fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/ctype.h>
  38 #include <linux/log2.h>
  39 #include <linux/crc16.h>
  40 #include <linux/cleancache.h>
  41 #include <asm/uaccess.h>
  42
  43 #include <linux/kthread.h>
  44 #include <linux/freezer.h>
  45
  46 #include "ext4.h"
  47 #include "ext4_extents.h"       /* Needed for trace points definition */
  48 #include "ext4_jbd2.h"
  49 #include "xattr.h"
  50 #include "acl.h"
  51 #include "mballoc.h"
  52
  53 #define CREATE_TRACE_POINTS
  54 #include <trace/events/ext4.h>
  55
  56 static struct ext4_lazy_init *ext4_li_info;
  57 static struct mutex ext4_li_mtx;
  58 static int ext4_mballoc_ready;
  59 static struct ratelimit_state ext4_mount_msg_ratelimit;
  60
  61 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  62                              unsigned long journal_devnum);
  63 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  64 static int ext4_commit_super(struct super_block *sb, int sync);
  65 static void ext4_mark_recovery_complete(struct super_block *sb,
  66                                         struct ext4_super_block *es);
  67 static void ext4_clear_journal_err(struct super_block *sb,
  68                                    struct ext4_super_block *es);
  69 static int ext4_sync_fs(struct super_block *sb, int wait);
  70 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  71 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  72 static int ext4_unfreeze(struct super_block *sb);
  73 static int ext4_freeze(struct super_block *sb);
  74 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  75                        const char *dev_name, void *data);
  76 static inline int ext2_feature_set_ok(struct super_block *sb);
  77 static inline int ext3_feature_set_ok(struct super_block *sb);
  78 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  79 static void ext4_destroy_lazyinit_thread(void);
  80 static void ext4_unregister_li_request(struct super_block *sb);
  81 static void ext4_clear_request_list(void);
  82
  83 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
  84 static struct file_system_type ext2_fs_type = {
  85         .owner          = THIS_MODULE,
  86         .name           = "ext2",
  87         .mount          = ext4_mount,
  88         .kill_sb        = kill_block_super,
  89         .fs_flags       = FS_REQUIRES_DEV,
  90 };
  91 MODULE_ALIAS_FS("ext2");
  92 MODULE_ALIAS("ext2");
  93 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  94 #else
  95 #define IS_EXT2_SB(sb) (0)
  96 #endif
  97
  98
  99 static struct file_system_type ext3_fs_type = {
 100         .owner          = THIS_MODULE,
 101         .name           = "ext3",
 102         .mount          = ext4_mount,
 103         .kill_sb        = kill_block_super,
 104         .fs_flags       = FS_REQUIRES_DEV,
 105 };
 106 MODULE_ALIAS_FS("ext3");
 107 MODULE_ALIAS("ext3");
 108 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 109
 110 static int ext4_verify_csum_type(struct super_block *sb,
 111                                  struct ext4_super_block *es)
 112 {
 113         if (!ext4_has_feature_metadata_csum(sb))
 114                 return 1;
 115
 116         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 117 }
 118
 119 static __le32 ext4_superblock_csum(struct super_block *sb,
 120                                    struct ext4_super_block *es)
 121 {
 122         struct ext4_sb_info *sbi = EXT4_SB(sb);
 123         int offset = offsetof(struct ext4_super_block, s_checksum);
 124         __u32 csum;
 125
 126         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 127
 128         return cpu_to_le32(csum);
 129 }
 130
 131 static int ext4_superblock_csum_verify(struct super_block *sb,
 132                                        struct ext4_super_block *es)
 133 {
 134         if (!ext4_has_metadata_csum(sb))
 135                 return 1;
 136
 137         return es->s_checksum == ext4_superblock_csum(sb, es);
 138 }
 139
 140 void ext4_superblock_csum_set(struct super_block *sb)
 141 {
 142         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 143
 144         if (!ext4_has_metadata_csum(sb))
 145                 return;
 146
 147         es->s_checksum = ext4_superblock_csum(sb, es);
 148 }
 149
 150 void *ext4_kvmalloc(size_t size, gfp_t flags)
 151 {
 152         void *ret;
 153
 154         ret = kmalloc(size, flags | __GFP_NOWARN);
 155         if (!ret)
 156                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 157         return ret;
 158 }
 159
 160 void *ext4_kvzalloc(size_t size, gfp_t flags)
 161 {
 162         void *ret;
 163
 164         ret = kzalloc(size, flags | __GFP_NOWARN);
 165         if (!ret)
 166                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 167         return ret;
 168 }
 169
 170 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 171                                struct ext4_group_desc *bg)
 172 {
 173         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 174                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 175                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 176 }
 177
 178 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 179                                struct ext4_group_desc *bg)
 180 {
 181         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 182                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 183                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 184 }
 185
 186 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 187                               struct ext4_group_desc *bg)
 188 {
 189         return le32_to_cpu(bg->bg_inode_table_lo) |
 190                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 191                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 192 }
 193
 194 __u32 ext4_free_group_clusters(struct super_block *sb,
 195                                struct ext4_group_desc *bg)
 196 {
 197         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 198                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 199                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 200 }
 201
 202 __u32 ext4_free_inodes_count(struct super_block *sb,
 203                               struct ext4_group_desc *bg)
 204 {
 205         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 206                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 207                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 208 }
 209
 210 __u32 ext4_used_dirs_count(struct super_block *sb,
 211                               struct ext4_group_desc *bg)
 212 {
 213         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 214                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 215                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 216 }
 217
 218 __u32 ext4_itable_unused_count(struct super_block *sb,
 219                               struct ext4_group_desc *bg)
 220 {
 221         return le16_to_cpu(bg->bg_itable_unused_lo) |
 222                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 223                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 224 }
 225
 226 void ext4_block_bitmap_set(struct super_block *sb,
 227                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 228 {
 229         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 230         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 231                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 232 }
 233
 234 void ext4_inode_bitmap_set(struct super_block *sb,
 235                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 236 {
 237         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 238         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 239                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 240 }
 241
 242 void ext4_inode_table_set(struct super_block *sb,
 243                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 244 {
 245         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 246         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 247                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 248 }
 249
 250 void ext4_free_group_clusters_set(struct super_block *sb,
 251                                   struct ext4_group_desc *bg, __u32 count)
 252 {
 253         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 254         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 255                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 256 }
 257
 258 void ext4_free_inodes_set(struct super_block *sb,
 259                           struct ext4_group_desc *bg, __u32 count)
 260 {
 261         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 262         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 263                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 264 }
 265
 266 void ext4_used_dirs_set(struct super_block *sb,
 267                           struct ext4_group_desc *bg, __u32 count)
 268 {
 269         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 270         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 271                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 272 }
 273
 274 void ext4_itable_unused_set(struct super_block *sb,
 275                           struct ext4_group_desc *bg, __u32 count)
 276 {
 277         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 278         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 279                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 280 }
 281
 282
 283 static void __save_error_info(struct super_block *sb, const char *func,
 284                             unsigned int line)
 285 {
 286         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 287
 288         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 289         if (bdev_read_only(sb->s_bdev))
 290                 return;
 291         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 292         es->s_last_error_time = cpu_to_le32(get_seconds());
 293         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 294         es->s_last_error_line = cpu_to_le32(line);
 295         if (!es->s_first_error_time) {
 296                 es->s_first_error_time = es->s_last_error_time;
 297                 strncpy(es->s_first_error_func, func,
 298                         sizeof(es->s_first_error_func));
 299                 es->s_first_error_line = cpu_to_le32(line);
 300                 es->s_first_error_ino = es->s_last_error_ino;
 301                 es->s_first_error_block = es->s_last_error_block;
 302         }
 303         /*
 304          * Start the daily error reporting function if it hasn't been
 305          * started already
 306          */
 307         if (!es->s_error_count)
 308                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 309         le32_add_cpu(&es->s_error_count, 1);
 310 }
 311
 312 static void save_error_info(struct super_block *sb, const char *func,
 313                             unsigned int line)
 314 {
 315         __save_error_info(sb, func, line);
 316         ext4_commit_super(sb, 1);
 317 }
 318
 319 /*
 320  * The del_gendisk() function uninitializes the disk-specific data
 321  * structures, including the bdi structure, without telling anyone
 322  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 323  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 324  * This is a kludge to prevent these oops until we can put in a proper
 325  * hook in del_gendisk() to inform the VFS and file system layers.
 326  */
 327 static int block_device_ejected(struct super_block *sb)
 328 {
 329         struct inode *bd_inode = sb->s_bdev->bd_inode;
 330         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 331
 332         return bdi->dev == NULL;
 333 }
 334
 335 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 336 {
 337         struct super_block              *sb = journal->j_private;
 338         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 339         int                             error = is_journal_aborted(journal);
 340         struct ext4_journal_cb_entry    *jce;
 341
 342         BUG_ON(txn->t_state == T_FINISHED);
 343         spin_lock(&sbi->s_md_lock);
 344         while (!list_empty(&txn->t_private_list)) {
 345                 jce = list_entry(txn->t_private_list.next,
 346                                  struct ext4_journal_cb_entry, jce_list);
 347                 list_del_init(&jce->jce_list);
 348                 spin_unlock(&sbi->s_md_lock);
 349                 jce->jce_func(sb, jce, error);
 350                 spin_lock(&sbi->s_md_lock);
 351         }
 352         spin_unlock(&sbi->s_md_lock);
 353 }
 354
 355 /* Deal with the reporting of failure conditions on a filesystem such as
 356  * inconsistencies detected or read IO failures.
 357  *
 358  * On ext2, we can store the error state of the filesystem in the
 359  * superblock.  That is not possible on ext4, because we may have other
 360  * write ordering constraints on the superblock which prevent us from
 361  * writing it out straight away; and given that the journal is about to
 362  * be aborted, we can't rely on the current, or future, transactions to
 363  * write out the superblock safely.
 364  *
 365  * We'll just use the jbd2_journal_abort() error code to record an error in
 366  * the journal instead.  On recovery, the journal will complain about
 367  * that error until we've noted it down and cleared it.
 368  */
 369
 370 static void ext4_handle_error(struct super_block *sb)
 371 {
 372         if (sb->s_flags & MS_RDONLY)
 373                 return;
 374
 375         if (!test_opt(sb, ERRORS_CONT)) {
 376                 journal_t *journal = EXT4_SB(sb)->s_journal;
 377
 378                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 379                 if (journal)
 380                         jbd2_journal_abort(journal, -EIO);
 381         }
 382         if (test_opt(sb, ERRORS_RO)) {
 383                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 384                 /*
 385                  * Make sure updated value of ->s_mount_flags will be visible
 386                  * before ->s_flags update
 387                  */
 388                 smp_wmb();
 389                 sb->s_flags |= MS_RDONLY;
 390         }
 391         if (test_opt(sb, ERRORS_PANIC)) {
 392                 if (EXT4_SB(sb)->s_journal &&
 393                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 394                         return;
 395                 panic("EXT4-fs (device %s): panic forced after error\n",
 396                         sb->s_id);
 397         }
 398 }
 399
 400 #define ext4_error_ratelimit(sb)                                        \
 401                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 402                              "EXT4-fs error")
 403
 404 void __ext4_error(struct super_block *sb, const char *function,
 405                   unsigned int line, const char *fmt, ...)
 406 {
 407         struct va_format vaf;
 408         va_list args;
 409
 410         if (ext4_error_ratelimit(sb)) {
 411                 va_start(args, fmt);
 412                 vaf.fmt = fmt;
 413                 vaf.va = &args;
 414                 printk(KERN_CRIT
 415                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 416                        sb->s_id, function, line, current->comm, &vaf);
 417                 va_end(args);
 418         }
 419         save_error_info(sb, function, line);
 420         ext4_handle_error(sb);
 421 }
 422
 423 void __ext4_error_inode(struct inode *inode, const char *function,
 424                         unsigned int line, ext4_fsblk_t block,
 425                         const char *fmt, ...)
 426 {
 427         va_list args;
 428         struct va_format vaf;
 429         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 430
 431         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 432         es->s_last_error_block = cpu_to_le64(block);
 433         if (ext4_error_ratelimit(inode->i_sb)) {
 434                 va_start(args, fmt);
 435                 vaf.fmt = fmt;
 436                 vaf.va = &args;
 437                 if (block)
 438                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 439                                "inode #%lu: block %llu: comm %s: %pV\n",
 440                                inode->i_sb->s_id, function, line, inode->i_ino,
 441                                block, current->comm, &vaf);
 442                 else
 443                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 444                                "inode #%lu: comm %s: %pV\n",
 445                                inode->i_sb->s_id, function, line, inode->i_ino,
 446                                current->comm, &vaf);
 447                 va_end(args);
 448         }
 449         save_error_info(inode->i_sb, function, line);
 450         ext4_handle_error(inode->i_sb);
 451 }
 452
 453 void __ext4_error_file(struct file *file, const char *function,
 454                        unsigned int line, ext4_fsblk_t block,
 455                        const char *fmt, ...)
 456 {
 457         va_list args;
 458         struct va_format vaf;
 459         struct ext4_super_block *es;
 460         struct inode *inode = file_inode(file);
 461         char pathname[80], *path;
 462
 463         es = EXT4_SB(inode->i_sb)->s_es;
 464         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 465         if (ext4_error_ratelimit(inode->i_sb)) {
 466                 path = file_path(file, pathname, sizeof(pathname));
 467                 if (IS_ERR(path))
 468                         path = "(unknown)";
 469                 va_start(args, fmt);
 470                 vaf.fmt = fmt;
 471                 vaf.va = &args;
 472                 if (block)
 473                         printk(KERN_CRIT
 474                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 475                                "block %llu: comm %s: path %s: %pV\n",
 476                                inode->i_sb->s_id, function, line, inode->i_ino,
 477                                block, current->comm, path, &vaf);
 478                 else
 479                         printk(KERN_CRIT
 480                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 481                                "comm %s: path %s: %pV\n",
 482                                inode->i_sb->s_id, function, line, inode->i_ino,
 483                                current->comm, path, &vaf);
 484                 va_end(args);
 485         }
 486         save_error_info(inode->i_sb, function, line);
 487         ext4_handle_error(inode->i_sb);
 488 }
 489
 490 const char *ext4_decode_error(struct super_block *sb, int errno,
 491                               char nbuf[16])
 492 {
 493         char *errstr = NULL;
 494
 495         switch (errno) {
 496         case -EFSCORRUPTED:
 497                 errstr = "Corrupt filesystem";
 498                 break;
 499         case -EFSBADCRC:
 500                 errstr = "Filesystem failed CRC";
 501                 break;
 502         case -EIO:
 503                 errstr = "IO failure";
 504                 break;
 505         case -ENOMEM:
 506                 errstr = "Out of memory";
 507                 break;
 508         case -EROFS:
 509                 if (!sb || (EXT4_SB(sb)->s_journal &&
 510                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 511                         errstr = "Journal has aborted";
 512                 else
 513                         errstr = "Readonly filesystem";
 514                 break;
 515         default:
 516                 /* If the caller passed in an extra buffer for unknown
 517                  * errors, textualise them now.  Else we just return
 518                  * NULL. */
 519                 if (nbuf) {
 520                         /* Check for truncated error codes... */
 521                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 522                                 errstr = nbuf;
 523                 }
 524                 break;
 525         }
 526
 527         return errstr;
 528 }
 529
 530 /* __ext4_std_error decodes expected errors from journaling functions
 531  * automatically and invokes the appropriate error response.  */
 532
 533 void __ext4_std_error(struct super_block *sb, const char *function,
 534                       unsigned int line, int errno)
 535 {
 536         char nbuf[16];
 537         const char *errstr;
 538
 539         /* Special case: if the error is EROFS, and we're not already
 540          * inside a transaction, then there's really no point in logging
 541          * an error. */
 542         if (errno == -EROFS && journal_current_handle() == NULL &&
 543             (sb->s_flags & MS_RDONLY))
 544                 return;
 545
 546         if (ext4_error_ratelimit(sb)) {
 547                 errstr = ext4_decode_error(sb, errno, nbuf);
 548                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 549                        sb->s_id, function, line, errstr);
 550         }
 551
 552         save_error_info(sb, function, line);
 553         ext4_handle_error(sb);
 554 }
 555
 556 /*
 557  * ext4_abort is a much stronger failure handler than ext4_error.  The
 558  * abort function may be used to deal with unrecoverable failures such
 559  * as journal IO errors or ENOMEM at a critical moment in log management.
 560  *
 561  * We unconditionally force the filesystem into an ABORT|READONLY state,
 562  * unless the error response on the fs has been set to panic in which
 563  * case we take the easy way out and panic immediately.
 564  */
 565
 566 void __ext4_abort(struct super_block *sb, const char *function,
 567                 unsigned int line, const char *fmt, ...)
 568 {
 569         va_list args;
 570
 571         save_error_info(sb, function, line);
 572         va_start(args, fmt);
 573         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 574                function, line);
 575         vprintk(fmt, args);
 576         printk("\n");
 577         va_end(args);
 578
 579         if ((sb->s_flags & MS_RDONLY) == 0) {
 580                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 581                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 582                 /*
 583                  * Make sure updated value of ->s_mount_flags will be visible
 584                  * before ->s_flags update
 585                  */
 586                 smp_wmb();
 587                 sb->s_flags |= MS_RDONLY;
 588                 if (EXT4_SB(sb)->s_journal)
 589                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 590                 save_error_info(sb, function, line);
 591         }
 592         if (test_opt(sb, ERRORS_PANIC)) {
 593                 if (EXT4_SB(sb)->s_journal &&
 594                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 595                         return;
 596                 panic("EXT4-fs panic from previous error\n");
 597         }
 598 }
 599
 600 void __ext4_msg(struct super_block *sb,
 601                 const char *prefix, const char *fmt, ...)
 602 {
 603         struct va_format vaf;
 604         va_list args;
 605
 606         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 607                 return;
 608
 609         va_start(args, fmt);
 610         vaf.fmt = fmt;
 611         vaf.va = &args;
 612         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 613         va_end(args);
 614 }
 615
 616 #define ext4_warning_ratelimit(sb)                                      \
 617                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 618                              "EXT4-fs warning")
 619
 620 void __ext4_warning(struct super_block *sb, const char *function,
 621                     unsigned int line, const char *fmt, ...)
 622 {
 623         struct va_format vaf;
 624         va_list args;
 625
 626         if (!ext4_warning_ratelimit(sb))
 627                 return;
 628
 629         va_start(args, fmt);
 630         vaf.fmt = fmt;
 631         vaf.va = &args;
 632         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 633                sb->s_id, function, line, &vaf);
 634         va_end(args);
 635 }
 636
 637 void __ext4_warning_inode(const struct inode *inode, const char *function,
 638                           unsigned int line, const char *fmt, ...)
 639 {
 640         struct va_format vaf;
 641         va_list args;
 642
 643         if (!ext4_warning_ratelimit(inode->i_sb))
 644                 return;
 645
 646         va_start(args, fmt);
 647         vaf.fmt = fmt;
 648         vaf.va = &args;
 649         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 650                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 651                function, line, inode->i_ino, current->comm, &vaf);
 652         va_end(args);
 653 }
 654
 655 void __ext4_grp_locked_error(const char *function, unsigned int line,
 656                              struct super_block *sb, ext4_group_t grp,
 657                              unsigned long ino, ext4_fsblk_t block,
 658                              const char *fmt, ...)
