desc *gdp)
2034 {
2035         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2036         __u16 crc = 0;
2037         __le32 le_group = cpu_to_le32(block_group);
2038         struct ext4_sb_info *sbi = EXT4_SB(sb);
2039
2040         if (ext4_has_metadata_csum(sbi->s_sb)) {
2041                 /* Use new metadata_csum algorithm */
2042                 __u32 csum32;
2043                 __u16 dummy_csum = 0;
2044
2045                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2046                                      sizeof(le_group));
2047                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2048                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2049                                      sizeof(dummy_csum));
2050                 offset += sizeof(dummy_csum);
2051                 if (offset < sbi->s_desc_size)
2052                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2053                                              sbi->s_desc_size - offset);
2054
2055                 crc = csum32 & 0xFFFF;
2056                 goto out;
2057         }
2058
2059         /* old crc16 code */
2060         if (!ext4_has_feature_gdt_csum(sb))
2061                 return 0;
2062
2063         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2064         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2065         crc = crc16(crc, (__u8 *)gdp, offset);
2066         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2067         /* for checksum of struct ext4_group_desc do the rest...*/
2068         if (ext4_has_feature_64bit(sb) &&
2069             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2070                 crc = crc16(crc, (__u8 *)gdp + offset,
2071                             le16_to_cpu(sbi->s_es->s_desc_size) -
2072                                 offset);
2073
2074 out:
2075         return cpu_to_le16(crc);
2076 }
2077
2078 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2079                                 struct ext4_group_desc *gdp)
2080 {
2081         if (ext4_has_group_desc_csum(sb) &&
2082             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2083                 return 0;
2084
2085         return 1;
2086 }
2087
2088 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2089                               struct ext4_group_desc *gdp)
2090 {
2091         if (!ext4_has_group_desc_csum(sb))
2092                 return;
2093         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2094 }
2095
2096 /* Called at mount-time, super-block is locked */
2097 static int ext4_check_descriptors(struct super_block *sb,
2098                                   ext4_fsblk_t sb_block,
2099                                   ext4_group_t *first_not_zeroed)
2100 {
2101         struct ext4_sb_info *sbi = EXT4_SB(sb);
2102         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2103         ext4_fsblk_t last_block;
2104         ext4_fsblk_t block_bitmap;
2105         ext4_fsblk_t inode_bitmap;
2106         ext4_fsblk_t inode_table;
2107         int flexbg_flag = 0;
2108         ext4_group_t i, grp = sbi->s_groups_count;
2109
2110         if (ext4_has_feature_flex_bg(sb))
2111                 flexbg_flag = 1;
2112
2113         ext4_debug("Checking group descriptors");
2114
2115         for (i = 0; i < sbi->s_groups_count; i++) {
2116                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2117
2118                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2119                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2120                 else
2121                         last_block = first_block +
2122                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2123
2124                 if ((grp == sbi->s_groups_count) &&
2125                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2126                         grp = i;
2127
2128                 block_bitmap = ext4_block_bitmap(sb, gdp);
2129                 if (block_bitmap == sb_block) {
2130                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2131                                  "Block bitmap for group %u overlaps "
2132                                  "superblock", i);
2133                 }
2134                 if (block_bitmap < first_block || block_bitmap > last_block) {
2135                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2136                                "Block bitmap for group %u not in group "
2137                                "(block %llu)!", i, block_bitmap);
2138                         return 0;
2139                 }
2140                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2141                 if (inode_bitmap == sb_block) {
2142                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2143                                  "Inode bitmap for group %u overlaps "
2144                                  "superblock", i);
2145                 }
2146                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2147                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2148                                "Inode bitmap for group %u not in group "
2149                                "(block %llu)!", i, inode_bitmap);
2150                         return 0;
2151                 }
2152                 inode_table = ext4_inode_table(sb, gdp);
2153                 if (inode_table == sb_block) {
2154                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2155                                  "Inode table for group %u overlaps "
2156                                  "superblock", i);
2157                 }
2158                 if (inode_table < first_block ||
2159                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2160                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2161                                "Inode table for group %u not in group "
2162                                "(block %llu)!", i, inode_table);
2163                         return 0;
2164                 }
2165                 ext4_lock_group(sb, i);
2166                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2167                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2168                                  "Checksum for group %u failed (%u!=%u)",
2169                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2170                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2171                         if (!(sb->s_flags & MS_RDONLY)) {
2172                                 ext4_unlock_group(sb, i);
2173                                 return 0;
2174                         }
2175                 }
2176                 ext4_unlock_group(sb, i);
2177                 if (!flexbg_flag)
2178                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2179         }
2180         if (NULL != first_not_zeroed)
2181                 *first_not_zeroed = grp;
2182         return 1;
2183 }
2184
2185 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2186  * the superblock) which were deleted from all directories, but held open by
2187  * a process at the time of a crash.  We walk the list and try to delete these
2188  * inodes at recovery time (only with a read-write filesystem).
2189  *
2190  * In order to keep the orphan inode chain consistent during traversal (in
2191  * case of crash during recovery), we link each inode into the superblock
2192  * orphan list_head and handle it the same way as an inode deletion during
2193  * normal operation (which journals the operations for us).
2194  *
2195  * We only do an iget() and an iput() on each inode, which is very safe if we
2196  * accidentally point at an in-use or already deleted inode.  The worst that
2197  * can happen in this case is that we get a "bit already cleared" message from
2198  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2199  * e2fsck was run on this filesystem, and it must have already done the orphan
2200  * inode cleanup for us, so we can safely abort without any further action.
2201  */
2202 static void ext4_orphan_cleanup(struct super_block *sb,
2203                                 struct ext4_super_block *es)
2204 {
2205         unsigned int s_flags = sb->s_flags;
2206         int nr_orphans = 0, nr_truncates = 0;
2207 #ifdef CONFIG_QUOTA
2208         int i;
2209 #endif
2210         if (!es->s_last_orphan) {
2211                 jbd_debug(4, "no orphan inodes to clean up\n");
2212                 return;
2213         }
2214
2215         if (bdev_read_only(sb->s_bdev)) {
2216                 ext4_msg(sb, KERN_ERR, "write access "
2217                         "unavailable, skipping orphan cleanup");
2218                 return;
2219         }
2220
2221         /* Check if feature set would not allow a r/w mount */
2222         if (!ext4_feature_set_ok(sb, 0)) {
2223                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2224                          "unknown ROCOMPAT features");
2225                 return;
2226         }
2227
2228         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2229                 /* don't clear list on RO mount w/ errors */
2230                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2231                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2232                                   "clearing orphan list.\n");
2233                         es->s_last_orphan = 0;
2234                 }
2235                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2236                 return;
2237         }
2238
2239         if (s_flags & MS_RDONLY) {
2240                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2241                 sb->s_flags &= ~MS_RDONLY;
2242         }
2243 #ifdef CONFIG_QUOTA
2244         /* Needed for iput() to work correctly and not trash data */
2245         sb->s_flags |= MS_ACTIVE;
2246         /* Turn on quotas so that they are updated correctly */
2247         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2248                 if (EXT4_SB(sb)->s_qf_names[i]) {
2249                         int ret = ext4_quota_on_mount(sb, i);
2250                         if (ret < 0)
2251                                 ext4_msg(sb, KERN_ERR,
2252                                         "Cannot turn on journaled "
2253                                         "quota: error %d", ret);
2254                 }
2255         }
2256 #endif
2257
2258         while (es->s_last_orphan) {
2259                 struct inode *inode;
2260
2261                 /*
2262                  * We may have encountered an error during cleanup; if
2263                  * so, skip the rest.
2264                  */
2265                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2266                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2267                         es->s_last_orphan = 0;
2268                         break;
2269                 }
2270
2271                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2272                 if (IS_ERR(inode)) {
2273                         es->s_last_orphan = 0;
2274                         break;
2275                 }
2276
2277                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2278                 dquot_initialize(inode);
2279                 if (inode->i_nlink) {
2280                         if (test_opt(sb, DEBUG))
2281                                 ext4_msg(sb, KERN_DEBUG,
2282                                         "%s: truncating inode %lu to %lld bytes",
2283                                         __func__, inode->i_ino, inode->i_size);
2284                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2285                                   inode->i_ino, inode->i_size);
2286                         mutex_lock(&inode->i_mutex);
2287                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2288                         ext4_truncate(inode);
2289                         mutex_unlock(&inode->i_mutex);
2290                         nr_truncates++;
2291                 } else {
2292                         if (test_opt(sb, DEBUG))
2293                                 ext4_msg(sb, KERN_DEBUG,
2294                                         "%s: deleting unreferenced inode %lu",
2295                                         __func__, inode->i_ino);
2296                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2297                                   inode->i_ino);
2298                         nr_orphans++;
2299                 }
2300                 iput(inode);  /* The delete magic happens here! */
2301         }
2302
2303 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2304
2305         if (nr_orphans)
2306                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2307                        PLURAL(nr_orphans));
2308         if (nr_truncates)
2309                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2310                        PLURAL(nr_truncates));
2311 #ifdef CONFIG_QUOTA
2312         /* Turn quotas off */
2313         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2314                 if (sb_dqopt(sb)->files[i])
2315                         dquot_quota_off(sb, i);
2316         }
2317 #endif
2318         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2319 }
2320
2321 /*
2322  * Maximal extent format file size.
2323  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2324  * extent format containers, within a sector_t, and within i_blocks
2325  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2326  * so that won't be a limiting factor.
2327  *
2328  * However there is other limiting factor. We do store extents in the form
2329  * of starting block and length, hence the resulting length of the extent
2330  * covering maximum file size must fit into on-disk format containers as
2331  * well. Given that length is always by 1 unit bigger than max unit (because
2332  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2333  *
2334  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2335  */
2336 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2337 {
2338         loff_t res;
2339         loff_t upper_limit = MAX_LFS_FILESIZE;
2340
2341         /* small i_blocks in vfs inode? */
2342         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2343                 /*
2344                  * CONFIG_LBDAF is not enabled implies the inode
2345                  * i_block represent total blocks in 512 bytes
2346                  * 32 == size of vfs inode i_blocks * 8
2347                  */
2348                 upper_limit = (1LL << 32) - 1;
2349
2350                 /* total blocks in file system block size */
2351                 upper_limit >>= (blkbits - 9);
2352                 upper_limit <<= blkbits;
2353         }
2354
2355         /*
2356          * 32-bit extent-start container, ee_block. We lower the maxbytes
2357          * by one fs block, so ee_len can cover the extent of maximum file
2358          * size
2359          */
2360         res = (1LL << 32) - 1;
2361         res <<= blkbits;
2362
2363         /* Sanity check against vm- & vfs- imposed limits */
2364         if (res > upper_limit)
2365                 res = upper_limit;
2366
2367         return res;
2368 }
2369
2370 /*
2371  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2372  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2373  * We need to be 1 filesystem block less than the 2^48 sector limit.
2374  */
2375 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2376 {
2377         loff_t res = EXT4_NDIR_BLOCKS;
2378         int meta_blocks;
2379         loff_t upper_limit;
2380         /* This is calculated to be the largest file size for a dense, block
2381          * mapped file such that the file's total number of 512-byte sectors,
2382          * including data and all indirect blocks, does not exceed (2^48 - 1).
2383          *
2384          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2385          * number of 512-byte sectors of the file.
2386          */
2387
2388         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2389                 /*
2390                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2391                  * the inode i_block field represents total file blocks in
2392                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2393                  */
2394                 upper_limit = (1LL << 32) - 1;
2395
2396                 /* total blocks in file system block size */
2397                 upper_limit >>= (bits - 9);
2398
2399         } else {
2400                 /*
2401                  * We use 48 bit ext4_inode i_blocks
2402                  * With EXT4_HUGE_FILE_FL set the i_blocks
2403                  * represent total number of blocks in
2404                  * file system block size
2405                  */
2406                 upper_limit = (1LL << 48) - 1;
2407
2408         }
2409
2410         /* indirect blocks */
2411         meta_blocks = 1;
2412         /* double indirect blocks */
2413         meta_blocks += 1 + (1LL << (bits-2));
2414         /* tripple indirect blocks */
2415         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2416
2417         upper_limit -= meta_blocks;
2418         upper_limit <<= bits;
2419
2420         res += 1LL << (bits-2);
2421         res += 1LL << (2*(bits-2));
2422         res += 1LL << (3*(bits-2));
2423         res <<= bits;
2424         if (res > upper_limit)
2425                 res = upper_limit;
2426
2427         if (res > MAX_LFS_FILESIZE)
2428                 res = MAX_LFS_FILESIZE;
2429
2430         return res;
2431 }
2432
2433 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2434                                    ext4_fsblk_t logical_sb_block, int nr)
2435 {
2436         struct ext4_sb_info *sbi = EXT4_SB(sb);
2437         ext4_group_t bg, first_meta_bg;
2438         int has_super = 0;
2439
2440         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2441
2442         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2443                 return logical_sb_block + nr + 1;
2444         bg = sbi->s_desc_per_block * nr;
2445         if (ext4_bg_has_super(sb, bg))
2446                 has_super = 1;
2447
2448         /*
2449          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2450          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2451          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2452          * compensate.
2453          */
2454         if (sb->s_blocksize == 1024 && nr == 0 &&
2455             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2456                 has_super++;
2457
2458         return (has_super + ext4_group_first_block_no(sb, bg));
2459 }
2460
2461 /**
2462  * ext4_get_stripe_size: Get the stripe size.
