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_commit_super(sb, 1);
 692         ext4_handle_error(sb);
 693         /*
 694          * We only get here in the ERRORS_RO case; relocking the group
 695          * may be dangerous, but nothing bad will happen since the
 696          * filesystem will have already been marked read/only and the
 697          * journal has been aborted.  We return 1 as a hint to callers
 698          * who might what to use the return value from
 699          * ext4_grp_locked_error() to distinguish between the
 700          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 701          * aggressively from the ext4 function in question, with a
 702          * more appropriate error code.
 703          */
 704         ext4_lock_group(sb, grp);
 705         return;
 706 }
 707
 708 void ext4_update_dynamic_rev(struct super_block *sb)
 709 {
 710         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 711
 712         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 713                 return;
 714
 715         ext4_warning(sb,
 716                      "updating to rev %d because of new feature flag, "
 717                      "running e2fsck is recommended",
 718                      EXT4_DYNAMIC_REV);
 719
 720         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 721         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 722         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 723         /* leave es->s_feature_*compat flags alone */
 724         /* es->s_uuid will be set by e2fsck if empty */
 725
 726         /*
 727          * The rest of the superblock fields should be zero, and if not it
 728          * means they are likely already in use, so leave them alone.  We
 729          * can leave it up to e2fsck to clean up any inconsistencies there.
 730          */
 731 }
 732
 733 /*
 734  * Open the external journal device
 735  */
 736 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 737 {
 738         struct block_device *bdev;
 739         char b[BDEVNAME_SIZE];
 740
 741         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 742         if (IS_ERR(bdev))
 743                 goto fail;
 744         return bdev;
 745
 746 fail:
 747         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 748                         __bdevname(dev, b), PTR_ERR(bdev));
 749         return NULL;
 750 }
 751
 752 /*
 753  * Release the journal device
 754  */
 755 static void ext4_blkdev_put(struct block_device *bdev)
 756 {
 757         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 758 }
 759
 760 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 761 {
 762         struct block_device *bdev;
 763         bdev = sbi->journal_bdev;
 764         if (bdev) {
 765                 ext4_blkdev_put(bdev);
 766                 sbi->journal_bdev = NULL;
 767         }
 768 }
 769
 770 static inline struct inode *orphan_list_entry(struct list_head *l)
 771 {
 772         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 773 }
 774
 775 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 776 {
 777         struct list_head *l;
 778
 779         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 780                  le32_to_cpu(sbi->s_es->s_last_orphan));
 781
 782         printk(KERN_ERR "sb_info orphan list:\n");
 783         list_for_each(l, &sbi->s_orphan) {
 784                 struct inode *inode = orphan_list_entry(l);
 785                 printk(KERN_ERR "  "
 786                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 787                        inode->i_sb->s_id, inode->i_ino, inode,
 788                        inode->i_mode, inode->i_nlink,
 789                        NEXT_ORPHAN(inode));
 790         }
 791 }
 792
 793 static void ext4_put_super(struct super_block *sb)
 794 {
 795         struct ext4_sb_info *sbi = EXT4_SB(sb);
 796         struct ext4_super_block *es = sbi->s_es;
 797         int aborted = 0;
 798         int i, err;
 799
 800         ext4_unregister_li_request(sb);
 801         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 802
 803         flush_workqueue(sbi->rsv_conversion_wq);
 804         destroy_workqueue(sbi->rsv_conversion_wq);
 805
 806         if (sbi->s_journal) {
 807                 aborted = is_journal_aborted(sbi->s_journal);
 808                 err = jbd2_journal_destroy(sbi->s_journal);
 809                 sbi->s_journal = NULL;
 810                 if ((err < 0) && !aborted)
 811                         ext4_abort(sb, "Couldn't clean up the journal");
 812         }
 813
 814         ext4_unregister_sysfs(sb);
 815         ext4_es_unregister_shrinker(sbi);
 816         del_timer_sync(&sbi->s_err_report);
 817         ext4_release_system_zone(sb);
 818         ext4_mb_release(sb);
 819         ext4_ext_release(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 quota_update = 0;
2209         int i;
2210 #endif
2211         if (!es->s_last_orphan) {
2212                 jbd_debug(4, "no orphan inodes to clean up\n");
2213                 return;
2214         }
2215
2216         if (bdev_read_only(sb->s_bdev)) {
2217                 ext4_msg(sb, KERN_ERR, "write access "
2218                         "unavailable, skipping orphan cleanup");
2219                 return;
2220         }
2221
2222         /* Check if feature set would not allow a r/w mount */
2223         if (!ext4_feature_set_ok(sb, 0)) {
2224                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2225                          "unknown ROCOMPAT features");
2226                 return;
2227         }
2228
2229         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2230                 /* don't clear list on RO mount w/ errors */
2231                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2232                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2233                                   "clearing orphan list.\n");
2234                         es->s_last_orphan = 0;
2235                 }
2236                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2237                 return;
2238         }
2239
2240         if (s_flags & MS_RDONLY) {
2241                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2242                 sb->s_flags &= ~MS_RDONLY;
2243         }
2244 #ifdef CONFIG_QUOTA
2245         /* Needed for iput() to work correctly and not trash data */
2246         sb->s_flags |= MS_ACTIVE;
2247
2248         /*
2249          * Turn on quotas which were not enabled for read-only mounts if
2250          * filesystem has quota feature, so that they are updated correctly.
2251          */
2252         if (ext4_has_feature_quota(sb) && (s_flags & MS_RDONLY)) {
2253                 int ret = ext4_enable_quotas(sb);
2254
2255                 if (!ret)
2256                         quota_update = 1;
2257                 else
2258                         ext4_msg(sb, KERN_ERR,
2259                                 "Cannot turn on quotas: error %d", ret);
2260         }
2261
2262         /* Turn on journaled quotas used for old sytle */
2263         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2264                 if (EXT4_SB(sb)->s_qf_names[i]) {
2265                         int ret = ext4_quota_on_mount(sb, i);
2266
2267                         if (!ret)
2268                                 quota_update = 1;
2269                         else
2270                                 ext4_msg(sb, KERN_ERR,
2271                                         "Cannot turn on journaled "
2272                                         "quota: type %d: error %d", i, ret);
2273                 }
2274         }
2275 #endif
2276
2277         while (es->s_last_orphan) {
2278                 struct inode *inode;
2279
2280                 /*
2281                  * We may have encountered an error during cleanup; if
2282                  * so, skip the rest.
2283                  */
2284                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2285                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2286                         es->s_last_orphan = 0;
2287                         break;
2288                 }
2289
2290                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2291                 if (IS_ERR(inode)) {
2292                         es->s_last_orphan = 0;
2293                         break;
2294                 }
2295
2296                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2297                 dquot_initialize(inode);
2298                 if (inode->i_nlink) {
2299                         if (test_opt(sb, DEBUG))
2300                                 ext4_msg(sb, KERN_DEBUG,
2301                                         "%s: truncating inode %lu to %lld bytes",
2302                                         __func__, inode->i_ino, inode->i_size);
2303                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2304                                   inode->i_ino, inode->i_size);
2305                         mutex_lock(&inode->i_mutex);
2306                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2307                         ext4_truncate(inode);
2308                         mutex_unlock(&inode->i_mutex);
2309                         nr_truncates++;
2310                 } else {
2311                         if (test_opt(sb, DEBUG))
2312                                 ext4_msg(sb, KERN_DEBUG,
2313                                         "%s: deleting unreferenced inode %lu",
2314                                         __func__, inode->i_ino);
2315                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2316                                   inode->i_ino);
2317                         nr_orphans++;
2318                 }
2319                 iput(inode);  /* The delete magic happens here! */
2320         }
2321
2322 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2323
2324         if (nr_orphans)
2325                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2326                        PLURAL(nr_orphans));
2327         if (nr_truncates)
2328                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2329                        PLURAL(nr_truncates));
2330 #ifdef CONFIG_QUOTA
2331         /* Turn off quotas if they were enabled for orphan cleanup */
2332         if (quota_update) {
2333                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2334                         if (sb_dqopt(sb)->files[i])
2335                                 dquot_quota_off(sb, i);
2336                 }
2337         }
2338 #endif
2339         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2340 }
2341
2342 /*
2343  * Maximal extent format file size.
2344  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2345  * extent format containers, within a sector_t, and within i_blocks
2346  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2347  * so that won't be a limiting factor.
2348  *
2349  * However there is other limiting factor. We do store extents in the form
2350  * of starting block and length, hence the resulting length of the extent
2351  * covering maximum file size must fit into on-disk format containers as
2352  * well. Given that length is always by 1 unit bigger than max unit (because
2353  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2354  *
2355  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2356  */
2357 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2358 {
2359         loff_t res;
2360         loff_t upper_limit = MAX_LFS_FILESIZE;
2361
2362         /* small i_blocks in vfs inode? */
2363         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2364                 /*
2365                  * CONFIG_LBDAF is not enabled implies the inode
2366                  * i_block represent total blocks in 512 bytes
2367                  * 32 == size of vfs inode i_blocks * 8
2368                  */
2369                 upper_limit = (1LL << 32) - 1;
2370
2371                 /* total blocks in file system block size */
2372                 upper_limit >>= (blkbits - 9);
2373                 upper_limit <<= blkbits;
2374         }
2375
2376         /*
2377          * 32-bit extent-start container, ee_block. We lower the maxbytes
2378          * by one fs block, so ee_len can cover the extent of maximum file
2379          * size
2380          */
2381         res = (1LL << 32) - 1;
2382         res <<= blkbits;
2383
2384         /* Sanity check against vm- & vfs- imposed limits */
2385         if (res > upper_limit)
2386                 res = upper_limit;
2387
2388         return res;
2389 }
2390
2391 /*
2392  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2393  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2394  * We need to be 1 filesystem block less than the 2^48 sector limit.
2395  */
2396 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2397 {
2398         loff_t res = EXT4_NDIR_BLOCKS;
2399         int meta_blocks;
2400         loff_t upper_limit;
2401         /* This is calculated to be the largest file size for a dense, block
2402          * mapped file such that the file's total number of 512-byte sectors,
2403          * including data and all indirect blocks, does not exceed (2^48 - 1).
2404          *
2405          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2406          * number of 512-byte sectors of the file.
2407          */
2408
2409         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2410                 /*
2411                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2412                  * the inode i_block field represents total file blocks in
2413                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2414                  */
2415                 upper_limit = (1LL << 32) - 1;
2416
2417                 /* total blocks in file system block size */
2418                 upper_limit >>= (bits - 9);
2419
2420         } else {
2421                 /*
2422                  * We use 48 bit ext4_inode i_blocks
2423                  * With EXT4_HUGE_FILE_FL set the i_blocks
2424                  * represent total number of blocks in
2425                  * file system block size
2426                  */
2427                 upper_limit = (1LL << 48) - 1;
2428
2429         }
2430
2431         /* indirect blocks */
2432         meta_blocks = 1;
2433         /* double indirect blocks */
2434         meta_blocks += 1 + (1LL << (bits-2));
2435         /* tripple indirect blocks */
2436         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2437
2438         upper_limit -= meta_blocks;
2439         upper_limit <<= bits;
2440
2441         res += 1LL << (bits-2);
2442         res += 1LL << (2*(bits-2));
2443         res += 1LL << (3*(bits-2));
2444         res <<= bits;
2445         if (res > upper_limit)
2446                 res = upper_limit;
2447
2448         if (res > MAX_LFS_FILESIZE)
2449                 res = MAX_LFS_FILESIZE;
2450
2451         return res;
2452 }
2453
2454 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2455                                    ext4_fsblk_t logical_sb_block, int nr)
2456 {
2457         struct ext4_sb_info *sbi = EXT4_SB(sb);
2458         ext4_group_t bg, first_meta_bg;
2459         int has_super = 0;
2460
2461         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2462
2463         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2464                 return logical_sb_block + nr + 1;
2465         bg = sbi->s_desc_per_block * nr;
2466         if (ext4_bg_has_super(sb, bg))
2467                 has_super = 1;
2468
2469         /*
2470          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2471          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2472          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2473          * compensate.
2474          */
2475         if (sb->s_blocksize == 1024 && nr == 0 &&
2476             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2477                 has_super++;
2478
2479         return (has_super + ext4_group_first_block_no(sb, bg));
2480 }
2481
2482 /**
2483  * ext4_get_stripe_size: Get the stripe size.
2484  * @sbi: In memory super block info
2485  *
2486  * If we have specified it via mount option, then
2487  * use the mount option value. If the value specified at mount time is
2488  * greater than the blocks per group use the super block value.
2489  * If the super block value is greater than blocks per group return 0.
2490  * Allocator needs it be less than blocks per group.
2491  *
2492  */
2493 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2494 {
2495         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2496         unsigned long stripe_width =
2497                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2498         int ret;
2499
2500         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2501                 ret = sbi->s_stripe;
2502         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2503                 ret = stripe_width;
2504         else if (stride && stride <= sbi->s_blocks_per_group)
2505                 ret = stride;
2506         else
2507                 ret = 0;
2508
2509         /*
2510          * If the stripe width is 1, this makes no sense and
2511          * we set it to 0 to turn off stripe handling code.