 659 __releases(bitlock)
 660 __acquires(bitlock)
 661 {
 662         struct va_format vaf;
 663         va_list args;
 664         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 665
 666         es->s_last_error_ino = cpu_to_le32(ino);
 667         es->s_last_error_block = cpu_to_le64(block);
 668         __save_error_info(sb, function, line);
 669
 670         if (ext4_error_ratelimit(sb)) {
 671                 va_start(args, fmt);
 672                 vaf.fmt = fmt;
 673                 vaf.va = &args;
 674                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 675                        sb->s_id, function, line, grp);
 676                 if (ino)
 677                         printk(KERN_CONT "inode %lu: ", ino);
 678                 if (block)
 679                         printk(KERN_CONT "block %llu:",
 680                                (unsigned long long) block);
 681                 printk(KERN_CONT "%pV\n", &vaf);
 682                 va_end(args);
 683         }
 684
 685         if (test_opt(sb, ERRORS_CONT)) {
 686                 ext4_commit_super(sb, 0);
 687                 return;
 688         }
 689
 690         ext4_unlock_group(sb, grp);
 691         ext4_handle_error(sb);
 692         /*
 693          * We only get here in the ERRORS_RO case; relocking the group
 694          * may be dangerous, but nothing bad will happen since the
 695          * filesystem will have already been marked read/only and the
 696          * journal has been aborted.  We return 1 as a hint to callers
 697          * who might what to use the return value from
 698          * ext4_grp_locked_error() to distinguish between the
 699          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 700          * aggressively from the ext4 function in question, with a
 701          * more appropriate error code.
 702          */
 703         ext4_lock_group(sb, grp);
 704         return;
 705 }
 706
 707 void ext4_update_dynamic_rev(struct super_block *sb)
 708 {
 709         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 710
 711         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 712                 return;
 713
 714         ext4_warning(sb,
 715                      "updating to rev %d because of new feature flag, "
 716                      "running e2fsck is recommended",
 717                      EXT4_DYNAMIC_REV);
 718
 719         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 720         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 721         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 722         /* leave es->s_feature_*compat flags alone */
 723         /* es->s_uuid will be set by e2fsck if empty */
 724
 725         /*
 726          * The rest of the superblock fields should be zero, and if not it
 727          * means they are likely already in use, so leave them alone.  We
 728          * can leave it up to e2fsck to clean up any inconsistencies there.
 729          */
 730 }
 731
 732 /*
 733  * Open the external journal device
 734  */
 735 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 736 {
 737         struct block_device *bdev;
 738         char b[BDEVNAME_SIZE];
 739
 740         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 741         if (IS_ERR(bdev))
 742                 goto fail;
 743         return bdev;
 744
 745 fail:
 746         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 747                         __bdevname(dev, b), PTR_ERR(bdev));
 748         return NULL;
 749 }
 750
 751 /*
 752  * Release the journal device
 753  */
 754 static void ext4_blkdev_put(struct block_device *bdev)
 755 {
 756         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 757 }
 758
 759 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 760 {
 761         struct block_device *bdev;
 762         bdev = sbi->journal_bdev;
 763         if (bdev) {
 764                 ext4_blkdev_put(bdev);
 765                 sbi->journal_bdev = NULL;
 766         }
 767 }
 768
 769 static inline struct inode *orphan_list_entry(struct list_head *l)
 770 {
 771         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 772 }
 773
 774 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 775 {
 776         struct list_head *l;
 777
 778         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 779                  le32_to_cpu(sbi->s_es->s_last_orphan));
 780
 781         printk(KERN_ERR "sb_info orphan list:\n");
 782         list_for_each(l, &sbi->s_orphan) {
 783                 struct inode *inode = orphan_list_entry(l);
 784                 printk(KERN_ERR "  "
 785                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 786                        inode->i_sb->s_id, inode->i_ino, inode,
 787                        inode->i_mode, inode->i_nlink,
 788                        NEXT_ORPHAN(inode));
 789         }
 790 }
 791
 792 static void ext4_put_super(struct super_block *sb)
 793 {
 794         struct ext4_sb_info *sbi = EXT4_SB(sb);
 795         struct ext4_super_block *es = sbi->s_es;
 796         int i, err;
 797
 798         ext4_unregister_li_request(sb);
 799         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 800
 801         flush_workqueue(sbi->rsv_conversion_wq);
 802         destroy_workqueue(sbi->rsv_conversion_wq);
 803
 804         if (sbi->s_journal) {
 805                 err = jbd2_journal_destroy(sbi->s_journal);
 806                 sbi->s_journal = NULL;
 807                 if (err < 0)
 808                         ext4_abort(sb, "Couldn't clean up the journal");
 809         }
 810
 811         ext4_unregister_sysfs(sb);
 812         ext4_es_unregister_shrinker(sbi);
 813         del_timer_sync(&sbi->s_err_report);
 814         ext4_release_system_zone(sb);
 815         ext4_mb_release(sb);
 816         ext4_ext_release(sb);
 817         ext4_xattr_put_super(sb);
 818
 819         if (!(sb->s_flags & MS_RDONLY)) {
 820                 ext4_clear_feature_journal_needs_recovery(sb);
 821                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 822         }
 823         if (!(sb->s_flags & MS_RDONLY))
 824                 ext4_commit_super(sb, 1);
 825
 826         for (i = 0; i < sbi->s_gdb_count; i++)
 827                 brelse(sbi->s_group_desc[i]);
 828         kvfree(sbi->s_group_desc);
 829         kvfree(sbi->s_flex_groups);
 830         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 831         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 832         percpu_counter_destroy(&sbi->s_dirs_counter);
 833         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 834         brelse(sbi->s_sbh);
 835 #ifdef CONFIG_QUOTA
 836         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 837                 kfree(sbi->s_qf_names[i]);
 838 #endif
 839
 840         /* Debugging code just in case the in-memory inode orphan list
 841          * isn't empty.  The on-disk one can be non-empty if we've
 842          * detected an error and taken the fs readonly, but the
 843          * in-memory list had better be clean by this point. */
 844         if (!list_empty(&sbi->s_orphan))
 845                 dump_orphan_list(sb, sbi);
 846         J_ASSERT(list_empty(&sbi->s_orphan));
 847
 848         sync_blockdev(sb->s_bdev);
 849         invalidate_bdev(sb->s_bdev);
 850         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 851                 /*
 852                  * Invalidate the journal device's buffers.  We don't want them
 853                  * floating about in memory - the physical journal device may
 854                  * hotswapped, and it breaks the `ro-after' testing code.
 855                  */
 856                 sync_blockdev(sbi->journal_bdev);
 857                 invalidate_bdev(sbi->journal_bdev);
 858                 ext4_blkdev_remove(sbi);
 859         }
 860         if (sbi->s_mb_cache) {
 861                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 862                 sbi->s_mb_cache = NULL;
 863         }
 864         if (sbi->s_mmp_tsk)
 865                 kthread_stop(sbi->s_mmp_tsk);
 866         sb->s_fs_info = NULL;
 867         /*
 868          * Now that we are completely done shutting down the
 869          * superblock, we need to actually destroy the kobject.
 870          */
 871         kobject_put(&sbi->s_kobj);
 872         wait_for_completion(&sbi->s_kobj_unregister);
 873         if (sbi->s_chksum_driver)
 874                 crypto_free_shash(sbi->s_chksum_driver);
 875         kfree(sbi->s_blockgroup_lock);
 876         kfree(sbi);
 877 }
 878
 879 static struct kmem_cache *ext4_inode_cachep;
 880
 881 /*
 882  * Called inside transaction, so use GFP_NOFS
 883  */
 884 static struct inode *ext4_alloc_inode(struct super_block *sb)
 885 {
 886         struct ext4_inode_info *ei;
 887
 888         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 889         if (!ei)
 890                 return NULL;
 891
 892         ei->vfs_inode.i_version = 1;
 893         spin_lock_init(&ei->i_raw_lock);
 894         INIT_LIST_HEAD(&ei->i_prealloc_list);
 895         spin_lock_init(&ei->i_prealloc_lock);
 896         ext4_es_init_tree(&ei->i_es_tree);
 897         rwlock_init(&ei->i_es_lock);
 898         INIT_LIST_HEAD(&ei->i_es_list);
 899         ei->i_es_all_nr = 0;
 900         ei->i_es_shk_nr = 0;
 901         ei->i_es_shrink_lblk = 0;
 902         ei->i_reserved_data_blocks = 0;
 903         ei->i_reserved_meta_blocks = 0;
 904         ei->i_allocated_meta_blocks = 0;
 905         ei->i_da_metadata_calc_len = 0;
 906         ei->i_da_metadata_calc_last_lblock = 0;
 907         spin_lock_init(&(ei->i_block_reservation_lock));
 908 #ifdef CONFIG_QUOTA
 909         ei->i_reserved_quota = 0;
 910         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 911 #endif
 912         ei->jinode = NULL;
 913         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 914         spin_lock_init(&ei->i_completed_io_lock);
 915         ei->i_sync_tid = 0;
 916         ei->i_datasync_tid = 0;
 917         atomic_set(&ei->i_ioend_count, 0);
 918         atomic_set(&ei->i_unwritten, 0);
 919         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 920 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 921         ei->i_crypt_info = NULL;
 922 #endif
 923         return &ei->vfs_inode;
 924 }
 925
 926 static int ext4_drop_inode(struct inode *inode)
 927 {
 928         int drop = generic_drop_inode(inode);
 929
 930         trace_ext4_drop_inode(inode, drop);
 931         return drop;
 932 }
 933
 934 static void ext4_i_callback(struct rcu_head *head)
 935 {
 936         struct inode *inode = container_of(head, struct inode, i_rcu);
 937         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 938 }
 939
 940 static void ext4_destroy_inode(struct inode *inode)
 941 {
 942         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 943                 ext4_msg(inode->i_sb, KERN_ERR,
 944                          "Inode %lu (%p): orphan list check failed!",
 945                          inode->i_ino, EXT4_I(inode));
 946                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 947                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 948                                 true);
 949                 dump_stack();
 950         }
 951         call_rcu(&inode->i_rcu, ext4_i_callback);
 952 }
 953
 954 static void init_once(void *foo)
 955 {
 956         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 957
 958         INIT_LIST_HEAD(&ei->i_orphan);
 959         init_rwsem(&ei->xattr_sem);
 960         init_rwsem(&ei->i_data_sem);
 961         init_rwsem(&ei->i_mmap_sem);
 962         inode_init_once(&ei->vfs_inode);
 963 }
 964
 965 static int __init init_inodecache(void)
 966 {
 967         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 968                                              sizeof(struct ext4_inode_info),
 969                                              0, (SLAB_RECLAIM_ACCOUNT|
 970                                                 SLAB_MEM_SPREAD),
 971                                              init_once);
 972         if (ext4_inode_cachep == NULL)
 973                 return -ENOMEM;
 974         return 0;
 975 }
 976
 977 static void destroy_inodecache(void)
 978 {
 979         /*
 980          * Make sure all delayed rcu free inodes are flushed before we
 981          * destroy cache.
 982          */
 983         rcu_barrier();
 984         kmem_cache_destroy(ext4_inode_cachep);
 985 }
 986
 987 void ext4_clear_inode(struct inode *inode)
 988 {
 989         invalidate_inode_buffers(inode);
 990         clear_inode(inode);
 991         dquot_drop(inode);
 992         ext4_discard_preallocations(inode);
 993         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 994         if (EXT4_I(inode)->jinode) {
 995                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 996                                                EXT4_I(inode)->jinode);
 997                 jbd2_free_inode(EXT4_I(inode)->jinode);
 998                 EXT4_I(inode)->jinode = NULL;
 999         }
1000 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1001         if (EXT4_I(inode)->i_crypt_info)
1002                 ext4_free_encryption_info(inode, EXT4_I(inode)->i_crypt_info);
1003 #endif
1004 }
1005
1006 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1007                                         u64 ino, u32 generation)
1008 {
1009         struct inode *inode;
1010
1011         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1012                 return ERR_PTR(-ESTALE);
1013         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1014                 return ERR_PTR(-ESTALE);
1015
1016         /* iget isn't really right if the inode is currently unallocated!!
1017          *
1018          * ext4_read_inode will return a bad_inode if the inode had been
1019          * deleted, so we should be safe.
1020          *
1021          * Currently we don't know the generation for parent directory, so
1022          * a generation of 0 means "accept any"
1023          */
1024         inode = ext4_iget_normal(sb, ino);
1025         if (IS_ERR(inode))
1026                 return ERR_CAST(inode);
1027         if (generation && inode->i_generation != generation) {
1028                 iput(inode);
1029                 return ERR_PTR(-ESTALE);
1030         }
1031
1032         return inode;
1033 }
1034
1035 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1036                                         int fh_len, int fh_type)
1037 {
1038         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1039                                     ext4_nfs_get_inode);
1040 }
1041
1042 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1043                                         int fh_len, int fh_type)
1044 {
1045         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1046                                     ext4_nfs_get_inode);
1047 }
1048
1049 /*
1050  * Try to release metadata pages (indirect blocks, directories) which are
1051  * mapped via the block device.  Since these pages could have journal heads
1052  * which would prevent try_to_free_buffers() from freeing them, we must use
1053  * jbd2 layer's try_to_free_buffers() function to release them.
1054  */
1055 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1056                                  gfp_t wait)
1057 {
1058         journal_t *journal = EXT4_SB(sb)->s_journal;
1059
1060         WARN_ON(PageChecked(page));
1061         if (!page_has_buffers(page))
1062                 return 0;
1063         if (journal)
1064                 return jbd2_journal_try_to_free_buffers(journal, page,
1065                                                 wait & ~__GFP_DIRECT_RECLAIM);
1066         return try_to_free_buffers(page);
1067 }
1068
1069 #ifdef CONFIG_QUOTA
1070 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1071 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1072
1073 static int ext4_write_dquot(struct dquot *dquot);
1074 static int ext4_acquire_dquot(struct dquot *dquot);
1075 static int ext4_release_dquot(struct dquot *dquot);
1076 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1077 static int ext4_write_info(struct super_block *sb, int type);
1078 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1079                          struct path *path);
1080 static int ext4_quota_off(struct super_block *sb, int type);
1081 static int ext4_quota_on_mount(struct super_block *sb, int type);
1082 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1083                                size_t len, loff_t off);
1084 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1085                                 const char *data, size_t len, loff_t off);
1086 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1087                              unsigned int flags);
1088 static int ext4_enable_quotas(struct super_block *sb);
1089
1090 static struct dquot **ext4_get_dquots(struct inode *inode)
1091 {
1092         return EXT4_I(inode)->i_dquot;
1093 }
1094
1095 static const struct dquot_operations ext4_quota_operations = {
1096         .get_reserved_space = ext4_get_reserved_space,
1097         .write_dquot    = ext4_write_dquot,
1098         .acquire_dquot  = ext4_acquire_dquot,
1099         .release_dquot  = ext4_release_dquot,
1100         .mark_dirty     = ext4_mark_dquot_dirty,
1101         .write_info     = ext4_write_info,
1102         .alloc_dquot    = dquot_alloc,
1103         .destroy_dquot  = dquot_destroy,
1104 };
1105
1106 static const struct quotactl_ops ext4_qctl_operations = {
1107         .quota_on       = ext4_quota_on,
1108         .quota_off      = ext4_quota_off,
1109         .quota_sync     = dquot_quota_sync,
1110         .get_state      = dquot_get_state,
1111         .set_info       = dquot_set_dqinfo,
1112         .get_dqblk      = dquot_get_dqblk,
1113         .set_dqblk      = dquot_set_dqblk
1114 };
1115 #endif
1116
1117 static const struct super_operations ext4_sops = {
1118         .alloc_inode    = ext4_alloc_inode,
1119         .destroy_inode  = ext4_destroy_inode,
1120         .write_inode    = ext4_write_inode,
1121         .dirty_inode    = ext4_dirty_inode,
1122         .drop_inode     = ext4_drop_inode,
1123         .evict_inode    = ext4_evict_inode,
1124         .put_super      = ext4_put_super,
1125         .sync_fs        = ext4_sync_fs,
1126         .freeze_fs      = ext4_freeze,
1127         .unfreeze_fs    = ext4_unfreeze,
1128         .statfs         = ext4_statfs,
1129         .remount_fs     = ext4_remount,
1130         .show_options   = ext4_show_options,
1131 #ifdef CONFIG_QUOTA
1132         .quota_read     = ext4_quota_read,
1133         .quota_write    = ext4_quota_write,
1134         .get_dquots     = ext4_get_dquots,
1135 #endif
1136         .bdev_try_to_free_page = bdev_try_to_free_page,
1137 };
1138
1139 static const struct export_operations ext4_export_ops = {
1140         .fh_to_dentry = ext4_fh_to_dentry,
1141         .fh_to_parent = ext4_fh_to_parent,
1142         .get_parent = ext4_get_parent,
1143 };
1144
1145 enum {
1146         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1147         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1148         Opt_nouid32, Opt_debug, Opt_removed,
1149         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1150         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1151         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1152         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1153         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1154         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1155         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1156         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1157         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1158         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1159         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1160         Opt_lazytime, Opt_nolazytime,
1161         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1162         Opt_inode_readahead_blks, Opt_journal_ioprio,
1163         Opt_dioread_nolock, Opt_dioread_lock,
1164         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1165         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1166 };
1167
1168 static const match_table_t tokens = {
1169         {Opt_bsd_df, "bsddf"},
1170         {Opt_minix_df, "minixdf"},
1171         {Opt_grpid, "grpid"},
1172         {Opt_grpid, "bsdgroups"},
1173         {Opt_nogrpid, "nogrpid"},
1174         {Opt_nogrpid, "sysvgroups"},
1175         {Opt_resgid, "resgid=%u"},
1176         {Opt_resuid, "resuid=%u"},
1177         {Opt_sb, "sb=%u"},
1178         {Opt_err_cont, "errors=continue"},
1179         {Opt_err_panic, "errors=panic"},
1180         {Opt_err_ro, "errors=remount-ro"},
1181         {Opt_nouid32, "nouid32"},
1182         {Opt_debug, "debug"},
1183         {Opt_removed, "oldalloc"},
1184         {Opt_removed, "orlov"},
1185         {Opt_user_xattr, "user_xattr"},
1186         {Opt_nouser_xattr, "nouser_xattr"},
1187         {Opt_acl, "acl"},
1188         {Opt_noacl, "noacl"},
1189         {Opt_noload, "norecovery"},
1190         {Opt_noload, "noload"},
1191         {Opt_removed, "nobh"},
1192         {Opt_removed, "bh"},
1193         {Opt_commit, "commit=%u"},
1194         {Opt_min_batch_time, "min_batch_time=%u"},
1195         {Opt_max_batch_time, "max_batch_time=%u"},
1196         {Opt_journal_dev, "journal_dev=%u"},
1197         {Opt_journal_path, "journal_path=%s"},
1198         {Opt_journal_checksum, "journal_checksum"},
1199         {Opt_nojournal_checksum, "nojournal_checksum"},
1200         {Opt_journal_async_commit, "journal_async_commit"},
1201         {Opt_abort, "abort"},
1202         {Opt_data_journal, "data=journal"},
1203         {Opt_data_ordered, "data=ordered"},
1204         {Opt_data_writeback, "data=writeback"},
1205         {Opt_data_err_abort, "data_err=abort"},
1206         {Opt_data_err_ignore, "data_err=ignore"},
1207         {Opt_offusrjquota, "usrjquota="},
1208         {Opt_usrjquota, "usrjquota=%s"},
1209         {Opt_offgrpjquota, "grpjquota="},
1210         {Opt_grpjquota, "grpjquota=%s"},
1211         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1212         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1213         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1214         {Opt_grpquota, "grpquota"},