2463  * @sbi: In memory super block info
2464  *
2465  * If we have specified it via mount option, then
2466  * use the mount option value. If the value specified at mount time is
2467  * greater than the blocks per group use the super block value.
2468  * If the super block value is greater than blocks per group return 0.
2469  * Allocator needs it be less than blocks per group.
2470  *
2471  */
2472 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2473 {
2474         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2475         unsigned long stripe_width =
2476                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2477         int ret;
2478
2479         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2480                 ret = sbi->s_stripe;
2481         else if (stripe_width <= sbi->s_blocks_per_group)
2482                 ret = stripe_width;
2483         else if (stride <= sbi->s_blocks_per_group)
2484                 ret = stride;
2485         else
2486                 ret = 0;
2487
2488         /*
2489          * If the stripe width is 1, this makes no sense and
2490          * we set it to 0 to turn off stripe handling code.
2491          */
2492         if (ret <= 1)
2493                 ret = 0;
2494
2495         return ret;
2496 }
2497
2498 /*
2499  * Check whether this filesystem can be mounted based on
2500  * the features present and the RDONLY/RDWR mount requested.
2501  * Returns 1 if this filesystem can be mounted as requested,
2502  * 0 if it cannot be.
2503  */
2504 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2505 {
2506         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2507                 ext4_msg(sb, KERN_ERR,
2508                         "Couldn't mount because of "
2509                         "unsupported optional features (%x)",
2510                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2511                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2512                 return 0;
2513         }
2514
2515         if (readonly)
2516                 return 1;
2517
2518         if (ext4_has_feature_readonly(sb)) {
2519                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2520                 sb->s_flags |= MS_RDONLY;
2521                 return 1;
2522         }
2523
2524         /* Check that feature set is OK for a read-write mount */
2525         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2526                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2527                          "unsupported optional features (%x)",
2528                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2529                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2530                 return 0;
2531         }
2532         /*
2533          * Large file size enabled file system can only be mounted
2534          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2535          */
2536         if (ext4_has_feature_huge_file(sb)) {
2537                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2538                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2539                                  "cannot be mounted RDWR without "
2540                                  "CONFIG_LBDAF");
2541                         return 0;
2542                 }
2543         }
2544         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2545                 ext4_msg(sb, KERN_ERR,
2546                          "Can't support bigalloc feature without "
2547                          "extents feature\n");
2548                 return 0;
2549         }
2550
2551 #ifndef CONFIG_QUOTA
2552         if (ext4_has_feature_quota(sb) && !readonly) {
2553                 ext4_msg(sb, KERN_ERR,
2554                          "Filesystem with quota feature cannot be mounted RDWR "
2555                          "without CONFIG_QUOTA");
2556                 return 0;
2557         }
2558 #endif  /* CONFIG_QUOTA */
2559         return 1;
2560 }
2561
2562 /*
2563  * This function is called once a day if we have errors logged
2564  * on the file system
2565  */
2566 static void print_daily_error_info(unsigned long arg)
2567 {
2568         struct super_block *sb = (struct super_block *) arg;
2569         struct ext4_sb_info *sbi;
2570         struct ext4_super_block *es;
2571
2572         sbi = EXT4_SB(sb);
2573         es = sbi->s_es;
2574
2575         if (es->s_error_count)
2576                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2577                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2578                          le32_to_cpu(es->s_error_count));
2579         if (es->s_first_error_time) {
2580                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2581                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2582                        (int) sizeof(es->s_first_error_func),
2583                        es->s_first_error_func,
2584                        le32_to_cpu(es->s_first_error_line));
2585                 if (es->s_first_error_ino)
2586                         printk(": inode %u",
2587                                le32_to_cpu(es->s_first_error_ino));
2588                 if (es->s_first_error_block)
2589                         printk(": block %llu", (unsigned long long)
2590                                le64_to_cpu(es->s_first_error_block));
2591                 printk("\n");
2592         }
2593         if (es->s_last_error_time) {
2594                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2595                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2596                        (int) sizeof(es->s_last_error_func),
2597                        es->s_last_error_func,
2598                        le32_to_cpu(es->s_last_error_line));
2599                 if (es->s_last_error_ino)
2600                         printk(": inode %u",
2601                                le32_to_cpu(es->s_last_error_ino));
2602                 if (es->s_last_error_block)
2603                         printk(": block %llu", (unsigned long long)
2604                                le64_to_cpu(es->s_last_error_block));
2605                 printk("\n");
2606         }
2607         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2608 }
2609
2610 /* Find next suitable group and run ext4_init_inode_table */
2611 static int ext4_run_li_request(struct ext4_li_request *elr)
2612 {
2613         struct ext4_group_desc *gdp = NULL;
2614         ext4_group_t group, ngroups;
2615         struct super_block *sb;
2616         unsigned long timeout = 0;
2617         int ret = 0;
2618
2619         sb = elr->lr_super;
2620         ngroups = EXT4_SB(sb)->s_groups_count;
2621
2622         sb_start_write(sb);
2623         for (group = elr->lr_next_group; group < ngroups; group++) {
2624                 gdp = ext4_get_group_desc(sb, group, NULL);
2625                 if (!gdp) {
2626                         ret = 1;
2627                         break;
2628                 }
2629
2630                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2631                         break;
2632         }
2633
2634         if (group >= ngroups)
2635                 ret = 1;
2636
2637         if (!ret) {
2638                 timeout = jiffies;
2639                 ret = ext4_init_inode_table(sb, group,
2640                                             elr->lr_timeout ? 0 : 1);
2641                 if (elr->lr_timeout == 0) {
2642                         timeout = (jiffies - timeout) *
2643                                   elr->lr_sbi->s_li_wait_mult;
2644                         elr->lr_timeout = timeout;
2645                 }
2646                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2647                 elr->lr_next_group = group + 1;
2648         }
2649         sb_end_write(sb);
2650
2651         return ret;
2652 }
2653
2654 /*
2655  * Remove lr_request from the list_request and free the
2656  * request structure. Should be called with li_list_mtx held
2657  */
2658 static void ext4_remove_li_request(struct ext4_li_request *elr)
2659 {
2660         struct ext4_sb_info *sbi;
2661
2662         if (!elr)
2663                 return;
2664
2665         sbi = elr->lr_sbi;
2666
2667         list_del(&elr->lr_request);
2668         sbi->s_li_request = NULL;
2669         kfree(elr);
2670 }
2671
2672 static void ext4_unregister_li_request(struct super_block *sb)
2673 {
2674         mutex_lock(&ext4_li_mtx);
2675         if (!ext4_li_info) {
2676                 mutex_unlock(&ext4_li_mtx);
2677                 return;
2678         }
2679
2680         mutex_lock(&ext4_li_info->li_list_mtx);
2681         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2682         mutex_unlock(&ext4_li_info->li_list_mtx);
2683         mutex_unlock(&ext4_li_mtx);
2684 }
2685
2686 static struct task_struct *ext4_lazyinit_task;
2687
2688 /*
2689  * This is the function where ext4lazyinit thread lives. It walks
2690  * through the request list searching for next scheduled filesystem.
2691  * When such a fs is found, run the lazy initialization request
2692  * (ext4_rn_li_request) and keep track of the time spend in this
2693  * function. Based on that time we compute next schedule time of
2694  * the request. When walking through the list is complete, compute
2695  * next waking time and put itself into sleep.
2696  */
2697 static int ext4_lazyinit_thread(void *arg)
2698 {
2699         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2700         struct list_head *pos, *n;
2701         struct ext4_li_request *elr;
2702         unsigned long next_wakeup, cur;
2703
2704         BUG_ON(NULL == eli);
2705
2706 cont_thread:
2707         while (true) {
2708                 next_wakeup = MAX_JIFFY_OFFSET;
2709
2710                 mutex_lock(&eli->li_list_mtx);
2711                 if (list_empty(&eli->li_request_list)) {
2712                         mutex_unlock(&eli->li_list_mtx);
2713                         goto exit_thread;
2714                 }
2715
2716                 list_for_each_safe(pos, n, &eli->li_request_list) {
2717                         elr = list_entry(pos, struct ext4_li_request,
2718                                          lr_request);
2719
2720                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2721                                 if (ext4_run_li_request(elr) != 0) {
2722                                         /* error, remove the lazy_init job */
2723                                         ext4_remove_li_request(elr);
2724                                         continue;
2725                                 }
2726                         }
2727
2728                         if (time_before(elr->lr_next_sched, next_wakeup))
2729                                 next_wakeup = elr->lr_next_sched;
2730                 }
2731                 mutex_unlock(&eli->li_list_mtx);
2732
2733                 try_to_freeze();
2734
2735                 cur = jiffies;
2736                 if ((time_after_eq(cur, next_wakeup)) ||
2737                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2738                         cond_resched();
2739                         continue;
2740                 }
2741
2742                 schedule_timeout_interruptible(next_wakeup - cur);
2743
2744                 if (kthread_should_stop()) {
2745                         ext4_clear_request_list();
2746                         goto exit_thread;
2747                 }
2748         }
2749
2750 exit_thread:
2751         /*
2752          * It looks like the request list is empty, but we need
2753          * to check it under the li_list_mtx lock, to prevent any
2754          * additions into it, and of course we should lock ext4_li_mtx
2755          * to atomically free the list and ext4_li_info, because at
2756          * this point another ext4 filesystem could be registering
2757          * new one.
2758          */
2759         mutex_lock(&ext4_li_mtx);
2760         mutex_lock(&eli->li_list_mtx);
2761         if (!list_empty(&eli->li_request_list)) {
2762                 mutex_unlock(&eli->li_list_mtx);
2763                 mutex_unlock(&ext4_li_mtx);
2764                 goto cont_thread;
2765         }
2766         mutex_unlock(&eli->li_list_mtx);
2767         kfree(ext4_li_info);
2768         ext4_li_info = NULL;
2769         mutex_unlock(&ext4_li_mtx);
2770
2771         return 0;
2772 }
2773
2774 static void ext4_clear_request_list(void)
2775 {
2776         struct list_head *pos, *n;
2777         struct ext4_li_request *elr;
2778
2779         mutex_lock(&ext4_li_info->li_list_mtx);
2780         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2781                 elr = list_entry(pos, struct ext4_li_request,
2782                                  lr_request);
2783                 ext4_remove_li_request(elr);
2784         }
2785         mutex_unlock(&ext4_li_info->li_list_mtx);
2786 }
2787
2788 static int ext4_run_lazyinit_thread(void)
2789 {
2790         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2791                                          ext4_li_info, "ext4lazyinit");
2792         if (IS_ERR(ext4_lazyinit_task)) {
2793                 int err = PTR_ERR(ext4_lazyinit_task);
2794                 ext4_clear_request_list();
2795                 kfree(ext4_li_info);
2796                 ext4_li_info = NULL;
2797                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2798                                  "initialization thread\n",
2799                                  err);
2800                 return err;
2801         }
2802         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2803         return 0;
2804 }
2805
2806 /*
2807  * Check whether it make sense to run itable init. thread or not.
2808  * If there is at least one uninitialized inode table, return
2809  * corresponding group number, else the loop goes through all
2810  * groups and return total number of groups.
2811  */
2812 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2813 {
2814         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2815         struct ext4_group_desc *gdp = NULL;
2816
2817         for (group = 0; group < ngroups; group++) {
2818                 gdp = ext4_get_group_desc(sb, group, NULL);
2819                 if (!gdp)
2820                         continue;
2821
2822                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2823                         break;
2824         }
2825
2826         return group;
2827 }
2828
2829 static int ext4_li_info_new(void)
2830 {
2831         struct ext4_lazy_init *eli = NULL;
2832
2833         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2834         if (!eli)
2835                 return -ENOMEM;
2836
2837         INIT_LIST_HEAD(&eli->li_request_list);
2838         mutex_init(&eli->li_list_mtx);
2839
2840         eli->li_state |= EXT4_LAZYINIT_QUIT;
2841
2842         ext4_li_info = eli;
2843
2844         return 0;
2845 }
2846
2847 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2848                                             ext4_group_t start)
2849 {
2850         struct ext4_sb_info *sbi = EXT4_SB(sb);
2851         struct ext4_li_request *elr;
2852
2853         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2854         if (!elr)
2855                 return NULL;
2856
2857         elr->lr_super = sb;
2858         elr->lr_sbi = sbi;
2859         elr->lr_next_group = start;
2860
2861         /*
2862          * Randomize first schedule time of the request to
2863          * spread the inode table initialization requests
2864          * better.
2865          */
2866         elr->lr_next_sched = jiffies + (prandom_u32() %
2867                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2868         return elr;
2869 }
2870
2871 int ext4_register_li_request(struct super_block *sb,
2872                              ext4_group_t first_not_zeroed)
2873 {
2874         struct ext4_sb_info *sbi = EXT4_SB(sb);
2875         struct ext4_li_request *elr = NULL;
2876         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2877         int ret = 0;
2878
2879         mutex_lock(&ext4_li_mtx);
2880         if (sbi->s_li_request != NULL) {
2881                 /*
2882                  * Reset timeout so it can be computed again, because
2883                  * s_li_wait_mult might have changed.