2512          */
2513         if (ret <= 1)
2514                 ret = 0;
2515
2516         return ret;
2517 }
2518
2519 /*
2520  * Check whether this filesystem can be mounted based on
2521  * the features present and the RDONLY/RDWR mount requested.
2522  * Returns 1 if this filesystem can be mounted as requested,
2523  * 0 if it cannot be.
2524  */
2525 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2526 {
2527         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2528                 ext4_msg(sb, KERN_ERR,
2529                         "Couldn't mount because of "
2530                         "unsupported optional features (%x)",
2531                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2532                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2533                 return 0;
2534         }
2535
2536         if (readonly)
2537                 return 1;
2538
2539         if (ext4_has_feature_readonly(sb)) {
2540                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2541                 sb->s_flags |= MS_RDONLY;
2542                 return 1;
2543         }
2544
2545         /* Check that feature set is OK for a read-write mount */
2546         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2547                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2548                          "unsupported optional features (%x)",
2549                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2550                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2551                 return 0;
2552         }
2553         /*
2554          * Large file size enabled file system can only be mounted
2555          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2556          */
2557         if (ext4_has_feature_huge_file(sb)) {
2558                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2559                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2560                                  "cannot be mounted RDWR without "
2561                                  "CONFIG_LBDAF");
2562                         return 0;
2563                 }
2564         }
2565         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2566                 ext4_msg(sb, KERN_ERR,
2567                          "Can't support bigalloc feature without "
2568                          "extents feature\n");
2569                 return 0;
2570         }
2571
2572 #ifndef CONFIG_QUOTA
2573         if (ext4_has_feature_quota(sb) && !readonly) {
2574                 ext4_msg(sb, KERN_ERR,
2575                          "Filesystem with quota feature cannot be mounted RDWR "
2576                          "without CONFIG_QUOTA");
2577                 return 0;
2578         }
2579 #endif  /* CONFIG_QUOTA */
2580         return 1;
2581 }
2582
2583 /*
2584  * This function is called once a day if we have errors logged
2585  * on the file system
2586  */
2587 static void print_daily_error_info(unsigned long arg)
2588 {
2589         struct super_block *sb = (struct super_block *) arg;
2590         struct ext4_sb_info *sbi;
2591         struct ext4_super_block *es;
2592
2593         sbi = EXT4_SB(sb);
2594         es = sbi->s_es;
2595
2596         if (es->s_error_count)
2597                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2598                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2599                          le32_to_cpu(es->s_error_count));
2600         if (es->s_first_error_time) {
2601                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2602                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2603                        (int) sizeof(es->s_first_error_func),
2604                        es->s_first_error_func,
2605                        le32_to_cpu(es->s_first_error_line));
2606                 if (es->s_first_error_ino)
2607                         printk(": inode %u",
2608                                le32_to_cpu(es->s_first_error_ino));
2609                 if (es->s_first_error_block)
2610                         printk(": block %llu", (unsigned long long)
2611                                le64_to_cpu(es->s_first_error_block));
2612                 printk("\n");
2613         }
2614         if (es->s_last_error_time) {
2615                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2616                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2617                        (int) sizeof(es->s_last_error_func),
2618                        es->s_last_error_func,
2619                        le32_to_cpu(es->s_last_error_line));
2620                 if (es->s_last_error_ino)
2621                         printk(": inode %u",
2622                                le32_to_cpu(es->s_last_error_ino));
2623                 if (es->s_last_error_block)
2624                         printk(": block %llu", (unsigned long long)
2625                                le64_to_cpu(es->s_last_error_block));
2626                 printk("\n");
2627         }
2628         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2629 }
2630
2631 /* Find next suitable group and run ext4_init_inode_table */
2632 static int ext4_run_li_request(struct ext4_li_request *elr)
2633 {
2634         struct ext4_group_desc *gdp = NULL;
2635         ext4_group_t group, ngroups;
2636         struct super_block *sb;
2637         unsigned long timeout = 0;
2638         int ret = 0;
2639
2640         sb = elr->lr_super;
2641         ngroups = EXT4_SB(sb)->s_groups_count;
2642
2643         sb_start_write(sb);
2644         for (group = elr->lr_next_group; group < ngroups; group++) {
2645                 gdp = ext4_get_group_desc(sb, group, NULL);
2646                 if (!gdp) {
2647                         ret = 1;
2648                         break;
2649                 }
2650
2651                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2652                         break;
2653         }
2654
2655         if (group >= ngroups)
2656                 ret = 1;
2657
2658         if (!ret) {
2659                 timeout = jiffies;
2660                 ret = ext4_init_inode_table(sb, group,
2661                                             elr->lr_timeout ? 0 : 1);
2662                 if (elr->lr_timeout == 0) {
2663                         timeout = (jiffies - timeout) *
2664                                   elr->lr_sbi->s_li_wait_mult;
2665                         elr->lr_timeout = timeout;
2666                 }
2667                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2668                 elr->lr_next_group = group + 1;
2669         }
2670         sb_end_write(sb);
2671
2672         return ret;
2673 }
2674
2675 /*
2676  * Remove lr_request from the list_request and free the
2677  * request structure. Should be called with li_list_mtx held
2678  */
2679 static void ext4_remove_li_request(struct ext4_li_request *elr)
2680 {
2681         struct ext4_sb_info *sbi;
2682
2683         if (!elr)
2684                 return;
2685
2686         sbi = elr->lr_sbi;
2687
2688         list_del(&elr->lr_request);
2689         sbi->s_li_request = NULL;
2690         kfree(elr);
2691 }
2692
2693 static void ext4_unregister_li_request(struct super_block *sb)
2694 {
2695         mutex_lock(&ext4_li_mtx);
2696         if (!ext4_li_info) {
2697                 mutex_unlock(&ext4_li_mtx);
2698                 return;
2699         }
2700
2701         mutex_lock(&ext4_li_info->li_list_mtx);
2702         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2703         mutex_unlock(&ext4_li_info->li_list_mtx);
2704         mutex_unlock(&ext4_li_mtx);
2705 }
2706
2707 static struct task_struct *ext4_lazyinit_task;
2708
2709 /*
2710  * This is the function where ext4lazyinit thread lives. It walks
2711  * through the request list searching for next scheduled filesystem.
2712  * When such a fs is found, run the lazy initialization request
2713  * (ext4_rn_li_request) and keep track of the time spend in this
2714  * function. Based on that time we compute next schedule time of
2715  * the request. When walking through the list is complete, compute
2716  * next waking time and put itself into sleep.
2717  */
2718 static int ext4_lazyinit_thread(void *arg)
2719 {
2720         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2721         struct list_head *pos, *n;
2722         struct ext4_li_request *elr;
2723         unsigned long next_wakeup, cur;
2724
2725         BUG_ON(NULL == eli);
2726
2727 cont_thread:
2728         while (true) {
2729                 next_wakeup = MAX_JIFFY_OFFSET;
2730
2731                 mutex_lock(&eli->li_list_mtx);
2732                 if (list_empty(&eli->li_request_list)) {
2733                         mutex_unlock(&eli->li_list_mtx);
2734                         goto exit_thread;
2735                 }
2736
2737                 list_for_each_safe(pos, n, &eli->li_request_list) {
2738                         elr = list_entry(pos, struct ext4_li_request,
2739                                          lr_request);
2740
2741                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2742                                 if (ext4_run_li_request(elr) != 0) {
2743                                         /* error, remove the lazy_init job */
2744                                         ext4_remove_li_request(elr);
2745                                         continue;
2746                                 }
2747                         }
2748
2749                         if (time_before(elr->lr_next_sched, next_wakeup))
2750                                 next_wakeup = elr->lr_next_sched;
2751                 }
2752                 mutex_unlock(&eli->li_list_mtx);
2753
2754                 try_to_freeze();
2755
2756                 cur = jiffies;
2757                 if ((time_after_eq(cur, next_wakeup)) ||
2758                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2759                         cond_resched();
2760                         continue;
2761                 }
2762
2763                 schedule_timeout_interruptible(next_wakeup - cur);
2764
2765                 if (kthread_should_stop()) {
2766                         ext4_clear_request_list();
2767                         goto exit_thread;
2768                 }
2769         }
2770
2771 exit_thread:
2772         /*
2773          * It looks like the request list is empty, but we need
2774          * to check it under the li_list_mtx lock, to prevent any
2775          * additions into it, and of course we should lock ext4_li_mtx
2776          * to atomically free the list and ext4_li_info, because at
2777          * this point another ext4 filesystem could be registering
2778          * new one.
2779          */
2780         mutex_lock(&ext4_li_mtx);
2781         mutex_lock(&eli->li_list_mtx);
2782         if (!list_empty(&eli->li_request_list)) {
2783                 mutex_unlock(&eli->li_list_mtx);
2784                 mutex_unlock(&ext4_li_mtx);
2785                 goto cont_thread;
2786         }
2787         mutex_unlock(&eli->li_list_mtx);
2788         kfree(ext4_li_info);
2789         ext4_li_info = NULL;
2790         mutex_unlock(&ext4_li_mtx);
2791
2792         return 0;
2793 }
2794
2795 static void ext4_clear_request_list(void)
2796 {
2797         struct list_head *pos, *n;
2798         struct ext4_li_request *elr;
2799
2800         mutex_lock(&ext4_li_info->li_list_mtx);
2801         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2802                 elr = list_entry(pos, struct ext4_li_request,
2803                                  lr_request);
2804                 ext4_remove_li_request(elr);
2805         }
2806         mutex_unlock(&ext4_li_info->li_list_mtx);
2807 }
2808
2809 static int ext4_run_lazyinit_thread(void)
2810 {
2811         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2812                                          ext4_li_info, "ext4lazyinit");
2813         if (IS_ERR(ext4_lazyinit_task)) {
2814                 int err = PTR_ERR(ext4_lazyinit_task);
2815                 ext4_clear_request_list();
2816                 kfree(ext4_li_info);
2817                 ext4_li_info = NULL;
2818                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2819                                  "initialization thread\n",
2820                                  err);
2821                 return err;
2822         }
2823         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2824         return 0;
2825 }
2826
2827 /*
2828  * Check whether it make sense to run itable init. thread or not.
2829  * If there is at least one uninitialized inode table, return
2830  * corresponding group number, else the loop goes through all
2831  * groups and return total number of groups.
2832  */
2833 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2834 {
2835         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2836         struct ext4_group_desc *gdp = NULL;
2837
2838         for (group = 0; group < ngroups; group++) {
2839                 gdp = ext4_get_group_desc(sb, group, NULL);
2840                 if (!gdp)
2841                         continue;
2842
2843                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2844                         break;
2845         }
2846
2847         return group;
2848 }
2849
2850 static int ext4_li_info_new(void)
2851 {
2852         struct ext4_lazy_init *eli = NULL;
2853
2854         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2855         if (!eli)
2856                 return -ENOMEM;
2857
2858         INIT_LIST_HEAD(&eli->li_request_list);
2859         mutex_init(&eli->li_list_mtx);
2860
2861         eli->li_state |= EXT4_LAZYINIT_QUIT;
2862
2863         ext4_li_info = eli;
2864
2865         return 0;
2866 }
2867
2868 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2869                                             ext4_group_t start)
2870 {
2871         struct ext4_sb_info *sbi = EXT4_SB(sb);
2872         struct ext4_li_request *elr;
2873
2874         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2875         if (!elr)
2876                 return NULL;
2877
2878         elr->lr_super = sb;
2879         elr->lr_sbi = sbi;
2880         elr->lr_next_group = start;
2881
2882         /*
2883          * Randomize first schedule time of the request to
2884          * spread the inode table initialization requests
2885          * better.
2886          */
2887         elr->lr_next_sched = jiffies + (prandom_u32() %
2888                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2889         return elr;
2890 }
2891
2892 int ext4_register_li_request(struct super_block *sb,
2893                              ext4_group_t first_not_zeroed)
2894 {
2895         struct ext4_sb_info *sbi = EXT4_SB(sb);
2896         struct ext4_li_request *elr = NULL;
2897         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2898         int ret = 0;
2899
2900         mutex_lock(&ext4_li_mtx);
2901         if (sbi->s_li_request != NULL) {
2902                 /*
2903                  * Reset timeout so it can be computed again, because
2904                  * s_li_wait_mult might have changed.
2905                  */
2906                 sbi->s_li_request->lr_timeout = 0;
2907                 goto out;
2908         }
2909
2910         if (first_not_zeroed == ngroups ||
2911             (sb->s_flags & MS_RDONLY) ||
2912             !test_opt(sb, INIT_INODE_TABLE))
2913                 goto out;
2914
2915         elr = ext4_li_request_new(sb, first_not_zeroed);
2916         if (!elr) {
2917                 ret = -ENOMEM;
2918                 goto out;
2919         }
2920
2921         if (NULL == ext4_li_info) {
2922                 ret = ext4_li_info_new();
2923                 if (ret)
2924                         goto out;
2925         }
2926
2927         mutex_lock(&ext4_li_info->li_list_mtx);
2928         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2929         mutex_unlock(&ext4_li_info->li_list_mtx);
2930
2931         sbi->s_li_request = elr;
2932         /*
2933          * set elr to NULL here since it has been inserted to
2934          * the request_list and the removal and free of it is
2935          * handled by ext4_clear_request_list from now on.