1215         {Opt_noquota, "noquota"},
1216         {Opt_quota, "quota"},
1217         {Opt_usrquota, "usrquota"},
1218         {Opt_barrier, "barrier=%u"},
1219         {Opt_barrier, "barrier"},
1220         {Opt_nobarrier, "nobarrier"},
1221         {Opt_i_version, "i_version"},
1222         {Opt_dax, "dax"},
1223         {Opt_stripe, "stripe=%u"},
1224         {Opt_delalloc, "delalloc"},
1225         {Opt_lazytime, "lazytime"},
1226         {Opt_nolazytime, "nolazytime"},
1227         {Opt_nodelalloc, "nodelalloc"},
1228         {Opt_removed, "mblk_io_submit"},
1229         {Opt_removed, "nomblk_io_submit"},
1230         {Opt_block_validity, "block_validity"},
1231         {Opt_noblock_validity, "noblock_validity"},
1232         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1233         {Opt_journal_ioprio, "journal_ioprio=%u"},
1234         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1235         {Opt_auto_da_alloc, "auto_da_alloc"},
1236         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1237         {Opt_dioread_nolock, "dioread_nolock"},
1238         {Opt_dioread_lock, "dioread_lock"},
1239         {Opt_discard, "discard"},
1240         {Opt_nodiscard, "nodiscard"},
1241         {Opt_init_itable, "init_itable=%u"},
1242         {Opt_init_itable, "init_itable"},
1243         {Opt_noinit_itable, "noinit_itable"},
1244         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1245         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1246         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1247         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1248         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1249         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1250         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1251         {Opt_err, NULL},
1252 };
1253
1254 static ext4_fsblk_t get_sb_block(void **data)
1255 {
1256         ext4_fsblk_t    sb_block;
1257         char            *options = (char *) *data;
1258
1259         if (!options || strncmp(options, "sb=", 3) != 0)
1260                 return 1;       /* Default location */
1261
1262         options += 3;
1263         /* TODO: use simple_strtoll with >32bit ext4 */
1264         sb_block = simple_strtoul(options, &options, 0);
1265         if (*options && *options != ',') {
1266                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1267                        (char *) *data);
1268                 return 1;
1269         }
1270         if (*options == ',')
1271                 options++;
1272         *data = (void *) options;
1273
1274         return sb_block;
1275 }
1276
1277 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1278 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1279         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1280
1281 #ifdef CONFIG_QUOTA
1282 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1283 {
1284         struct ext4_sb_info *sbi = EXT4_SB(sb);
1285         char *qname;
1286         int ret = -1;
1287
1288         if (sb_any_quota_loaded(sb) &&
1289                 !sbi->s_qf_names[qtype]) {
1290                 ext4_msg(sb, KERN_ERR,
1291                         "Cannot change journaled "
1292                         "quota options when quota turned on");
1293                 return -1;
1294         }
1295         if (ext4_has_feature_quota(sb)) {
1296                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1297                          "ignored when QUOTA feature is enabled");
1298                 return 1;
1299         }
1300         qname = match_strdup(args);
1301         if (!qname) {
1302                 ext4_msg(sb, KERN_ERR,
1303                         "Not enough memory for storing quotafile name");
1304                 return -1;
1305         }
1306         if (sbi->s_qf_names[qtype]) {
1307                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1308                         ret = 1;
1309                 else
1310                         ext4_msg(sb, KERN_ERR,
1311                                  "%s quota file already specified",
1312                                  QTYPE2NAME(qtype));
1313                 goto errout;
1314         }
1315         if (strchr(qname, '/')) {
1316                 ext4_msg(sb, KERN_ERR,
1317                         "quotafile must be on filesystem root");
1318                 goto errout;
1319         }
1320         sbi->s_qf_names[qtype] = qname;
1321         set_opt(sb, QUOTA);
1322         return 1;
1323 errout:
1324         kfree(qname);
1325         return ret;
1326 }
1327
1328 static int clear_qf_name(struct super_block *sb, int qtype)
1329 {
1330
1331         struct ext4_sb_info *sbi = EXT4_SB(sb);
1332
1333         if (sb_any_quota_loaded(sb) &&
1334                 sbi->s_qf_names[qtype]) {
1335                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1336                         " when quota turned on");
1337                 return -1;
1338         }
1339         kfree(sbi->s_qf_names[qtype]);
1340         sbi->s_qf_names[qtype] = NULL;
1341         return 1;
1342 }
1343 #endif
1344
1345 #define MOPT_SET        0x0001
1346 #define MOPT_CLEAR      0x0002
1347 #define MOPT_NOSUPPORT  0x0004
1348 #define MOPT_EXPLICIT   0x0008
1349 #define MOPT_CLEAR_ERR  0x0010
1350 #define MOPT_GTE0       0x0020
1351 #ifdef CONFIG_QUOTA
1352 #define MOPT_Q          0
1353 #define MOPT_QFMT       0x0040
1354 #else
1355 #define MOPT_Q          MOPT_NOSUPPORT
1356 #define MOPT_QFMT       MOPT_NOSUPPORT
1357 #endif
1358 #define MOPT_DATAJ      0x0080
1359 #define MOPT_NO_EXT2    0x0100
1360 #define MOPT_NO_EXT3    0x0200
1361 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1362 #define MOPT_STRING     0x0400
1363
1364 static const struct mount_opts {
1365         int     token;
1366         int     mount_opt;
1367         int     flags;
1368 } ext4_mount_opts[] = {
1369         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1370         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1371         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1372         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1373         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1374         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1375         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1376          MOPT_EXT4_ONLY | MOPT_SET},
1377         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1378          MOPT_EXT4_ONLY | MOPT_CLEAR},
1379         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1380         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1381         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1382          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1383         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1384          MOPT_EXT4_ONLY | MOPT_CLEAR},
1385         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1386          MOPT_EXT4_ONLY | MOPT_CLEAR},
1387         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1388          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1389         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1390                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1391          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1392         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1393         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1394         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1395         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1396         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1397          MOPT_NO_EXT2 | MOPT_SET},
1398         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1399          MOPT_NO_EXT2 | MOPT_CLEAR},
1400         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1401         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1402         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1403         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1404         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1405         {Opt_commit, 0, MOPT_GTE0},
1406         {Opt_max_batch_time, 0, MOPT_GTE0},
1407         {Opt_min_batch_time, 0, MOPT_GTE0},
1408         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1409         {Opt_init_itable, 0, MOPT_GTE0},
1410         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1411         {Opt_stripe, 0, MOPT_GTE0},
1412         {Opt_resuid, 0, MOPT_GTE0},
1413         {Opt_resgid, 0, MOPT_GTE0},
1414         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1415         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1416         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1417         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1418         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1419         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1420          MOPT_NO_EXT2 | MOPT_DATAJ},
1421         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1422         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1423 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1424         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1425         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1426 #else
1427         {Opt_acl, 0, MOPT_NOSUPPORT},
1428         {Opt_noacl, 0, MOPT_NOSUPPORT},
1429 #endif
1430         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1431         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1432         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1433         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1434                                                         MOPT_SET | MOPT_Q},
1435         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1436                                                         MOPT_SET | MOPT_Q},
1437         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1438                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1439         {Opt_usrjquota, 0, MOPT_Q},
1440         {Opt_grpjquota, 0, MOPT_Q},
1441         {Opt_offusrjquota, 0, MOPT_Q},
1442         {Opt_offgrpjquota, 0, MOPT_Q},
1443         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1444         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1445         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1446         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1447         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1448         {Opt_err, 0, 0}
1449 };
1450
1451 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1452                             substring_t *args, unsigned long *journal_devnum,
1453                             unsigned int *journal_ioprio, int is_remount)
1454 {
1455         struct ext4_sb_info *sbi = EXT4_SB(sb);
1456         const struct mount_opts *m;
1457         kuid_t uid;
1458         kgid_t gid;
1459         int arg = 0;
1460
1461 #ifdef CONFIG_QUOTA
1462         if (token == Opt_usrjquota)
1463                 return set_qf_name(sb, USRQUOTA, &args[0]);
1464         else if (token == Opt_grpjquota)
1465                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1466         else if (token == Opt_offusrjquota)
1467                 return clear_qf_name(sb, USRQUOTA);
1468         else if (token == Opt_offgrpjquota)
1469                 return clear_qf_name(sb, GRPQUOTA);
1470 #endif
1471         switch (token) {
1472         case Opt_noacl:
1473         case Opt_nouser_xattr:
1474                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1475                 break;
1476         case Opt_sb:
1477                 return 1;       /* handled by get_sb_block() */
1478         case Opt_removed:
1479                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1480                 return 1;
1481         case Opt_abort:
1482                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1483                 return 1;
1484         case Opt_i_version:
1485                 sb->s_flags |= MS_I_VERSION;
1486                 return 1;
1487         case Opt_lazytime:
1488                 sb->s_flags |= MS_LAZYTIME;
1489                 return 1;
1490         case Opt_nolazytime:
1491                 sb->s_flags &= ~MS_LAZYTIME;
1492                 return 1;
1493         }
1494
1495         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1496                 if (token == m->token)
1497                         break;
1498
1499         if (m->token == Opt_err) {
1500                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1501                          "or missing value", opt);
1502                 return -1;
1503         }
1504
1505         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1506                 ext4_msg(sb, KERN_ERR,
1507                          "Mount option \"%s\" incompatible with ext2", opt);
1508                 return -1;
1509         }
1510         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1511                 ext4_msg(sb, KERN_ERR,
1512                          "Mount option \"%s\" incompatible with ext3", opt);
1513                 return -1;
1514         }
1515
1516         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1517                 return -1;
1518         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1519                 return -1;
1520         if (m->flags & MOPT_EXPLICIT) {
1521                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1522                         set_opt2(sb, EXPLICIT_DELALLOC);
1523                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1524                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1525                 } else
1526                         return -1;
1527         }
1528         if (m->flags & MOPT_CLEAR_ERR)
1529                 clear_opt(sb, ERRORS_MASK);
1530         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1531                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1532                          "options when quota turned on");
1533                 return -1;
1534         }
1535
1536         if (m->flags & MOPT_NOSUPPORT) {
1537                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1538         } else if (token == Opt_commit) {
1539                 if (arg == 0)
1540                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1541                 sbi->s_commit_interval = HZ * arg;
1542         } else if (token == Opt_max_batch_time) {
1543                 sbi->s_max_batch_time = arg;
1544         } else if (token == Opt_min_batch_time) {
1545                 sbi->s_min_batch_time = arg;
1546         } else if (token == Opt_inode_readahead_blks) {
1547                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1548                         ext4_msg(sb, KERN_ERR,
1549                                  "EXT4-fs: inode_readahead_blks must be "
1550                                  "0 or a power of 2 smaller than 2^31");
1551                         return -1;
1552                 }
1553                 sbi->s_inode_readahead_blks = arg;
1554         } else if (token == Opt_init_itable) {
1555                 set_opt(sb, INIT_INODE_TABLE);
1556                 if (!args->from)
1557                         arg = EXT4_DEF_LI_WAIT_MULT;
1558                 sbi->s_li_wait_mult = arg;
1559         } else if (token == Opt_max_dir_size_kb) {
1560                 sbi->s_max_dir_size_kb = arg;
1561         } else if (token == Opt_stripe) {
1562                 sbi->s_stripe = arg;
1563         } else if (token == Opt_resuid) {
1564                 uid = make_kuid(current_user_ns(), arg);
1565                 if (!uid_valid(uid)) {
1566                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1567                         return -1;
1568                 }
1569                 sbi->s_resuid = uid;
1570         } else if (token == Opt_resgid) {
1571                 gid = make_kgid(current_user_ns(), arg);
1572                 if (!gid_valid(gid)) {
1573                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1574                         return -1;
1575                 }
1576                 sbi->s_resgid = gid;
1577         } else if (token == Opt_journal_dev) {
1578                 if (is_remount) {
1579                         ext4_msg(sb, KERN_ERR,
1580                                  "Cannot specify journal on remount");
1581                         return -1;
1582                 }
1583                 *journal_devnum = arg;
1584         } else if (token == Opt_journal_path) {
1585                 char *journal_path;
1586                 struct inode *journal_inode;
1587                 struct path path;
1588                 int error;
1589
1590                 if (is_remount) {
1591                         ext4_msg(sb, KERN_ERR,
1592                                  "Cannot specify journal on remount");
1593                         return -1;
1594                 }
1595                 journal_path = match_strdup(&args[0]);
1596                 if (!journal_path) {
1597                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1598                                 "journal device string");
1599                         return -1;
1600                 }
1601
1602                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1603                 if (error) {
1604                         ext4_msg(sb, KERN_ERR, "error: could not find "
1605                                 "journal device path: error %d", error);
1606                         kfree(journal_path);
1607                         return -1;
1608                 }
1609
1610                 journal_inode = d_inode(path.dentry);
1611                 if (!S_ISBLK(journal_inode->i_mode)) {
1612                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1613                                 "is not a block device", journal_path);
1614                         path_put(&path);
1615                         kfree(journal_path);
1616                         return -1;
1617                 }
1618
1619                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1620                 path_put(&path);
1621                 kfree(journal_path);
1622         } else if (token == Opt_journal_ioprio) {
1623                 if (arg > 7) {
1624                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1625                                  " (must be 0-7)");
1626                         return -1;
1627                 }
1628                 *journal_ioprio =
1629                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1630         } else if (token == Opt_test_dummy_encryption) {
1631 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1632                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1633                 ext4_msg(sb, KERN_WARNING,
1634                          "Test dummy encryption mode enabled");
1635 #else
1636                 ext4_msg(sb, KERN_WARNING,
1637                          "Test dummy encryption mount option ignored");
1638 #endif
1639         } else if (m->flags & MOPT_DATAJ) {
1640                 if (is_remount) {
1641                         if (!sbi->s_journal)
1642                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1643                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1644                                 ext4_msg(sb, KERN_ERR,
1645                                          "Cannot change data mode on remount");
1646                                 return -1;
1647                         }
1648                 } else {
1649                         clear_opt(sb, DATA_FLAGS);
1650                         sbi->s_mount_opt |= m->mount_opt;
1651                 }
1652 #ifdef CONFIG_QUOTA
1653         } else if (m->flags & MOPT_QFMT) {
1654                 if (sb_any_quota_loaded(sb) &&
1655                     sbi->s_jquota_fmt != m->mount_opt) {
1656                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1657                                  "quota options when quota turned on");
1658                         return -1;
1659                 }
1660                 if (ext4_has_feature_quota(sb)) {
1661                         ext4_msg(sb, KERN_INFO,
1662                                  "Quota format mount options ignored "
1663                                  "when QUOTA feature is enabled");
1664                         return 1;
1665                 }
1666                 sbi->s_jquota_fmt = m->mount_opt;
1667 #endif
1668         } else if (token == Opt_dax) {
1669 #ifdef CONFIG_FS_DAX
1670                 ext4_msg(sb, KERN_WARNING,
1671                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1672                         sbi->s_mount_opt |= m->mount_opt;
1673 #else
1674                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1675                 return -1;
1676 #endif
1677         } else {
1678                 if (!args->from)
1679                         arg = 1;
1680                 if (m->flags & MOPT_CLEAR)
1681                         arg = !arg;
1682                 else if (unlikely(!(m->flags & MOPT_SET))) {
1683                         ext4_msg(sb, KERN_WARNING,
1684                                  "buggy handling of option %s", opt);
1685                         WARN_ON(1);
1686                         return -1;
1687                 }
1688                 if (arg != 0)
1689                         sbi->s_mount_opt |= m->mount_opt;
1690                 else
1691                         sbi->s_mount_opt &= ~m->mount_opt;
1692         }
1693         return 1;
1694 }
1695
1696 static int parse_options(char *options, struct super_block *sb,
1697                          unsigned long *journal_devnum,
1698                          unsigned int *journal_ioprio,
1699                          int is_remount)
1700 {
1701         struct ext4_sb_info *sbi = EXT4_SB(sb);
1702         char *p;
1703         substring_t args[MAX_OPT_ARGS];
1704         int token;
1705
1706         if (!options)
1707                 return 1;
1708
1709         while ((p = strsep(&options, ",")) != NULL) {
1710                 if (!*p)
1711                         continue;
1712                 /*
1713                  * Initialize args struct so we know whether arg was
1714                  * found; some options take optional arguments.