2884                  */
2885                 sbi->s_li_request->lr_timeout = 0;
2886                 goto out;
2887         }
2888
2889         if (first_not_zeroed == ngroups ||
2890             (sb->s_flags & MS_RDONLY) ||
2891             !test_opt(sb, INIT_INODE_TABLE))
2892                 goto out;
2893
2894         elr = ext4_li_request_new(sb, first_not_zeroed);
2895         if (!elr) {
2896                 ret = -ENOMEM;
2897                 goto out;
2898         }
2899
2900         if (NULL == ext4_li_info) {
2901                 ret = ext4_li_info_new();
2902                 if (ret)
2903                         goto out;
2904         }
2905
2906         mutex_lock(&ext4_li_info->li_list_mtx);
2907         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2908         mutex_unlock(&ext4_li_info->li_list_mtx);
2909
2910         sbi->s_li_request = elr;
2911         /*
2912          * set elr to NULL here since it has been inserted to
2913          * the request_list and the removal and free of it is
2914          * handled by ext4_clear_request_list from now on.
2915          */
2916         elr = NULL;
2917
2918         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2919                 ret = ext4_run_lazyinit_thread();
2920                 if (ret)
2921                         goto out;
2922         }
2923 out:
2924         mutex_unlock(&ext4_li_mtx);
2925         if (ret)
2926                 kfree(elr);
2927         return ret;
2928 }
2929
2930 /*
2931  * We do not need to lock anything since this is called on
2932  * module unload.
2933  */
2934 static void ext4_destroy_lazyinit_thread(void)
2935 {
2936         /*
2937          * If thread exited earlier
2938          * there's nothing to be done.
2939          */
2940         if (!ext4_li_info || !ext4_lazyinit_task)
2941                 return;
2942
2943         kthread_stop(ext4_lazyinit_task);
2944 }
2945
2946 static int set_journal_csum_feature_set(struct super_block *sb)
2947 {
2948         int ret = 1;
2949         int compat, incompat;
2950         struct ext4_sb_info *sbi = EXT4_SB(sb);
2951
2952         if (ext4_has_metadata_csum(sb)) {
2953                 /* journal checksum v3 */
2954                 compat = 0;
2955                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
2956         } else {
2957                 /* journal checksum v1 */
2958                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
2959                 incompat = 0;
2960         }
2961
2962         jbd2_journal_clear_features(sbi->s_journal,
2963                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2964                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
2965                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
2966         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2967                 ret = jbd2_journal_set_features(sbi->s_journal,
2968                                 compat, 0,
2969                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
2970                                 incompat);
2971         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2972                 ret = jbd2_journal_set_features(sbi->s_journal,
2973                                 compat, 0,
2974                                 incompat);
2975                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2976                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2977         } else {
2978                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2979                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2980         }
2981
2982         return ret;
2983 }
2984
2985 /*
2986  * Note: calculating the overhead so we can be compatible with
2987  * historical BSD practice is quite difficult in the face of
2988  * clusters/bigalloc.  This is because multiple metadata blocks from
2989  * different block group can end up in the same allocation cluster.
2990  * Calculating the exact overhead in the face of clustered allocation
2991  * requires either O(all block bitmaps) in memory or O(number of block
2992  * groups**2) in time.  We will still calculate the superblock for
2993  * older file systems --- and if we come across with a bigalloc file
2994  * system with zero in s_overhead_clusters the estimate will be close to
2995  * correct especially for very large cluster sizes --- but for newer
2996  * file systems, it's better to calculate this figure once at mkfs
2997  * time, and store it in the superblock.  If the superblock value is
2998  * present (even for non-bigalloc file systems), we will use it.
2999  */
3000 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3001                           char *buf)
3002 {
3003         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3004         struct ext4_group_desc  *gdp;
3005         ext4_fsblk_t            first_block, last_block, b;
3006         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3007         int                     s, j, count = 0;
3008
3009         if (!ext4_has_feature_bigalloc(sb))
3010                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3011                         sbi->s_itb_per_group + 2);
3012
3013         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3014                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3015         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3016         for (i = 0; i < ngroups; i++) {
3017                 gdp = ext4_get_group_desc(sb, i, NULL);
3018                 b = ext4_block_bitmap(sb, gdp);
3019                 if (b >= first_block && b <= last_block) {
3020                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3021                         count++;
3022                 }
3023                 b = ext4_inode_bitmap(sb, gdp);
3024                 if (b >= first_block && b <= last_block) {
3025                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3026                         count++;
3027                 }
3028                 b = ext4_inode_table(sb, gdp);
3029                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3030                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3031                                 int c = EXT4_B2C(sbi, b - first_block);
3032                                 ext4_set_bit(c, buf);
3033                                 count++;
3034                         }
3035                 if (i != grp)
3036                         continue;
3037                 s = 0;
3038                 if (ext4_bg_has_super(sb, grp)) {
3039                         ext4_set_bit(s++, buf);
3040                         count++;
3041                 }
3042                 j = ext4_bg_num_gdb(sb, grp);
3043                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3044                         ext4_error(sb, "Invalid number of block group "
3045                                    "descriptor blocks: %d", j);
3046                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3047                 }
3048                 count += j;
3049                 for (; j > 0; j--)
3050                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3051         }
3052         if (!count)
3053                 return 0;
3054         return EXT4_CLUSTERS_PER_GROUP(sb) -
3055                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3056 }
3057
3058 /*
3059  * Compute the overhead and stash it in sbi->s_overhead
3060  */
3061 int ext4_calculate_overhead(struct super_block *sb)
3062 {
3063         struct ext4_sb_info *sbi = EXT4_SB(sb);
3064         struct ext4_super_block *es = sbi->s_es;
3065         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3066         ext4_fsblk_t overhead = 0;
3067         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3068
3069         if (!buf)
3070                 return -ENOMEM;
3071
3072         /*
3073          * Compute the overhead (FS structures).  This is constant
3074          * for a given filesystem unless the number of block groups
3075          * changes so we cache the previous value until it does.
3076          */
3077
3078         /*
3079          * All of the blocks before first_data_block are overhead
3080          */
3081         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3082
3083         /*
3084          * Add the overhead found in each block group
3085          */
3086         for (i = 0; i < ngroups; i++) {
3087                 int blks;
3088
3089                 blks = count_overhead(sb, i, buf);
3090                 overhead += blks;
3091                 if (blks)
3092                         memset(buf, 0, PAGE_SIZE);
3093                 cond_resched();
3094         }
3095         /* Add the internal journal blocks as well */
3096         if (sbi->s_journal && !sbi->journal_bdev)
3097                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3098
3099         sbi->s_overhead = overhead;
3100         smp_wmb();
3101         free_page((unsigned long) buf);
3102         return 0;
3103 }
3104
3105 static void ext4_set_resv_clusters(struct super_block *sb)
3106 {
3107         ext4_fsblk_t resv_clusters;
3108         struct ext4_sb_info *sbi = EXT4_SB(sb);
3109
3110         /*
3111          * There's no need to reserve anything when we aren't using extents.
3112          * The space estimates are exact, there are no unwritten extents,
3113          * hole punching doesn't need new metadata... This is needed especially
3114          * to keep ext2/3 backward compatibility.
3115          */
3116         if (!ext4_has_feature_extents(sb))
3117                 return;
3118         /*
3119          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3120          * This should cover the situations where we can not afford to run
3121          * out of space like for example punch hole, or converting
3122          * unwritten extents in delalloc path. In most cases such
3123          * allocation would require 1, or 2 blocks, higher numbers are
3124          * very rare.
3125          */
3126         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3127                          sbi->s_cluster_bits);
3128
3129         do_div(resv_clusters, 50);
3130         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3131
3132         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3133 }
3134
3135 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3136 {
3137         char *orig_data = kstrdup(data, GFP_KERNEL);
3138         struct buffer_head *bh;
3139         struct ext4_super_block *es = NULL;
3140         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3141         ext4_fsblk_t block;
3142         ext4_fsblk_t sb_block = get_sb_block(&data);
3143         ext4_fsblk_t logical_sb_block;
3144         unsigned long offset = 0;
3145         unsigned long journal_devnum = 0;
3146         unsigned long def_mount_opts;
3147         struct inode *root;
3148         const char *descr;
3149         int ret = -ENOMEM;
3150         int blocksize, clustersize;
3151         unsigned int db_count;
3152         unsigned int i;
3153         int needs_recovery, has_huge_files, has_bigalloc;
3154         __u64 blocks_count;
3155         int err = 0;
3156         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3157         ext4_group_t first_not_zeroed;
3158
3159         if ((data && !orig_data) || !sbi)
3160                 goto out_free_base;
3161
3162         sbi->s_blockgroup_lock =
3163                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3164         if (!sbi->s_blockgroup_lock)
3165                 goto out_free_base;
3166
3167         sb->s_fs_info = sbi;
3168         sbi->s_sb = sb;
3169         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3170         sbi->s_sb_block = sb_block;
3171         if (sb->s_bdev->bd_part)
3172                 sbi->s_sectors_written_start =
3173                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3174
3175         /* Cleanup superblock name */
3176         strreplace(sb->s_id, '/', '!');
3177
3178         /* -EINVAL is default */
3179         ret = -EINVAL;
3180         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3181         if (!blocksize) {
3182                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3183                 goto out_fail;
3184         }
3185
3186         /*
3187          * The ext4 superblock will not be buffer aligned for other than 1kB
3188          * block sizes.  We need to calculate the offset from buffer start.
3189          */
3190         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3191                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3192                 offset = do_div(logical_sb_block, blocksize);
3193         } else {
3194                 logical_sb_block = sb_block;
3195         }
3196
3197         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3198                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3199                 goto out_fail;
3200         }
3201         /*
3202          * Note: s_es must be initialized as soon as possible because
3203          *       some ext4 macro-instructions depend on its value
3204          */
3205         es = (struct ext4_super_block *) (bh->b_data + offset);
3206         sbi->s_es = es;
3207         sb->s_magic = le16_to_cpu(es->s_magic);
3208         if (sb->s_magic != EXT4_SUPER_MAGIC)
3209                 goto cantfind_ext4;
3210         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3211
3212         /* Warn if metadata_csum and gdt_csum are both set. */
3213         if (ext4_has_feature_metadata_csum(sb) &&
3214             ext4_has_feature_gdt_csum(sb))
3215                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3216                              "redundant flags; please run fsck.");
3217
3218         /* Check for a known checksum algorithm */
3219         if (!ext4_verify_csum_type(sb, es)) {
3220                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3221                          "unknown checksum algorithm.");
3222                 silent = 1;
3223                 goto cantfind_ext4;
3224         }
3225
3226         /* Load the checksum driver */
3227         if (ext4_has_feature_metadata_csum(sb)) {
3228                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3229                 if (IS_ERR(sbi->s_chksum_driver)) {
3230                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3231                         ret = PTR_ERR(sbi->s_chksum_driver);
3232                         sbi->s_chksum_driver = NULL;
3233                         goto failed_mount;
3234                 }
3235         }
3236
3237         /* Check superblock checksum */
3238         if (!ext4_superblock_csum_verify(sb, es)) {
3239                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3240                          "invalid superblock checksum.  Run e2fsck?");
3241                 silent = 1;
3242                 ret = -EFSBADCRC;
3243                 goto cantfind_ext4;
3244         }
3245
3246         /* Precompute checksum seed for all metadata */
3247         if (ext4_has_feature_csum_seed(sb))
3248                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3249         else if (ext4_has_metadata_csum(sb))
3250                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3251                                                sizeof(es->s_uuid));
3252
3253         /* Set defaults before we parse the mount options */
3254         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3255         set_opt(sb, INIT_INODE_TABLE);
3256         if (def_mount_opts & EXT4_DEFM_DEBUG)
3257                 set_opt(sb, DEBUG);
3258         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3259                 set_opt(sb, GRPID);
3260         if (def_mount_opts & EXT4_DEFM_UID16)
3261                 set_opt(sb, NO_UID32);
3262         /* xattr user namespace & acls are now defaulted on */
3263         set_opt(sb, XATTR_USER);
3264 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3265         set_opt(sb, POSIX_ACL);
3266 #endif
3267         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3268         if (ext4_has_metadata_csum(sb))
3269                 set_opt(sb, JOURNAL_CHECKSUM);
3270
3271         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3272                 set_opt(sb, JOURNAL_DATA);
3273         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3274                 set_opt(sb, ORDERED_DATA);
3275         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3276                 set_opt(sb, WRITEBACK_DATA);
3277
3278         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3279                 set_opt(sb, ERRORS_PANIC);
3280         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3281                 set_opt(sb, ERRORS_CONT);
3282         else
3283                 set_opt(sb, ERRORS_RO);
3284         /* block_validity enabled by default; disable with noblock_validity */
3285         set_opt(sb, BLOCK_VALIDITY);
3286         if (def_mount_opts & EXT4_DEFM_DISCARD)
3287                 set_opt(sb, DISCARD);
3288
3289         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3290         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3291         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3292         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3293         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3294
3295         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3296                 set_opt(sb, BARRIER);
3297
3298         /*
3299          * enable delayed allocation by default
3300          * Use -o nodelalloc to turn it off
3301          */
3302         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3303             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3304                 set_opt(sb, DELALLOC);
3305
3306         /*
3307          * set default s_li_wait_mult for lazyinit, for the case there is
3308          * no mount option specified.