2936          */
2937         elr = NULL;
2938
2939         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2940                 ret = ext4_run_lazyinit_thread();
2941                 if (ret)
2942                         goto out;
2943         }
2944 out:
2945         mutex_unlock(&ext4_li_mtx);
2946         if (ret)
2947                 kfree(elr);
2948         return ret;
2949 }
2950
2951 /*
2952  * We do not need to lock anything since this is called on
2953  * module unload.
2954  */
2955 static void ext4_destroy_lazyinit_thread(void)
2956 {
2957         /*
2958          * If thread exited earlier
2959          * there's nothing to be done.
2960          */
2961         if (!ext4_li_info || !ext4_lazyinit_task)
2962                 return;
2963
2964         kthread_stop(ext4_lazyinit_task);
2965 }
2966
2967 static int set_journal_csum_feature_set(struct super_block *sb)
2968 {
2969         int ret = 1;
2970         int compat, incompat;
2971         struct ext4_sb_info *sbi = EXT4_SB(sb);
2972
2973         if (ext4_has_metadata_csum(sb)) {
2974                 /* journal checksum v3 */
2975                 compat = 0;
2976                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
2977         } else {
2978                 /* journal checksum v1 */
2979                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
2980                 incompat = 0;
2981         }
2982
2983         jbd2_journal_clear_features(sbi->s_journal,
2984                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2985                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
2986                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
2987         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2988                 ret = jbd2_journal_set_features(sbi->s_journal,
2989                                 compat, 0,
2990                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
2991                                 incompat);
2992         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2993                 ret = jbd2_journal_set_features(sbi->s_journal,
2994                                 compat, 0,
2995                                 incompat);
2996                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2997                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2998         } else {
2999                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
3000                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3001         }
3002
3003         return ret;
3004 }
3005
3006 /*
3007  * Note: calculating the overhead so we can be compatible with
3008  * historical BSD practice is quite difficult in the face of
3009  * clusters/bigalloc.  This is because multiple metadata blocks from
3010  * different block group can end up in the same allocation cluster.
3011  * Calculating the exact overhead in the face of clustered allocation
3012  * requires either O(all block bitmaps) in memory or O(number of block
3013  * groups**2) in time.  We will still calculate the superblock for
3014  * older file systems --- and if we come across with a bigalloc file
3015  * system with zero in s_overhead_clusters the estimate will be close to
3016  * correct especially for very large cluster sizes --- but for newer
3017  * file systems, it's better to calculate this figure once at mkfs
3018  * time, and store it in the superblock.  If the superblock value is
3019  * present (even for non-bigalloc file systems), we will use it.
3020  */
3021 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3022                           char *buf)
3023 {
3024         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3025         struct ext4_group_desc  *gdp;
3026         ext4_fsblk_t            first_block, last_block, b;
3027         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3028         int                     s, j, count = 0;
3029
3030         if (!ext4_has_feature_bigalloc(sb))
3031                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3032                         sbi->s_itb_per_group + 2);
3033
3034         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3035                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3036         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3037         for (i = 0; i < ngroups; i++) {
3038                 gdp = ext4_get_group_desc(sb, i, NULL);
3039                 b = ext4_block_bitmap(sb, gdp);
3040                 if (b >= first_block && b <= last_block) {
3041                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3042                         count++;
3043                 }
3044                 b = ext4_inode_bitmap(sb, gdp);
3045                 if (b >= first_block && b <= last_block) {
3046                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3047                         count++;
3048                 }
3049                 b = ext4_inode_table(sb, gdp);
3050                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3051                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3052                                 int c = EXT4_B2C(sbi, b - first_block);
3053                                 ext4_set_bit(c, buf);
3054                                 count++;
3055                         }
3056                 if (i != grp)
3057                         continue;
3058                 s = 0;
3059                 if (ext4_bg_has_super(sb, grp)) {
3060                         ext4_set_bit(s++, buf);
3061                         count++;
3062                 }
3063                 j = ext4_bg_num_gdb(sb, grp);
3064                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3065                         ext4_error(sb, "Invalid number of block group "
3066                                    "descriptor blocks: %d", j);
3067                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3068                 }
3069                 count += j;
3070                 for (; j > 0; j--)
3071                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3072         }
3073         if (!count)
3074                 return 0;
3075         return EXT4_CLUSTERS_PER_GROUP(sb) -
3076                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3077 }
3078
3079 /*
3080  * Compute the overhead and stash it in sbi->s_overhead
3081  */
3082 int ext4_calculate_overhead(struct super_block *sb)
3083 {
3084         struct ext4_sb_info *sbi = EXT4_SB(sb);
3085         struct ext4_super_block *es = sbi->s_es;
3086         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3087         ext4_fsblk_t overhead = 0;
3088         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3089
3090         if (!buf)
3091                 return -ENOMEM;
3092
3093         /*
3094          * Compute the overhead (FS structures).  This is constant
3095          * for a given filesystem unless the number of block groups
3096          * changes so we cache the previous value until it does.
3097          */
3098
3099         /*
3100          * All of the blocks before first_data_block are overhead
3101          */
3102         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3103
3104         /*
3105          * Add the overhead found in each block group
3106          */
3107         for (i = 0; i < ngroups; i++) {
3108                 int blks;
3109
3110                 blks = count_overhead(sb, i, buf);
3111                 overhead += blks;
3112                 if (blks)
3113                         memset(buf, 0, PAGE_SIZE);
3114                 cond_resched();
3115         }
3116         /* Add the internal journal blocks as well */
3117         if (sbi->s_journal && !sbi->journal_bdev)
3118                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3119
3120         sbi->s_overhead = overhead;
3121         smp_wmb();
3122         free_page((unsigned long) buf);
3123         return 0;
3124 }
3125
3126 static void ext4_set_resv_clusters(struct super_block *sb)
3127 {
3128         ext4_fsblk_t resv_clusters;
3129         struct ext4_sb_info *sbi = EXT4_SB(sb);
3130
3131         /*
3132          * There's no need to reserve anything when we aren't using extents.
3133          * The space estimates are exact, there are no unwritten extents,
3134          * hole punching doesn't need new metadata... This is needed especially
3135          * to keep ext2/3 backward compatibility.
3136          */
3137         if (!ext4_has_feature_extents(sb))
3138                 return;
3139         /*
3140          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3141          * This should cover the situations where we can not afford to run
3142          * out of space like for example punch hole, or converting
3143          * unwritten extents in delalloc path. In most cases such
3144          * allocation would require 1, or 2 blocks, higher numbers are
3145          * very rare.
3146          */
3147         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3148                          sbi->s_cluster_bits);
3149
3150         do_div(resv_clusters, 50);
3151         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3152
3153         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3154 }
3155
3156 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3157 {
3158         char *orig_data = kstrdup(data, GFP_KERNEL);
3159         struct buffer_head *bh;
3160         struct ext4_super_block *es = NULL;
3161         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3162         ext4_fsblk_t block;
3163         ext4_fsblk_t sb_block = get_sb_block(&data);
3164         ext4_fsblk_t logical_sb_block;
3165         unsigned long offset = 0;
3166         unsigned long journal_devnum = 0;
3167         unsigned long def_mount_opts;
3168         struct inode *root;
3169         const char *descr;
3170         int ret = -ENOMEM;
3171         int blocksize, clustersize;
3172         unsigned int db_count;
3173         unsigned int i;
3174         int needs_recovery, has_huge_files, has_bigalloc;
3175         __u64 blocks_count;
3176         int err = 0;
3177         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3178         ext4_group_t first_not_zeroed;
3179
3180         if ((data && !orig_data) || !sbi)
3181                 goto out_free_base;
3182
3183         sbi->s_blockgroup_lock =
3184                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3185         if (!sbi->s_blockgroup_lock)
3186                 goto out_free_base;
3187
3188         sb->s_fs_info = sbi;
3189         sbi->s_sb = sb;
3190         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3191         sbi->s_sb_block = sb_block;
3192         if (sb->s_bdev->bd_part)
3193                 sbi->s_sectors_written_start =
3194                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3195
3196         /* Cleanup superblock name */
3197         strreplace(sb->s_id, '/', '!');
3198
3199         /* -EINVAL is default */
3200         ret = -EINVAL;
3201         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3202         if (!blocksize) {
3203                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3204                 goto out_fail;
3205         }
3206
3207         /*
3208          * The ext4 superblock will not be buffer aligned for other than 1kB
3209          * block sizes.  We need to calculate the offset from buffer start.
3210          */
3211         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3212                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3213                 offset = do_div(logical_sb_block, blocksize);
3214         } else {
3215                 logical_sb_block = sb_block;
3216         }
3217
3218         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3219                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3220                 goto out_fail;
3221         }
3222         /*
3223          * Note: s_es must be initialized as soon as possible because
3224          *       some ext4 macro-instructions depend on its value
3225          */
3226         es = (struct ext4_super_block *) (bh->b_data + offset);
3227         sbi->s_es = es;
3228         sb->s_magic = le16_to_cpu(es->s_magic);
3229         if (sb->s_magic != EXT4_SUPER_MAGIC)
3230                 goto cantfind_ext4;
3231         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3232
3233         /* Warn if metadata_csum and gdt_csum are both set. */
3234         if (ext4_has_feature_metadata_csum(sb) &&
3235             ext4_has_feature_gdt_csum(sb))
3236                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3237                              "redundant flags; please run fsck.");
3238
3239         /* Check for a known checksum algorithm */
3240         if (!ext4_verify_csum_type(sb, es)) {
3241                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3242                          "unknown checksum algorithm.");
3243                 silent = 1;
3244                 goto cantfind_ext4;
3245         }
3246
3247         /* Load the checksum driver */
3248         if (ext4_has_feature_metadata_csum(sb)) {
3249                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3250                 if (IS_ERR(sbi->s_chksum_driver)) {
3251                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3252                         ret = PTR_ERR(sbi->s_chksum_driver);
3253                         sbi->s_chksum_driver = NULL;
3254                         goto failed_mount;
3255                 }
3256         }
3257
3258         /* Check superblock checksum */
3259         if (!ext4_superblock_csum_verify(sb, es)) {
3260                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3261                          "invalid superblock checksum.  Run e2fsck?");
3262                 silent = 1;
3263                 ret = -EFSBADCRC;
3264                 goto cantfind_ext4;
3265         }
3266
3267         /* Precompute checksum seed for all metadata */
3268         if (ext4_has_feature_csum_seed(sb))
3269                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3270         else if (ext4_has_metadata_csum(sb))
3271                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3272                                                sizeof(es->s_uuid));
3273
3274         /* Set defaults before we parse the mount options */
3275         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3276         set_opt(sb, INIT_INODE_TABLE);
3277         if (def_mount_opts & EXT4_DEFM_DEBUG)
3278                 set_opt(sb, DEBUG);
3279         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3280                 set_opt(sb, GRPID);
3281         if (def_mount_opts & EXT4_DEFM_UID16)
3282                 set_opt(sb, NO_UID32);
3283         /* xattr user namespace & acls are now defaulted on */
3284         set_opt(sb, XATTR_USER);
3285 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3286         set_opt(sb, POSIX_ACL);
3287 #endif
3288         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3289         if (ext4_has_metadata_csum(sb))
3290                 set_opt(sb, JOURNAL_CHECKSUM);
3291
3292         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3293                 set_opt(sb, JOURNAL_DATA);
3294         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3295                 set_opt(sb, ORDERED_DATA);
3296         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3297                 set_opt(sb, WRITEBACK_DATA);
3298
3299         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3300                 set_opt(sb, ERRORS_PANIC);
3301         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3302                 set_opt(sb, ERRORS_CONT);
3303         else
3304                 set_opt(sb, ERRORS_RO);
3305         /* block_validity enabled by default; disable with noblock_validity */
3306         set_opt(sb, BLOCK_VALIDITY);
3307         if (def_mount_opts & EXT4_DEFM_DISCARD)
3308                 set_opt(sb, DISCARD);
3309
3310         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3311         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3312         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3313         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3314         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3315
3316         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3317                 set_opt(sb, BARRIER);
3318
3319         /*
3320          * enable delayed allocation by default
3321          * Use -o nodelalloc to turn it off
3322          */
3323         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3324             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3325                 set_opt(sb, DELALLOC);
3326
3327         /*
3328          * set default s_li_wait_mult for lazyinit, for the case there is
3329          * no mount option specified.