1715                  */
1716                 args[0].to = args[0].from = NULL;
1717                 token = match_token(p, tokens, args);
1718                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1719                                      journal_ioprio, is_remount) < 0)
1720                         return 0;
1721         }
1722 #ifdef CONFIG_QUOTA
1723         if (ext4_has_feature_quota(sb) &&
1724             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1725                 ext4_msg(sb, KERN_INFO, "Quota feature enabled, usrquota and grpquota "
1726                          "mount options ignored.");
1727                 clear_opt(sb, USRQUOTA);
1728                 clear_opt(sb, GRPQUOTA);
1729         } else if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1730                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1731                         clear_opt(sb, USRQUOTA);
1732
1733                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1734                         clear_opt(sb, GRPQUOTA);
1735
1736                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1737                         ext4_msg(sb, KERN_ERR, "old and new quota "
1738                                         "format mixing");
1739                         return 0;
1740                 }
1741
1742                 if (!sbi->s_jquota_fmt) {
1743                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1744                                         "not specified");
1745                         return 0;
1746                 }
1747         }
1748 #endif
1749         if (test_opt(sb, DIOREAD_NOLOCK)) {
1750                 int blocksize =
1751                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1752
1753                 if (blocksize < PAGE_CACHE_SIZE) {
1754                         ext4_msg(sb, KERN_ERR, "can't mount with "
1755                                  "dioread_nolock if block size != PAGE_SIZE");
1756                         return 0;
1757                 }
1758         }
1759         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1760             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1761                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1762                          "in data=ordered mode");
1763                 return 0;
1764         }
1765         return 1;
1766 }
1767
1768 static inline void ext4_show_quota_options(struct seq_file *seq,
1769                                            struct super_block *sb)
1770 {
1771 #if defined(CONFIG_QUOTA)
1772         struct ext4_sb_info *sbi = EXT4_SB(sb);
1773
1774         if (sbi->s_jquota_fmt) {
1775                 char *fmtname = "";
1776
1777                 switch (sbi->s_jquota_fmt) {
1778                 case QFMT_VFS_OLD:
1779                         fmtname = "vfsold";
1780                         break;
1781                 case QFMT_VFS_V0:
1782                         fmtname = "vfsv0";
1783                         break;
1784                 case QFMT_VFS_V1:
1785                         fmtname = "vfsv1";
1786                         break;
1787                 }
1788                 seq_printf(seq, ",jqfmt=%s", fmtname);
1789         }
1790
1791         if (sbi->s_qf_names[USRQUOTA])
1792                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1793
1794         if (sbi->s_qf_names[GRPQUOTA])
1795                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1796 #endif
1797 }
1798
1799 static const char *token2str(int token)
1800 {
1801         const struct match_token *t;
1802
1803         for (t = tokens; t->token != Opt_err; t++)
1804                 if (t->token == token && !strchr(t->pattern, '='))
1805                         break;
1806         return t->pattern;
1807 }
1808
1809 /*
1810  * Show an option if
1811  *  - it's set to a non-default value OR
1812  *  - if the per-sb default is different from the global default
1813  */
1814 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1815                               int nodefs)
1816 {
1817         struct ext4_sb_info *sbi = EXT4_SB(sb);
1818         struct ext4_super_block *es = sbi->s_es;
1819         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1820         const struct mount_opts *m;
1821         char sep = nodefs ? '\n' : ',';
1822
1823 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1824 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1825
1826         if (sbi->s_sb_block != 1)
1827                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1828
1829         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1830                 int want_set = m->flags & MOPT_SET;
1831                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1832                     (m->flags & MOPT_CLEAR_ERR))
1833                         continue;
1834                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1835                         continue; /* skip if same as the default */
1836                 if ((want_set &&
1837                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1838                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1839                         continue; /* select Opt_noFoo vs Opt_Foo */
1840                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1841         }
1842
1843         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1844             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1845                 SEQ_OPTS_PRINT("resuid=%u",
1846                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1847         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1848             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1849                 SEQ_OPTS_PRINT("resgid=%u",
1850                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1851         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1852         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1853                 SEQ_OPTS_PUTS("errors=remount-ro");
1854         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1855                 SEQ_OPTS_PUTS("errors=continue");
1856         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1857                 SEQ_OPTS_PUTS("errors=panic");
1858         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1859                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1860         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1861                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1862         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1863                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1864         if (sb->s_flags & MS_I_VERSION)
1865                 SEQ_OPTS_PUTS("i_version");
1866         if (nodefs || sbi->s_stripe)
1867                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1868         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1869                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1870                         SEQ_OPTS_PUTS("data=journal");
1871                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1872                         SEQ_OPTS_PUTS("data=ordered");
1873                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1874                         SEQ_OPTS_PUTS("data=writeback");
1875         }
1876         if (nodefs ||
1877             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1878                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1879                                sbi->s_inode_readahead_blks);
1880
1881         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1882                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1883                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1884         if (nodefs || sbi->s_max_dir_size_kb)
1885                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1886
1887         ext4_show_quota_options(seq, sb);
1888         return 0;
1889 }
1890
1891 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1892 {
1893         return _ext4_show_options(seq, root->d_sb, 0);
1894 }
1895
1896 int ext4_seq_options_show(struct seq_file *seq, void *offset)
1897 {
1898         struct super_block *sb = seq->private;
1899         int rc;
1900
1901         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1902         rc = _ext4_show_options(seq, sb, 1);
1903         seq_puts(seq, "\n");
1904         return rc;
1905 }
1906
1907 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1908                             int read_only)
1909 {
1910         struct ext4_sb_info *sbi = EXT4_SB(sb);
1911         int res = 0;
1912
1913         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1914                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1915                          "forcing read-only mode");
1916                 res = MS_RDONLY;
1917         }
1918         if (read_only)
1919                 goto done;
1920         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1921                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1922                          "running e2fsck is recommended");
1923         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1924                 ext4_msg(sb, KERN_WARNING,
1925                          "warning: mounting fs with errors, "
1926                          "running e2fsck is recommended");
1927         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1928                  le16_to_cpu(es->s_mnt_count) >=
1929                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1930                 ext4_msg(sb, KERN_WARNING,
1931                          "warning: maximal mount count reached, "
1932                          "running e2fsck is recommended");
1933         else if (le32_to_cpu(es->s_checkinterval) &&
1934                 (le32_to_cpu(es->s_lastcheck) +
1935                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1936                 ext4_msg(sb, KERN_WARNING,
1937                          "warning: checktime reached, "
1938                          "running e2fsck is recommended");
1939         if (!sbi->s_journal)
1940                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1941         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1942                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1943         le16_add_cpu(&es->s_mnt_count, 1);
1944         es->s_mtime = cpu_to_le32(get_seconds());
1945         ext4_update_dynamic_rev(sb);
1946         if (sbi->s_journal)
1947                 ext4_set_feature_journal_needs_recovery(sb);
1948
1949         ext4_commit_super(sb, 1);
1950 done:
1951         if (test_opt(sb, DEBUG))
1952                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1953                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1954                         sb->s_blocksize,
1955                         sbi->s_groups_count,
1956                         EXT4_BLOCKS_PER_GROUP(sb),
1957                         EXT4_INODES_PER_GROUP(sb),
1958                         sbi->s_mount_opt, sbi->s_mount_opt2);
1959
1960         cleancache_init_fs(sb);
1961         return res;
1962 }
1963
1964 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1965 {
1966         struct ext4_sb_info *sbi = EXT4_SB(sb);
1967         struct flex_groups *new_groups;
1968         int size;
1969
1970         if (!sbi->s_log_groups_per_flex)
1971                 return 0;
1972
1973         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1974         if (size <= sbi->s_flex_groups_allocated)
1975                 return 0;
1976
1977         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1978         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1979         if (!new_groups) {
1980                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1981                          size / (int) sizeof(struct flex_groups));
1982                 return -ENOMEM;
1983         }
1984
1985         if (sbi->s_flex_groups) {
1986                 memcpy(new_groups, sbi->s_flex_groups,
1987                        (sbi->s_flex_groups_allocated *
1988                         sizeof(struct flex_groups)));
1989                 kvfree(sbi->s_flex_groups);
1990         }
1991         sbi->s_flex_groups = new_groups;
1992         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1993         return 0;
1994 }
1995
1996 static int ext4_fill_flex_info(struct super_block *sb)
1997 {
1998         struct ext4_sb_info *sbi = EXT4_SB(sb);
1999         struct ext4_group_desc *gdp = NULL;
2000         ext4_group_t flex_group;
2001         int i, err;
2002
2003         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2004         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2005                 sbi->s_log_groups_per_flex = 0;
2006                 return 1;
2007         }
2008
2009         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2010         if (err)
2011                 goto failed;
2012
2013         for (i = 0; i < sbi->s_groups_count; i++) {
2014                 gdp = ext4_get_group_desc(sb, i, NULL);
2015
2016                 flex_group = ext4_flex_group(sbi, i);
2017                 atomic_add(ext4_free_inodes_count(sb, gdp),
2018                            &sbi->s_flex_groups[flex_group].free_inodes);
2019                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2020                              &sbi->s_flex_groups[flex_group].free_clusters);
2021                 atomic_add(ext4_used_dirs_count(sb, gdp),
2022                            &sbi->s_flex_groups[flex_group].used_dirs);
2023         }
2024
2025         return 1;
2026 failed:
2027         return 0;
2028 }
2029
2030 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2031                                    struct ext4_group_desc *gdp)
2032 {
2033         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2034         __u16 crc = 0;
2035         __le32 le_group = cpu_to_le32(block_group);
2036         struct ext4_sb_info *sbi = EXT4_SB(sb);
2037
2038         if (ext4_has_metadata_csum(sbi->s_sb)) {
2039                 /* Use new metadata_csum algorithm */
2040                 __u32 csum32;
2041                 __u16 dummy_csum = 0;
2042
2043                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2044                                      sizeof(le_group));
2045                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2046                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2047                                      sizeof(dummy_csum));
2048                 offset += sizeof(dummy_csum);
2049                 if (offset < sbi->s_desc_size)
2050                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2051                                              sbi->s_desc_size - offset);
2052
2053                 crc = csum32 & 0xFFFF;
2054                 goto out;
2055         }
2056
2057         /* old crc16 code */
2058         if (!ext4_has_feature_gdt_csum(sb))
2059                 return 0;
2060
2061         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2062         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2063         crc = crc16(crc, (__u8 *)gdp, offset);
2064         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2065         /* for checksum of struct ext4_group_desc do the rest...*/
2066         if (ext4_has_feature_64bit(sb) &&
2067             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2068                 crc = crc16(crc, (__u8 *)gdp + offset,
2069                             le16_to_cpu(sbi->s_es->s_desc_size) -
2070                                 offset);
2071
2072 out:
2073         return cpu_to_le16(crc);
2074 }
2075
2076 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2077                                 struct ext4_group_desc *gdp)
2078 {
2079         if (ext4_has_group_desc_csum(sb) &&
2080             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2081                 return 0;
2082
2083         return 1;
2084 }
2085
2086 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2087                               struct ext4_group_desc *gdp)
2088 {
2089         if (!ext4_has_group_desc_csum(sb))
2090                 return;
2091         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2092 }
2093
2094 /* Called at mount-time, super-block is locked */
2095 static int ext4_check_descriptors(struct super_block *sb,
2096                                   ext4_fsblk_t sb_block,
2097                                   ext4_group_t *first_not_zeroed)
2098 {
2099         struct ext4_sb_info *sbi = EXT4_SB(sb);
2100         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2101         ext4_fsblk_t last_block;
2102         ext4_fsblk_t block_bitmap;
2103         ext4_fsblk_t inode_bitmap;
2104         ext4_fsblk_t inode_table;
2105         int flexbg_flag = 0;
2106         ext4_group_t i, grp = sbi->s_groups_count;
2107
2108         if (ext4_has_feature_flex_bg(sb))
2109                 flexbg_flag = 1;
2110
2111         ext4_debug("Checking group descriptors");
2112
2113         for (i = 0; i < sbi->s_groups_count; i++) {
2114                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2115
2116                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2117                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2118                 else
2119                         last_block = first_block +
2120                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2121
2122                 if ((grp == sbi->s_groups_count) &&
2123                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2124                         grp = i;
2125
2126                 block_bitmap = ext4_block_bitmap(sb, gdp);
2127                 if (block_bitmap == sb_block) {
2128                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2129                                  "Block bitmap for group %u overlaps "
2130                                  "superblock", i);
2131                 }
2132                 if (block_bitmap < first_block || block_bitmap > last_block) {
2133                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2134                                "Block bitmap for group %u not in group "
2135                                "(block %llu)!", i, block_bitmap);
2136                         return 0;
2137                 }
2138                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2139                 if (inode_bitmap == sb_block) {
2140                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2141                                  "Inode bitmap for group %u overlaps "
2142                                  "superblock", i);
2143                 }
2144                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2145                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2146                                "Inode bitmap for group %u not in group "
2147                                "(block %llu)!", i, inode_bitmap);
2148                         return 0;
2149                 }
2150                 inode_table = ext4_inode_table(sb, gdp);
2151                 if (inode_table == sb_block) {
2152                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2153                                  "Inode table for group %u overlaps "
2154                                  "superblock", i);
2155                 }
2156                 if (inode_table < first_block ||
2157                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2158                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2159                                "Inode table for group %u not in group "
2160                                "(block %llu)!", i, inode_table);
2161                         return 0;
2162                 }
2163                 ext4_lock_group(sb, i);
2164                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2165                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2166                                  "Checksum for group %u failed (%u!=%u)",
2167                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2168                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2169                         if (!(sb->s_flags & MS_RDONLY)) {
2170                                 ext4_unlock_group(sb, i);
2171                                 return 0;
2172                         }
2173                 }
2174                 ext4_unlock_group(sb, i);
2175                 if (!flexbg_flag)
2176                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2177         }
2178         if (NULL != first_not_zeroed)
2179                 *first_not_zeroed = grp;
2180         return 1;
2181 }
2182
2183 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2184  * the superblock) which were deleted from all directories, but held open by
2185  * a process at the time of a crash.  We walk the list and try to delete these
2186  * inodes at recovery time (only with a read-write filesystem).
2187  *
2188  * In order to keep the orphan inode chain consistent during traversal (in
2189  * case of crash during recovery), we link each inode into the superblock
2190  * orphan list_head and handle it the same way as an inode deletion during
2191  * normal operation (which journals the operations for us).
2192  *
2193  * We only do an iget() and an iput() on each inode, which is very safe if we
2194  * accidentally point at an in-use or already deleted inode.  The worst that
2195  * can happen in this case is that we get a "bit already cleared" message from
2196  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2197  * e2fsck was run on this filesystem, and it must have already done the orphan
2198  * inode cleanup for us, so we can safely abort without any further action.
2199  */
2200 static void ext4_orphan_cleanup(struct super_block *sb,
2201                                 struct ext4_super_block *es)
2202 {
2203         unsigned int s_flags = sb->s_flags;
2204         int nr_orphans = 0, nr_truncates = 0;
2205 #ifdef CONFIG_QUOTA
2206         int i;
2207 #endif
2208         if (!es->s_last_orphan) {
2209                 jbd_debug(4, "no orphan inodes to clean up\n");
2210                 return;
2211         }
2212
2213         if (bdev_read_only(sb->s_bdev)) {
2214                 ext4_msg(sb, KERN_ERR, "write access "
2215                         "unavailable, skipping orphan cleanup");
2216                 return;
2217         }
2218
2219         /* Check if feature set would not allow a r/w mount */
2220         if (!ext4_feature_set_ok(sb, 0)) {
2221                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2222                          "unknown ROCOMPAT features");
2223                 return;
2224         }
2225
2226         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2227                 /* don't clear list on RO mount w/ errors */
2228                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2229                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2230                                   "clearing orphan list.\n");
2231                         es->s_last_orphan = 0;
2232                 }
2233                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2234                 return;
2235         }
2236
2237         if (s_flags & MS_RDONLY) {
2238                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2239                 sb->s_flags &= ~MS_RDONLY;
2240         }
2241 #ifdef CONFIG_QUOTA
2242         /* Needed for iput() to work correctly and not trash data */
2243         sb->s_flags |= MS_ACTIVE;
2244         /* Turn on quotas so that they are updated correctly */
2245         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2246                 if (EXT4_SB(sb)->s_qf_names[i]) {
2247                         int ret = ext4_quota_on_mount(sb, i);
2248                         if (ret < 0)
2249                                 ext4_msg(sb, KERN_ERR,
2250                                         "Cannot turn on journaled "
2251                                         "quota: error %d", ret);
2252                 }
2253         }
2254 #endif
2255
2256         while (es->s_last_orphan) {
2257                 struct inode *inode;
2258
2259                 /*
2260                  * We may have encountered an error during cleanup; if
2261                  * so, skip the rest.
2262                  */
2263                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2264                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2265                         es->s_last_orphan = 0;
2266                         break;
2267                 }
2268
2269                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2270                 if (IS_ERR(inode)) {
2271                         es->s_last_orphan = 0;
2272                         break;
2273                 }
2274
2275                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2276                 dquot_initialize(inode);
2277                 if (inode->i_nlink) {
2278                         if (test_opt(sb, DEBUG))
2279                                 ext4_msg(sb, KERN_DEBUG,
2280                                         "%s: truncating inode %lu to %lld bytes",
2281                                         __func__, inode->i_ino, inode->i_size);
2282                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2283                                   inode->i_ino, inode->i_size);
2284                         mutex_lock(&inode->i_mutex);
2285                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2286                         ext4_truncate(inode);
2287                         mutex_unlock(&inode->i_mutex);
2288                         nr_truncates++;
2289                 } else {
2290                         if (test_opt(sb, DEBUG))
2291                                 ext4_msg(sb, KERN_DEBUG,
2292                                         "%s: deleting unreferenced inode %lu",
2293                                         __func__, inode->i_ino);
2294                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2295                                   inode->i_ino);
2296                         nr_orphans++;
2297                 }
2298                 iput(inode);  /* The delete magic happens here! */
2299         }
2300
2301 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2302
2303         if (nr_orphans)
2304                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2305                        PLURAL(nr_orphans));
2306         if (nr_truncates)
2307                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2308                        PLURAL(nr_truncates));
2309 #ifdef CONFIG_QUOTA
2310         /* Turn quotas off */
2311         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2312                 if (sb_dqopt(sb)->files[i])
2313                         dquot_quota_off(sb, i);
2314         }
2315 #endif
2316         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2317 }
2318
2319 /*
2320  * Maximal extent format file size.
2321  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2322  * extent format containers, within a sector_t, and within i_blocks
2323  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2324  * so that won't be a limiting factor.
2325  *
2326  * However there is other limiting factor. We do store extents in the form
2327  * of starting block and length, hence the resulting length of the extent
2328  * covering maximum file size must fit into on-disk format containers as
2329  * well. Given that length is always by 1 unit bigger than max unit (because
2330  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2331  *
2332  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2333  */
2334 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2335 {
2336         loff_t res;
2337         loff_t upper_limit = MAX_LFS_FILESIZE;
2338
2339         /* small i_blocks in vfs inode? */
2340         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2341                 /*
2342                  * CONFIG_LBDAF is not enabled implies the inode
2343                  * i_block represent total blocks in 512 bytes
2344                  * 32 == size of vfs inode i_blocks * 8
2345                  */
2346                 upper_limit = (1LL << 32) - 1;
2347
2348                 /* total blocks in file system block size */
2349                 upper_limit >>= (blkbits - 9);
2350                 upper_limit <<= blkbits;
2351         }
2352
2353         /*
2354          * 32-bit extent-start container, ee_block. We lower the maxbytes
2355          * by one fs block, so ee_len can cover the extent of maximum file
2356          * size
2357          */
2358         res = (1LL << 32) - 1;
2359         res <<= blkbits;
2360
2361         /* Sanity check against vm- & vfs- imposed limits */
2362         if (res > upper_limit)
2363                 res = upper_limit;
2364
2365         return res;
2366 }
2367
2368 /*
2369  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2370  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2371  * We need to be 1 filesystem block less than the 2^48 sector limit.
2372  */
2373 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2374 {
2375         loff_t res = EXT4_NDIR_BLOCKS;
2376         int meta_blocks;
2377         loff_t upper_limit;
2378         /* This is calculated to be the largest file size for a dense, block
2379          * mapped file such that the file's total number of 512-byte sectors,
2380          * including data and all indirect blocks, does not exceed (2^48 - 1).
2381          *
2382          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2383          * number of 512-byte sectors of the file.
2384          */
2385
2386         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2387                 /*
2388                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2389                  * the inode i_block field represents total file blocks in
2390                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2391                  */
2392                 upper_limit = (1LL << 32) - 1;
2393
2394                 /* total blocks in file system block size */
2395                 upper_limit >>= (bits - 9);
2396
2397         } else {
2398                 /*
2399                  * We use 48 bit ext4_inode i_blocks
2400                  * With EXT4_HUGE_FILE_FL set the i_blocks
2401                  * represent total number of blocks in
2402                  * file system block size
2403                  */
2404                 upper_limit = (1LL << 48) - 1;
2405
2406         }
2407
2408         /* indirect blocks */
2409         meta_blocks = 1;
2410         /* double indirect blocks */
2411         meta_blocks += 1 + (1LL << (bits-2));
2412         /* tripple indirect blocks */
2413         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2414
2415         upper_limit -= meta_blocks;
2416         upper_limit <<= bits;
2417
2418         res += 1LL << (bits-2);
2419         res += 1LL << (2*(bits-2));
2420         res += 1LL << (3*(bits-2));
2421         res <<= bits;
2422         if (res > upper_limit)
2423                 res = upper_limit;
2424
2425         if (res > MAX_LFS_FILESIZE)
2426                 res = MAX_LFS_FILESIZE;
2427
2428         return res;
2429 }
2430
2431 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2432                                    ext4_fsblk_t logical_sb_block, int nr)
2433 {
2434         struct ext4_sb_info *sbi = EXT4_SB(sb);
2435         ext4_group_t bg, first_meta_bg;
2436         int has_super = 0;
2437
2438         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2439
2440         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2441                 return logical_sb_block + nr + 1;
2442         bg = sbi->s_desc_per_block * nr;
2443         if (ext4_bg_has_super(sb, bg))
2444                 has_super = 1;
2445
2446         /*
2447          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2448          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2449          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2450          * compensate.