3309          */
3310         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3311
3312         if (sbi->s_es->s_mount_opts[0]) {
3313                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3314                                               sizeof(sbi->s_es->s_mount_opts),
3315                                               GFP_KERNEL);
3316                 if (!s_mount_opts)
3317                         goto failed_mount;
3318                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3319                                    &journal_ioprio, 0)) {
3320                         ext4_msg(sb, KERN_WARNING,
3321                                  "failed to parse options in superblock: %s",
3322                                  s_mount_opts);
3323                 }
3324                 kfree(s_mount_opts);
3325         }
3326         sbi->s_def_mount_opt = sbi->s_mount_opt;
3327         if (!parse_options((char *) data, sb, &journal_devnum,
3328                            &journal_ioprio, 0))
3329                 goto failed_mount;
3330
3331         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3332                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3333                             "with data=journal disables delayed "
3334                             "allocation and O_DIRECT support!\n");
3335                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3336                         ext4_msg(sb, KERN_ERR, "can't mount with "
3337                                  "both data=journal and delalloc");
3338                         goto failed_mount;
3339                 }
3340                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3341                         ext4_msg(sb, KERN_ERR, "can't mount with "
3342                                  "both data=journal and dioread_nolock");
3343                         goto failed_mount;
3344                 }
3345                 if (test_opt(sb, DAX)) {
3346                         ext4_msg(sb, KERN_ERR, "can't mount with "
3347                                  "both data=journal and dax");
3348                         goto failed_mount;
3349                 }
3350                 if (ext4_has_feature_encrypt(sb)) {
3351                         ext4_msg(sb, KERN_WARNING,
3352                                  "encrypted files will use data=ordered "
3353                                  "instead of data journaling mode");
3354                 }
3355                 if (test_opt(sb, DELALLOC))
3356                         clear_opt(sb, DELALLOC);
3357         } else {
3358                 sb->s_iflags |= SB_I_CGROUPWB;
3359         }
3360
3361         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3362                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3363
3364         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3365             (ext4_has_compat_features(sb) ||
3366              ext4_has_ro_compat_features(sb) ||
3367              ext4_has_incompat_features(sb)))
3368                 ext4_msg(sb, KERN_WARNING,
3369                        "feature flags set on rev 0 fs, "
3370                        "running e2fsck is recommended");
3371
3372         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3373                 set_opt2(sb, HURD_COMPAT);
3374                 if (ext4_has_feature_64bit(sb)) {
3375                         ext4_msg(sb, KERN_ERR,
3376                                  "The Hurd can't support 64-bit file systems");
3377                         goto failed_mount;
3378                 }
3379         }
3380
3381         if (IS_EXT2_SB(sb)) {
3382                 if (ext2_feature_set_ok(sb))
3383                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3384                                  "using the ext4 subsystem");
3385                 else {
3386                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3387                                  "to feature incompatibilities");
3388                         goto failed_mount;
3389                 }
3390         }
3391
3392         if (IS_EXT3_SB(sb)) {
3393                 if (ext3_feature_set_ok(sb))
3394                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3395                                  "using the ext4 subsystem");
3396                 else {
3397                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3398                                  "to feature incompatibilities");
3399                         goto failed_mount;
3400                 }
3401         }
3402
3403         /*
3404          * Check feature flags regardless of the revision level, since we
3405          * previously didn't change the revision level when setting the flags,
3406          * so there is a chance incompat flags are set on a rev 0 filesystem.
3407          */
3408         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3409                 goto failed_mount;
3410
3411         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3412         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3413             blocksize > EXT4_MAX_BLOCK_SIZE) {
3414                 ext4_msg(sb, KERN_ERR,
3415                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3416                          blocksize, le32_to_cpu(es->s_log_block_size));
3417                 goto failed_mount;
3418         }
3419         if (le32_to_cpu(es->s_log_block_size) >
3420             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3421                 ext4_msg(sb, KERN_ERR,
3422                          "Invalid log block size: %u",
3423                          le32_to_cpu(es->s_log_block_size));
3424                 goto failed_mount;
3425         }
3426
3427         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3428                 ext4_msg(sb, KERN_ERR,
3429                          "Number of reserved GDT blocks insanely large: %d",
3430                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3431                 goto failed_mount;
3432         }
3433
3434         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3435                 if (blocksize != PAGE_SIZE) {
3436                         ext4_msg(sb, KERN_ERR,
3437                                         "error: unsupported blocksize for dax");
3438                         goto failed_mount;
3439                 }
3440                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3441                         ext4_msg(sb, KERN_ERR,
3442                                         "error: device does not support dax");
3443                         goto failed_mount;
3444                 }
3445         }
3446
3447         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3448                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3449                          es->s_encryption_level);
3450                 goto failed_mount;
3451         }
3452
3453         if (sb->s_blocksize != blocksize) {
3454                 /* Validate the filesystem blocksize */
3455                 if (!sb_set_blocksize(sb, blocksize)) {
3456                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3457                                         blocksize);
3458                         goto failed_mount;
3459                 }
3460
3461                 brelse(bh);
3462                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3463                 offset = do_div(logical_sb_block, blocksize);
3464                 bh = sb_bread_unmovable(sb, logical_sb_block);
3465                 if (!bh) {
3466                         ext4_msg(sb, KERN_ERR,
3467                                "Can't read superblock on 2nd try");
3468                         goto failed_mount;
3469                 }
3470                 es = (struct ext4_super_block *)(bh->b_data + offset);
3471                 sbi->s_es = es;
3472                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3473                         ext4_msg(sb, KERN_ERR,
3474                                "Magic mismatch, very weird!");
3475                         goto failed_mount;
3476                 }
3477         }
3478
3479         has_huge_files = ext4_has_feature_huge_file(sb);
3480         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3481                                                       has_huge_files);
3482         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3483
3484         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3485                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3486                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3487         } else {
3488                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3489                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3490                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3491                     (!is_power_of_2(sbi->s_inode_size)) ||
3492                     (sbi->s_inode_size > blocksize)) {
3493                         ext4_msg(sb, KERN_ERR,
3494                                "unsupported inode size: %d",
3495                                sbi->s_inode_size);
3496                         goto failed_mount;
3497                 }
3498                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3499                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3500         }
3501
3502         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3503         if (ext4_has_feature_64bit(sb)) {
3504                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3505                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3506                     !is_power_of_2(sbi->s_desc_size)) {
3507                         ext4_msg(sb, KERN_ERR,
3508                                "unsupported descriptor size %lu",
3509                                sbi->s_desc_size);
3510                         goto failed_mount;
3511                 }
3512         } else
3513                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3514
3515         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3516         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3517
3518         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3519         if (sbi->s_inodes_per_block == 0)
3520                 goto cantfind_ext4;
3521         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3522             sbi->s_inodes_per_group > blocksize * 8) {
3523                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3524                          sbi->s_blocks_per_group);
3525                 goto failed_mount;
3526         }
3527         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3528                                         sbi->s_inodes_per_block;
3529         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3530         sbi->s_sbh = bh;
3531         sbi->s_mount_state = le16_to_cpu(es->s_state);
3532         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3533         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3534
3535         for (i = 0; i < 4; i++)
3536                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3537         sbi->s_def_hash_version = es->s_def_hash_version;
3538         if (ext4_has_feature_dir_index(sb)) {
3539                 i = le32_to_cpu(es->s_flags);
3540                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3541                         sbi->s_hash_unsigned = 3;
3542                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3543 #ifdef __CHAR_UNSIGNED__
3544                         if (!(sb->s_flags & MS_RDONLY))
3545                                 es->s_flags |=
3546                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3547                         sbi->s_hash_unsigned = 3;
3548 #else
3549                         if (!(sb->s_flags & MS_RDONLY))
3550                                 es->s_flags |=
3551                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3552 #endif
3553                 }
3554         }
3555
3556         /* Handle clustersize */
3557         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3558         has_bigalloc = ext4_has_feature_bigalloc(sb);
3559         if (has_bigalloc) {
3560                 if (clustersize < blocksize) {
3561                         ext4_msg(sb, KERN_ERR,
3562                                  "cluster size (%d) smaller than "
3563                                  "block size (%d)", clustersize, blocksize);
3564                         goto failed_mount;
3565                 }
3566                 if (le32_to_cpu(es->s_log_cluster_size) >
3567                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3568                         ext4_msg(sb, KERN_ERR,
3569                                  "Invalid log cluster size: %u",
3570                                  le32_to_cpu(es->s_log_cluster_size));
3571                         goto failed_mount;
3572                 }
3573                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3574                         le32_to_cpu(es->s_log_block_size);
3575                 sbi->s_clusters_per_group =
3576                         le32_to_cpu(es->s_clusters_per_group);
3577                 if (sbi->s_clusters_per_group > blocksize * 8) {
3578                         ext4_msg(sb, KERN_ERR,
3579                                  "#clusters per group too big: %lu",
3580                                  sbi->s_clusters_per_group);
3581                         goto failed_mount;
3582                 }
3583                 if (sbi->s_blocks_per_group !=
3584                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3585                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3586                                  "clusters per group (%lu) inconsistent",
3587                                  sbi->s_blocks_per_group,
3588                                  sbi->s_clusters_per_group);
3589                         goto failed_mount;
3590                 }
3591         } else {
3592                 if (clustersize != blocksize) {
3593                         ext4_warning(sb, "fragment/cluster size (%d) != "
3594                                      "block size (%d)", clustersize,
3595                                      blocksize);
3596                         clustersize = blocksize;
3597                 }
3598                 if (sbi->s_blocks_per_group > blocksize * 8) {
3599                         ext4_msg(sb, KERN_ERR,
3600                                  "#blocks per group too big: %lu",
3601                                  sbi->s_blocks_per_group);
3602                         goto failed_mount;
3603                 }
3604                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3605                 sbi->s_cluster_bits = 0;
3606         }
3607         sbi->s_cluster_ratio = clustersize / blocksize;
3608
3609         /* Do we have standard group size of clustersize * 8 blocks ? */
3610         if (sbi->s_blocks_per_group == clustersize << 3)
3611                 set_opt2(sb, STD_GROUP_SIZE);
3612
3613         /*
3614          * Test whether we have more sectors than will fit in sector_t,
3615          * and whether the max offset is addressable by the page cache.
3616          */
3617         err = generic_check_addressable(sb->s_blocksize_bits,
3618                                         ext4_blocks_count(es));
3619         if (err) {
3620                 ext4_msg(sb, KERN_ERR, "filesystem"
3621                          " too large to mount safely on this system");
3622                 if (sizeof(sector_t) < 8)
3623                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3624                 goto failed_mount;
3625         }
3626
3627         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3628                 goto cantfind_ext4;
3629
3630         /* check blocks count against device size */
3631         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3632         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3633                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3634                        "exceeds size of device (%llu blocks)",
3635                        ext4_blocks_count(es), blocks_count);
3636                 goto failed_mount;
3637         }
3638
3639         /*
3640          * It makes no sense for the first data block to be beyond the end
3641          * of the filesystem.
3642          */
3643         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3644                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3645                          "block %u is beyond end of filesystem (%llu)",
3646                          le32_to_cpu(es->s_first_data_block),
3647                          ext4_blocks_count(es));
3648                 goto failed_mount;
3649         }
3650         blocks_count = (ext4_blocks_count(es) -
3651                         le32_to_cpu(es->s_first_data_block) +
3652                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3653         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3654         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3655                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3656                        "(block count %llu, first data block %u, "
3657                        "blocks per group %lu)", sbi->s_groups_count,
3658                        ext4_blocks_count(es),
3659                        le32_to_cpu(es->s_first_data_block),
3660                        EXT4_BLOCKS_PER_GROUP(sb));
3661                 goto failed_mount;
3662         }
3663         sbi->s_groups_count = blocks_count;
3664         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3665                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3666         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3667                    EXT4_DESC_PER_BLOCK(sb);
3668         if (ext4_has_feature_meta_bg(sb)) {
3669                 if (le32_to_cpu(es->s_first_meta_bg) >= db_count) {
3670                         ext4_msg(sb, KERN_WARNING,
3671                                  "first meta block group too large: %u "
3672                                  "(group descriptor block count %u)",
3673                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3674                         goto failed_mount;
3675                 }
3676         }
3677         sbi->s_group_desc = ext4_kvmalloc(db_count *
3678                                           sizeof(struct buffer_head *),
3679                                           GFP_KERNEL);
3680         if (sbi->s_group_desc == NULL) {
3681                 ext4_msg(sb, KERN_ERR, "not enough memory");
3682                 ret = -ENOMEM;
3683                 goto failed_mount;
3684         }
3685
3686         bgl_lock_init(sbi->s_blockgroup_lock);
3687
3688         for (i = 0; i < db_count; i++) {
3689                 block = descriptor_loc(sb, logical_sb_block, i);
3690                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3691                 if (!sbi->s_group_desc[i]) {
3692                         ext4_msg(sb, KERN_ERR,
3693                                "can't read group descriptor %d", i);
3694                         db_count = i;
3695                         goto failed_mount2;
3696                 }
3697         }
3698         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3699                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3700                 ret = -EFSCORRUPTED;
3701                 goto failed_mount2;
3702         }
3703
3704         sbi->s_gdb_count = db_count;
3705         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3706         spin_lock_init(&sbi->s_next_gen_lock);
3707
3708         setup_timer(&sbi->s_err_report, print_daily_error_info,
3709                 (unsigned long) sb);
3710
3711         /* Register extent status tree shrinker */
3712         if (ext4_es_register_shrinker(sbi))
3713                 goto failed_mount3;
3714
3715         sbi->s_stripe = ext4_get_stripe_size(sbi);
3716         sbi->s_extent_max_zeroout_kb = 32;
3717
3718         /*
3719          * set up enough so that it can read an inode
3720          */
3721         sb->s_op = &ext4_sops;
3722         sb->s_export_op = &ext4_export_ops;
3723         sb->s_xattr = ext4_xattr_handlers;
3724 #ifdef CONFIG_QUOTA
3725         sb->dq_op = &ext4_quota_operations;
3726         if (ext4_has_feature_quota(sb))
3727                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3728         else
3729                 sb->s_qcop = &ext4_qctl_operations;
3730         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
3731 #endif
3732         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3733
3734         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3735         mutex_init(&sbi->s_orphan_lock);
3736
3737         sb->s_root = NULL;
3738
3739         needs_recovery = (es->s_last_orphan != 0 ||
3740                           ext4_has_feature_journal_needs_recovery(sb));
3741
3742         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3743                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3744                         goto failed_mount3a;
3745
3746         /*
3747          * The first inode we look at is the journal inode.  Don't try
3748          * root first: it may be modified in the journal!