3330          */
3331         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3332
3333         if (sbi->s_es->s_mount_opts[0]) {
3334                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3335                                               sizeof(sbi->s_es->s_mount_opts),
3336                                               GFP_KERNEL);
3337                 if (!s_mount_opts)
3338                         goto failed_mount;
3339                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3340                                    &journal_ioprio, 0)) {
3341                         ext4_msg(sb, KERN_WARNING,
3342                                  "failed to parse options in superblock: %s",
3343                                  s_mount_opts);
3344                 }
3345                 kfree(s_mount_opts);
3346         }
3347         sbi->s_def_mount_opt = sbi->s_mount_opt;
3348         if (!parse_options((char *) data, sb, &journal_devnum,
3349                            &journal_ioprio, 0))
3350                 goto failed_mount;
3351
3352         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3353                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3354                             "with data=journal disables delayed "
3355                             "allocation and O_DIRECT support!\n");
3356                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3357                         ext4_msg(sb, KERN_ERR, "can't mount with "
3358                                  "both data=journal and delalloc");
3359                         goto failed_mount;
3360                 }
3361                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3362                         ext4_msg(sb, KERN_ERR, "can't mount with "
3363                                  "both data=journal and dioread_nolock");
3364                         goto failed_mount;
3365                 }
3366                 if (test_opt(sb, DAX)) {
3367                         ext4_msg(sb, KERN_ERR, "can't mount with "
3368                                  "both data=journal and dax");
3369                         goto failed_mount;
3370                 }
3371                 if (ext4_has_feature_encrypt(sb)) {
3372                         ext4_msg(sb, KERN_WARNING,
3373                                  "encrypted files will use data=ordered "
3374                                  "instead of data journaling mode");
3375                 }
3376                 if (test_opt(sb, DELALLOC))
3377                         clear_opt(sb, DELALLOC);
3378         } else {
3379                 sb->s_iflags |= SB_I_CGROUPWB;
3380         }
3381
3382         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3383                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3384
3385         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3386             (ext4_has_compat_features(sb) ||
3387              ext4_has_ro_compat_features(sb) ||
3388              ext4_has_incompat_features(sb)))
3389                 ext4_msg(sb, KERN_WARNING,
3390                        "feature flags set on rev 0 fs, "
3391                        "running e2fsck is recommended");
3392
3393         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3394                 set_opt2(sb, HURD_COMPAT);
3395                 if (ext4_has_feature_64bit(sb)) {
3396                         ext4_msg(sb, KERN_ERR,
3397                                  "The Hurd can't support 64-bit file systems");
3398                         goto failed_mount;
3399                 }
3400         }
3401
3402         if (IS_EXT2_SB(sb)) {
3403                 if (ext2_feature_set_ok(sb))
3404                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3405                                  "using the ext4 subsystem");
3406                 else {
3407                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3408                                  "to feature incompatibilities");
3409                         goto failed_mount;
3410                 }
3411         }
3412
3413         if (IS_EXT3_SB(sb)) {
3414                 if (ext3_feature_set_ok(sb))
3415                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3416                                  "using the ext4 subsystem");
3417                 else {
3418                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3419                                  "to feature incompatibilities");
3420                         goto failed_mount;
3421                 }
3422         }
3423
3424         /*
3425          * Check feature flags regardless of the revision level, since we
3426          * previously didn't change the revision level when setting the flags,
3427          * so there is a chance incompat flags are set on a rev 0 filesystem.
3428          */
3429         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3430                 goto failed_mount;
3431
3432         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3433         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3434             blocksize > EXT4_MAX_BLOCK_SIZE) {
3435                 ext4_msg(sb, KERN_ERR,
3436                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3437                          blocksize, le32_to_cpu(es->s_log_block_size));
3438                 goto failed_mount;
3439         }
3440         if (le32_to_cpu(es->s_log_block_size) >
3441             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3442                 ext4_msg(sb, KERN_ERR,
3443                          "Invalid log block size: %u",
3444                          le32_to_cpu(es->s_log_block_size));
3445                 goto failed_mount;
3446         }
3447
3448         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3449                 ext4_msg(sb, KERN_ERR,
3450                          "Number of reserved GDT blocks insanely large: %d",
3451                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3452                 goto failed_mount;
3453         }
3454
3455         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3456                 if (blocksize != PAGE_SIZE) {
3457                         ext4_msg(sb, KERN_ERR,
3458                                         "error: unsupported blocksize for dax");
3459                         goto failed_mount;
3460                 }
3461                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3462                         ext4_msg(sb, KERN_ERR,
3463                                         "error: device does not support dax");
3464                         goto failed_mount;
3465                 }
3466         }
3467
3468         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3469                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3470                          es->s_encryption_level);
3471                 goto failed_mount;
3472         }
3473
3474         if (sb->s_blocksize != blocksize) {
3475                 /* Validate the filesystem blocksize */
3476                 if (!sb_set_blocksize(sb, blocksize)) {
3477                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3478                                         blocksize);
3479                         goto failed_mount;
3480                 }
3481
3482                 brelse(bh);
3483                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3484                 offset = do_div(logical_sb_block, blocksize);
3485                 bh = sb_bread_unmovable(sb, logical_sb_block);
3486                 if (!bh) {
3487                         ext4_msg(sb, KERN_ERR,
3488                                "Can't read superblock on 2nd try");
3489                         goto failed_mount;
3490                 }
3491                 es = (struct ext4_super_block *)(bh->b_data + offset);
3492                 sbi->s_es = es;
3493                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3494                         ext4_msg(sb, KERN_ERR,
3495                                "Magic mismatch, very weird!");
3496                         goto failed_mount;
3497                 }
3498         }
3499
3500         has_huge_files = ext4_has_feature_huge_file(sb);
3501         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3502                                                       has_huge_files);
3503         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3504
3505         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3506                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3507                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3508         } else {
3509                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3510                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3511                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3512                     (!is_power_of_2(sbi->s_inode_size)) ||
3513                     (sbi->s_inode_size > blocksize)) {
3514                         ext4_msg(sb, KERN_ERR,
3515                                "unsupported inode size: %d",
3516                                sbi->s_inode_size);
3517                         goto failed_mount;
3518                 }
3519                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3520                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3521         }
3522
3523         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3524         if (ext4_has_feature_64bit(sb)) {
3525                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3526                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3527                     !is_power_of_2(sbi->s_desc_size)) {
3528                         ext4_msg(sb, KERN_ERR,
3529                                "unsupported descriptor size %lu",
3530                                sbi->s_desc_size);
3531                         goto failed_mount;
3532                 }
3533         } else
3534                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3535
3536         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3537         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3538
3539         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3540         if (sbi->s_inodes_per_block == 0)
3541                 goto cantfind_ext4;
3542         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3543             sbi->s_inodes_per_group > blocksize * 8) {
3544                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3545                          sbi->s_blocks_per_group);
3546                 goto failed_mount;
3547         }
3548         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3549                                         sbi->s_inodes_per_block;
3550         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3551         sbi->s_sbh = bh;
3552         sbi->s_mount_state = le16_to_cpu(es->s_state);
3553         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3554         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3555
3556         for (i = 0; i < 4; i++)
3557                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3558         sbi->s_def_hash_version = es->s_def_hash_version;
3559         if (ext4_has_feature_dir_index(sb)) {
3560                 i = le32_to_cpu(es->s_flags);
3561                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3562                         sbi->s_hash_unsigned = 3;
3563                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3564 #ifdef __CHAR_UNSIGNED__
3565                         if (!(sb->s_flags & MS_RDONLY))
3566                                 es->s_flags |=
3567                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3568                         sbi->s_hash_unsigned = 3;
3569 #else
3570                         if (!(sb->s_flags & MS_RDONLY))
3571                                 es->s_flags |=
3572                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3573 #endif
3574                 }
3575         }
3576
3577         /* Handle clustersize */
3578         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3579         has_bigalloc = ext4_has_feature_bigalloc(sb);
3580         if (has_bigalloc) {
3581                 if (clustersize < blocksize) {
3582                         ext4_msg(sb, KERN_ERR,
3583                                  "cluster size (%d) smaller than "
3584                                  "block size (%d)", clustersize, blocksize);
3585                         goto failed_mount;
3586                 }
3587                 if (le32_to_cpu(es->s_log_cluster_size) >
3588                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3589                         ext4_msg(sb, KERN_ERR,
3590                                  "Invalid log cluster size: %u",
3591                                  le32_to_cpu(es->s_log_cluster_size));
3592                         goto failed_mount;
3593                 }
3594                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3595                         le32_to_cpu(es->s_log_block_size);
3596                 sbi->s_clusters_per_group =
3597                         le32_to_cpu(es->s_clusters_per_group);
3598                 if (sbi->s_clusters_per_group > blocksize * 8) {
3599                         ext4_msg(sb, KERN_ERR,
3600                                  "#clusters per group too big: %lu",
3601                                  sbi->s_clusters_per_group);
3602                         goto failed_mount;
3603                 }
3604                 if (sbi->s_blocks_per_group !=
3605                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3606                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3607                                  "clusters per group (%lu) inconsistent",
3608                                  sbi->s_blocks_per_group,
3609                                  sbi->s_clusters_per_group);
3610                         goto failed_mount;
3611                 }
3612         } else {
3613                 if (clustersize != blocksize) {
3614                         ext4_warning(sb, "fragment/cluster size (%d) != "
3615                                      "block size (%d)", clustersize,
3616                                      blocksize);
3617                         clustersize = blocksize;
3618                 }
3619                 if (sbi->s_blocks_per_group > blocksize * 8) {
3620                         ext4_msg(sb, KERN_ERR,
3621                                  "#blocks per group too big: %lu",
3622                                  sbi->s_blocks_per_group);
3623                         goto failed_mount;
3624                 }
3625                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3626                 sbi->s_cluster_bits = 0;
3627         }
3628         sbi->s_cluster_ratio = clustersize / blocksize;
3629
3630         /* Do we have standard group size of clustersize * 8 blocks ? */
3631         if (sbi->s_blocks_per_group == clustersize << 3)
3632                 set_opt2(sb, STD_GROUP_SIZE);
3633
3634         /*
3635          * Test whether we have more sectors than will fit in sector_t,
3636          * and whether the max offset is addressable by the page cache.
3637          */
3638         err = generic_check_addressable(sb->s_blocksize_bits,
3639                                         ext4_blocks_count(es));
3640         if (err) {
3641                 ext4_msg(sb, KERN_ERR, "filesystem"
3642                          " too large to mount safely on this system");
3643                 if (sizeof(sector_t) < 8)
3644                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3645                 goto failed_mount;
3646         }
3647
3648         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3649                 goto cantfind_ext4;
3650
3651         /* check blocks count against device size */
3652         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3653         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3654                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3655                        "exceeds size of device (%llu blocks)",
3656                        ext4_blocks_count(es), blocks_count);
3657                 goto failed_mount;
3658         }
3659
3660         /*
3661          * It makes no sense for the first data block to be beyond the end
3662          * of the filesystem.
3663          */
3664         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3665                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3666                          "block %u is beyond end of filesystem (%llu)",
3667                          le32_to_cpu(es->s_first_data_block),
3668                          ext4_blocks_count(es));
3669                 goto failed_mount;
3670         }
3671         blocks_count = (ext4_blocks_count(es) -
3672                         le32_to_cpu(es->s_first_data_block) +
3673                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3674         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3675         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3676                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3677                        "(block count %llu, first data block %u, "
3678                        "blocks per group %lu)", sbi->s_groups_count,
3679                        ext4_blocks_count(es),
3680                        le32_to_cpu(es->s_first_data_block),
3681                        EXT4_BLOCKS_PER_GROUP(sb));
3682                 goto failed_mount;
3683         }
3684         sbi->s_groups_count = blocks_count;
3685         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3686                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3687         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3688                    EXT4_DESC_PER_BLOCK(sb);
3689         if (ext4_has_feature_meta_bg(sb)) {
3690                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3691                         ext4_msg(sb, KERN_WARNING,
3692                                  "first meta block group too large: %u "
3693                                  "(group descriptor block count %u)",
3694                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3695                         goto failed_mount;
3696                 }
3697         }
3698         sbi->s_group_desc = ext4_kvmalloc(db_count *
3699                                           sizeof(struct buffer_head *),
3700                                           GFP_KERNEL);
3701         if (sbi->s_group_desc == NULL) {
3702                 ext4_msg(sb, KERN_ERR, "not enough memory");
3703                 ret = -ENOMEM;
3704                 goto failed_mount;
3705         }
3706
3707         bgl_lock_init(sbi->s_blockgroup_lock);
3708
3709         for (i = 0; i < db_count; i++) {
3710                 block = descriptor_loc(sb, logical_sb_block, i);
3711                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3712                 if (!sbi->s_group_desc[i]) {
3713                         ext4_msg(sb, KERN_ERR,
3714                                "can't read group descriptor %d", i);
3715                         db_count = i;
3716                         goto failed_mount2;
3717                 }
3718         }
3719         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3720                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3721                 ret = -EFSCORRUPTED;
3722                 goto failed_mount2;
3723         }
3724
3725         sbi->s_gdb_count = db_count;
3726         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3727         spin_lock_init(&sbi->s_next_gen_lock);
3728
3729         setup_timer(&sbi->s_err_report, print_daily_error_info,
3730                 (unsigned long) sb);
3731
3732         /* Register extent status tree shrinker */
3733         if (ext4_es_register_shrinker(sbi))
3734                 goto failed_mount3;
3735
3736         sbi->s_stripe = ext4_get_stripe_size(sbi);
3737         sbi->s_extent_max_zeroout_kb = 32;
3738
3739         /*
3740          * set up enough so that it can read an inode
3741          */
3742         sb->s_op = &ext4_sops;
3743         sb->s_export_op = &ext4_export_ops;
3744         sb->s_xattr = ext4_xattr_handlers;
3745 #ifdef CONFIG_QUOTA
3746         sb->dq_op = &ext4_quota_operations;
3747         if (ext4_has_feature_quota(sb))
3748                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3749         else
3750                 sb->s_qcop = &ext4_qctl_operations;
3751         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
3752 #endif
3753         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3754
3755         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3756         mutex_init(&sbi->s_orphan_lock);
3757
3758         sb->s_root = NULL;
3759
3760         needs_recovery = (es->s_last_orphan != 0 ||
3761                           ext4_has_feature_journal_needs_recovery(sb));
3762
3763         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3764                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3765                         goto failed_mount3a;
3766
3767         /*
3768          * The first inode we look at is the journal inode.  Don't try
3769          * root first: it may be modified in the journal!