2451          */
2452         if (sb->s_blocksize == 1024 && nr == 0 &&
2453             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2454                 has_super++;
2455
2456         return (has_super + ext4_group_first_block_no(sb, bg));
2457 }
2458
2459 /**
2460  * ext4_get_stripe_size: Get the stripe size.
2461  * @sbi: In memory super block info
2462  *
2463  * If we have specified it via mount option, then
2464  * use the mount option value. If the value specified at mount time is
2465  * greater than the blocks per group use the super block value.
2466  * If the super block value is greater than blocks per group return 0.
2467  * Allocator needs it be less than blocks per group.
2468  *
2469  */
2470 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2471 {
2472         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2473         unsigned long stripe_width =
2474                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2475         int ret;
2476
2477         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2478                 ret = sbi->s_stripe;
2479         else if (stripe_width <= sbi->s_blocks_per_group)
2480                 ret = stripe_width;
2481         else if (stride <= sbi->s_blocks_per_group)
2482                 ret = stride;
2483         else
2484                 ret = 0;
2485
2486         /*
2487          * If the stripe width is 1, this makes no sense and
2488          * we set it to 0 to turn off stripe handling code.
2489          */
2490         if (ret <= 1)
2491                 ret = 0;
2492
2493         return ret;
2494 }
2495
2496 /*
2497  * Check whether this filesystem can be mounted based on
2498  * the features present and the RDONLY/RDWR mount requested.
2499  * Returns 1 if this filesystem can be mounted as requested,
2500  * 0 if it cannot be.
2501  */
2502 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2503 {
2504         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2505                 ext4_msg(sb, KERN_ERR,
2506                         "Couldn't mount because of "
2507                         "unsupported optional features (%x)",
2508                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2509                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2510                 return 0;
2511         }
2512
2513         if (readonly)
2514                 return 1;
2515
2516         if (ext4_has_feature_readonly(sb)) {
2517                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2518                 sb->s_flags |= MS_RDONLY;
2519                 return 1;
2520         }
2521
2522         /* Check that feature set is OK for a read-write mount */
2523         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2524                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2525                          "unsupported optional features (%x)",
2526                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2527                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2528                 return 0;
2529         }
2530         /*
2531          * Large file size enabled file system can only be mounted
2532          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2533          */
2534         if (ext4_has_feature_huge_file(sb)) {
2535                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2536                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2537                                  "cannot be mounted RDWR without "
2538                                  "CONFIG_LBDAF");
2539                         return 0;
2540                 }
2541         }
2542         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2543                 ext4_msg(sb, KERN_ERR,
2544                          "Can't support bigalloc feature without "
2545                          "extents feature\n");
2546                 return 0;
2547         }
2548
2549 #ifndef CONFIG_QUOTA
2550         if (ext4_has_feature_quota(sb) && !readonly) {
2551                 ext4_msg(sb, KERN_ERR,
2552                          "Filesystem with quota feature cannot be mounted RDWR "
2553                          "without CONFIG_QUOTA");
2554                 return 0;
2555         }
2556 #endif  /* CONFIG_QUOTA */
2557         return 1;
2558 }
2559
2560 /*
2561  * This function is called once a day if we have errors logged
2562  * on the file system
2563  */
2564 static void print_daily_error_info(unsigned long arg)
2565 {
2566         struct super_block *sb = (struct super_block *) arg;
2567         struct ext4_sb_info *sbi;
2568         struct ext4_super_block *es;
2569
2570         sbi = EXT4_SB(sb);
2571         es = sbi->s_es;
2572
2573         if (es->s_error_count)
2574                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2575                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2576                          le32_to_cpu(es->s_error_count));
2577         if (es->s_first_error_time) {
2578                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2579                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2580                        (int) sizeof(es->s_first_error_func),
2581                        es->s_first_error_func,
2582                        le32_to_cpu(es->s_first_error_line));
2583                 if (es->s_first_error_ino)
2584                         printk(": inode %u",
2585                                le32_to_cpu(es->s_first_error_ino));
2586                 if (es->s_first_error_block)
2587                         printk(": block %llu", (unsigned long long)
2588                                le64_to_cpu(es->s_first_error_block));
2589                 printk("\n");
2590         }
2591         if (es->s_last_error_time) {
2592                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2593                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2594                        (int) sizeof(es->s_last_error_func),
2595                        es->s_last_error_func,
2596                        le32_to_cpu(es->s_last_error_line));
2597                 if (es->s_last_error_ino)
2598                         printk(": inode %u",
2599                                le32_to_cpu(es->s_last_error_ino));
2600                 if (es->s_last_error_block)
2601                         printk(": block %llu", (unsigned long long)
2602                                le64_to_cpu(es->s_last_error_block));
2603                 printk("\n");
2604         }
2605         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2606 }
2607
2608 /* Find next suitable group and run ext4_init_inode_table */
2609 static int ext4_run_li_request(struct ext4_li_request *elr)
2610 {
2611         struct ext4_group_desc *gdp = NULL;
2612         ext4_group_t group, ngroups;
2613         struct super_block *sb;
2614         unsigned long timeout = 0;
2615         int ret = 0;
2616
2617         sb = elr->lr_super;
2618         ngroups = EXT4_SB(sb)->s_groups_count;
2619
2620         sb_start_write(sb);
2621         for (group = elr->lr_next_group; group < ngroups; group++) {
2622                 gdp = ext4_get_group_desc(sb, group, NULL);
2623                 if (!gdp) {
2624                         ret = 1;
2625                         break;
2626                 }
2627
2628                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2629                         break;
2630         }
2631
2632         if (group >= ngroups)
2633                 ret = 1;
2634
2635         if (!ret) {
2636                 timeout = jiffies;
2637                 ret = ext4_init_inode_table(sb, group,
2638                                             elr->lr_timeout ? 0 : 1);
2639                 if (elr->lr_timeout == 0) {
2640                         timeout = (jiffies - timeout) *
2641                                   elr->lr_sbi->s_li_wait_mult;
2642                         elr->lr_timeout = timeout;
2643                 }
2644                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2645                 elr->lr_next_group = group + 1;
2646         }
2647         sb_end_write(sb);
2648
2649         return ret;
2650 }
2651
2652 /*
2653  * Remove lr_request from the list_request and free the
2654  * request structure. Should be called with li_list_mtx held
2655  */
2656 static void ext4_remove_li_request(struct ext4_li_request *elr)
2657 {
2658         struct ext4_sb_info *sbi;
2659
2660         if (!elr)
2661                 return;
2662
2663         sbi = elr->lr_sbi;
2664
2665         list_del(&elr->lr_request);
2666         sbi->s_li_request = NULL;
2667         kfree(elr);
2668 }
2669
2670 static void ext4_unregister_li_request(struct super_block *sb)
2671 {
2672         mutex_lock(&ext4_li_mtx);
2673         if (!ext4_li_info) {
2674                 mutex_unlock(&ext4_li_mtx);
2675                 return;
2676         }
2677
2678         mutex_lock(&ext4_li_info->li_list_mtx);
2679         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2680         mutex_unlock(&ext4_li_info->li_list_mtx);
2681         mutex_unlock(&ext4_li_mtx);
2682 }
2683
2684 static struct task_struct *ext4_lazyinit_task;
2685
2686 /*
2687  * This is the function where ext4lazyinit thread lives. It walks
2688  * through the request list searching for next scheduled filesystem.
2689  * When such a fs is found, run the lazy initialization request
2690  * (ext4_rn_li_request) and keep track of the time spend in this
2691  * function. Based on that time we compute next schedule time of
2692  * the request. When walking through the list is complete, compute
2693  * next waking time and put itself into sleep.
2694  */
2695 static int ext4_lazyinit_thread(void *arg)
2696 {
2697         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2698         struct list_head *pos, *n;
2699         struct ext4_li_request *elr;
2700         unsigned long next_wakeup, cur;
2701
2702         BUG_ON(NULL == eli);
2703
2704 cont_thread:
2705         while (true) {
2706                 next_wakeup = MAX_JIFFY_OFFSET;
2707
2708                 mutex_lock(&eli->li_list_mtx);
2709                 if (list_empty(&eli->li_request_list)) {
2710                         mutex_unlock(&eli->li_list_mtx);
2711                         goto exit_thread;
2712                 }
2713
2714                 list_for_each_safe(pos, n, &eli->li_request_list) {
2715                         elr = list_entry(pos, struct ext4_li_request,
2716                                          lr_request);
2717
2718                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2719                                 if (ext4_run_li_request(elr) != 0) {
2720                                         /* error, remove the lazy_init job */
2721                                         ext4_remove_li_request(elr);
2722                                         continue;
2723                                 }
2724                         }
2725
2726                         if (time_before(elr->lr_next_sched, next_wakeup))
2727                                 next_wakeup = elr->lr_next_sched;
2728                 }
2729                 mutex_unlock(&eli->li_list_mtx);
2730
2731                 try_to_freeze();
2732
2733                 cur = jiffies;
2734                 if ((time_after_eq(cur, next_wakeup)) ||
2735                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2736                         cond_resched();
2737                         continue;
2738                 }
2739
2740                 schedule_timeout_interruptible(next_wakeup - cur);
2741
2742                 if (kthread_should_stop()) {
2743                         ext4_clear_request_list();
2744                         goto exit_thread;
2745                 }
2746         }
2747
2748 exit_thread:
2749         /*
2750          * It looks like the request list is empty, but we need
2751          * to check it under the li_list_mtx lock, to prevent any
2752          * additions into it, and of course we should lock ext4_li_mtx
2753          * to atomically free the list and ext4_li_info, because at
2754          * this point another ext4 filesystem could be registering
2755          * new one.
2756          */
2757         mutex_lock(&ext4_li_mtx);
2758         mutex_lock(&eli->li_list_mtx);
2759         if (!list_empty(&eli->li_request_list)) {
2760                 mutex_unlock(&eli->li_list_mtx);
2761                 mutex_unlock(&ext4_li_mtx);
2762                 goto cont_thread;
2763         }
2764         mutex_unlock(&eli->li_list_mtx);
2765         kfree(ext4_li_info);
2766         ext4_li_info = NULL;
2767         mutex_unlock(&ext4_li_mtx);
2768
2769         return 0;
2770 }
2771
2772 static void ext4_clear_request_list(void)
2773 {
2774         struct list_head *pos, *n;
2775         struct ext4_li_request *elr;
2776
2777         mutex_lock(&ext4_li_info->li_list_mtx);
2778         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2779                 elr = list_entry(pos, struct ext4_li_request,
2780                                  lr_request);
2781                 ext4_remove_li_request(elr);
2782         }
2783         mutex_unlock(&ext4_li_info->li_list_mtx);
2784 }
2785
2786 static int ext4_run_lazyinit_thread(void)
2787 {
2788         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2789                                          ext4_li_info, "ext4lazyinit");
2790         if (IS_ERR(ext4_lazyinit_task)) {
2791                 int err = PTR_ERR(ext4_lazyinit_task);
2792                 ext4_clear_request_list();
2793                 kfree(ext4_li_info);
2794                 ext4_li_info = NULL;
2795                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2796                                  "initialization thread\n",
2797                                  err);
2798                 return err;
2799         }
2800         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2801         return 0;
2802 }
2803
2804 /*
2805  * Check whether it make sense to run itable init. thread or not.
2806  * If there is at least one uninitialized inode table, return
2807  * corresponding group number, else the loop goes through all
2808  * groups and return total number of groups.
2809  */
2810 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2811 {
2812         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2813         struct ext4_group_desc *gdp = NULL;
2814
2815         for (group = 0; group < ngroups; group++) {
2816                 gdp = ext4_get_group_desc(sb, group, NULL);
2817                 if (!gdp)
2818                         continue;
2819
2820                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2821                         break;
2822         }
2823
2824         return group;
2825 }
2826
2827 static int ext4_li_info_new(void)
2828 {
2829         struct ext4_lazy_init *eli = NULL;
2830
2831         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2832         if (!eli)
2833                 return -ENOMEM;
2834
2835         INIT_LIST_HEAD(&eli->li_request_list);
2836         mutex_init(&eli->li_list_mtx);
2837
2838         eli->li_state |= EXT4_LAZYINIT_QUIT;
2839
2840         ext4_li_info = eli;
2841
2842         return 0;
2843 }
2844
2845 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2846                                             ext4_group_t start)
2847 {
2848         struct ext4_sb_info *sbi = EXT4_SB(sb);
2849         struct ext4_li_request *elr;
2850
2851         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2852         if (!elr)
2853                 return NULL;
2854
2855         elr->lr_super = sb;
2856         elr->lr_sbi = sbi;
2857         elr->lr_next_group = start;
2858
2859         /*
2860          * Randomize first schedule time of the request to
2861          * spread the inode table initialization requests
2862          * better.
2863          */
2864         elr->lr_next_sched = jiffies + (prandom_u32() %
2865                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2866         return elr;
2867 }
2868
2869 int ext4_register_li_request(struct super_block *sb,
2870                              ext4_group_t first_not_zeroed)
2871 {
2872         struct ext4_sb_info *sbi = EXT4_SB(sb);
2873         struct ext4_li_request *elr = NULL;
2874         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2875         int ret = 0;
2876
2877         mutex_lock(&ext4_li_mtx);
2878         if (sbi->s_li_request != NULL) {
2879                 /*
2880                  * Reset timeout so it can be computed again, because
2881                  * s_li_wait_mult might have changed.
2882                  */
2883                 sbi->s_li_request->lr_timeout = 0;
2884                 goto out;
2885         }
2886
2887         if (first_not_zeroed == ngroups ||
2888             (sb->s_flags & MS_RDONLY) ||
2889             !test_opt(sb, INIT_INODE_TABLE))
2890                 goto out;
2891
2892         elr = ext4_li_request_new(sb, first_not_zeroed);
2893         if (!elr) {
2894                 ret = -ENOMEM;
2895                 goto out;
2896         }
2897
2898         if (NULL == ext4_li_info) {
2899                 ret = ext4_li_info_new();
2900                 if (ret)
2901                         goto out;
2902         }
2903
2904         mutex_lock(&ext4_li_info->li_list_mtx);
2905         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2906         mutex_unlock(&ext4_li_info->li_list_mtx);
2907
2908         sbi->s_li_request = elr;
2909         /*
2910          * set elr to NULL here since it has been inserted to
2911          * the request_list and the removal and free of it is
2912          * handled by ext4_clear_request_list from now on.
2913          */
2914         elr = NULL;
2915
2916         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2917                 ret = ext4_run_lazyinit_thread();
2918                 if (ret)
2919                         goto out;
2920         }
2921 out:
2922         mutex_unlock(&ext4_li_mtx);
2923         if (ret)
2924                 kfree(elr);
2925         return ret;
2926 }
2927
2928 /*
2929  * We do not need to lock anything since this is called on
2930  * module unload.
2931  */
2932 static void ext4_destroy_lazyinit_thread(void)
2933 {
2934         /*
2935          * If thread exited earlier
2936          * there's nothing to be done.
2937          */
2938         if (!ext4_li_info || !ext4_lazyinit_task)
2939                 return;
2940
2941         kthread_stop(ext4_lazyinit_task);
2942 }
2943
2944 static int set_journal_csum_feature_set(struct super_block *sb)
2945 {
2946         int ret = 1;
2947         int compat, incompat;
2948         struct ext4_sb_info *sbi = EXT4_SB(sb);
2949
2950         if (ext4_has_metadata_csum(sb)) {
2951                 /* journal checksum v3 */
2952                 compat = 0;
2953                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
2954         } else {
2955                 /* journal checksum v1 */
2956                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
2957                 incompat = 0;
2958         }
2959
2960         jbd2_journal_clear_features(sbi->s_journal,
2961                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2962                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
2963                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
2964         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2965                 ret = jbd2_journal_set_features(sbi->s_journal,
2966                                 compat, 0,
2967                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
2968                                 incompat);
2969         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2970                 ret = jbd2_journal_set_features(sbi->s_journal,
2971                                 compat, 0,
2972                                 incompat);
2973                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2974                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2975         } else {
2976                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2977                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2978         }
2979
2980         return ret;
2981 }
2982
2983 /*
2984  * Note: calculating the overhead so we can be compatible with
2985  * historical BSD practice is quite difficult in the face of
2986  * clusters/bigalloc.  This is because multiple metadata blocks from
2987  * different block group can end up in the same allocation cluster.
2988  * Calculating the exact overhead in the face of clustered allocation
2989  * requires either O(all block bitmaps) in memory or O(number of block
2990  * groups**2) in time.  We will still calculate the superblock for
2991  * older file systems --- and if we come across with a bigalloc file
2992  * system with zero in s_overhead_clusters the estimate will be close to
2993  * correct especially for very large cluster sizes --- but for newer
2994  * file systems, it's better to calculate this figure once at mkfs
2995  * time, and store it in the superblock.  If the superblock value is
2996  * present (even for non-bigalloc file systems), we will use it.
2997  */
2998 static int count_overhead(struct super_block *sb, ext4_group_t grp,
2999                           char *buf)
3000 {
3001         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3002         struct ext4_group_desc  *gdp;
3003         ext4_fsblk_t            first_block, last_block, b;
3004         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3005         int                     s, j, count = 0;
3006
3007         if (!ext4_has_feature_bigalloc(sb))
3008                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3009                         sbi->s_itb_per_group + 2);
3010
3011         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3012                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3013         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3014         for (i = 0; i < ngroups; i++) {
3015                 gdp = ext4_get_group_desc(sb, i, NULL);
3016                 b = ext4_block_bitmap(sb, gdp);
3017                 if (b >= first_block && b <= last_block) {
3018                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3019                         count++;
3020                 }
3021                 b = ext4_inode_bitmap(sb, gdp);
3022                 if (b >= first_block && b <= last_block) {
3023                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3024                         count++;
3025                 }
3026                 b = ext4_inode_table(sb, gdp);
3027                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3028                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3029                                 int c = EXT4_B2C(sbi, b - first_block);
3030                                 ext4_set_bit(c, buf);
3031                                 count++;
3032                         }
3033                 if (i != grp)
3034                         continue;
3035                 s = 0;
3036                 if (ext4_bg_has_super(sb, grp)) {
3037                         ext4_set_bit(s++, buf);
3038                         count++;
3039                 }
3040                 for (j = ext4_bg_num_gdb(sb, grp); j > 0; j--) {
3041                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3042                         count++;
3043                 }
3044         }
3045         if (!count)
3046                 return 0;
3047         return EXT4_CLUSTERS_PER_GROUP(sb) -
3048                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3049 }
3050
3051 /*
3052  * Compute the overhead and stash it in sbi->s_overhead
3053  */
3054 int ext4_calculate_overhead(struct super_block *sb)
3055 {
3056         struct ext4_sb_info *sbi = EXT4_SB(sb);
3057         struct ext4_super_block *es = sbi->s_es;
3058         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3059         ext4_fsblk_t overhead = 0;
3060         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3061
3062         if (!buf)
3063                 return -ENOMEM;
3064
3065         /*
3066          * Compute the overhead (FS structures).  This is constant
3067          * for a given filesystem unless the number of block groups
3068          * changes so we cache the previous value until it does.