3749          */
3750         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3751                 err = ext4_load_journal(sb, es, journal_devnum);
3752                 if (err)
3753                         goto failed_mount3a;
3754         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3755                    ext4_has_feature_journal_needs_recovery(sb)) {
3756                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3757                        "suppressed and not mounted read-only");
3758                 goto failed_mount_wq;
3759         } else {
3760                 /* Nojournal mode, all journal mount options are illegal */
3761                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3762                         ext4_msg(sb, KERN_ERR, "can't mount with "
3763                                  "journal_checksum, fs mounted w/o journal");
3764                         goto failed_mount_wq;
3765                 }
3766                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3767                         ext4_msg(sb, KERN_ERR, "can't mount with "
3768                                  "journal_async_commit, fs mounted w/o journal");
3769                         goto failed_mount_wq;
3770                 }
3771                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3772                         ext4_msg(sb, KERN_ERR, "can't mount with "
3773                                  "commit=%lu, fs mounted w/o journal",
3774                                  sbi->s_commit_interval / HZ);
3775                         goto failed_mount_wq;
3776                 }
3777                 if (EXT4_MOUNT_DATA_FLAGS &
3778                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3779                         ext4_msg(sb, KERN_ERR, "can't mount with "
3780                                  "data=, fs mounted w/o journal");
3781                         goto failed_mount_wq;
3782                 }
3783                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
3784                 clear_opt(sb, JOURNAL_CHECKSUM);
3785                 clear_opt(sb, DATA_FLAGS);
3786                 sbi->s_journal = NULL;
3787                 needs_recovery = 0;
3788                 goto no_journal;
3789         }
3790
3791         if (ext4_has_feature_64bit(sb) &&
3792             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3793                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3794                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3795                 goto failed_mount_wq;
3796         }
3797
3798         if (!set_journal_csum_feature_set(sb)) {
3799                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3800                          "feature set");
3801                 goto failed_mount_wq;
3802         }
3803
3804         /* We have now updated the journal if required, so we can
3805          * validate the data journaling mode. */
3806         switch (test_opt(sb, DATA_FLAGS)) {
3807         case 0:
3808                 /* No mode set, assume a default based on the journal
3809                  * capabilities: ORDERED_DATA if the journal can
3810                  * cope, else JOURNAL_DATA
3811                  */
3812                 if (jbd2_journal_check_available_features
3813                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3814                         set_opt(sb, ORDERED_DATA);
3815                 else
3816                         set_opt(sb, JOURNAL_DATA);
3817                 break;
3818
3819         case EXT4_MOUNT_ORDERED_DATA:
3820         case EXT4_MOUNT_WRITEBACK_DATA:
3821                 if (!jbd2_journal_check_available_features
3822                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3823                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3824                                "requested data journaling mode");
3825                         goto failed_mount_wq;
3826                 }
3827         default:
3828                 break;
3829         }
3830         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3831
3832         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3833
3834 no_journal:
3835         if (ext4_mballoc_ready) {
3836                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
3837                 if (!sbi->s_mb_cache) {
3838                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
3839                         goto failed_mount_wq;
3840                 }
3841         }
3842
3843         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
3844             (blocksize != PAGE_CACHE_SIZE)) {
3845                 ext4_msg(sb, KERN_ERR,
3846                          "Unsupported blocksize for fs encryption");
3847                 goto failed_mount_wq;
3848         }
3849
3850         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
3851             !ext4_has_feature_encrypt(sb)) {
3852                 ext4_set_feature_encrypt(sb);
3853                 ext4_commit_super(sb, 1);
3854         }
3855
3856         /*
3857          * Get the # of file system overhead blocks from the
3858          * superblock if present.
3859          */
3860         if (es->s_overhead_clusters)
3861                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3862         else {
3863                 err = ext4_calculate_overhead(sb);
3864                 if (err)
3865                         goto failed_mount_wq;
3866         }
3867
3868         /*
3869          * The maximum number of concurrent works can be high and
3870          * concurrency isn't really necessary.  Limit it to 1.
3871          */
3872         EXT4_SB(sb)->rsv_conversion_wq =
3873                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3874         if (!EXT4_SB(sb)->rsv_conversion_wq) {
3875                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3876                 ret = -ENOMEM;
3877                 goto failed_mount4;
3878         }
3879
3880         /*
3881          * The jbd2_journal_load will have done any necessary log recovery,
3882          * so we can safely mount the rest of the filesystem now.
3883          */
3884
3885         root = ext4_iget(sb, EXT4_ROOT_INO);
3886         if (IS_ERR(root)) {
3887                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3888                 ret = PTR_ERR(root);
3889                 root = NULL;
3890                 goto failed_mount4;
3891         }
3892         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3893                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3894                 iput(root);
3895                 goto failed_mount4;
3896         }
3897         sb->s_root = d_make_root(root);
3898         if (!sb->s_root) {
3899                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3900                 ret = -ENOMEM;
3901                 goto failed_mount4;
3902         }
3903
3904         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3905                 sb->s_flags |= MS_RDONLY;
3906
3907         /* determine the minimum size of new large inodes, if present */
3908         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3909                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3910                                                      EXT4_GOOD_OLD_INODE_SIZE;
3911                 if (ext4_has_feature_extra_isize(sb)) {
3912                         if (sbi->s_want_extra_isize <
3913                             le16_to_cpu(es->s_want_extra_isize))
3914                                 sbi->s_want_extra_isize =
3915                                         le16_to_cpu(es->s_want_extra_isize);
3916                         if (sbi->s_want_extra_isize <
3917                             le16_to_cpu(es->s_min_extra_isize))
3918                                 sbi->s_want_extra_isize =
3919                                         le16_to_cpu(es->s_min_extra_isize);
3920                 }
3921         }
3922         /* Check if enough inode space is available */
3923         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3924                                                         sbi->s_inode_size) {
3925                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3926                                                        EXT4_GOOD_OLD_INODE_SIZE;
3927                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3928                          "available");
3929         }
3930
3931         ext4_set_resv_clusters(sb);
3932
3933         err = ext4_setup_system_zone(sb);
3934         if (err) {
3935                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3936                          "zone (%d)", err);
3937                 goto failed_mount4a;
3938         }
3939
3940         ext4_ext_init(sb);
3941         err = ext4_mb_init(sb);
3942         if (err) {
3943                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3944                          err);
3945                 goto failed_mount5;
3946         }
3947
3948         block = ext4_count_free_clusters(sb);
3949         ext4_free_blocks_count_set(sbi->s_es,
3950                                    EXT4_C2B(sbi, block));
3951         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
3952                                   GFP_KERNEL);
3953         if (!err) {
3954                 unsigned long freei = ext4_count_free_inodes(sb);
3955                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
3956                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
3957                                           GFP_KERNEL);
3958         }
3959         if (!err)
3960                 err = percpu_counter_init(&sbi->s_dirs_counter,
3961                                           ext4_count_dirs(sb), GFP_KERNEL);
3962         if (!err)
3963                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
3964                                           GFP_KERNEL);
3965         if (err) {
3966                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3967                 goto failed_mount6;
3968         }
3969
3970         if (ext4_has_feature_flex_bg(sb))
3971                 if (!ext4_fill_flex_info(sb)) {
3972                         ext4_msg(sb, KERN_ERR,
3973                                "unable to initialize "
3974                                "flex_bg meta info!");
3975                         goto failed_mount6;
3976                 }
3977
3978         err = ext4_register_li_request(sb, first_not_zeroed);
3979         if (err)
3980                 goto failed_mount6;
3981
3982         err = ext4_register_sysfs(sb);
3983         if (err)
3984                 goto failed_mount7;
3985
3986 #ifdef CONFIG_QUOTA
3987         /* Enable quota usage during mount. */
3988         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
3989                 err = ext4_enable_quotas(sb);
3990                 if (err)
3991                         goto failed_mount8;
3992         }
3993 #endif  /* CONFIG_QUOTA */
3994
3995         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3996         ext4_orphan_cleanup(sb, es);
3997         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3998         if (needs_recovery) {
3999                 ext4_msg(sb, KERN_INFO, "recovery complete");
4000                 ext4_mark_recovery_complete(sb, es);
4001         }
4002         if (EXT4_SB(sb)->s_journal) {
4003                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4004                         descr = " journalled data mode";
4005                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4006                         descr = " ordered data mode";
4007                 else
4008                         descr = " writeback data mode";
4009         } else
4010                 descr = "out journal";
4011
4012         if (test_opt(sb, DISCARD)) {
4013                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4014                 if (!blk_queue_discard(q))
4015                         ext4_msg(sb, KERN_WARNING,
4016                                  "mounting with \"discard\" option, but "
4017                                  "the device does not support discard");
4018         }
4019
4020         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4021                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4022                          "Opts: %.*s%s%s", descr,
4023                          (int) sizeof(sbi->s_es->s_mount_opts),
4024                          sbi->s_es->s_mount_opts,
4025                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4026
4027         if (es->s_error_count)
4028                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4029
4030         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4031         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4032         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4033         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4034
4035         kfree(orig_data);
4036         return 0;
4037
4038 cantfind_ext4:
4039         if (!silent)
4040                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4041         goto failed_mount;
4042
4043 #ifdef CONFIG_QUOTA
4044 failed_mount8:
4045         ext4_unregister_sysfs(sb);
4046 #endif
4047 failed_mount7:
4048         ext4_unregister_li_request(sb);
4049 failed_mount6:
4050         ext4_mb_release(sb);
4051         if (sbi->s_flex_groups)
4052                 kvfree(sbi->s_flex_groups);
4053         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4054         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4055         percpu_counter_destroy(&sbi->s_dirs_counter);
4056         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4057 failed_mount5:
4058         ext4_ext_release(sb);
4059         ext4_release_system_zone(sb);
4060 failed_mount4a:
4061         dput(sb->s_root);
4062         sb->s_root = NULL;
4063 failed_mount4:
4064         ext4_msg(sb, KERN_ERR, "mount failed");
4065         if (EXT4_SB(sb)->rsv_conversion_wq)
4066                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4067 failed_mount_wq:
4068         if (sbi->s_journal) {
4069                 jbd2_journal_destroy(sbi->s_journal);
4070                 sbi->s_journal = NULL;
4071         }
4072 failed_mount3a:
4073         ext4_es_unregister_shrinker(sbi);
4074 failed_mount3:
4075         del_timer_sync(&sbi->s_err_report);
4076         if (sbi->s_mmp_tsk)
4077                 kthread_stop(sbi->s_mmp_tsk);
4078 failed_mount2:
4079         for (i = 0; i < db_count; i++)
4080                 brelse(sbi->s_group_desc[i]);
4081         kvfree(sbi->s_group_desc);
4082 failed_mount:
4083         if (sbi->s_chksum_driver)
4084                 crypto_free_shash(sbi->s_chksum_driver);
4085 #ifdef CONFIG_QUOTA
4086         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4087                 kfree(sbi->s_qf_names[i]);
4088 #endif
4089         ext4_blkdev_remove(sbi);
4090         brelse(bh);
4091 out_fail:
4092         sb->s_fs_info = NULL;
4093         kfree(sbi->s_blockgroup_lock);
4094 out_free_base:
4095         kfree(sbi);
4096         kfree(orig_data);
4097         return err ? err : ret;
4098 }
4099
4100 /*
4101  * Setup any per-fs journal parameters now.  We'll do this both on
4102  * initial mount, once the journal has been initialised but before we've
4103  * done any recovery; and again on any subsequent remount.