3770          */
3771         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3772                 err = ext4_load_journal(sb, es, journal_devnum);
3773                 if (err)
3774                         goto failed_mount3a;
3775         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3776                    ext4_has_feature_journal_needs_recovery(sb)) {
3777                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3778                        "suppressed and not mounted read-only");
3779                 goto failed_mount_wq;
3780         } else {
3781                 /* Nojournal mode, all journal mount options are illegal */
3782                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3783                         ext4_msg(sb, KERN_ERR, "can't mount with "
3784                                  "journal_checksum, fs mounted w/o journal");
3785                         goto failed_mount_wq;
3786                 }
3787                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3788                         ext4_msg(sb, KERN_ERR, "can't mount with "
3789                                  "journal_async_commit, fs mounted w/o journal");
3790                         goto failed_mount_wq;
3791                 }
3792                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3793                         ext4_msg(sb, KERN_ERR, "can't mount with "
3794                                  "commit=%lu, fs mounted w/o journal",
3795                                  sbi->s_commit_interval / HZ);
3796                         goto failed_mount_wq;
3797                 }
3798                 if (EXT4_MOUNT_DATA_FLAGS &
3799                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3800                         ext4_msg(sb, KERN_ERR, "can't mount with "
3801                                  "data=, fs mounted w/o journal");
3802                         goto failed_mount_wq;
3803                 }
3804                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
3805                 clear_opt(sb, JOURNAL_CHECKSUM);
3806                 clear_opt(sb, DATA_FLAGS);
3807                 sbi->s_journal = NULL;
3808                 needs_recovery = 0;
3809                 goto no_journal;
3810         }
3811
3812         if (ext4_has_feature_64bit(sb) &&
3813             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3814                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3815                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3816                 goto failed_mount_wq;
3817         }
3818
3819         if (!set_journal_csum_feature_set(sb)) {
3820                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3821                          "feature set");
3822                 goto failed_mount_wq;
3823         }
3824
3825         /* We have now updated the journal if required, so we can
3826          * validate the data journaling mode. */
3827         switch (test_opt(sb, DATA_FLAGS)) {
3828         case 0:
3829                 /* No mode set, assume a default based on the journal
3830                  * capabilities: ORDERED_DATA if the journal can
3831                  * cope, else JOURNAL_DATA
3832                  */
3833                 if (jbd2_journal_check_available_features
3834                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3835                         set_opt(sb, ORDERED_DATA);
3836                 else
3837                         set_opt(sb, JOURNAL_DATA);
3838                 break;
3839
3840         case EXT4_MOUNT_ORDERED_DATA:
3841         case EXT4_MOUNT_WRITEBACK_DATA:
3842                 if (!jbd2_journal_check_available_features
3843                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3844                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3845                                "requested data journaling mode");
3846                         goto failed_mount_wq;
3847                 }
3848         default:
3849                 break;
3850         }
3851         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3852
3853         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3854
3855 no_journal:
3856         if (ext4_mballoc_ready) {
3857                 sbi->s_mb_cache = ext4_xattr_create_cache();
3858                 if (!sbi->s_mb_cache) {
3859                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
3860                         goto failed_mount_wq;
3861                 }
3862         }
3863
3864         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
3865             (blocksize != PAGE_CACHE_SIZE)) {
3866                 ext4_msg(sb, KERN_ERR,
3867                          "Unsupported blocksize for fs encryption");
3868                 goto failed_mount_wq;
3869         }
3870
3871         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
3872             !ext4_has_feature_encrypt(sb)) {
3873                 ext4_set_feature_encrypt(sb);
3874                 ext4_commit_super(sb, 1);
3875         }
3876
3877         /*
3878          * Get the # of file system overhead blocks from the
3879          * superblock if present.
3880          */
3881         if (es->s_overhead_clusters)
3882                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3883         else {
3884                 err = ext4_calculate_overhead(sb);
3885                 if (err)
3886                         goto failed_mount_wq;
3887         }
3888
3889         /*
3890          * The maximum number of concurrent works can be high and
3891          * concurrency isn't really necessary.  Limit it to 1.
3892          */
3893         EXT4_SB(sb)->rsv_conversion_wq =
3894                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3895         if (!EXT4_SB(sb)->rsv_conversion_wq) {
3896                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3897                 ret = -ENOMEM;
3898                 goto failed_mount4;
3899         }
3900
3901         /*
3902          * The jbd2_journal_load will have done any necessary log recovery,
3903          * so we can safely mount the rest of the filesystem now.
3904          */
3905
3906         root = ext4_iget(sb, EXT4_ROOT_INO);
3907         if (IS_ERR(root)) {
3908                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3909                 ret = PTR_ERR(root);
3910                 root = NULL;
3911                 goto failed_mount4;
3912         }
3913         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3914                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3915                 iput(root);
3916                 goto failed_mount4;
3917         }
3918         sb->s_root = d_make_root(root);
3919         if (!sb->s_root) {
3920                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3921                 ret = -ENOMEM;
3922                 goto failed_mount4;
3923         }
3924
3925         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3926                 sb->s_flags |= MS_RDONLY;
3927
3928         /* determine the minimum size of new large inodes, if present */
3929         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3930                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3931                                                      EXT4_GOOD_OLD_INODE_SIZE;
3932                 if (ext4_has_feature_extra_isize(sb)) {
3933                         if (sbi->s_want_extra_isize <
3934                             le16_to_cpu(es->s_want_extra_isize))
3935                                 sbi->s_want_extra_isize =
3936                                         le16_to_cpu(es->s_want_extra_isize);
3937                         if (sbi->s_want_extra_isize <
3938                             le16_to_cpu(es->s_min_extra_isize))
3939                                 sbi->s_want_extra_isize =
3940                                         le16_to_cpu(es->s_min_extra_isize);
3941                 }
3942         }
3943         /* Check if enough inode space is available */
3944         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3945                                                         sbi->s_inode_size) {
3946                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3947                                                        EXT4_GOOD_OLD_INODE_SIZE;
3948                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3949                          "available");
3950         }
3951
3952         ext4_set_resv_clusters(sb);
3953
3954         err = ext4_setup_system_zone(sb);
3955         if (err) {
3956                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3957                          "zone (%d)", err);
3958                 goto failed_mount4a;
3959         }
3960
3961         ext4_ext_init(sb);
3962         err = ext4_mb_init(sb);
3963         if (err) {
3964                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3965                          err);
3966                 goto failed_mount5;
3967         }
3968
3969         block = ext4_count_free_clusters(sb);
3970         ext4_free_blocks_count_set(sbi->s_es,
3971                                    EXT4_C2B(sbi, block));
3972         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
3973                                   GFP_KERNEL);
3974         if (!err) {
3975                 unsigned long freei = ext4_count_free_inodes(sb);
3976                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
3977                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
3978                                           GFP_KERNEL);
3979         }
3980         if (!err)
3981                 err = percpu_counter_init(&sbi->s_dirs_counter,
3982                                           ext4_count_dirs(sb), GFP_KERNEL);
3983         if (!err)
3984                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
3985                                           GFP_KERNEL);
3986         if (err) {
3987                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3988                 goto failed_mount6;
3989         }
3990
3991         if (ext4_has_feature_flex_bg(sb))
3992                 if (!ext4_fill_flex_info(sb)) {
3993                         ext4_msg(sb, KERN_ERR,
3994                                "unable to initialize "
3995                                "flex_bg meta info!");
3996                         goto failed_mount6;
3997                 }
3998
3999         err = ext4_register_li_request(sb, first_not_zeroed);
4000         if (err)
4001                 goto failed_mount6;
4002
4003         err = ext4_register_sysfs(sb);
4004         if (err)
4005                 goto failed_mount7;
4006
4007 #ifdef CONFIG_QUOTA
4008         /* Enable quota usage during mount. */
4009         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
4010                 err = ext4_enable_quotas(sb);
4011                 if (err)
4012                         goto failed_mount8;
4013         }
4014 #endif  /* CONFIG_QUOTA */
4015
4016         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4017         ext4_orphan_cleanup(sb, es);
4018         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4019         if (needs_recovery) {
4020                 ext4_msg(sb, KERN_INFO, "recovery complete");
4021                 ext4_mark_recovery_complete(sb, es);
4022         }
4023         if (EXT4_SB(sb)->s_journal) {
4024                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4025                         descr = " journalled data mode";
4026                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4027                         descr = " ordered data mode";
4028                 else
4029                         descr = " writeback data mode";
4030         } else
4031                 descr = "out journal";
4032
4033         if (test_opt(sb, DISCARD)) {
4034                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4035                 if (!blk_queue_discard(q))
4036                         ext4_msg(sb, KERN_WARNING,
4037                                  "mounting with \"discard\" option, but "
4038                                  "the device does not support discard");
4039         }
4040
4041         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4042                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4043                          "Opts: %.*s%s%s", descr,
4044                          (int) sizeof(sbi->s_es->s_mount_opts),
4045                          sbi->s_es->s_mount_opts,
4046                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4047
4048         if (es->s_error_count)
4049                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4050
4051         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4052         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4053         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4054         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4055
4056         kfree(orig_data);
4057         return 0;
4058
4059 cantfind_ext4:
4060         if (!silent)
4061                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4062         goto failed_mount;
4063
4064 #ifdef CONFIG_QUOTA
4065 failed_mount8:
4066         ext4_unregister_sysfs(sb);
4067 #endif
4068 failed_mount7:
4069         ext4_unregister_li_request(sb);
4070 failed_mount6:
4071         ext4_mb_release(sb);
4072         if (sbi->s_flex_groups)
4073                 kvfree(sbi->s_flex_groups);
4074         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4075         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4076         percpu_counter_destroy(&sbi->s_dirs_counter);
4077         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4078 failed_mount5:
4079         ext4_ext_release(sb);
4080         ext4_release_system_zone(sb);
4081 failed_mount4a:
4082         dput(sb->s_root);
4083         sb->s_root = NULL;
4084 failed_mount4:
4085         ext4_msg(sb, KERN_ERR, "mount failed");
4086         if (EXT4_SB(sb)->rsv_conversion_wq)
4087                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4088 failed_mount_wq:
4089         if (sbi->s_mb_cache) {
4090                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
4091                 sbi->s_mb_cache = NULL;
4092         }
4093         if (sbi->s_journal) {
4094                 jbd2_journal_destroy(sbi->s_journal);
4095                 sbi->s_journal = NULL;
4096         }
4097 failed_mount3a:
4098         ext4_es_unregister_shrinker(sbi);
4099 failed_mount3:
4100         del_timer_sync(&sbi->s_err_report);
4101         if (sbi->s_mmp_tsk)
4102                 kthread_stop(sbi->s_mmp_tsk);
4103 failed_mount2:
4104         for (i = 0; i < db_count; i++)
4105                 brelse(sbi->s_group_desc[i]);
4106         kvfree(sbi->s_group_desc);
4107 failed_mount:
4108         if (sbi->s_chksum_driver)
4109                 crypto_free_shash(sbi->s_chksum_driver);
4110 #ifdef CONFIG_QUOTA
4111         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4112                 kfree(sbi->s_qf_names[i]);
4113 #endif
4114         ext4_blkdev_remove(sbi);
4115         brelse(bh);
4116 out_fail:
4117         sb->s_fs_info = NULL;
4118         kfree(sbi->s_blockgroup_lock);
4119 out_free_base:
4120         kfree(sbi);
4121         kfree(orig_data);
4122         return err ? err : ret;
4123 }
4124
4125 /*
4126  * Setup any per-fs journal parameters now.  We'll do this both on
4127  * initial mount, once the journal has been initialised but before we've
4128  * done any recovery; and again on any subsequent remount.