3069          */
3070
3071         /*
3072          * All of the blocks before first_data_block are overhead
3073          */
3074         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3075
3076         /*
3077          * Add the overhead found in each block group
3078          */
3079         for (i = 0; i < ngroups; i++) {
3080                 int blks;
3081
3082                 blks = count_overhead(sb, i, buf);
3083                 overhead += blks;
3084                 if (blks)
3085                         memset(buf, 0, PAGE_SIZE);
3086                 cond_resched();
3087         }
3088         /* Add the internal journal blocks as well */
3089         if (sbi->s_journal && !sbi->journal_bdev)
3090                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3091
3092         sbi->s_overhead = overhead;
3093         smp_wmb();
3094         free_page((unsigned long) buf);
3095         return 0;
3096 }
3097
3098 static void ext4_set_resv_clusters(struct super_block *sb)
3099 {
3100         ext4_fsblk_t resv_clusters;
3101         struct ext4_sb_info *sbi = EXT4_SB(sb);
3102
3103         /*
3104          * There's no need to reserve anything when we aren't using extents.
3105          * The space estimates are exact, there are no unwritten extents,
3106          * hole punching doesn't need new metadata... This is needed especially
3107          * to keep ext2/3 backward compatibility.
3108          */
3109         if (!ext4_has_feature_extents(sb))
3110                 return;
3111         /*
3112          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3113          * This should cover the situations where we can not afford to run
3114          * out of space like for example punch hole, or converting
3115          * unwritten extents in delalloc path. In most cases such
3116          * allocation would require 1, or 2 blocks, higher numbers are
3117          * very rare.
3118          */
3119         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3120                          sbi->s_cluster_bits);
3121
3122         do_div(resv_clusters, 50);
3123         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3124
3125         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3126 }
3127
3128 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3129 {
3130         char *orig_data = kstrdup(data, GFP_KERNEL);
3131         struct buffer_head *bh;
3132         struct ext4_super_block *es = NULL;
3133         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3134         ext4_fsblk_t block;
3135         ext4_fsblk_t sb_block = get_sb_block(&data);
3136         ext4_fsblk_t logical_sb_block;
3137         unsigned long offset = 0;
3138         unsigned long journal_devnum = 0;
3139         unsigned long def_mount_opts;
3140         struct inode *root;
3141         const char *descr;
3142         int ret = -ENOMEM;
3143         int blocksize, clustersize;
3144         unsigned int db_count;
3145         unsigned int i;
3146         int needs_recovery, has_huge_files, has_bigalloc;
3147         __u64 blocks_count;
3148         int err = 0;
3149         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3150         ext4_group_t first_not_zeroed;
3151
3152         if ((data && !orig_data) || !sbi)
3153                 goto out_free_base;
3154
3155         sbi->s_blockgroup_lock =
3156                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3157         if (!sbi->s_blockgroup_lock)
3158                 goto out_free_base;
3159
3160         sb->s_fs_info = sbi;
3161         sbi->s_sb = sb;
3162         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3163         sbi->s_sb_block = sb_block;
3164         if (sb->s_bdev->bd_part)
3165                 sbi->s_sectors_written_start =
3166                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3167
3168         /* Cleanup superblock name */
3169         strreplace(sb->s_id, '/', '!');
3170
3171         /* -EINVAL is default */
3172         ret = -EINVAL;
3173         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3174         if (!blocksize) {
3175                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3176                 goto out_fail;
3177         }
3178
3179         /*
3180          * The ext4 superblock will not be buffer aligned for other than 1kB
3181          * block sizes.  We need to calculate the offset from buffer start.
3182          */
3183         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3184                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3185                 offset = do_div(logical_sb_block, blocksize);
3186         } else {
3187                 logical_sb_block = sb_block;
3188         }
3189
3190         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3191                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3192                 goto out_fail;
3193         }
3194         /*
3195          * Note: s_es must be initialized as soon as possible because
3196          *       some ext4 macro-instructions depend on its value
3197          */
3198         es = (struct ext4_super_block *) (bh->b_data + offset);
3199         sbi->s_es = es;
3200         sb->s_magic = le16_to_cpu(es->s_magic);
3201         if (sb->s_magic != EXT4_SUPER_MAGIC)
3202                 goto cantfind_ext4;
3203         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3204
3205         /* Warn if metadata_csum and gdt_csum are both set. */
3206         if (ext4_has_feature_metadata_csum(sb) &&
3207             ext4_has_feature_gdt_csum(sb))
3208                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3209                              "redundant flags; please run fsck.");
3210
3211         /* Check for a known checksum algorithm */
3212         if (!ext4_verify_csum_type(sb, es)) {
3213                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3214                          "unknown checksum algorithm.");
3215                 silent = 1;
3216                 goto cantfind_ext4;
3217         }
3218
3219         /* Load the checksum driver */
3220         if (ext4_has_feature_metadata_csum(sb)) {
3221                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3222                 if (IS_ERR(sbi->s_chksum_driver)) {
3223                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3224                         ret = PTR_ERR(sbi->s_chksum_driver);
3225                         sbi->s_chksum_driver = NULL;
3226                         goto failed_mount;
3227                 }
3228         }
3229
3230         /* Check superblock checksum */
3231         if (!ext4_superblock_csum_verify(sb, es)) {
3232                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3233                          "invalid superblock checksum.  Run e2fsck?");
3234                 silent = 1;
3235                 ret = -EFSBADCRC;
3236                 goto cantfind_ext4;
3237         }
3238
3239         /* Precompute checksum seed for all metadata */
3240         if (ext4_has_feature_csum_seed(sb))
3241                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3242         else if (ext4_has_metadata_csum(sb))
3243                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3244                                                sizeof(es->s_uuid));
3245
3246         /* Set defaults before we parse the mount options */
3247         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3248         set_opt(sb, INIT_INODE_TABLE);
3249         if (def_mount_opts & EXT4_DEFM_DEBUG)
3250                 set_opt(sb, DEBUG);
3251         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3252                 set_opt(sb, GRPID);
3253         if (def_mount_opts & EXT4_DEFM_UID16)
3254                 set_opt(sb, NO_UID32);
3255         /* xattr user namespace & acls are now defaulted on */
3256         set_opt(sb, XATTR_USER);
3257 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3258         set_opt(sb, POSIX_ACL);
3259 #endif
3260         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3261         if (ext4_has_metadata_csum(sb))
3262                 set_opt(sb, JOURNAL_CHECKSUM);
3263
3264         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3265                 set_opt(sb, JOURNAL_DATA);
3266         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3267                 set_opt(sb, ORDERED_DATA);
3268         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3269                 set_opt(sb, WRITEBACK_DATA);
3270
3271         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3272                 set_opt(sb, ERRORS_PANIC);
3273         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3274                 set_opt(sb, ERRORS_CONT);
3275         else
3276                 set_opt(sb, ERRORS_RO);
3277         /* block_validity enabled by default; disable with noblock_validity */
3278         set_opt(sb, BLOCK_VALIDITY);
3279         if (def_mount_opts & EXT4_DEFM_DISCARD)
3280                 set_opt(sb, DISCARD);
3281
3282         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3283         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3284         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3285         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3286         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3287
3288         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3289                 set_opt(sb, BARRIER);
3290
3291         /*
3292          * enable delayed allocation by default
3293          * Use -o nodelalloc to turn it off
3294          */
3295         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3296             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3297                 set_opt(sb, DELALLOC);
3298
3299         /*
3300          * set default s_li_wait_mult for lazyinit, for the case there is
3301          * no mount option specified.
3302          */
3303         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3304
3305         if (sbi->s_es->s_mount_opts[0]) {
3306                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3307                                               sizeof(sbi->s_es->s_mount_opts),
3308                                               GFP_KERNEL);
3309                 if (!s_mount_opts)
3310                         goto failed_mount;
3311                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3312                                    &journal_ioprio, 0)) {
3313                         ext4_msg(sb, KERN_WARNING,
3314                                  "failed to parse options in superblock: %s",
3315                                  s_mount_opts);
3316                 }
3317                 kfree(s_mount_opts);
3318         }
3319         sbi->s_def_mount_opt = sbi->s_mount_opt;
3320         if (!parse_options((char *) data, sb, &journal_devnum,
3321                            &journal_ioprio, 0))
3322                 goto failed_mount;
3323
3324         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3325                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3326                             "with data=journal disables delayed "
3327                             "allocation and O_DIRECT support!\n");
3328                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3329                         ext4_msg(sb, KERN_ERR, "can't mount with "
3330                                  "both data=journal and delalloc");
3331                         goto failed_mount;
3332                 }
3333                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3334                         ext4_msg(sb, KERN_ERR, "can't mount with "
3335                                  "both data=journal and dioread_nolock");
3336                         goto failed_mount;
3337                 }
3338                 if (test_opt(sb, DAX)) {
3339                         ext4_msg(sb, KERN_ERR, "can't mount with "
3340                                  "both data=journal and dax");
3341                         goto failed_mount;
3342                 }
3343                 if (test_opt(sb, DELALLOC))
3344                         clear_opt(sb, DELALLOC);
3345         } else {
3346                 sb->s_iflags |= SB_I_CGROUPWB;
3347         }
3348
3349         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3350                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3351
3352         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3353             (ext4_has_compat_features(sb) ||
3354              ext4_has_ro_compat_features(sb) ||
3355              ext4_has_incompat_features(sb)))
3356                 ext4_msg(sb, KERN_WARNING,
3357                        "feature flags set on rev 0 fs, "
3358                        "running e2fsck is recommended");
3359
3360         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3361                 set_opt2(sb, HURD_COMPAT);
3362                 if (ext4_has_feature_64bit(sb)) {
3363                         ext4_msg(sb, KERN_ERR,
3364                                  "The Hurd can't support 64-bit file systems");
3365                         goto failed_mount;
3366                 }
3367         }
3368
3369         if (IS_EXT2_SB(sb)) {
3370                 if (ext2_feature_set_ok(sb))
3371                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3372                                  "using the ext4 subsystem");
3373                 else {
3374                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3375                                  "to feature incompatibilities");
3376                         goto failed_mount;
3377                 }
3378         }
3379
3380         if (IS_EXT3_SB(sb)) {
3381                 if (ext3_feature_set_ok(sb))
3382                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3383                                  "using the ext4 subsystem");
3384                 else {
3385                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3386                                  "to feature incompatibilities");
3387                         goto failed_mount;
3388                 }
3389         }
3390
3391         /*
3392          * Check feature flags regardless of the revision level, since we
3393          * previously didn't change the revision level when setting the flags,
3394          * so there is a chance incompat flags are set on a rev 0 filesystem.
3395          */
3396         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3397                 goto failed_mount;
3398
3399         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3400         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3401             blocksize > EXT4_MAX_BLOCK_SIZE) {
3402                 ext4_msg(sb, KERN_ERR,
3403                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3404                          blocksize, le32_to_cpu(es->s_log_block_size));
3405                 goto failed_mount;
3406         }
3407         if (le32_to_cpu(es->s_log_block_size) >
3408             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3409                 ext4_msg(sb, KERN_ERR,
3410                          "Invalid log block size: %u",
3411                          le32_to_cpu(es->s_log_block_size));
3412                 goto failed_mount;
3413         }
3414
3415         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3416                 ext4_msg(sb, KERN_ERR,
3417                          "Number of reserved GDT blocks insanely large: %d",
3418                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3419                 goto failed_mount;
3420         }
3421
3422         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3423                 if (blocksize != PAGE_SIZE) {
3424                         ext4_msg(sb, KERN_ERR,
3425                                         "error: unsupported blocksize for dax");
3426                         goto failed_mount;
3427                 }
3428                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3429                         ext4_msg(sb, KERN_ERR,
3430                                         "error: device does not support dax");
3431                         goto failed_mount;
3432                 }
3433         }
3434
3435         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3436                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3437                          es->s_encryption_level);
3438                 goto failed_mount;
3439         }
3440
3441         if (sb->s_blocksize != blocksize) {
3442                 /* Validate the filesystem blocksize */
3443                 if (!sb_set_blocksize(sb, blocksize)) {
3444                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3445                                         blocksize);
3446                         goto failed_mount;
3447                 }
3448
3449                 brelse(bh);
3450                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3451                 offset = do_div(logical_sb_block, blocksize);
3452                 bh = sb_bread_unmovable(sb, logical_sb_block);
3453                 if (!bh) {
3454                         ext4_msg(sb, KERN_ERR,
3455                                "Can't read superblock on 2nd try");
3456                         goto failed_mount;
3457                 }
3458                 es = (struct ext4_super_block *)(bh->b_data + offset);
3459                 sbi->s_es = es;
3460                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3461                         ext4_msg(sb, KERN_ERR,
3462                                "Magic mismatch, very weird!");
3463                         goto failed_mount;
3464                 }
3465         }
3466
3467         has_huge_files = ext4_has_feature_huge_file(sb);
3468         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3469                                                       has_huge_files);
3470         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3471
3472         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3473                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3474                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3475         } else {
3476                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3477                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3478                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3479                     (!is_power_of_2(sbi->s_inode_size)) ||
3480                     (sbi->s_inode_size > blocksize)) {
3481                         ext4_msg(sb, KERN_ERR,
3482                                "unsupported inode size: %d",
3483                                sbi->s_inode_size);
3484                         goto failed_mount;
3485                 }
3486                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3487                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3488         }
3489
3490         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3491         if (ext4_has_feature_64bit(sb)) {
3492                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3493                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3494                     !is_power_of_2(sbi->s_desc_size)) {
3495                         ext4_msg(sb, KERN_ERR,
3496                                "unsupported descriptor size %lu",
3497                                sbi->s_desc_size);
3498                         goto failed_mount;
3499                 }
3500         } else
3501                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3502
3503         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3504         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3505
3506         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3507         if (sbi->s_inodes_per_block == 0)
3508                 goto cantfind_ext4;
3509         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3510             sbi->s_inodes_per_group > blocksize * 8) {
3511                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3512                          sbi->s_blocks_per_group);
3513                 goto failed_mount;
3514         }
3515         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3516                                         sbi->s_inodes_per_block;
3517         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3518         sbi->s_sbh = bh;
3519         sbi->s_mount_state = le16_to_cpu(es->s_state);
3520         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3521         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3522
3523         for (i = 0; i < 4; i++)
3524                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3525         sbi->s_def_hash_version = es->s_def_hash_version;
3526         if (ext4_has_feature_dir_index(sb)) {
3527                 i = le32_to_cpu(es->s_flags);
3528                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3529                         sbi->s_hash_unsigned = 3;
3530                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3531 #ifdef __CHAR_UNSIGNED__
3532                         if (!(sb->s_flags & MS_RDONLY))
3533                                 es->s_flags |=
3534                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3535                         sbi->s_hash_unsigned = 3;
3536 #else
3537                         if (!(sb->s_flags & MS_RDONLY))
3538                                 es->s_flags |=
3539                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3540 #endif
3541                 }
3542         }
3543
3544         /* Handle clustersize */
3545         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3546         has_bigalloc = ext4_has_feature_bigalloc(sb);
3547         if (has_bigalloc) {
3548                 if (clustersize < blocksize) {
3549                         ext4_msg(sb, KERN_ERR,
3550                                  "cluster size (%d) smaller than "
3551                                  "block size (%d)", clustersize, blocksize);
3552                         goto failed_mount;
3553                 }
3554                 if (le32_to_cpu(es->s_log_cluster_size) >
3555                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3556                         ext4_msg(sb, KERN_ERR,
3557                                  "Invalid log cluster size: %u",
3558                                  le32_to_cpu(es->s_log_cluster_size));
3559                         goto failed_mount;
3560                 }
3561                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3562                         le32_to_cpu(es->s_log_block_size);
3563                 sbi->s_clusters_per_group =
3564                         le32_to_cpu(es->s_clusters_per_group);
3565                 if (sbi->s_clusters_per_group > blocksize * 8) {
3566                         ext4_msg(sb, KERN_ERR,
3567                                  "#clusters per group too big: %lu",
3568                                  sbi->s_clusters_per_group);
3569                         goto failed_mount;
3570                 }
3571                 if (sbi->s_blocks_per_group !=
3572                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3573                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3574                                  "clusters per group (%lu) inconsistent",
3575                                  sbi->s_blocks_per_group,
3576                                  sbi->s_clusters_per_group);
3577                         goto failed_mount;
3578                 }
3579         } else {
3580                 if (clustersize != blocksize) {
3581                         ext4_warning(sb, "fragment/cluster size (%d) != "
3582                                      "block size (%d)", clustersize,
3583                                      blocksize);
3584                         clustersize = blocksize;
3585                 }
3586                 if (sbi->s_blocks_per_group > blocksize * 8) {
3587                         ext4_msg(sb, KERN_ERR,
3588                                  "#blocks per group too big: %lu",
3589                                  sbi->s_blocks_per_group);
3590                         goto failed_mount;
3591                 }
3592                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3593                 sbi->s_cluster_bits = 0;
3594         }
3595         sbi->s_cluster_ratio = clustersize / blocksize;
3596
3597         /* Do we have standard group size of clustersize * 8 blocks ? */
3598         if (sbi->s_blocks_per_group == clustersize << 3)
3599                 set_opt2(sb, STD_GROUP_SIZE);
3600
3601         /*
3602          * Test whether we have more sectors than will fit in sector_t,
3603          * and whether the max offset is addressable by the page cache.
3604          */
3605         err = generic_check_addressable(sb->s_blocksize_bits,
3606                                         ext4_blocks_count(es));
3607         if (err) {
3608                 ext4_msg(sb, KERN_ERR, "filesystem"
3609                          " too large to mount safely on this system");
3610                 if (sizeof(sector_t) < 8)
3611                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3612                 goto failed_mount;
3613         }
3614
3615         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3616                 goto cantfind_ext4;
3617
3618         /* check blocks count against device size */
3619         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3620         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3621                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3622                        "exceeds size of device (%llu blocks)",
3623                        ext4_blocks_count(es), blocks_count);
3624                 goto failed_mount;
3625         }
3626
3627         /*
3628          * It makes no sense for the first data block to be beyond the end
3629          * of the filesystem.
3630          */
3631         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3632                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3633                          "block %u is beyond end of filesystem (%llu)",
3634                          le32_to_cpu(es->s_first_data_block),
3635                          ext4_blocks_count(es));
3636                 goto failed_mount;
3637         }
3638         blocks_count = (ext4_blocks_count(es) -
3639                         le32_to_cpu(es->s_first_data_block) +
3640                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3641         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3642         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3643                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3644                        "(block count %llu, first data block %u, "
3645                        "blocks per group %lu)", sbi->s_groups_count,
3646                        ext4_blocks_count(es),
3647                        le32_to_cpu(es->s_first_data_block),
3648                        EXT4_BLOCKS_PER_GROUP(sb));
3649                 goto failed_mount;
3650         }
3651         sbi->s_groups_count = blocks_count;
3652         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3653                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3654         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3655                    EXT4_DESC_PER_BLOCK(sb);
3656         sbi->s_group_desc = ext4_kvmalloc(db_count *
3657                                           sizeof(struct buffer_head *),
3658                                           GFP_KERNEL);
3659         if (sbi->s_group_desc == NULL) {
3660                 ext4_msg(sb, KERN_ERR, "not enough memory");
3661                 ret = -ENOMEM;
3662                 goto failed_mount;
3663         }
3664
3665         bgl_lock_init(sbi->s_blockgroup_lock);
3666
3667         for (i = 0; i < db_count; i++) {
3668                 block = descriptor_loc(sb, logical_sb_block, i);
3669                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3670                 if (!sbi->s_group_desc[i]) {
3671                         ext4_msg(sb, KERN_ERR,
3672                                "can't read group descriptor %d", i);
3673                         db_count = i;
3674                         goto failed_mount2;
3675                 }
3676         }
3677         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3678                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3679                 ret = -EFSCORRUPTED;
3680                 goto failed_mount2;
3681         }
3682
3683         sbi->s_gdb_count = db_count;
3684         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3685         spin_lock_init(&sbi->s_next_gen_lock);
3686
3687         setup_timer(&sbi->s_err_report, print_daily_error_info,
3688                 (unsigned long) sb);
3689
3690         /* Register extent status tree shrinker */
3691         if (ext4_es_register_shrinker(sbi))
3692                 goto failed_mount3;
3693
3694         sbi->s_stripe = ext4_get_stripe_size(sbi);
3695         sbi->s_extent_max_zeroout_kb = 32;
3696
3697         /*
3698          * set up enough so that it can read an inode
3699          */
3700         sb->s_op = &ext4_sops;
3701         sb->s_export_op = &ext4_export_ops;
3702         sb->s_xattr = ext4_xattr_handlers;
3703 #ifdef CONFIG_QUOTA
3704         sb->dq_op = &ext4_quota_operations;
3705         if (ext4_has_feature_quota(sb))
3706                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3707         else
3708                 sb->s_qcop = &ext4_qctl_operations;
3709         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
3710 #endif
3711         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3712
3713         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3714         mutex_init(&sbi->s_orphan_lock);
3715
3716         sb->s_root = NULL;
3717
3718         needs_recovery = (es->s_last_orphan != 0 ||
3719                           ext4_has_feature_journal_needs_recovery(sb));
3720
3721         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3722                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3723                         goto failed_mount3a;
3724
3725         /*
3726          * The first inode we look at is the journal inode.  Don't try
3727          * root first: it may be modified in the journal!