4104  */
4105 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4106 {
4107         struct ext4_sb_info *sbi = EXT4_SB(sb);
4108
4109         journal->j_commit_interval = sbi->s_commit_interval;
4110         journal->j_min_batch_time = sbi->s_min_batch_time;
4111         journal->j_max_batch_time = sbi->s_max_batch_time;
4112
4113         write_lock(&journal->j_state_lock);
4114         if (test_opt(sb, BARRIER))
4115                 journal->j_flags |= JBD2_BARRIER;
4116         else
4117                 journal->j_flags &= ~JBD2_BARRIER;
4118         if (test_opt(sb, DATA_ERR_ABORT))
4119                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4120         else
4121                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4122         write_unlock(&journal->j_state_lock);
4123 }
4124
4125 static journal_t *ext4_get_journal(struct super_block *sb,
4126                                    unsigned int journal_inum)
4127 {
4128         struct inode *journal_inode;
4129         journal_t *journal;
4130
4131         BUG_ON(!ext4_has_feature_journal(sb));
4132
4133         /* First, test for the existence of a valid inode on disk.  Bad
4134          * things happen if we iget() an unused inode, as the subsequent
4135          * iput() will try to delete it. */
4136
4137         journal_inode = ext4_iget(sb, journal_inum);
4138         if (IS_ERR(journal_inode)) {
4139                 ext4_msg(sb, KERN_ERR, "no journal found");
4140                 return NULL;
4141         }
4142         if (!journal_inode->i_nlink) {
4143                 make_bad_inode(journal_inode);
4144                 iput(journal_inode);
4145                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4146                 return NULL;
4147         }
4148
4149         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4150                   journal_inode, journal_inode->i_size);
4151         if (!S_ISREG(journal_inode->i_mode)) {
4152                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4153                 iput(journal_inode);
4154                 return NULL;
4155         }
4156
4157         journal = jbd2_journal_init_inode(journal_inode);
4158         if (!journal) {
4159                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4160                 iput(journal_inode);
4161                 return NULL;
4162         }
4163         journal->j_private = sb;
4164         ext4_init_journal_params(sb, journal);
4165         return journal;
4166 }
4167
4168 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4169                                        dev_t j_dev)
4170 {
4171         struct buffer_head *bh;
4172         journal_t *journal;
4173         ext4_fsblk_t start;
4174         ext4_fsblk_t len;
4175         int hblock, blocksize;
4176         ext4_fsblk_t sb_block;
4177         unsigned long offset;
4178         struct ext4_super_block *es;
4179         struct block_device *bdev;
4180
4181         BUG_ON(!ext4_has_feature_journal(sb));
4182
4183         bdev = ext4_blkdev_get(j_dev, sb);
4184         if (bdev == NULL)
4185                 return NULL;
4186
4187         blocksize = sb->s_blocksize;
4188         hblock = bdev_logical_block_size(bdev);
4189         if (blocksize < hblock) {
4190                 ext4_msg(sb, KERN_ERR,
4191                         "blocksize too small for journal device");
4192                 goto out_bdev;
4193         }
4194
4195         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4196         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4197         set_blocksize(bdev, blocksize);
4198         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4199                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4200                        "external journal");
4201                 goto out_bdev;
4202         }
4203
4204         es = (struct ext4_super_block *) (bh->b_data + offset);
4205         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4206             !(le32_to_cpu(es->s_feature_incompat) &
4207               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4208                 ext4_msg(sb, KERN_ERR, "external journal has "
4209                                         "bad superblock");
4210                 brelse(bh);
4211                 goto out_bdev;
4212         }
4213
4214         if ((le32_to_cpu(es->s_feature_ro_compat) &
4215              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4216             es->s_checksum != ext4_superblock_csum(sb, es)) {
4217                 ext4_msg(sb, KERN_ERR, "external journal has "
4218                                        "corrupt superblock");
4219                 brelse(bh);
4220                 goto out_bdev;
4221         }
4222
4223         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4224                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4225                 brelse(bh);
4226                 goto out_bdev;
4227         }
4228
4229         len = ext4_blocks_count(es);
4230         start = sb_block + 1;
4231         brelse(bh);     /* we're done with the superblock */
4232
4233         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4234                                         start, len, blocksize);
4235         if (!journal) {
4236                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4237                 goto out_bdev;
4238         }
4239         journal->j_private = sb;
4240         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4241         wait_on_buffer(journal->j_sb_buffer);
4242         if (!buffer_uptodate(journal->j_sb_buffer)) {
4243                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4244                 goto out_journal;
4245         }
4246         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4247                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4248                                         "user (unsupported) - %d",
4249                         be32_to_cpu(journal->j_superblock->s_nr_users));
4250                 goto out_journal;
4251         }
4252         EXT4_SB(sb)->journal_bdev = bdev;
4253         ext4_init_journal_params(sb, journal);
4254         return journal;
4255
4256 out_journal:
4257         jbd2_journal_destroy(journal);
4258 out_bdev:
4259         ext4_blkdev_put(bdev);
4260         return NULL;
4261 }
4262
4263 static int ext4_load_journal(struct super_block *sb,
4264                              struct ext4_super_block *es,
4265                              unsigned long journal_devnum)
4266 {
4267         journal_t *journal;
4268         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4269         dev_t journal_dev;
4270         int err = 0;
4271         int really_read_only;
4272
4273         BUG_ON(!ext4_has_feature_journal(sb));
4274
4275         if (journal_devnum &&
4276             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4277                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4278                         "numbers have changed");
4279                 journal_dev = new_decode_dev(journal_devnum);
4280         } else
4281                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4282
4283         really_read_only = bdev_read_only(sb->s_bdev);
4284
4285         /*
4286          * Are we loading a blank journal or performing recovery after a
4287          * crash?  For recovery, we need to check in advance whether we
4288          * can get read-write access to the device.
4289          */
4290         if (ext4_has_feature_journal_needs_recovery(sb)) {
4291                 if (sb->s_flags & MS_RDONLY) {
4292                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4293                                         "required on readonly filesystem");
4294                         if (really_read_only) {
4295                                 ext4_msg(sb, KERN_ERR, "write access "
4296                                         "unavailable, cannot proceed");
4297                                 return -EROFS;
4298                         }
4299                         ext4_msg(sb, KERN_INFO, "write access will "
4300                                "be enabled during recovery");
4301                 }
4302         }
4303
4304         if (journal_inum && journal_dev) {
4305                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4306                        "and inode journals!");
4307                 return -EINVAL;
4308         }
4309
4310         if (journal_inum) {
4311                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4312                         return -EINVAL;
4313         } else {
4314                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4315                         return -EINVAL;
4316         }
4317
4318         if (!(journal->j_flags & JBD2_BARRIER))
4319                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4320
4321         if (!ext4_has_feature_journal_needs_recovery(sb))
4322                 err = jbd2_journal_wipe(journal, !really_read_only);
4323         if (!err) {
4324                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4325                 if (save)
4326                         memcpy(save, ((char *) es) +
4327                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4328                 err = jbd2_journal_load(journal);
4329                 if (save)
4330                         memcpy(((char *) es) + EXT4_S_ERR_START,
4331                                save, EXT4_S_ERR_LEN);
4332                 kfree(save);
4333         }
4334
4335         if (err) {
4336                 ext4_msg(sb, KERN_ERR, "error loading journal");
4337                 jbd2_journal_destroy(journal);
4338                 return err;
4339         }
4340
4341         EXT4_SB(sb)->s_journal = journal;
4342         ext4_clear_journal_err(sb, es);
4343
4344         if (!really_read_only && journal_devnum &&
4345             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4346                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4347
4348                 /* Make sure we flush the recovery flag to disk. */
4349                 ext4_commit_super(sb, 1);
4350         }
4351
4352         return 0;
4353 }
4354
4355 static int ext4_commit_super(struct super_block *sb, int sync)
4356 {
4357         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4358         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4359         int error = 0;
4360
4361         if (!sbh || block_device_ejected(sb))
4362                 return error;
4363         if (buffer_write_io_error(sbh)) {
4364                 /*
4365                  * Oh, dear.  A previous attempt to write the
4366                  * superblock failed.  This could happen because the
4367                  * USB device was yanked out.  Or it could happen to
4368                  * be a transient write error and maybe the block will
4369                  * be remapped.  Nothing we can do but to retry the
4370                  * write and hope for the best.
4371                  */
4372                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4373                        "superblock detected");
4374                 clear_buffer_write_io_error(sbh);
4375                 set_buffer_uptodate(sbh);
4376         }
4377         /*
4378          * If the file system is mounted read-only, don't update the
4379          * superblock write time.  This avoids updating the superblock
4380          * write time when we are mounting the root file system
4381          * read/only but we need to replay the journal; at that point,
4382          * for people who are east of GMT and who make their clock
4383          * tick in localtime for Windows bug-for-bug compatibility,
4384          * the clock is set in the future, and this will cause e2fsck
4385          * to complain and force a full file system check.
4386          */
4387         if (!(sb->s_flags & MS_RDONLY))
4388                 es->s_wtime = cpu_to_le32(get_seconds());
4389         if (sb->s_bdev->bd_part)
4390                 es->s_kbytes_written =
4391                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4392                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4393                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4394         else
4395                 es->s_kbytes_written =
4396                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4397         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4398                 ext4_free_blocks_count_set(es,
4399                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4400                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4401         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4402                 es->s_free_inodes_count =
4403                         cpu_to_le32(percpu_counter_sum_positive(
4404                                 &EXT4_SB(sb)->s_freeinodes_counter));
4405         BUFFER_TRACE(sbh, "marking dirty");
4406         ext4_superblock_csum_set(sb);
4407         mark_buffer_dirty(sbh);
4408         if (sync) {
4409                 error = __sync_dirty_buffer(sbh,
4410                         test_opt(sb, BARRIER) ? WRITE_FUA : WRITE_SYNC);
4411                 if (error)
4412                         return error;
4413
4414                 error = buffer_write_io_error(sbh);
4415                 if (error) {
4416                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4417                                "superblock");
4418                         clear_buffer_write_io_error(sbh);
4419                         set_buffer_uptodate(sbh);
4420                 }
4421         }
4422         return error;
4423 }
4424
4425 /*
4426  * Have we just finished recovery?  If so, and if we are mounting (or
4427  * remounting) the filesystem readonly, then we will end up with a
4428  * consistent fs on disk.  Record that fact.
4429  */
4430 static void ext4_mark_recovery_complete(struct super_block *sb,
4431                                         struct ext4_super_block *es)
4432 {
4433         journal_t *journal = EXT4_SB(sb)->s_journal;
4434
4435         if (!ext4_has_feature_journal(sb)) {
4436                 BUG_ON(journal != NULL);
4437                 return;
4438         }
4439         jbd2_journal_lock_updates(journal);
4440         if (jbd2_journal_flush(journal) < 0)
4441                 goto out;
4442
4443         if (ext4_has_feature_journal_needs_recovery(sb) &&
4444             sb->s_flags & MS_RDONLY) {
4445                 ext4_clear_feature_journal_needs_recovery(sb);
4446                 ext4_commit_super(sb, 1);
4447         }
4448
4449 out:
4450         jbd2_journal_unlock_updates(journal);
4451 }
4452
4453 /*
4454  * If we are mounting (or read-write remounting) a filesystem whose journal
4455  * has recorded an error from a previous lifetime, move that error to the
4456  * main filesystem now.
4457  */
4458 static void ext4_clear_journal_err(struct super_block *sb,
4459                                    struct ext4_super_block *es)
4460 {
4461         journal_t *journal;
4462         int j_errno;
4463         const char *errstr;
4464
4465         BUG_ON(!ext4_has_feature_journal(sb));
4466
4467         journal = EXT4_SB(sb)->s_journal;
4468
4469         /*
4470          * Now check for any error status which may have been recorded in the
4471          * journal by a prior ext4_error() or ext4_abort()
4472          */
4473
4474         j_errno = jbd2_journal_errno(journal);
4475         if (j_errno) {
4476                 char nbuf[16];
4477
4478                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4479                 ext4_warning(sb, "Filesystem error recorded "
4480                              "from previous mount: %s", errstr);
4481                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4482
4483                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4484                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4485                 ext4_commit_super(sb, 1);
4486
4487                 jbd2_journal_clear_err(journal);
4488                 jbd2_journal_update_sb_errno(journal);
4489         }
4490 }
4491
4492 /*
4493  * Force the running and committing transactions to commit,
4494  * and wait on the commit.
4495  */
4496 int ext4_force_commit(struct super_block *sb)
4497 {
4498         journal_t *journal;
4499
4500         if (sb->s_flags & MS_RDONLY)
4501                 return 0;
4502
4503         journal = EXT4_SB(sb)->s_journal;
4504         return ext4_journal_force_commit(journal);
4505 }
4506
4507 static int ext4_sync_fs(struct super_block *sb, int wait)
4508 {
4509         int ret = 0;
4510         tid_t target;
4511         bool needs_barrier = false;
4512         struct ext4_sb_info *sbi = EXT4_SB(sb);
4513
4514         trace_ext4_sync_fs(sb, wait);
4515         flush_workqueue(sbi->rsv_conversion_wq);
4516         /*
4517          * Writeback quota in non-journalled quota case - journalled quota has
4518          * no dirty dquots
4519          */
4520         dquot_writeback_dquots(sb, -1);
4521         /*
4522          * Data writeback is possible w/o journal transaction, so barrier must
4523          * being sent at the end of the function. But we can skip it if
4524          * transaction_commit will do it for us.
4525          */
4526         if (sbi->s_journal) {
4527                 target = jbd2_get_latest_transaction(sbi->s_journal);
4528                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4529                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4530                         needs_barrier = true;
4531
4532                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4533                         if (wait)
4534                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4535                                                            target);
4536                 }
4537         } else if (wait && test_opt(sb, BARRIER))
4538                 needs_barrier = true;
4539         if (needs_barrier) {
4540                 int err;
4541                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4542                 if (!ret)
4543                         ret = err;
4544         }
4545
4546         return ret;
4547 }
4548
4549 /*
4550  * LVM calls this function before a (read-only) snapshot is created.  This
4551  * gives us a chance to flush the journal completely and mark the fs clean.