4129  */
4130 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4131 {
4132         struct ext4_sb_info *sbi = EXT4_SB(sb);
4133
4134         journal->j_commit_interval = sbi->s_commit_interval;
4135         journal->j_min_batch_time = sbi->s_min_batch_time;
4136         journal->j_max_batch_time = sbi->s_max_batch_time;
4137
4138         write_lock(&journal->j_state_lock);
4139         if (test_opt(sb, BARRIER))
4140                 journal->j_flags |= JBD2_BARRIER;
4141         else
4142                 journal->j_flags &= ~JBD2_BARRIER;
4143         if (test_opt(sb, DATA_ERR_ABORT))
4144                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4145         else
4146                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4147         write_unlock(&journal->j_state_lock);
4148 }
4149
4150 static journal_t *ext4_get_journal(struct super_block *sb,
4151                                    unsigned int journal_inum)
4152 {
4153         struct inode *journal_inode;
4154         journal_t *journal;
4155
4156         BUG_ON(!ext4_has_feature_journal(sb));
4157
4158         /* First, test for the existence of a valid inode on disk.  Bad
4159          * things happen if we iget() an unused inode, as the subsequent
4160          * iput() will try to delete it. */
4161
4162         journal_inode = ext4_iget(sb, journal_inum);
4163         if (IS_ERR(journal_inode)) {
4164                 ext4_msg(sb, KERN_ERR, "no journal found");
4165                 return NULL;
4166         }
4167         if (!journal_inode->i_nlink) {
4168                 make_bad_inode(journal_inode);
4169                 iput(journal_inode);
4170                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4171                 return NULL;
4172         }
4173
4174         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4175                   journal_inode, journal_inode->i_size);
4176         if (!S_ISREG(journal_inode->i_mode)) {
4177                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4178                 iput(journal_inode);
4179                 return NULL;
4180         }
4181
4182         journal = jbd2_journal_init_inode(journal_inode);
4183         if (!journal) {
4184                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4185                 iput(journal_inode);
4186                 return NULL;
4187         }
4188         journal->j_private = sb;
4189         ext4_init_journal_params(sb, journal);
4190         return journal;
4191 }
4192
4193 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4194                                        dev_t j_dev)
4195 {
4196         struct buffer_head *bh;
4197         journal_t *journal;
4198         ext4_fsblk_t start;
4199         ext4_fsblk_t len;
4200         int hblock, blocksize;
4201         ext4_fsblk_t sb_block;
4202         unsigned long offset;
4203         struct ext4_super_block *es;
4204         struct block_device *bdev;
4205
4206         BUG_ON(!ext4_has_feature_journal(sb));
4207
4208         bdev = ext4_blkdev_get(j_dev, sb);
4209         if (bdev == NULL)
4210                 return NULL;
4211
4212         blocksize = sb->s_blocksize;
4213         hblock = bdev_logical_block_size(bdev);
4214         if (blocksize < hblock) {
4215                 ext4_msg(sb, KERN_ERR,
4216                         "blocksize too small for journal device");
4217                 goto out_bdev;
4218         }
4219
4220         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4221         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4222         set_blocksize(bdev, blocksize);
4223         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4224                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4225                        "external journal");
4226                 goto out_bdev;
4227         }
4228
4229         es = (struct ext4_super_block *) (bh->b_data + offset);
4230         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4231             !(le32_to_cpu(es->s_feature_incompat) &
4232               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4233                 ext4_msg(sb, KERN_ERR, "external journal has "
4234                                         "bad superblock");
4235                 brelse(bh);
4236                 goto out_bdev;
4237         }
4238
4239         if ((le32_to_cpu(es->s_feature_ro_compat) &
4240              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4241             es->s_checksum != ext4_superblock_csum(sb, es)) {
4242                 ext4_msg(sb, KERN_ERR, "external journal has "
4243                                        "corrupt superblock");
4244                 brelse(bh);
4245                 goto out_bdev;
4246         }
4247
4248         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4249                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4250                 brelse(bh);
4251                 goto out_bdev;
4252         }
4253
4254         len = ext4_blocks_count(es);
4255         start = sb_block + 1;
4256         brelse(bh);     /* we're done with the superblock */
4257
4258         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4259                                         start, len, blocksize);
4260         if (!journal) {
4261                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4262                 goto out_bdev;
4263         }
4264         journal->j_private = sb;
4265         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4266         wait_on_buffer(journal->j_sb_buffer);
4267         if (!buffer_uptodate(journal->j_sb_buffer)) {
4268                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4269                 goto out_journal;
4270         }
4271         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4272                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4273                                         "user (unsupported) - %d",
4274                         be32_to_cpu(journal->j_superblock->s_nr_users));
4275                 goto out_journal;
4276         }
4277         EXT4_SB(sb)->journal_bdev = bdev;
4278         ext4_init_journal_params(sb, journal);
4279         return journal;
4280
4281 out_journal:
4282         jbd2_journal_destroy(journal);
4283 out_bdev:
4284         ext4_blkdev_put(bdev);
4285         return NULL;
4286 }
4287
4288 static int ext4_load_journal(struct super_block *sb,
4289                              struct ext4_super_block *es,
4290                              unsigned long journal_devnum)
4291 {
4292         journal_t *journal;
4293         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4294         dev_t journal_dev;
4295         int err = 0;
4296         int really_read_only;
4297
4298         BUG_ON(!ext4_has_feature_journal(sb));
4299
4300         if (journal_devnum &&
4301             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4302                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4303                         "numbers have changed");
4304                 journal_dev = new_decode_dev(journal_devnum);
4305         } else
4306                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4307
4308         really_read_only = bdev_read_only(sb->s_bdev);
4309
4310         /*
4311          * Are we loading a blank journal or performing recovery after a
4312          * crash?  For recovery, we need to check in advance whether we
4313          * can get read-write access to the device.
4314          */
4315         if (ext4_has_feature_journal_needs_recovery(sb)) {
4316                 if (sb->s_flags & MS_RDONLY) {
4317                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4318                                         "required on readonly filesystem");
4319                         if (really_read_only) {
4320                                 ext4_msg(sb, KERN_ERR, "write access "
4321                                         "unavailable, cannot proceed");
4322                                 return -EROFS;
4323                         }
4324                         ext4_msg(sb, KERN_INFO, "write access will "
4325                                "be enabled during recovery");
4326                 }
4327         }
4328
4329         if (journal_inum && journal_dev) {
4330                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4331                        "and inode journals!");
4332                 return -EINVAL;
4333         }
4334
4335         if (journal_inum) {
4336                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4337                         return -EINVAL;
4338         } else {
4339                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4340                         return -EINVAL;
4341         }
4342
4343         if (!(journal->j_flags & JBD2_BARRIER))
4344                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4345
4346         if (!ext4_has_feature_journal_needs_recovery(sb))
4347                 err = jbd2_journal_wipe(journal, !really_read_only);
4348         if (!err) {
4349                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4350                 if (save)
4351                         memcpy(save, ((char *) es) +
4352                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4353                 err = jbd2_journal_load(journal);
4354                 if (save)
4355                         memcpy(((char *) es) + EXT4_S_ERR_START,
4356                                save, EXT4_S_ERR_LEN);
4357                 kfree(save);
4358         }
4359
4360         if (err) {
4361                 ext4_msg(sb, KERN_ERR, "error loading journal");
4362                 jbd2_journal_destroy(journal);
4363                 return err;
4364         }
4365
4366         EXT4_SB(sb)->s_journal = journal;
4367         ext4_clear_journal_err(sb, es);
4368
4369         if (!really_read_only && journal_devnum &&
4370             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4371                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4372
4373                 /* Make sure we flush the recovery flag to disk. */
4374                 ext4_commit_super(sb, 1);
4375         }
4376
4377         return 0;
4378 }
4379
4380 static int ext4_commit_super(struct super_block *sb, int sync)
4381 {
4382         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4383         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4384         int error = 0;
4385
4386         if (!sbh || block_device_ejected(sb))
4387                 return error;
4388         if (buffer_write_io_error(sbh)) {
4389                 /*
4390                  * Oh, dear.  A previous attempt to write the
4391                  * superblock failed.  This could happen because the
4392                  * USB device was yanked out.  Or it could happen to
4393                  * be a transient write error and maybe the block will
4394                  * be remapped.  Nothing we can do but to retry the
4395                  * write and hope for the best.
4396                  */
4397                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4398                        "superblock detected");
4399                 clear_buffer_write_io_error(sbh);
4400                 set_buffer_uptodate(sbh);
4401         }
4402         /*
4403          * If the file system is mounted read-only, don't update the
4404          * superblock write time.  This avoids updating the superblock
4405          * write time when we are mounting the root file system
4406          * read/only but we need to replay the journal; at that point,
4407          * for people who are east of GMT and who make their clock
4408          * tick in localtime for Windows bug-for-bug compatibility,
4409          * the clock is set in the future, and this will cause e2fsck
4410          * to complain and force a full file system check.
4411          */
4412         if (!(sb->s_flags & MS_RDONLY))
4413                 es->s_wtime = cpu_to_le32(get_seconds());
4414         if (sb->s_bdev->bd_part)
4415                 es->s_kbytes_written =
4416                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4417                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4418                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4419         else
4420                 es->s_kbytes_written =
4421                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4422         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4423                 ext4_free_blocks_count_set(es,
4424                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4425                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4426         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4427                 es->s_free_inodes_count =
4428                         cpu_to_le32(percpu_counter_sum_positive(
4429                                 &EXT4_SB(sb)->s_freeinodes_counter));
4430         BUFFER_TRACE(sbh, "marking dirty");
4431         ext4_superblock_csum_set(sb);
4432         mark_buffer_dirty(sbh);
4433         if (sync) {
4434                 error = __sync_dirty_buffer(sbh,
4435                         test_opt(sb, BARRIER) ? WRITE_FUA : WRITE_SYNC);
4436                 if (error)
4437                         return error;
4438
4439                 error = buffer_write_io_error(sbh);
4440                 if (error) {
4441                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4442                                "superblock");
4443                         clear_buffer_write_io_error(sbh);
4444                         set_buffer_uptodate(sbh);
4445                 }
4446         }
4447         return error;
4448 }
4449
4450 /*
4451  * Have we just finished recovery?  If so, and if we are mounting (or
4452  * remounting) the filesystem readonly, then we will end up with a
4453  * consistent fs on disk.  Record that fact.
4454  */
4455 static void ext4_mark_recovery_complete(struct super_block *sb,
4456                                         struct ext4_super_block *es)
4457 {
4458         journal_t *journal = EXT4_SB(sb)->s_journal;
4459
4460         if (!ext4_has_feature_journal(sb)) {
4461                 BUG_ON(journal != NULL);
4462                 return;
4463         }
4464         jbd2_journal_lock_updates(journal);
4465         if (jbd2_journal_flush(journal) < 0)
4466                 goto out;
4467
4468         if (ext4_has_feature_journal_needs_recovery(sb) &&
4469             sb->s_flags & MS_RDONLY) {
4470                 ext4_clear_feature_journal_needs_recovery(sb);
4471                 ext4_commit_super(sb, 1);
4472         }
4473
4474 out:
4475         jbd2_journal_unlock_updates(journal);
4476 }
4477
4478 /*
4479  * If we are mounting (or read-write remounting) a filesystem whose journal
4480  * has recorded an error from a previous lifetime, move that error to the
4481  * main filesystem now.
4482  */
4483 static void ext4_clear_journal_err(struct super_block *sb,
4484                                    struct ext4_super_block *es)
4485 {
4486         journal_t *journal;
4487         int j_errno;
4488         const char *errstr;
4489
4490         BUG_ON(!ext4_has_feature_journal(sb));
4491
4492         journal = EXT4_SB(sb)->s_journal;
4493
4494         /*
4495          * Now check for any error status which may have been recorded in the
4496          * journal by a prior ext4_error() or ext4_abort()
4497          */
4498
4499         j_errno = jbd2_journal_errno(journal);
4500         if (j_errno) {
4501                 char nbuf[16];
4502
4503                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4504                 ext4_warning(sb, "Filesystem error recorded "
4505                              "from previous mount: %s", errstr);
4506                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4507
4508                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4509                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4510                 ext4_commit_super(sb, 1);
4511
4512                 jbd2_journal_clear_err(journal);
4513                 jbd2_journal_update_sb_errno(journal);
4514         }
4515 }
4516
4517 /*
4518  * Force the running and committing transactions to commit,
4519  * and wait on the commit.
4520  */
4521 int ext4_force_commit(struct super_block *sb)
4522 {
4523         journal_t *journal;
4524
4525         if (sb->s_flags & MS_RDONLY)
4526                 return 0;
4527
4528         journal = EXT4_SB(sb)->s_journal;
4529         return ext4_journal_force_commit(journal);
4530 }
4531
4532 static int ext4_sync_fs(struct super_block *sb, int wait)
4533 {
4534         int ret = 0;
4535         tid_t target;
4536         bool needs_barrier = false;
4537         struct ext4_sb_info *sbi = EXT4_SB(sb);
4538
4539         trace_ext4_sync_fs(sb, wait);
4540         flush_workqueue(sbi->rsv_conversion_wq);
4541         /*
4542          * Writeback quota in non-journalled quota case - journalled quota has
4543          * no dirty dquots
4544          */
4545         dquot_writeback_dquots(sb, -1);
4546         /*
4547          * Data writeback is possible w/o journal transaction, so barrier must
4548          * being sent at the end of the function. But we can skip it if
4549          * transaction_commit will do it for us.
4550          */
4551         if (sbi->s_journal) {
4552                 target = jbd2_get_latest_transaction(sbi->s_journal);
4553                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4554                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4555                         needs_barrier = true;
4556
4557                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4558                         if (wait)
4559                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4560                                                            target);
4561                 }
4562         } else if (wait && test_opt(sb, BARRIER))
4563                 needs_barrier = true;
4564         if (needs_barrier) {
4565                 int err;
4566                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4567                 if (!ret)
4568                         ret = err;
4569         }
4570
4571         return ret;
4572 }
4573
4574 /*
4575  * LVM calls this function before a (read-only) snapshot is created.  This
4576  * gives us a chance to flush the journal completely and mark the fs clean.
4577  *
4578  * Note that only this function cannot bring a filesystem to be in a clean
4579  * state independently. It relies on upper layer to stop all data & metadata
4580  * modifications.