3728          */
3729         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3730                 if (ext4_load_journal(sb, es, journal_devnum))
3731                         goto failed_mount3a;
3732         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3733                    ext4_has_feature_journal_needs_recovery(sb)) {
3734                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3735                        "suppressed and not mounted read-only");
3736                 goto failed_mount_wq;
3737         } else {
3738                 /* Nojournal mode, all journal mount options are illegal */
3739                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3740                         ext4_msg(sb, KERN_ERR, "can't mount with "
3741                                  "journal_checksum, fs mounted w/o journal");
3742                         goto failed_mount_wq;
3743                 }
3744                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3745                         ext4_msg(sb, KERN_ERR, "can't mount with "
3746                                  "journal_async_commit, fs mounted w/o journal");
3747                         goto failed_mount_wq;
3748                 }
3749                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3750                         ext4_msg(sb, KERN_ERR, "can't mount with "
3751                                  "commit=%lu, fs mounted w/o journal",
3752                                  sbi->s_commit_interval / HZ);
3753                         goto failed_mount_wq;
3754                 }
3755                 if (EXT4_MOUNT_DATA_FLAGS &
3756                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3757                         ext4_msg(sb, KERN_ERR, "can't mount with "
3758                                  "data=, fs mounted w/o journal");
3759                         goto failed_mount_wq;
3760                 }
3761                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
3762                 clear_opt(sb, JOURNAL_CHECKSUM);
3763                 clear_opt(sb, DATA_FLAGS);
3764                 sbi->s_journal = NULL;
3765                 needs_recovery = 0;
3766                 goto no_journal;
3767         }
3768
3769         if (ext4_has_feature_64bit(sb) &&
3770             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3771                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3772                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3773                 goto failed_mount_wq;
3774         }
3775
3776         if (!set_journal_csum_feature_set(sb)) {
3777                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3778                          "feature set");
3779                 goto failed_mount_wq;
3780         }
3781
3782         /* We have now updated the journal if required, so we can
3783          * validate the data journaling mode. */
3784         switch (test_opt(sb, DATA_FLAGS)) {
3785         case 0:
3786                 /* No mode set, assume a default based on the journal
3787                  * capabilities: ORDERED_DATA if the journal can
3788                  * cope, else JOURNAL_DATA
3789                  */
3790                 if (jbd2_journal_check_available_features
3791                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3792                         set_opt(sb, ORDERED_DATA);
3793                 else
3794                         set_opt(sb, JOURNAL_DATA);
3795                 break;
3796
3797         case EXT4_MOUNT_ORDERED_DATA:
3798         case EXT4_MOUNT_WRITEBACK_DATA:
3799                 if (!jbd2_journal_check_available_features
3800                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3801                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3802                                "requested data journaling mode");
3803                         goto failed_mount_wq;
3804                 }
3805         default:
3806                 break;
3807         }
3808         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3809
3810         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3811
3812 no_journal:
3813         if (ext4_mballoc_ready) {
3814                 sbi->s_mb_cache = ext4_xattr_create_cache(sb->s_id);
3815                 if (!sbi->s_mb_cache) {
3816                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
3817                         goto failed_mount_wq;
3818                 }
3819         }
3820
3821         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
3822             (blocksize != PAGE_CACHE_SIZE)) {
3823                 ext4_msg(sb, KERN_ERR,
3824                          "Unsupported blocksize for fs encryption");
3825                 goto failed_mount_wq;
3826         }
3827
3828         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
3829             !ext4_has_feature_encrypt(sb)) {
3830                 ext4_set_feature_encrypt(sb);
3831                 ext4_commit_super(sb, 1);
3832         }
3833
3834         /*
3835          * Get the # of file system overhead blocks from the
3836          * superblock if present.
3837          */
3838         if (es->s_overhead_clusters)
3839                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3840         else {
3841                 err = ext4_calculate_overhead(sb);
3842                 if (err)
3843                         goto failed_mount_wq;
3844         }
3845
3846         /*
3847          * The maximum number of concurrent works can be high and
3848          * concurrency isn't really necessary.  Limit it to 1.
3849          */
3850         EXT4_SB(sb)->rsv_conversion_wq =
3851                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3852         if (!EXT4_SB(sb)->rsv_conversion_wq) {
3853                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3854                 ret = -ENOMEM;
3855                 goto failed_mount4;
3856         }
3857
3858         /*
3859          * The jbd2_journal_load will have done any necessary log recovery,
3860          * so we can safely mount the rest of the filesystem now.
3861          */
3862
3863         root = ext4_iget(sb, EXT4_ROOT_INO);
3864         if (IS_ERR(root)) {
3865                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3866                 ret = PTR_ERR(root);
3867                 root = NULL;
3868                 goto failed_mount4;
3869         }
3870         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3871                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3872                 iput(root);
3873                 goto failed_mount4;
3874         }
3875         sb->s_root = d_make_root(root);
3876         if (!sb->s_root) {
3877                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3878                 ret = -ENOMEM;
3879                 goto failed_mount4;
3880         }
3881
3882         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3883                 sb->s_flags |= MS_RDONLY;
3884
3885         /* determine the minimum size of new large inodes, if present */
3886         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3887                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3888                                                      EXT4_GOOD_OLD_INODE_SIZE;
3889                 if (ext4_has_feature_extra_isize(sb)) {
3890                         if (sbi->s_want_extra_isize <
3891                             le16_to_cpu(es->s_want_extra_isize))
3892                                 sbi->s_want_extra_isize =
3893                                         le16_to_cpu(es->s_want_extra_isize);
3894                         if (sbi->s_want_extra_isize <
3895                             le16_to_cpu(es->s_min_extra_isize))
3896                                 sbi->s_want_extra_isize =
3897                                         le16_to_cpu(es->s_min_extra_isize);
3898                 }
3899         }
3900         /* Check if enough inode space is available */
3901         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3902                                                         sbi->s_inode_size) {
3903                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3904                                                        EXT4_GOOD_OLD_INODE_SIZE;
3905                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3906                          "available");
3907         }
3908
3909         ext4_set_resv_clusters(sb);
3910
3911         err = ext4_setup_system_zone(sb);
3912         if (err) {
3913                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3914                          "zone (%d)", err);
3915                 goto failed_mount4a;
3916         }
3917
3918         ext4_ext_init(sb);
3919         err = ext4_mb_init(sb);
3920         if (err) {
3921                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3922                          err);
3923                 goto failed_mount5;
3924         }
3925
3926         block = ext4_count_free_clusters(sb);
3927         ext4_free_blocks_count_set(sbi->s_es,
3928                                    EXT4_C2B(sbi, block));
3929         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
3930                                   GFP_KERNEL);
3931         if (!err) {
3932                 unsigned long freei = ext4_count_free_inodes(sb);
3933                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
3934                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
3935                                           GFP_KERNEL);
3936         }
3937         if (!err)
3938                 err = percpu_counter_init(&sbi->s_dirs_counter,
3939                                           ext4_count_dirs(sb), GFP_KERNEL);
3940         if (!err)
3941                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
3942                                           GFP_KERNEL);
3943         if (err) {
3944                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3945                 goto failed_mount6;
3946         }
3947
3948         if (ext4_has_feature_flex_bg(sb))
3949                 if (!ext4_fill_flex_info(sb)) {
3950                         ext4_msg(sb, KERN_ERR,
3951                                "unable to initialize "
3952                                "flex_bg meta info!");
3953                         goto failed_mount6;
3954                 }
3955
3956         err = ext4_register_li_request(sb, first_not_zeroed);
3957         if (err)
3958                 goto failed_mount6;
3959
3960         err = ext4_register_sysfs(sb);
3961         if (err)
3962                 goto failed_mount7;
3963
3964 #ifdef CONFIG_QUOTA
3965         /* Enable quota usage during mount. */
3966         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
3967                 err = ext4_enable_quotas(sb);
3968                 if (err)
3969                         goto failed_mount8;
3970         }
3971 #endif  /* CONFIG_QUOTA */
3972
3973         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
3974         ext4_orphan_cleanup(sb, es);
3975         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
3976         if (needs_recovery) {
3977                 ext4_msg(sb, KERN_INFO, "recovery complete");
3978                 ext4_mark_recovery_complete(sb, es);
3979         }
3980         if (EXT4_SB(sb)->s_journal) {
3981                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
3982                         descr = " journalled data mode";
3983                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
3984                         descr = " ordered data mode";
3985                 else
3986                         descr = " writeback data mode";
3987         } else
3988                 descr = "out journal";
3989
3990         if (test_opt(sb, DISCARD)) {
3991                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
3992                 if (!blk_queue_discard(q))
3993                         ext4_msg(sb, KERN_WARNING,
3994                                  "mounting with \"discard\" option, but "
3995                                  "the device does not support discard");
3996         }
3997
3998         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
3999                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4000                          "Opts: %.*s%s%s", descr,
4001                          (int) sizeof(sbi->s_es->s_mount_opts),
4002                          sbi->s_es->s_mount_opts,
4003                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4004
4005         if (es->s_error_count)
4006                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4007
4008         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4009         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4010         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4011         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4012
4013         kfree(orig_data);
4014         return 0;
4015
4016 cantfind_ext4:
4017         if (!silent)
4018                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4019         goto failed_mount;
4020
4021 #ifdef CONFIG_QUOTA
4022 failed_mount8:
4023         ext4_unregister_sysfs(sb);
4024 #endif
4025 failed_mount7:
4026         ext4_unregister_li_request(sb);
4027 failed_mount6:
4028         ext4_mb_release(sb);
4029         if (sbi->s_flex_groups)
4030                 kvfree(sbi->s_flex_groups);
4031         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4032         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4033         percpu_counter_destroy(&sbi->s_dirs_counter);
4034         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4035 failed_mount5:
4036         ext4_ext_release(sb);
4037         ext4_release_system_zone(sb);
4038 failed_mount4a:
4039         dput(sb->s_root);
4040         sb->s_root = NULL;
4041 failed_mount4:
4042         ext4_msg(sb, KERN_ERR, "mount failed");
4043         if (EXT4_SB(sb)->rsv_conversion_wq)
4044                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4045 failed_mount_wq:
4046         if (sbi->s_journal) {
4047                 jbd2_journal_destroy(sbi->s_journal);
4048                 sbi->s_journal = NULL;
4049         }
4050 failed_mount3a:
4051         ext4_es_unregister_shrinker(sbi);
4052 failed_mount3:
4053         del_timer_sync(&sbi->s_err_report);
4054         if (sbi->s_mmp_tsk)
4055                 kthread_stop(sbi->s_mmp_tsk);
4056 failed_mount2:
4057         for (i = 0; i < db_count; i++)
4058                 brelse(sbi->s_group_desc[i]);
4059         kvfree(sbi->s_group_desc);
4060 failed_mount:
4061         if (sbi->s_chksum_driver)
4062                 crypto_free_shash(sbi->s_chksum_driver);
4063 #ifdef CONFIG_QUOTA
4064         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4065                 kfree(sbi->s_qf_names[i]);
4066 #endif
4067         ext4_blkdev_remove(sbi);
4068         brelse(bh);
4069 out_fail:
4070         sb->s_fs_info = NULL;
4071         kfree(sbi->s_blockgroup_lock);
4072 out_free_base:
4073         kfree(sbi);
4074         kfree(orig_data);
4075         return err ? err : ret;
4076 }
4077
4078 /*
4079  * Setup any per-fs journal parameters now.  We'll do this both on
4080  * initial mount, once the journal has been initialised but before we've
4081  * done any recovery; and again on any subsequent remount.
4082  */
4083 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4084 {
4085         struct ext4_sb_info *sbi = EXT4_SB(sb);
4086
4087         journal->j_commit_interval = sbi->s_commit_interval;
4088         journal->j_min_batch_time = sbi->s_min_batch_time;
4089         journal->j_max_batch_time = sbi->s_max_batch_time;
4090
4091         write_lock(&journal->j_state_lock);
4092         if (test_opt(sb, BARRIER))
4093                 journal->j_flags |= JBD2_BARRIER;
4094         else
4095                 journal->j_flags &= ~JBD2_BARRIER;
4096         if (test_opt(sb, DATA_ERR_ABORT))
4097                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4098         else
4099                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4100         write_unlock(&journal->j_state_lock);
4101 }
4102
4103 static journal_t *ext4_get_journal(struct super_block *sb,
4104                                    unsigned int journal_inum)
4105 {
4106         struct inode *journal_inode;
4107         journal_t *journal;
4108
4109         BUG_ON(!ext4_has_feature_journal(sb));
4110
4111         /* First, test for the existence of a valid inode on disk.  Bad
4112          * things happen if we iget() an unused inode, as the subsequent
4113          * iput() will try to delete it. */
4114
4115         journal_inode = ext4_iget(sb, journal_inum);
4116         if (IS_ERR(journal_inode)) {
4117                 ext4_msg(sb, KERN_ERR, "no journal found");
4118                 return NULL;
4119         }
4120         if (!journal_inode->i_nlink) {
4121                 make_bad_inode(journal_inode);
4122                 iput(journal_inode);
4123                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4124                 return NULL;
4125         }
4126
4127         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4128                   journal_inode, journal_inode->i_size);
4129         if (!S_ISREG(journal_inode->i_mode)) {
4130                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4131                 iput(journal_inode);
4132                 return NULL;
4133         }
4134
4135         journal = jbd2_journal_init_inode(journal_inode);
4136         if (!journal) {
4137                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4138                 iput(journal_inode);
4139                 return NULL;
4140         }
4141         journal->j_private = sb;
4142         ext4_init_journal_params(sb, journal);
4143         return journal;
4144 }
4145
4146 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4147                                        dev_t j_dev)
4148 {
4149         struct buffer_head *bh;
4150         journal_t *journal;
4151         ext4_fsblk_t start;
4152         ext4_fsblk_t len;
4153         int hblock, blocksize;
4154         ext4_fsblk_t sb_block;
4155         unsigned long offset;
4156         struct ext4_super_block *es;
4157         struct block_device *bdev;
4158
4159         BUG_ON(!ext4_has_feature_journal(sb));
4160
4161         bdev = ext4_blkdev_get(j_dev, sb);
4162         if (bdev == NULL)
4163                 return NULL;
4164
4165         blocksize = sb->s_blocksize;
4166         hblock = bdev_logical_block_size(bdev);
4167         if (blocksize < hblock) {
4168                 ext4_msg(sb, KERN_ERR,
4169                         "blocksize too small for journal device");
4170                 goto out_bdev;
4171         }
4172
4173         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4174         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4175         set_blocksize(bdev, blocksize);
4176         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4177                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4178                        "external journal");
4179                 goto out_bdev;
4180         }
4181
4182         es = (struct ext4_super_block *) (bh->b_data + offset);
4183         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4184             !(le32_to_cpu(es->s_feature_incompat) &
4185               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4186                 ext4_msg(sb, KERN_ERR, "external journal has "
4187                                         "bad superblock");
4188                 brelse(bh);
4189                 goto out_bdev;
4190         }
4191
4192         if ((le32_to_cpu(es->s_feature_ro_compat) &
4193              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4194             es->s_checksum != ext4_superblock_csum(sb, es)) {
4195                 ext4_msg(sb, KERN_ERR, "external journal has "
4196                                        "corrupt superblock");
4197                 brelse(bh);
4198                 goto out_bdev;
4199         }
4200
4201         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4202                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4203                 brelse(bh);
4204                 goto out_bdev;
4205         }
4206
4207         len = ext4_blocks_count(es);
4208         start = sb_block + 1;
4209         brelse(bh);     /* we're done with the superblock */
4210
4211         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4212                                         start, len, blocksize);
4213         if (!journal) {
4214                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4215                 goto out_bdev;
4216         }
4217         journal->j_private = sb;
4218         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4219         wait_on_buffer(journal->j_sb_buffer);
4220         if (!buffer_uptodate(journal->j_sb_buffer)) {
4221                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4222                 goto out_journal;
4223         }
4224         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4225                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4226                                         "user (unsupported) - %d",
4227                         be32_to_cpu(journal->j_superblock->s_nr_users));
4228                 goto out_journal;
4229         }
4230         EXT4_SB(sb)->journal_bdev = bdev;
4231         ext4_init_journal_params(sb, journal);
4232         return journal;
4233
4234 out_journal:
4235         jbd2_journal_destroy(journal);
4236 out_bdev:
4237         ext4_blkdev_put(bdev);
4238         return NULL;
4239 }
4240
4241 static int ext4_load_journal(struct super_block *sb,
4242                              struct ext4_super_block *es,
4243                              unsigned long journal_devnum)
4244 {
4245         journal_t *journal;
4246         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4247         dev_t journal_dev;
4248         int err = 0;
4249         int really_read_only;
4250
4251         BUG_ON(!ext4_has_feature_journal(sb));
4252
4253         if (journal_devnum &&
4254             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4255                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4256                         "numbers have changed");
4257                 journal_dev = new_decode_dev(journal_devnum);
4258         } else
4259                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4260
4261         really_read_only = bdev_read_only(sb->s_bdev);
4262
4263         /*
4264          * Are we loading a blank journal or performing recovery after a
4265          * crash?  For recovery, we need to check in advance whether we
4266          * can get read-write access to the device.