4552  *
4553  * Note that only this function cannot bring a filesystem to be in a clean
4554  * state independently. It relies on upper layer to stop all data & metadata
4555  * modifications.
4556  */
4557 static int ext4_freeze(struct super_block *sb)
4558 {
4559         int error = 0;
4560         journal_t *journal;
4561
4562         if (sb->s_flags & MS_RDONLY)
4563                 return 0;
4564
4565         journal = EXT4_SB(sb)->s_journal;
4566
4567         if (journal) {
4568                 /* Now we set up the journal barrier. */
4569                 jbd2_journal_lock_updates(journal);
4570
4571                 /*
4572                  * Don't clear the needs_recovery flag if we failed to
4573                  * flush the journal.
4574                  */
4575                 error = jbd2_journal_flush(journal);
4576                 if (error < 0)
4577                         goto out;
4578
4579                 /* Journal blocked and flushed, clear needs_recovery flag. */
4580                 ext4_clear_feature_journal_needs_recovery(sb);
4581         }
4582
4583         error = ext4_commit_super(sb, 1);
4584 out:
4585         if (journal)
4586                 /* we rely on upper layer to stop further updates */
4587                 jbd2_journal_unlock_updates(journal);
4588         return error;
4589 }
4590
4591 /*
4592  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4593  * flag here, even though the filesystem is not technically dirty yet.
4594  */
4595 static int ext4_unfreeze(struct super_block *sb)
4596 {
4597         if (sb->s_flags & MS_RDONLY)
4598                 return 0;
4599
4600         if (EXT4_SB(sb)->s_journal) {
4601                 /* Reset the needs_recovery flag before the fs is unlocked. */
4602                 ext4_set_feature_journal_needs_recovery(sb);
4603         }
4604
4605         ext4_commit_super(sb, 1);
4606         return 0;
4607 }
4608
4609 /*
4610  * Structure to save mount options for ext4_remount's benefit
4611  */
4612 struct ext4_mount_options {
4613         unsigned long s_mount_opt;
4614         unsigned long s_mount_opt2;
4615         kuid_t s_resuid;
4616         kgid_t s_resgid;
4617         unsigned long s_commit_interval;
4618         u32 s_min_batch_time, s_max_batch_time;
4619 #ifdef CONFIG_QUOTA
4620         int s_jquota_fmt;
4621         char *s_qf_names[EXT4_MAXQUOTAS];
4622 #endif
4623 };
4624
4625 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4626 {
4627         struct ext4_super_block *es;
4628         struct ext4_sb_info *sbi = EXT4_SB(sb);
4629         unsigned long old_sb_flags;
4630         struct ext4_mount_options old_opts;
4631         int enable_quota = 0;
4632         ext4_group_t g;
4633         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4634         int err = 0;
4635 #ifdef CONFIG_QUOTA
4636         int i, j;
4637 #endif
4638         char *orig_data = kstrdup(data, GFP_KERNEL);
4639
4640         /* Store the original options */
4641         old_sb_flags = sb->s_flags;
4642         old_opts.s_mount_opt = sbi->s_mount_opt;
4643         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4644         old_opts.s_resuid = sbi->s_resuid;
4645         old_opts.s_resgid = sbi->s_resgid;
4646         old_opts.s_commit_interval = sbi->s_commit_interval;
4647         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4648         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4649 #ifdef CONFIG_QUOTA
4650         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4651         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4652                 if (sbi->s_qf_names[i]) {
4653                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4654                                                          GFP_KERNEL);
4655                         if (!old_opts.s_qf_names[i]) {
4656                                 for (j = 0; j < i; j++)
4657                                         kfree(old_opts.s_qf_names[j]);
4658                                 kfree(orig_data);
4659                                 return -ENOMEM;
4660                         }
4661                 } else
4662                         old_opts.s_qf_names[i] = NULL;
4663 #endif
4664         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4665                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4666
4667         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4668                 err = -EINVAL;
4669                 goto restore_opts;
4670         }
4671
4672         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4673             test_opt(sb, JOURNAL_CHECKSUM)) {
4674                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4675                          "during remount not supported; ignoring");
4676                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4677         }
4678
4679         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4680                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4681                         ext4_msg(sb, KERN_ERR, "can't mount with "
4682                                  "both data=journal and delalloc");
4683                         err = -EINVAL;
4684                         goto restore_opts;
4685                 }
4686                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4687                         ext4_msg(sb, KERN_ERR, "can't mount with "
4688                                  "both data=journal and dioread_nolock");
4689                         err = -EINVAL;
4690                         goto restore_opts;
4691                 }
4692                 if (test_opt(sb, DAX)) {
4693                         ext4_msg(sb, KERN_ERR, "can't mount with "
4694                                  "both data=journal and dax");
4695                         err = -EINVAL;
4696                         goto restore_opts;
4697                 }
4698         }
4699
4700         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4701                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4702                         "dax flag with busy inodes while remounting");
4703                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4704         }
4705
4706         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4707                 ext4_abort(sb, "Abort forced by user");
4708
4709         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4710                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4711
4712         es = sbi->s_es;
4713
4714         if (sbi->s_journal) {
4715                 ext4_init_journal_params(sb, sbi->s_journal);
4716                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4717         }
4718
4719         if (*flags & MS_LAZYTIME)
4720                 sb->s_flags |= MS_LAZYTIME;
4721
4722         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4723                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4724                         err = -EROFS;
4725                         goto restore_opts;
4726                 }
4727
4728                 if (*flags & MS_RDONLY) {
4729                         err = sync_filesystem(sb);
4730                         if (err < 0)
4731                                 goto restore_opts;
4732                         err = dquot_suspend(sb, -1);
4733                         if (err < 0)
4734                                 goto restore_opts;
4735
4736                         /*
4737                          * First of all, the unconditional stuff we have to do
4738                          * to disable replay of the journal when we next remount
4739                          */
4740                         sb->s_flags |= MS_RDONLY;
4741
4742                         /*
4743                          * OK, test if we are remounting a valid rw partition
4744                          * readonly, and if so set the rdonly flag and then
4745                          * mark the partition as valid again.
4746                          */
4747                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4748                             (sbi->s_mount_state & EXT4_VALID_FS))
4749                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4750
4751                         if (sbi->s_journal)
4752                                 ext4_mark_recovery_complete(sb, es);
4753                 } else {
4754                         /* Make sure we can mount this feature set readwrite */
4755                         if (ext4_has_feature_readonly(sb) ||
4756                             !ext4_feature_set_ok(sb, 0)) {
4757                                 err = -EROFS;
4758                                 goto restore_opts;
4759                         }
4760                         /*
4761                          * Make sure the group descriptor checksums
4762                          * are sane.  If they aren't, refuse to remount r/w.
4763                          */
4764                         for (g = 0; g < sbi->s_groups_count; g++) {
4765                                 struct ext4_group_desc *gdp =
4766                                         ext4_get_group_desc(sb, g, NULL);
4767
4768                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4769                                         ext4_msg(sb, KERN_ERR,
4770                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4771                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
4772                                                le16_to_cpu(gdp->bg_checksum));
4773                                         err = -EFSBADCRC;
4774                                         goto restore_opts;
4775                                 }
4776                         }
4777
4778                         /*
4779                          * If we have an unprocessed orphan list hanging
4780                          * around from a previously readonly bdev mount,
4781                          * require a full umount/remount for now.
4782                          */
4783                         if (es->s_last_orphan) {
4784                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4785                                        "remount RDWR because of unprocessed "
4786                                        "orphan inode list.  Please "
4787                                        "umount/remount instead");
4788                                 err = -EINVAL;
4789                                 goto restore_opts;
4790                         }
4791
4792                         /*
4793                          * Mounting a RDONLY partition read-write, so reread
4794                          * and store the current valid flag.  (It may have
4795                          * been changed by e2fsck since we originally mounted
4796                          * the partition.)
4797                          */
4798                         if (sbi->s_journal)
4799                                 ext4_clear_journal_err(sb, es);
4800                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4801                         if (!ext4_setup_super(sb, es, 0))
4802                                 sb->s_flags &= ~MS_RDONLY;
4803                         if (ext4_has_feature_mmp(sb))
4804                                 if (ext4_multi_mount_protect(sb,
4805                                                 le64_to_cpu(es->s_mmp_block))) {
4806                                         err = -EROFS;
4807                                         goto restore_opts;
4808                                 }
4809                         enable_quota = 1;
4810                 }
4811         }
4812
4813         /*
4814          * Reinitialize lazy itable initialization thread based on
4815          * current settings
4816          */
4817         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4818                 ext4_unregister_li_request(sb);
4819         else {
4820                 ext4_group_t first_not_zeroed;
4821                 first_not_zeroed = ext4_has_uninit_itable(sb);
4822                 ext4_register_li_request(sb, first_not_zeroed);
4823         }
4824
4825         ext4_setup_system_zone(sb);
4826         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4827                 ext4_commit_super(sb, 1);
4828
4829 #ifdef CONFIG_QUOTA
4830         /* Release old quota file names */
4831         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4832                 kfree(old_opts.s_qf_names[i]);
4833         if (enable_quota) {
4834                 if (sb_any_quota_suspended(sb))
4835                         dquot_resume(sb, -1);
4836                 else if (ext4_has_feature_quota(sb)) {
4837                         err = ext4_enable_quotas(sb);
4838                         if (err)
4839                                 goto restore_opts;
4840                 }
4841         }
4842 #endif
4843
4844         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
4845         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4846         kfree(orig_data);
4847         return 0;
4848
4849 restore_opts:
4850         sb->s_flags = old_sb_flags;
4851         sbi->s_mount_opt = old_opts.s_mount_opt;
4852         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4853         sbi->s_resuid = old_opts.s_resuid;
4854         sbi->s_resgid = old_opts.s_resgid;
4855         sbi->s_commit_interval = old_opts.s_commit_interval;
4856         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4857         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4858 #ifdef CONFIG_QUOTA
4859         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4860         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
4861                 kfree(sbi->s_qf_names[i]);
4862                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4863         }
4864 #endif
4865         kfree(orig_data);
4866         return err;
4867 }
4868
4869 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4870 {
4871         struct super_block *sb = dentry->d_sb;
4872         struct ext4_sb_info *sbi = EXT4_SB(sb);
4873         struct ext4_super_block *es = sbi->s_es;
4874         ext4_fsblk_t overhead = 0, resv_blocks;
4875         u64 fsid;
4876         s64 bfree;
4877         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4878
4879         if (!test_opt(sb, MINIX_DF))
4880                 overhead = sbi->s_overhead;
4881
4882         buf->f_type = EXT4_SUPER_MAGIC;
4883         buf->f_bsize = sb->s_blocksize;
4884         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4885         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4886                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4887         /* prevent underflow in case that few free space is available */
4888         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4889         buf->f_bavail = buf->f_bfree -
4890                         (ext4_r_blocks_count(es) + resv_blocks);
4891         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4892                 buf->f_bavail = 0;
4893         buf->f_files = le32_to_cpu(es->s_inodes_count);
4894         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4895         buf->f_namelen = EXT4_NAME_LEN;
4896         fsid = le64_to_cpup((void *)es->s_uuid) ^
4897                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4898         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4899         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4900
4901         return 0;
4902 }
4903
4904 /* Helper function for writing quotas on sync - we need to start transaction
4905  * before quota file is locked for write. Otherwise the are possible deadlocks:
4906  * Process 1                         Process 2
4907  * ext4_create()                     quota_sync()
4908  *   jbd2_journal_start()                  write_dquot()
4909  *   dquot_initialize()                         down(dqio_mutex)
4910  *     down(dqio_mutex)                    jbd2_journal_start()
4911  *
4912  */
4913
4914 #ifdef CONFIG_QUOTA
4915
4916 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4917 {
4918         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4919 }
4920
4921 static int ext4_write_dquot(struct dquot *dquot)
4922 {
4923         int ret, err;
4924         handle_t *handle;
4925         struct inode *inode;
4926
4927         inode = dquot_to_inode(dquot);
4928         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4929                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4930         if (IS_ERR(handle))
4931                 return PTR_ERR(handle);
4932         ret = dquot_commit(dquot);
4933         err = ext4_journal_stop(handle);
4934         if (!ret)
4935                 ret = err;
4936         return ret;
4937 }
4938
4939 static int ext4_acquire_dquot(struct dquot *dquot)
4940 {
4941         int ret, err;
4942         handle_t *handle;
4943
4944         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4945                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4946         if (IS_ERR(handle))
4947                 return PTR_ERR(handle);
4948         ret = dquot_acquire(dquot);
4949         err = ext4_journal_stop(handle);
4950         if (!ret)
4951                 ret = err;
4952         return ret;
4953 }
4954
4955 static int ext4_release_dquot(struct dquot *dquot)
4956 {
4957         int ret, err;
4958         handle_t *handle;
4959
4960         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4961                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4962         if (IS_ERR(handle)) {
4963                 /* Release dquot anyway to avoid endless cycle in dqput() */
4964                 dquot_release(dquot);
4965                 return PTR_ERR(handle);
4966         }
4967         ret = dquot_release(dquot);
4968         err = ext4_journal_stop(handle);
4969         if (!ret)
4970                 ret = err;
4971         return ret;
4972 }
4973
4974 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4975 {
4976         struct super_block *sb = dquot->dq_sb;
4977         struct ext4_sb_info *sbi = EXT4_SB(sb);
4978
4979         /* Are we journaling quotas? */
4980         if (ext4_has_feature_quota(sb) ||
4981             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4982                 dquot_mark_dquot_dirty(dquot);
4983                 return ext4_write_dquot(dquot);
4984         } else {
4985                 return dquot_mark_dquot_dirty(dquot);
4986         }
4987 }
4988
4989 static int ext4_write_info(struct super_block *sb, int type)
4990 {
4991         int ret, err;
4992         handle_t *handle;
4993
4994         /* Data block + inode block */
4995         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
4996         if (IS_ERR(handle))
4997                 return PTR_ERR(handle);
4998         ret = dquot_commit_info(sb, type);
4999         err = ext4_journal_stop(handle);
5000         if (!ret)
5001                 ret = err;
5002         return ret;
5003 }
5004
5005 /*
5006  * Turn on quotas during mount time - we need to find
5007  * the quota file and such...