4581  */
4582 static int ext4_freeze(struct super_block *sb)
4583 {
4584         int error = 0;
4585         journal_t *journal;
4586
4587         if (sb->s_flags & MS_RDONLY)
4588                 return 0;
4589
4590         journal = EXT4_SB(sb)->s_journal;
4591
4592         if (journal) {
4593                 /* Now we set up the journal barrier. */
4594                 jbd2_journal_lock_updates(journal);
4595
4596                 /*
4597                  * Don't clear the needs_recovery flag if we failed to
4598                  * flush the journal.
4599                  */
4600                 error = jbd2_journal_flush(journal);
4601                 if (error < 0)
4602                         goto out;
4603
4604                 /* Journal blocked and flushed, clear needs_recovery flag. */
4605                 ext4_clear_feature_journal_needs_recovery(sb);
4606         }
4607
4608         error = ext4_commit_super(sb, 1);
4609 out:
4610         if (journal)
4611                 /* we rely on upper layer to stop further updates */
4612                 jbd2_journal_unlock_updates(journal);
4613         return error;
4614 }
4615
4616 /*
4617  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4618  * flag here, even though the filesystem is not technically dirty yet.
4619  */
4620 static int ext4_unfreeze(struct super_block *sb)
4621 {
4622         if (sb->s_flags & MS_RDONLY)
4623                 return 0;
4624
4625         if (EXT4_SB(sb)->s_journal) {
4626                 /* Reset the needs_recovery flag before the fs is unlocked. */
4627                 ext4_set_feature_journal_needs_recovery(sb);
4628         }
4629
4630         ext4_commit_super(sb, 1);
4631         return 0;
4632 }
4633
4634 /*
4635  * Structure to save mount options for ext4_remount's benefit
4636  */
4637 struct ext4_mount_options {
4638         unsigned long s_mount_opt;
4639         unsigned long s_mount_opt2;
4640         kuid_t s_resuid;
4641         kgid_t s_resgid;
4642         unsigned long s_commit_interval;
4643         u32 s_min_batch_time, s_max_batch_time;
4644 #ifdef CONFIG_QUOTA
4645         int s_jquota_fmt;
4646         char *s_qf_names[EXT4_MAXQUOTAS];
4647 #endif
4648 };
4649
4650 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4651 {
4652         struct ext4_super_block *es;
4653         struct ext4_sb_info *sbi = EXT4_SB(sb);
4654         unsigned long old_sb_flags;
4655         struct ext4_mount_options old_opts;
4656         int enable_quota = 0;
4657         ext4_group_t g;
4658         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4659         int err = 0;
4660 #ifdef CONFIG_QUOTA
4661         int i, j;
4662 #endif
4663         char *orig_data = kstrdup(data, GFP_KERNEL);
4664
4665         /* Store the original options */
4666         old_sb_flags = sb->s_flags;
4667         old_opts.s_mount_opt = sbi->s_mount_opt;
4668         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4669         old_opts.s_resuid = sbi->s_resuid;
4670         old_opts.s_resgid = sbi->s_resgid;
4671         old_opts.s_commit_interval = sbi->s_commit_interval;
4672         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4673         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4674 #ifdef CONFIG_QUOTA
4675         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4676         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4677                 if (sbi->s_qf_names[i]) {
4678                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4679                                                          GFP_KERNEL);
4680                         if (!old_opts.s_qf_names[i]) {
4681                                 for (j = 0; j < i; j++)
4682                                         kfree(old_opts.s_qf_names[j]);
4683                                 kfree(orig_data);
4684                                 return -ENOMEM;
4685                         }
4686                 } else
4687                         old_opts.s_qf_names[i] = NULL;
4688 #endif
4689         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4690                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4691
4692         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4693                 err = -EINVAL;
4694                 goto restore_opts;
4695         }
4696
4697         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4698             test_opt(sb, JOURNAL_CHECKSUM)) {
4699                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4700                          "during remount not supported; ignoring");
4701                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4702         }
4703
4704         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4705                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4706                         ext4_msg(sb, KERN_ERR, "can't mount with "
4707                                  "both data=journal and delalloc");
4708                         err = -EINVAL;
4709                         goto restore_opts;
4710                 }
4711                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4712                         ext4_msg(sb, KERN_ERR, "can't mount with "
4713                                  "both data=journal and dioread_nolock");
4714                         err = -EINVAL;
4715                         goto restore_opts;
4716                 }
4717                 if (test_opt(sb, DAX)) {
4718                         ext4_msg(sb, KERN_ERR, "can't mount with "
4719                                  "both data=journal and dax");
4720                         err = -EINVAL;
4721                         goto restore_opts;
4722                 }
4723         }
4724
4725         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4726                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4727                         "dax flag with busy inodes while remounting");
4728                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4729         }
4730
4731         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4732                 ext4_abort(sb, "Abort forced by user");
4733
4734         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4735                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4736
4737         es = sbi->s_es;
4738
4739         if (sbi->s_journal) {
4740                 ext4_init_journal_params(sb, sbi->s_journal);
4741                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4742         }
4743
4744         if (*flags & MS_LAZYTIME)
4745                 sb->s_flags |= MS_LAZYTIME;
4746
4747         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4748                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4749                         err = -EROFS;
4750                         goto restore_opts;
4751                 }
4752
4753                 if (*flags & MS_RDONLY) {
4754                         err = sync_filesystem(sb);
4755                         if (err < 0)
4756                                 goto restore_opts;
4757                         err = dquot_suspend(sb, -1);
4758                         if (err < 0)
4759                                 goto restore_opts;
4760
4761                         /*
4762                          * First of all, the unconditional stuff we have to do
4763                          * to disable replay of the journal when we next remount
4764                          */
4765                         sb->s_flags |= MS_RDONLY;
4766
4767                         /*
4768                          * OK, test if we are remounting a valid rw partition
4769                          * readonly, and if so set the rdonly flag and then
4770                          * mark the partition as valid again.
4771                          */
4772                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4773                             (sbi->s_mount_state & EXT4_VALID_FS))
4774                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4775
4776                         if (sbi->s_journal)
4777                                 ext4_mark_recovery_complete(sb, es);
4778                 } else {
4779                         /* Make sure we can mount this feature set readwrite */
4780                         if (ext4_has_feature_readonly(sb) ||
4781                             !ext4_feature_set_ok(sb, 0)) {
4782                                 err = -EROFS;
4783                                 goto restore_opts;
4784                         }
4785                         /*
4786                          * Make sure the group descriptor checksums
4787                          * are sane.  If they aren't, refuse to remount r/w.
4788                          */
4789                         for (g = 0; g < sbi->s_groups_count; g++) {
4790                                 struct ext4_group_desc *gdp =
4791                                         ext4_get_group_desc(sb, g, NULL);
4792
4793                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4794                                         ext4_msg(sb, KERN_ERR,
4795                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4796                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
4797                                                le16_to_cpu(gdp->bg_checksum));
4798                                         err = -EFSBADCRC;
4799                                         goto restore_opts;
4800                                 }
4801                         }
4802
4803                         /*
4804                          * If we have an unprocessed orphan list hanging
4805                          * around from a previously readonly bdev mount,
4806                          * require a full umount/remount for now.
4807                          */
4808                         if (es->s_last_orphan) {
4809                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4810                                        "remount RDWR because of unprocessed "
4811                                        "orphan inode list.  Please "
4812                                        "umount/remount instead");
4813                                 err = -EINVAL;
4814                                 goto restore_opts;
4815                         }
4816
4817                         /*
4818                          * Mounting a RDONLY partition read-write, so reread
4819                          * and store the current valid flag.  (It may have
4820                          * been changed by e2fsck since we originally mounted
4821                          * the partition.)
4822                          */
4823                         if (sbi->s_journal)
4824                                 ext4_clear_journal_err(sb, es);
4825                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4826                         if (!ext4_setup_super(sb, es, 0))
4827                                 sb->s_flags &= ~MS_RDONLY;
4828                         if (ext4_has_feature_mmp(sb))
4829                                 if (ext4_multi_mount_protect(sb,
4830                                                 le64_to_cpu(es->s_mmp_block))) {
4831                                         err = -EROFS;
4832                                         goto restore_opts;
4833                                 }
4834                         enable_quota = 1;
4835                 }
4836         }
4837
4838         /*
4839          * Reinitialize lazy itable initialization thread based on
4840          * current settings
4841          */
4842         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4843                 ext4_unregister_li_request(sb);
4844         else {
4845                 ext4_group_t first_not_zeroed;
4846                 first_not_zeroed = ext4_has_uninit_itable(sb);
4847                 ext4_register_li_request(sb, first_not_zeroed);
4848         }
4849
4850         ext4_setup_system_zone(sb);
4851         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4852                 ext4_commit_super(sb, 1);
4853
4854 #ifdef CONFIG_QUOTA
4855         /* Release old quota file names */
4856         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4857                 kfree(old_opts.s_qf_names[i]);
4858         if (enable_quota) {
4859                 if (sb_any_quota_suspended(sb))
4860                         dquot_resume(sb, -1);
4861                 else if (ext4_has_feature_quota(sb)) {
4862                         err = ext4_enable_quotas(sb);
4863                         if (err)
4864                                 goto restore_opts;
4865                 }
4866         }
4867 #endif
4868
4869         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
4870         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4871         kfree(orig_data);
4872         return 0;
4873
4874 restore_opts:
4875         sb->s_flags = old_sb_flags;
4876         sbi->s_mount_opt = old_opts.s_mount_opt;
4877         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4878         sbi->s_resuid = old_opts.s_resuid;
4879         sbi->s_resgid = old_opts.s_resgid;
4880         sbi->s_commit_interval = old_opts.s_commit_interval;
4881         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4882         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4883 #ifdef CONFIG_QUOTA
4884         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4885         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
4886                 kfree(sbi->s_qf_names[i]);
4887                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4888         }
4889 #endif
4890         kfree(orig_data);
4891         return err;
4892 }
4893
4894 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4895 {
4896         struct super_block *sb = dentry->d_sb;
4897         struct ext4_sb_info *sbi = EXT4_SB(sb);
4898         struct ext4_super_block *es = sbi->s_es;
4899         ext4_fsblk_t overhead = 0, resv_blocks;
4900         u64 fsid;
4901         s64 bfree;
4902         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4903
4904         if (!test_opt(sb, MINIX_DF))
4905                 overhead = sbi->s_overhead;
4906
4907         buf->f_type = EXT4_SUPER_MAGIC;
4908         buf->f_bsize = sb->s_blocksize;
4909         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4910         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4911                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4912         /* prevent underflow in case that few free space is available */
4913         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4914         buf->f_bavail = buf->f_bfree -
4915                         (ext4_r_blocks_count(es) + resv_blocks);
4916         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4917                 buf->f_bavail = 0;
4918         buf->f_files = le32_to_cpu(es->s_inodes_count);
4919         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4920         buf->f_namelen = EXT4_NAME_LEN;
4921         fsid = le64_to_cpup((void *)es->s_uuid) ^
4922                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4923         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4924         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4925
4926         return 0;
4927 }
4928
4929 /* Helper function for writing quotas on sync - we need to start transaction
4930  * before quota file is locked for write. Otherwise the are possible deadlocks:
4931  * Process 1                         Process 2
4932  * ext4_create()                     quota_sync()
4933  *   jbd2_journal_start()                  write_dquot()
4934  *   dquot_initialize()                         down(dqio_mutex)
4935  *     down(dqio_mutex)                    jbd2_journal_start()
4936  *
4937  */
4938
4939 #ifdef CONFIG_QUOTA
4940
4941 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4942 {
4943         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4944 }
4945
4946 static int ext4_write_dquot(struct dquot *dquot)
4947 {
4948         int ret, err;
4949         handle_t *handle;
4950         struct inode *inode;
4951
4952         inode = dquot_to_inode(dquot);
4953         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4954                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4955         if (IS_ERR(handle))
4956                 return PTR_ERR(handle);
4957         ret = dquot_commit(dquot);
4958         err = ext4_journal_stop(handle);
4959         if (!ret)
4960                 ret = err;
4961         return ret;
4962 }
4963
4964 static int ext4_acquire_dquot(struct dquot *dquot)
4965 {
4966         int ret, err;
4967         handle_t *handle;
4968
4969         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4970                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4971         if (IS_ERR(handle))
4972                 return PTR_ERR(handle);
4973         ret = dquot_acquire(dquot);
4974         err = ext4_journal_stop(handle);
4975         if (!ret)
4976                 ret = err;
4977         return ret;
4978 }
4979
4980 static int ext4_release_dquot(struct dquot *dquot)
4981 {
4982         int ret, err;
4983         handle_t *handle;
4984
4985         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4986                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4987         if (IS_ERR(handle)) {
4988                 /* Release dquot anyway to avoid endless cycle in dqput() */
4989                 dquot_release(dquot);
4990                 return PTR_ERR(handle);
4991         }
4992         ret = dquot_release(dquot);
4993         err = ext4_journal_stop(handle);
4994         if (!ret)
4995                 ret = err;
4996         return ret;
4997 }
4998
4999 static int ext4_mark_dquot_dirty(struct dquot *dquot)
5000 {
5001         struct super_block *sb = dquot->dq_sb;
5002         struct ext4_sb_info *sbi = EXT4_SB(sb);
5003
5004         /* Are we journaling quotas? */
5005         if (ext4_has_feature_quota(sb) ||
5006             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5007                 dquot_mark_dquot_dirty(dquot);
5008                 return ext4_write_dquot(dquot);
5009         } else {
5010                 return dquot_mark_dquot_dirty(dquot);
5011         }
5012 }
5013
5014 static int ext4_write_info(struct super_block *sb, int type)
5015 {
5016         int ret, err;
5017         handle_t *handle;
5018
5019         /* Data block + inode block */
5020         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5021         if (IS_ERR(handle))
5022                 return PTR_ERR(handle);
5023         ret = dquot_commit_info(sb, type);
5024         err = ext4_journal_stop(handle);
5025         if (!ret)
5026                 ret = err;
5027         return ret;
5028 }
5029
5030 /*
5031  * Turn on quotas during mount time - we need to find
5032  * the quota file and such...
5033  */
5034 static int ext4_quota_on_mount(struct super_block *sb, int type)
5035 {
5036         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5037                                         EXT4_SB(sb)->s_jquota_fmt, type);
5038 }
5039
5040 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5041 {
5042         struct ext4_inode_info *ei = EXT4_I(inode);
5043
5044         /* The first argument of lockdep_set_subclass has to be
5045          * *exactly* the same as the argument to init_rwsem() --- in
5046          * this case, in init_once() --- or lockdep gets unhappy
5047          * because the name of the lock is set using the
5048          * stringification of the argument to init_rwsem().
5049          */
5050         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5051         lockdep_set_subclass(&ei->i_data_sem, subclass);
5052 }
5053
5054 /*
5055  * Standard function to be called on quota_on
5056  */
5057 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5058                          struct path *path)
5059 {
5060         int err;
5061
5062         if (!test_opt(sb, QUOTA))
5063                 return -EINVAL;
5064
5065         /* Quotafile not on the same filesystem? */
5066         if (path->dentry->d_sb != sb)
5067                 return -EXDEV;
5068         /* Journaling quota? */
5069         if (EXT4_SB(sb)->s_qf_names[type]) {
5070                 /* Quotafile not in fs root? */
5071                 if (path->dentry->d_parent != sb->s_root)
5072                         ext4_msg(sb, KERN_WARNING,
5073                                 "Quota file not on filesystem root. "
5074                                 "Journaled quota will not work");
5075         }
5076
5077         /*
5078          * When we journal data on quota file, we have to flush journal to see
5079          * all updates to the file when we bypass pagecache...
5080          */
5081         if (EXT4_SB(sb)->s_journal &&
5082             ext4_should_journal_data(d_inode(path->dentry))) {
5083                 /*
5084                  * We don't need to lock updates but journal_flush() could
5085                  * otherwise be livelocked...
5086                  */
5087                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5088                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5089                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5090                 if (err)
5091                         return err;
5092         }
5093         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5094         err = dquot_quota_on(sb, type, format_id, path);
5095         if (err)
5096                 lockdep_set_quota_inode(path->dentry->d_inode,
5097                                              I_DATA_SEM_NORMAL);
5098         return err;
5099 }
5100
5101 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5102                              unsigned int flags)
5103 {
5104         int err;
5105         struct inode *qf_inode;
5106         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5107                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5108                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5109         };
5110
5111         BUG_ON(!ext4_has_feature_quota(sb));
5112
5113         if (!qf_inums[type])
5114                 return -EPERM;
5115
5116         qf_inode = ext4_iget(sb, qf_inums[type]);
5117         if (IS_ERR(qf_inode)) {
5118                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5119                 return PTR_ERR(qf_inode);
5120         }
5121
5122         /* Don't account quota for quota files to avoid recursion */
5123         qf_inode->i_flags |= S_NOQUOTA;
5124         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5125         err = dquot_enable(qf_inode, type, format_id, flags);
5126         iput(qf_inode);
5127         if (err)
5128                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5129
5130         return err;
5131 }
5132
5133 /* Enable usage tracking for all quota types. */
5134 static int ext4_enable_quotas(struct super_block *sb)
5135 {
5136         int type, err = 0;
5137         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5138                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5139                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5140         };
5141
5142         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5143         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5144                 if (qf_inums[type]) {
5145                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5146                                                 DQUOT_USAGE_ENABLED);
5147                         if (err) {
5148                                 for (type--; type >= 0; type--)
5149                                         dquot_quota_off(sb, type);
5150
5151                                 ext4_warning(sb,
5152                                         "Failed to enable quota tracking "
5153                                         "(type=%d, err=%d). Please run "
5154                                         "e2fsck to fix.", type, err);
5155                                 return err;
5156                         }
5157                 }
5158         }
5159         return 0;
5160 }
5161
5162 static int ext4_quota_off(struct super_block *sb, int type)
5163 {
5164         struct inode *inode = sb_dqopt(sb)->files[type];
5165         handle_t *handle;
5166
5167         /* Force all delayed allocation blocks to be allocated.
5168          * Caller already holds s_umount sem */
5169         if (test_opt(sb, DELALLOC))
5170                 sync_filesystem(sb);
5171
5172         if (!inode)
5173                 goto out;
5174
5175         /* Update modification times of quota files when userspace can
5176          * start looking at them */
5177         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5178         if (IS_ERR(handle))
5179                 goto out;
5180         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5181         ext4_mark_inode_dirty(handle, inode);
5182         ext4_journal_stop(handle);
5183
5184 out:
5185         return dquot_quota_off(sb, type);
5186 }
5187
5188 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5189  * acquiring the locks... As quota files are never truncated and quota code
5190  * itself serializes the operations (and no one else should touch the files)
5191  * we don't have to be afraid of races */
5192 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5193                                size_t len, loff_t off)
5194 {
5195         struct inode *inode = sb_dqopt(sb)->files[type];
5196         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5197         int offset = off & (sb->s_blocksize - 1);
5198         int tocopy;
5199         size_t toread;
5200         struct buffer_head *bh;
5201         loff_t i_size = i_size_read(inode);
5202
5203         if (off > i_size)
5204                 return 0;
5205         if (off+len > i_size)
5206                 len = i_size-off;
5207         toread = len;
5208         while (toread > 0) {
5209                 tocopy = sb->s_blocksize - offset < toread ?
5210                                 sb->s_blocksize - offset : toread;
5211                 bh = ext4_bread(NULL, inode, blk, 0);
5212                 if (IS_ERR(bh))
5213                         return PTR_ERR(bh);
5214                 if (!bh)        /* A hole? */
5215                         memset(data, 0, tocopy);
5216                 else
5217                         memcpy(data, bh->b_data+offset, tocopy);
5218                 brelse(bh);
5219                 offset = 0;
5220                 toread -= tocopy;
5221                 data += tocopy;
5222                 blk++;
5223         }
5224         return len;
5225 }
5226
5227 /* Write to quotafile (we know the transaction is already started and has
5228  * enough credits) */
5229 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5230                                 const char *data, size_t len, loff_t off)
5231 {
5232         struct inode *inode = sb_dqopt(sb)->files[type];
5233         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5234         int err, offset = off & (sb->s_blocksize - 1);
5235         int retries = 0;
5236         struct buffer_head *bh;
5237         handle_t *handle = journal_current_handle();
5238
5239         if (EXT4_SB(sb)->s_journal && !handle) {
5240                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5241                         " cancelled because transaction is not started",
5242                         (unsigned long long)off, (unsigned long long)len);
5243                 return -EIO;
5244         }
5245         /*
5246          * Since we account only one data block in transaction credits,
5247          * then it is impossible to cross a block boundary.
5248          */
5249         if (sb->s_blocksize - offset < len) {
5250                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5251                         " cancelled because not block aligned",
5252                         (unsigned long long)off, (unsigned long long)len);
5253                 return -EIO;
5254         }
5255
5256         do {
5257                 bh = ext4_bread(handle, inode, blk,
5258                                 EXT4_GET_BLOCKS_CREATE |
5259                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5260         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5261                  ext4_should_retry_alloc(inode->i_sb, &retries));
5262         if (IS_ERR(bh))
5263                 return PTR_ERR(bh);
5264         if (!bh)
5265                 goto out;
5266         BUFFER_TRACE(bh, "get write access");
5267         err = ext4_journal_get_write_access(handle, bh);
5268         if (err) {
5269                 brelse(bh);
5270                 return err;
5271         }
5272         lock_buffer(bh);
5273         memcpy(bh->b_data+offset, data, len);
5274         flush_dcache_page(bh->b_page);
5275         unlock_buffer(bh);
5276         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5277         brelse(bh);
5278 out:
5279         if (inode->i_size < off + len) {
5280                 i_size_write(inode, off + len);
5281                 EXT4_I(inode)->i_disksize = inode->i_size;
5282                 ext4_mark_inode_dirty(handle, inode);
5283         }
5284         return len;
5285 }
5286
5287 #endif
5288
5289 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5290                        const char *dev_name, void *data)
5291 {
5292         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5293 }
5294
5295 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5296 static inline void register_as_ext2(void)
5297 {
5298         int err = register_filesystem(&ext2_fs_type);
5299         if (err)
5300                 printk(KERN_WARNING
5301                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5302 }
5303
5304 static inline void unregister_as_ext2(void)
5305 {
5306         unregister_filesystem(&ext2_fs_type);
5307 }
5308
5309 static inline int ext2_feature_set_ok(struct super_block *sb)
5310 {
5311         if (ext4_has_unknown_ext2_incompat_features(sb))
5312                 return 0;
5313         if (sb->s_flags & MS_RDONLY)
5314                 return 1;
5315         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5316                 return 0;
5317         return 1;
5318 }
5319 #else
5320 static inline void register_as_ext2(void) { }
5321 static inline void unregister_as_ext2(void) { }
5322 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5323 #endif
5324
5325 static inline void register_as_ext3(void)
5326 {
5327         int err = register_filesystem(&ext3_fs_type);
5328         if (err)
5329                 printk(KERN_WARNING
5330                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5331 }
5332
5333 static inline void unregister_as_ext3(void)
5334 {
5335         unregister_filesystem(&ext3_fs_type);
5336 }
5337
5338 static inline int ext3_feature_set_ok(struct super_block *sb)
5339 {
5340         if (ext4_has_unknown_ext3_incompat_features(sb))
5341                 return 0;
5342         if (!ext4_has_feature_journal(sb))
5343                 return 0;
5344         if (sb->s_flags & MS_RDONLY)
5345                 return 1;
5346         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5347                 return 0;
5348         return 1;
5349 }
5350
5351 static struct file_system_type ext4_fs_type = {
5352         .owner          = THIS_MODULE,
5353         .name           = "ext4",
5354         .mount          = ext4_mount,
5355         .kill_sb        = kill_block_super,
5356         .fs_flags       = FS_REQUIRES_DEV,
5357 };
5358 MODULE_ALIAS_FS("ext4");
5359
5360 /* Shared across all ext4 file systems */
5361 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5362 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5363
5364 static int __init ext4_init_fs(void)
5365 {
5366         int i, err;
5367
5368         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5369         ext4_li_info = NULL;
5370         mutex_init(&ext4_li_mtx);
5371
5372         /* Build-time check for flags consistency */
5373         ext4_check_flag_values();
5374
5375         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5376                 mutex_init(&ext4__aio_mutex[i]);
5377                 init_waitqueue_head(&ext4__ioend_wq[i]);
5378         }
5379
5380         err = ext4_init_es();
5381         if (err)
5382                 return err;
5383
5384         err = ext4_init_pageio();
5385         if (err)
5386                 goto out5;
5387
5388         err = ext4_init_system_zone();
5389         if (err)
5390                 goto out4;
5391
5392         err = ext4_init_sysfs();
5393         if (err)
5394                 goto out3;
5395
5396         err = ext4_init_mballoc();
5397         if (err)
5398                 goto out2;
5399         else
5400                 ext4_mballoc_ready = 1;
5401         err = init_inodecache();
5402         if (err)
5403                 goto out1;
5404         register_as_ext3();
5405         register_as_ext2();
5406         err = register_filesystem(&ext4_fs_type);
5407         if (err)
5408                 goto out;
5409
5410         return 0;
5411 out:
5412         unregister_as_ext2();
5413         unregister_as_ext3();
5414         destroy_inodecache();
5415 out1:
5416         ext4_mballoc_ready = 0;
5417         ext4_exit_mballoc();
5418 out2:
5419         ext4_exit_sysfs();
5420 out3:
5421         ext4_exit_system_zone();
5422 out4:
5423         ext4_exit_pageio();
5424 out5:
5425         ext4_exit_es();
5426
5427         return err;
5428 }
5429
5430 static void __exit ext4_exit_fs(void)
5431 {
5432         ext4_exit_crypto();
5433         ext4_destroy_lazyinit_thread();
5434         unregister_as_ext2();
5435         unregister_as_ext3();
5436         unregister_filesystem(&ext4_fs_type);
5437         destroy_inodecache();
5438         ext4_exit_mballoc();
5439         ext4_exit_sysfs();
5440         ext4_exit_system_zone();
5441         ext4_exit_pageio();
5442         ext4_exit_es();
5443 }
5444
5445 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5446 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5447 MODULE_LICENSE("GPL");
5448 module_init(ext4_init_fs)
5449 module_exit(ext4_exit_fs)