5008  */
5009 static int ext4_quota_on_mount(struct super_block *sb, int type)
5010 {
5011         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5012                                         EXT4_SB(sb)->s_jquota_fmt, type);
5013 }
5014
5015 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5016 {
5017         struct ext4_inode_info *ei = EXT4_I(inode);
5018
5019         /* The first argument of lockdep_set_subclass has to be
5020          * *exactly* the same as the argument to init_rwsem() --- in
5021          * this case, in init_once() --- or lockdep gets unhappy
5022          * because the name of the lock is set using the
5023          * stringification of the argument to init_rwsem().
5024          */
5025         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5026         lockdep_set_subclass(&ei->i_data_sem, subclass);
5027 }
5028
5029 /*
5030  * Standard function to be called on quota_on
5031  */
5032 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5033                          struct path *path)
5034 {
5035         int err;
5036
5037         if (!test_opt(sb, QUOTA))
5038                 return -EINVAL;
5039
5040         /* Quotafile not on the same filesystem? */
5041         if (path->dentry->d_sb != sb)
5042                 return -EXDEV;
5043         /* Journaling quota? */
5044         if (EXT4_SB(sb)->s_qf_names[type]) {
5045                 /* Quotafile not in fs root? */
5046                 if (path->dentry->d_parent != sb->s_root)
5047                         ext4_msg(sb, KERN_WARNING,
5048                                 "Quota file not on filesystem root. "
5049                                 "Journaled quota will not work");
5050         }
5051
5052         /*
5053          * When we journal data on quota file, we have to flush journal to see
5054          * all updates to the file when we bypass pagecache...
5055          */
5056         if (EXT4_SB(sb)->s_journal &&
5057             ext4_should_journal_data(d_inode(path->dentry))) {
5058                 /*
5059                  * We don't need to lock updates but journal_flush() could
5060                  * otherwise be livelocked...
5061                  */
5062                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5063                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5064                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5065                 if (err)
5066                         return err;
5067         }
5068         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5069         err = dquot_quota_on(sb, type, format_id, path);
5070         if (err)
5071                 lockdep_set_quota_inode(path->dentry->d_inode,
5072                                              I_DATA_SEM_NORMAL);
5073         return err;
5074 }
5075
5076 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5077                              unsigned int flags)
5078 {
5079         int err;
5080         struct inode *qf_inode;
5081         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5082                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5083                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5084         };
5085
5086         BUG_ON(!ext4_has_feature_quota(sb));
5087
5088         if (!qf_inums[type])
5089                 return -EPERM;
5090
5091         qf_inode = ext4_iget(sb, qf_inums[type]);
5092         if (IS_ERR(qf_inode)) {
5093                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5094                 return PTR_ERR(qf_inode);
5095         }
5096
5097         /* Don't account quota for quota files to avoid recursion */
5098         qf_inode->i_flags |= S_NOQUOTA;
5099         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5100         err = dquot_enable(qf_inode, type, format_id, flags);
5101         iput(qf_inode);
5102         if (err)
5103                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5104
5105         return err;
5106 }
5107
5108 /* Enable usage tracking for all quota types. */
5109 static int ext4_enable_quotas(struct super_block *sb)
5110 {
5111         int type, err = 0;
5112         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5113                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5114                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5115         };
5116
5117         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5118         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5119                 if (qf_inums[type]) {
5120                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5121                                                 DQUOT_USAGE_ENABLED);
5122                         if (err) {
5123                                 ext4_warning(sb,
5124                                         "Failed to enable quota tracking "
5125                                         "(type=%d, err=%d). Please run "
5126                                         "e2fsck to fix.", type, err);
5127                                 return err;
5128                         }
5129                 }
5130         }
5131         return 0;
5132 }
5133
5134 static int ext4_quota_off(struct super_block *sb, int type)
5135 {
5136         struct inode *inode = sb_dqopt(sb)->files[type];
5137         handle_t *handle;
5138
5139         /* Force all delayed allocation blocks to be allocated.
5140          * Caller already holds s_umount sem */
5141         if (test_opt(sb, DELALLOC))
5142                 sync_filesystem(sb);
5143
5144         if (!inode)
5145                 goto out;
5146
5147         /* Update modification times of quota files when userspace can
5148          * start looking at them */
5149         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5150         if (IS_ERR(handle))
5151                 goto out;
5152         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5153         ext4_mark_inode_dirty(handle, inode);
5154         ext4_journal_stop(handle);
5155
5156 out:
5157         return dquot_quota_off(sb, type);
5158 }
5159
5160 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5161  * acquiring the locks... As quota files are never truncated and quota code
5162  * itself serializes the operations (and no one else should touch the files)
5163  * we don't have to be afraid of races */
5164 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5165                                size_t len, loff_t off)
5166 {
5167         struct inode *inode = sb_dqopt(sb)->files[type];
5168         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5169         int offset = off & (sb->s_blocksize - 1);
5170         int tocopy;
5171         size_t toread;
5172         struct buffer_head *bh;
5173         loff_t i_size = i_size_read(inode);
5174
5175         if (off > i_size)
5176                 return 0;
5177         if (off+len > i_size)
5178                 len = i_size-off;
5179         toread = len;
5180         while (toread > 0) {
5181                 tocopy = sb->s_blocksize - offset < toread ?
5182                                 sb->s_blocksize - offset : toread;
5183                 bh = ext4_bread(NULL, inode, blk, 0);
5184                 if (IS_ERR(bh))
5185                         return PTR_ERR(bh);
5186                 if (!bh)        /* A hole? */
5187                         memset(data, 0, tocopy);
5188                 else
5189                         memcpy(data, bh->b_data+offset, tocopy);
5190                 brelse(bh);
5191                 offset = 0;
5192                 toread -= tocopy;
5193                 data += tocopy;
5194                 blk++;
5195         }
5196         return len;
5197 }
5198
5199 /* Write to quotafile (we know the transaction is already started and has
5200  * enough credits) */
5201 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5202                                 const char *data, size_t len, loff_t off)
5203 {
5204         struct inode *inode = sb_dqopt(sb)->files[type];
5205         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5206         int err, offset = off & (sb->s_blocksize - 1);
5207         int retries = 0;
5208         struct buffer_head *bh;
5209         handle_t *handle = journal_current_handle();
5210
5211         if (EXT4_SB(sb)->s_journal && !handle) {
5212                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5213                         " cancelled because transaction is not started",
5214                         (unsigned long long)off, (unsigned long long)len);
5215                 return -EIO;
5216         }
5217         /*
5218          * Since we account only one data block in transaction credits,
5219          * then it is impossible to cross a block boundary.
5220          */
5221         if (sb->s_blocksize - offset < len) {
5222                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5223                         " cancelled because not block aligned",
5224                         (unsigned long long)off, (unsigned long long)len);
5225                 return -EIO;
5226         }
5227
5228         do {
5229                 bh = ext4_bread(handle, inode, blk,
5230                                 EXT4_GET_BLOCKS_CREATE |
5231                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5232         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5233                  ext4_should_retry_alloc(inode->i_sb, &retries));
5234         if (IS_ERR(bh))
5235                 return PTR_ERR(bh);
5236         if (!bh)
5237                 goto out;
5238         BUFFER_TRACE(bh, "get write access");
5239         err = ext4_journal_get_write_access(handle, bh);
5240         if (err) {
5241                 brelse(bh);
5242                 return err;
5243         }
5244         lock_buffer(bh);
5245         memcpy(bh->b_data+offset, data, len);
5246         flush_dcache_page(bh->b_page);
5247         unlock_buffer(bh);
5248         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5249         brelse(bh);
5250 out:
5251         if (inode->i_size < off + len) {
5252                 i_size_write(inode, off + len);
5253                 EXT4_I(inode)->i_disksize = inode->i_size;
5254                 ext4_mark_inode_dirty(handle, inode);
5255         }
5256         return len;
5257 }
5258
5259 #endif
5260
5261 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5262                        const char *dev_name, void *data)
5263 {
5264         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5265 }
5266
5267 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5268 static inline void register_as_ext2(void)
5269 {
5270         int err = register_filesystem(&ext2_fs_type);
5271         if (err)
5272                 printk(KERN_WARNING
5273                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5274 }
5275
5276 static inline void unregister_as_ext2(void)
5277 {
5278         unregister_filesystem(&ext2_fs_type);
5279 }
5280
5281 static inline int ext2_feature_set_ok(struct super_block *sb)
5282 {
5283         if (ext4_has_unknown_ext2_incompat_features(sb))
5284                 return 0;
5285         if (sb->s_flags & MS_RDONLY)
5286                 return 1;
5287         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5288                 return 0;
5289         return 1;
5290 }
5291 #else
5292 static inline void register_as_ext2(void) { }
5293 static inline void unregister_as_ext2(void) { }
5294 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5295 #endif
5296
5297 static inline void register_as_ext3(void)
5298 {
5299         int err = register_filesystem(&ext3_fs_type);
5300         if (err)
5301                 printk(KERN_WARNING
5302                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5303 }
5304
5305 static inline void unregister_as_ext3(void)
5306 {
5307         unregister_filesystem(&ext3_fs_type);
5308 }
5309
5310 static inline int ext3_feature_set_ok(struct super_block *sb)
5311 {
5312         if (ext4_has_unknown_ext3_incompat_features(sb))
5313                 return 0;
5314         if (!ext4_has_feature_journal(sb))
5315                 return 0;
5316         if (sb->s_flags & MS_RDONLY)
5317                 return 1;
5318         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5319                 return 0;
5320         return 1;
5321 }
5322
5323 static struct file_system_type ext4_fs_type = {
5324         .owner          = THIS_MODULE,
5325         .name           = "ext4",
5326         .mount          = ext4_mount,
5327         .kill_sb        = kill_block_super,
5328         .fs_flags       = FS_REQUIRES_DEV,
5329 };
5330 MODULE_ALIAS_FS("ext4");
5331
5332 /* Shared across all ext4 file systems */
5333 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5334 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5335
5336 static int __init ext4_init_fs(void)
5337 {
5338         int i, err;
5339
5340         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5341         ext4_li_info = NULL;
5342         mutex_init(&ext4_li_mtx);
5343
5344         /* Build-time check for flags consistency */
5345         ext4_check_flag_values();
5346
5347         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5348                 mutex_init(&ext4__aio_mutex[i]);
5349                 init_waitqueue_head(&ext4__ioend_wq[i]);
5350         }
5351
5352         err = ext4_init_es();
5353         if (err)
5354                 return err;
5355
5356         err = ext4_init_pageio();
5357         if (err)
5358                 goto out5;
5359
5360         err = ext4_init_system_zone();
5361         if (err)
5362                 goto out4;
5363
5364         err = ext4_init_sysfs();
5365         if (err)
5366                 goto out3;
5367
5368         err = ext4_init_mballoc();
5369         if (err)
5370                 goto out2;
5371         else
5372                 ext4_mballoc_ready = 1;
5373         err = init_inodecache();
5374         if (err)
5375                 goto out1;
5376         register_as_ext3();
5377         register_as_ext2();
5378         err = register_filesystem(&ext4_fs_type);
5379         if (err)
5380                 goto out;
5381
5382         return 0;
5383 out:
5384         unregister_as_ext2();
5385         unregister_as_ext3();
5386         destroy_inodecache();
5387 out1:
5388         ext4_mballoc_ready = 0;
5389         ext4_exit_mballoc();
5390 out2:
5391         ext4_exit_sysfs();
5392 out3:
5393         ext4_exit_system_zone();
5394 out4:
5395         ext4_exit_pageio();
5396 out5:
5397         ext4_exit_es();
5398
5399         return err;
5400 }
5401
5402 static void __exit ext4_exit_fs(void)
5403 {
5404         ext4_exit_crypto();
5405         ext4_destroy_lazyinit_thread();
5406         unregister_as_ext2();
5407         unregister_as_ext3();
5408         unregister_filesystem(&ext4_fs_type);
5409         destroy_inodecache();
5410         ext4_exit_mballoc();
5411         ext4_exit_sysfs();
5412         ext4_exit_system_zone();
5413         ext4_exit_pageio();
5414         ext4_exit_es();
5415 }
5416
5417 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5418 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5419 MODULE_LICENSE("GPL");
5420 module_init(ext4_init_fs)
5421 module_exit(ext4_exit_fs)