4267          */
4268         if (ext4_has_feature_journal_needs_recovery(sb)) {
4269                 if (sb->s_flags & MS_RDONLY) {
4270                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4271                                         "required on readonly filesystem");
4272                         if (really_read_only) {
4273                                 ext4_msg(sb, KERN_ERR, "write access "
4274                                         "unavailable, cannot proceed");
4275                                 return -EROFS;
4276                         }
4277                         ext4_msg(sb, KERN_INFO, "write access will "
4278                                "be enabled during recovery");
4279                 }
4280         }
4281
4282         if (journal_inum && journal_dev) {
4283                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4284                        "and inode journals!");
4285                 return -EINVAL;
4286         }
4287
4288         if (journal_inum) {
4289                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4290                         return -EINVAL;
4291         } else {
4292                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4293                         return -EINVAL;
4294         }
4295
4296         if (!(journal->j_flags & JBD2_BARRIER))
4297                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4298
4299         if (!ext4_has_feature_journal_needs_recovery(sb))
4300                 err = jbd2_journal_wipe(journal, !really_read_only);
4301         if (!err) {
4302                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4303                 if (save)
4304                         memcpy(save, ((char *) es) +
4305                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4306                 err = jbd2_journal_load(journal);
4307                 if (save)
4308                         memcpy(((char *) es) + EXT4_S_ERR_START,
4309                                save, EXT4_S_ERR_LEN);
4310                 kfree(save);
4311         }
4312
4313         if (err) {
4314                 ext4_msg(sb, KERN_ERR, "error loading journal");
4315                 jbd2_journal_destroy(journal);
4316                 return err;
4317         }
4318
4319         EXT4_SB(sb)->s_journal = journal;
4320         ext4_clear_journal_err(sb, es);
4321
4322         if (!really_read_only && journal_devnum &&
4323             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4324                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4325
4326                 /* Make sure we flush the recovery flag to disk. */
4327                 ext4_commit_super(sb, 1);
4328         }
4329
4330         return 0;
4331 }
4332
4333 static int ext4_commit_super(struct super_block *sb, int sync)
4334 {
4335         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4336         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4337         int error = 0;
4338
4339         if (!sbh || block_device_ejected(sb))
4340                 return error;
4341         if (buffer_write_io_error(sbh)) {
4342                 /*
4343                  * Oh, dear.  A previous attempt to write the
4344                  * superblock failed.  This could happen because the
4345                  * USB device was yanked out.  Or it could happen to
4346                  * be a transient write error and maybe the block will
4347                  * be remapped.  Nothing we can do but to retry the
4348                  * write and hope for the best.
4349                  */
4350                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4351                        "superblock detected");
4352                 clear_buffer_write_io_error(sbh);
4353                 set_buffer_uptodate(sbh);
4354         }
4355         /*
4356          * If the file system is mounted read-only, don't update the
4357          * superblock write time.  This avoids updating the superblock
4358          * write time when we are mounting the root file system
4359          * read/only but we need to replay the journal; at that point,
4360          * for people who are east of GMT and who make their clock
4361          * tick in localtime for Windows bug-for-bug compatibility,
4362          * the clock is set in the future, and this will cause e2fsck
4363          * to complain and force a full file system check.
4364          */
4365         if (!(sb->s_flags & MS_RDONLY))
4366                 es->s_wtime = cpu_to_le32(get_seconds());
4367         if (sb->s_bdev->bd_part)
4368                 es->s_kbytes_written =
4369                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4370                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4371                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4372         else
4373                 es->s_kbytes_written =
4374                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4375         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4376                 ext4_free_blocks_count_set(es,
4377                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4378                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4379         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4380                 es->s_free_inodes_count =
4381                         cpu_to_le32(percpu_counter_sum_positive(
4382                                 &EXT4_SB(sb)->s_freeinodes_counter));
4383         BUFFER_TRACE(sbh, "marking dirty");
4384         ext4_superblock_csum_set(sb);
4385         mark_buffer_dirty(sbh);
4386         if (sync) {
4387                 error = __sync_dirty_buffer(sbh,
4388                         test_opt(sb, BARRIER) ? WRITE_FUA : WRITE_SYNC);
4389                 if (error)
4390                         return error;
4391
4392                 error = buffer_write_io_error(sbh);
4393                 if (error) {
4394                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4395                                "superblock");
4396                         clear_buffer_write_io_error(sbh);
4397                         set_buffer_uptodate(sbh);
4398                 }
4399         }
4400         return error;
4401 }
4402
4403 /*
4404  * Have we just finished recovery?  If so, and if we are mounting (or
4405  * remounting) the filesystem readonly, then we will end up with a
4406  * consistent fs on disk.  Record that fact.
4407  */
4408 static void ext4_mark_recovery_complete(struct super_block *sb,
4409                                         struct ext4_super_block *es)
4410 {
4411         journal_t *journal = EXT4_SB(sb)->s_journal;
4412
4413         if (!ext4_has_feature_journal(sb)) {
4414                 BUG_ON(journal != NULL);
4415                 return;
4416         }
4417         jbd2_journal_lock_updates(journal);
4418         if (jbd2_journal_flush(journal) < 0)
4419                 goto out;
4420
4421         if (ext4_has_feature_journal_needs_recovery(sb) &&
4422             sb->s_flags & MS_RDONLY) {
4423                 ext4_clear_feature_journal_needs_recovery(sb);
4424                 ext4_commit_super(sb, 1);
4425         }
4426
4427 out:
4428         jbd2_journal_unlock_updates(journal);
4429 }
4430
4431 /*
4432  * If we are mounting (or read-write remounting) a filesystem whose journal
4433  * has recorded an error from a previous lifetime, move that error to the
4434  * main filesystem now.
4435  */
4436 static void ext4_clear_journal_err(struct super_block *sb,
4437                                    struct ext4_super_block *es)
4438 {
4439         journal_t *journal;
4440         int j_errno;
4441         const char *errstr;
4442
4443         BUG_ON(!ext4_has_feature_journal(sb));
4444
4445         journal = EXT4_SB(sb)->s_journal;
4446
4447         /*
4448          * Now check for any error status which may have been recorded in the
4449          * journal by a prior ext4_error() or ext4_abort()
4450          */
4451
4452         j_errno = jbd2_journal_errno(journal);
4453         if (j_errno) {
4454                 char nbuf[16];
4455
4456                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4457                 ext4_warning(sb, "Filesystem error recorded "
4458                              "from previous mount: %s", errstr);
4459                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4460
4461                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4462                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4463                 ext4_commit_super(sb, 1);
4464
4465                 jbd2_journal_clear_err(journal);
4466                 jbd2_journal_update_sb_errno(journal);
4467         }
4468 }
4469
4470 /*
4471  * Force the running and committing transactions to commit,
4472  * and wait on the commit.
4473  */
4474 int ext4_force_commit(struct super_block *sb)
4475 {
4476         journal_t *journal;
4477
4478         if (sb->s_flags & MS_RDONLY)
4479                 return 0;
4480
4481         journal = EXT4_SB(sb)->s_journal;
4482         return ext4_journal_force_commit(journal);
4483 }
4484
4485 static int ext4_sync_fs(struct super_block *sb, int wait)
4486 {
4487         int ret = 0;
4488         tid_t target;
4489         bool needs_barrier = false;
4490         struct ext4_sb_info *sbi = EXT4_SB(sb);
4491
4492         trace_ext4_sync_fs(sb, wait);
4493         flush_workqueue(sbi->rsv_conversion_wq);
4494         /*
4495          * Writeback quota in non-journalled quota case - journalled quota has
4496          * no dirty dquots
4497          */
4498         dquot_writeback_dquots(sb, -1);
4499         /*
4500          * Data writeback is possible w/o journal transaction, so barrier must
4501          * being sent at the end of the function. But we can skip it if
4502          * transaction_commit will do it for us.
4503          */
4504         if (sbi->s_journal) {
4505                 target = jbd2_get_latest_transaction(sbi->s_journal);
4506                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4507                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4508                         needs_barrier = true;
4509
4510                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4511                         if (wait)
4512                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4513                                                            target);
4514                 }
4515         } else if (wait && test_opt(sb, BARRIER))
4516                 needs_barrier = true;
4517         if (needs_barrier) {
4518                 int err;
4519                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4520                 if (!ret)
4521                         ret = err;
4522         }
4523
4524         return ret;
4525 }
4526
4527 /*
4528  * LVM calls this function before a (read-only) snapshot is created.  This
4529  * gives us a chance to flush the journal completely and mark the fs clean.
4530  *
4531  * Note that only this function cannot bring a filesystem to be in a clean
4532  * state independently. It relies on upper layer to stop all data & metadata
4533  * modifications.
4534  */
4535 static int ext4_freeze(struct super_block *sb)
4536 {
4537         int error = 0;
4538         journal_t *journal;
4539
4540         if (sb->s_flags & MS_RDONLY)
4541                 return 0;
4542
4543         journal = EXT4_SB(sb)->s_journal;
4544
4545         if (journal) {
4546                 /* Now we set up the journal barrier. */
4547                 jbd2_journal_lock_updates(journal);
4548
4549                 /*
4550                  * Don't clear the needs_recovery flag if we failed to
4551                  * flush the journal.
4552                  */
4553                 error = jbd2_journal_flush(journal);
4554                 if (error < 0)
4555                         goto out;
4556
4557                 /* Journal blocked and flushed, clear needs_recovery flag. */
4558                 ext4_clear_feature_journal_needs_recovery(sb);
4559         }
4560
4561         error = ext4_commit_super(sb, 1);
4562 out:
4563         if (journal)
4564                 /* we rely on upper layer to stop further updates */
4565                 jbd2_journal_unlock_updates(journal);
4566         return error;
4567 }
4568
4569 /*
4570  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4571  * flag here, even though the filesystem is not technically dirty yet.
4572  */
4573 static int ext4_unfreeze(struct super_block *sb)
4574 {
4575         if (sb->s_flags & MS_RDONLY)
4576                 return 0;
4577
4578         if (EXT4_SB(sb)->s_journal) {
4579                 /* Reset the needs_recovery flag before the fs is unlocked. */
4580                 ext4_set_feature_journal_needs_recovery(sb);
4581         }
4582
4583         ext4_commit_super(sb, 1);
4584         return 0;
4585 }
4586
4587 /*
4588  * Structure to save mount options for ext4_remount's benefit
4589  */
4590 struct ext4_mount_options {
4591         unsigned long s_mount_opt;
4592         unsigned long s_mount_opt2;
4593         kuid_t s_resuid;
4594         kgid_t s_resgid;
4595         unsigned long s_commit_interval;
4596         u32 s_min_batch_time, s_max_batch_time;
4597 #ifdef CONFIG_QUOTA
4598         int s_jquota_fmt;
4599         char *s_qf_names[EXT4_MAXQUOTAS];
4600 #endif
4601 };
4602
4603 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4604 {
4605         struct ext4_super_block *es;
4606         struct ext4_sb_info *sbi = EXT4_SB(sb);
4607         unsigned long old_sb_flags;
4608         struct ext4_mount_options old_opts;
4609         int enable_quota = 0;
4610         ext4_group_t g;
4611         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4612         int err = 0;
4613 #ifdef CONFIG_QUOTA
4614         int i, j;
4615 #endif
4616         char *orig_data = kstrdup(data, GFP_KERNEL);
4617
4618         /* Store the original options */
4619         old_sb_flags = sb->s_flags;
4620         old_opts.s_mount_opt = sbi->s_mount_opt;
4621         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4622         old_opts.s_resuid = sbi->s_resuid;
4623         old_opts.s_resgid = sbi->s_resgid;
4624         old_opts.s_commit_interval = sbi->s_commit_interval;
4625         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4626         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4627 #ifdef CONFIG_QUOTA
4628         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4629         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4630                 if (sbi->s_qf_names[i]) {
4631                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4632                                                          GFP_KERNEL);
4633                         if (!old_opts.s_qf_names[i]) {
4634                                 for (j = 0; j < i; j++)
4635                                         kfree(old_opts.s_qf_names[j]);
4636                                 kfree(orig_data);
4637                                 return -ENOMEM;
4638                         }
4639                 } else
4640                         old_opts.s_qf_names[i] = NULL;
4641 #endif
4642         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4643                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4644
4645         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4646                 err = -EINVAL;
4647                 goto restore_opts;
4648         }
4649
4650         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4651             test_opt(sb, JOURNAL_CHECKSUM)) {
4652                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4653                          "during remount not supported; ignoring");
4654                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4655         }
4656
4657         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4658                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4659                         ext4_msg(sb, KERN_ERR, "can't mount with "
4660                                  "both data=journal and delalloc");
4661                         err = -EINVAL;
4662                         goto restore_opts;
4663                 }
4664                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4665                         ext4_msg(sb, KERN_ERR, "can't mount with "
4666                                  "both data=journal and dioread_nolock");
4667                         err = -EINVAL;
4668                         goto restore_opts;
4669                 }
4670                 if (test_opt(sb, DAX)) {
4671                         ext4_msg(sb, KERN_ERR, "can't mount with "
4672                                  "both data=journal and dax");
4673                         err = -EINVAL;
4674                         goto restore_opts;
4675                 }
4676         }
4677
4678         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4679                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4680                         "dax flag with busy inodes while remounting");
4681                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4682         }
4683
4684         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4685                 ext4_abort(sb, "Abort forced by user");
4686
4687         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4688                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4689
4690         es = sbi->s_es;
4691
4692         if (sbi->s_journal) {
4693                 ext4_init_journal_params(sb, sbi->s_journal);
4694                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4695         }
4696
4697         if (*flags & MS_LAZYTIME)
4698                 sb->s_flags |= MS_LAZYTIME;
4699
4700         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4701                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4702                         err = -EROFS;
4703                         goto restore_opts;
4704                 }
4705
4706                 if (*flags & MS_RDONLY) {
4707                         err = sync_filesystem(sb);
4708                         if (err < 0)
4709                                 goto restore_opts;
4710                         err = dquot_suspend(sb, -1);
4711                         if (err < 0)
4712                                 goto restore_opts;
4713
4714                         /*
4715                          * First of all, the unconditional stuff we have to do
4716                          * to disable replay of the journal when we next remount
4717                          */
4718                         sb->s_flags |= MS_RDONLY;
4719
4720                         /*
4721                          * OK, test if we are remounting a valid rw partition
4722                          * readonly, and if so set the rdonly flag and then
4723                          * mark the partition as valid again.
4724                          */
4725                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4726                             (sbi->s_mount_state & EXT4_VALID_FS))
4727                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4728
4729                         if (sbi->s_journal)
4730                                 ext4_mark_recovery_complete(sb, es);
4731                 } else {
4732                         /* Make sure we can mount this feature set readwrite */
4733                         if (ext4_has_feature_readonly(sb) ||
4734                             !ext4_feature_set_ok(sb, 0)) {
4735                                 err = -EROFS;
4736                                 goto restore_opts;
4737                         }
4738                         /*
4739                          * Make sure the group descriptor checksums
4740                          * are sane.  If they aren't, refuse to remount r/w.
4741                          */
4742                         for (g = 0; g < sbi->s_groups_count; g++) {
4743                                 struct ext4_group_desc *gdp =
4744                                         ext4_get_group_desc(sb, g, NULL);
4745
4746                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4747                                         ext4_msg(sb, KERN_ERR,
4748                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4749                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
4750                                                le16_to_cpu(gdp->bg_checksum));
4751                                         err = -EFSBADCRC;
4752                                         goto restore_opts;
4753                                 }
4754                         }
4755
4756                         /*
4757                          * If we have an unprocessed orphan list hanging
4758                          * around from a previously readonly bdev mount,
4759                          * require a full umount/remount for now.
4760                          */
4761                         if (es->s_last_orphan) {
4762                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4763                                        "remount RDWR because of unprocessed "
4764                                        "orphan inode list.  Please "
4765                                        "umount/remount instead");
4766                                 err = -EINVAL;
4767                                 goto restore_opts;
4768                         }
4769
4770                         /*
4771                          * Mounting a RDONLY partition read-write, so reread
4772                          * and store the current valid flag.  (It may have
4773                          * been changed by e2fsck since we originally mounted
4774                          * the partition.)
4775                          */
4776                         if (sbi->s_journal)
4777                                 ext4_clear_journal_err(sb, es);
4778                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4779                         if (!ext4_setup_super(sb, es, 0))
4780                                 sb->s_flags &= ~MS_RDONLY;
4781                         if (ext4_has_feature_mmp(sb))
4782                                 if (ext4_multi_mount_protect(sb,
4783                                                 le64_to_cpu(es->s_mmp_block))) {
4784                                         err = -EROFS;
4785                                         goto restore_opts;
4786                                 }
4787                         enable_quota = 1;
4788                 }
4789         }
4790
4791         /*
4792          * Reinitialize lazy itable initialization thread based on
4793          * current settings
4794          */
4795         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4796                 ext4_unregister_li_request(sb);
4797         else {
4798                 ext4_group_t first_not_zeroed;
4799                 first_not_zeroed = ext4_has_uninit_itable(sb);
4800                 ext4_register_li_request(sb, first_not_zeroed);
4801         }
4802
4803         ext4_setup_system_zone(sb);
4804         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4805                 ext4_commit_super(sb, 1);
4806
4807 #ifdef CONFIG_QUOTA
4808         /* Release old quota file names */
4809         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4810                 kfree(old_opts.s_qf_names[i]);
4811         if (enable_quota) {
4812                 if (sb_any_quota_suspended(sb))
4813                         dquot_resume(sb, -1);
4814                 else if (ext4_has_feature_quota(sb)) {
4815                         err = ext4_enable_quotas(sb);
4816                         if (err)
4817                                 goto restore_opts;
4818                 }
4819         }
4820 #endif
4821
4822         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
4823         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4824         kfree(orig_data);
4825         return 0;
4826
4827 restore_opts:
4828         sb->s_flags = old_sb_flags;
4829         sbi->s_mount_opt = old_opts.s_mount_opt;
4830         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4831         sbi->s_resuid = old_opts.s_resuid;
4832         sbi->s_resgid = old_opts.s_resgid;
4833         sbi->s_commit_interval = old_opts.s_commit_interval;
4834         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4835         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4836 #ifdef CONFIG_QUOTA
4837         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4838         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
4839                 kfree(sbi->s_qf_names[i]);
4840                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4841         }
4842 #endif
4843         kfree(orig_data);
4844         return err;
4845 }
4846
4847 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4848 {
4849         struct super_block *sb = dentry->d_sb;
4850         struct ext4_sb_info *sbi = EXT4_SB(sb);
4851         struct ext4_super_block *es = sbi->s_es;
4852         ext4_fsblk_t overhead = 0, resv_blocks;
4853         u64 fsid;
4854         s64 bfree;
4855         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4856
4857         if (!test_opt(sb, MINIX_DF))
4858                 overhead = sbi->s_overhead;
4859
4860         buf->f_type = EXT4_SUPER_MAGIC;
4861         buf->f_bsize = sb->s_blocksize;
4862         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4863         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4864                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4865         /* prevent underflow in case that few free space is available */
4866         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4867         buf->f_bavail = buf->f_bfree -
4868                         (ext4_r_blocks_count(es) + resv_blocks);
4869         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4870                 buf->f_bavail = 0;
4871         buf->f_files = le32_to_cpu(es->s_inodes_count);
4872         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4873         buf->f_namelen = EXT4_NAME_LEN;
4874         fsid = le64_to_cpup((void *)es->s_uuid) ^
4875                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4876         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4877         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4878
4879         return 0;
4880 }
4881
4882 /* Helper function for writing quotas on sync - we need to start transaction
4883  * before quota file is locked for write. Otherwise the are possible deadlocks:
4884  * Process 1                         Process 2
4885  * ext4_create()                     quota_sync()
4886  *   jbd2_journal_start()                  write_dquot()
4887  *   dquot_initialize()                         down(dqio_mutex)
4888  *     down(dqio_mutex)                    jbd2_journal_start()
4889  *
4890  */
4891
4892 #ifdef CONFIG_QUOTA
4893
4894 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4895 {
4896         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4897 }
4898
4899 static int ext4_write_dquot(struct dquot *dquot)
4900 {
4901         int ret, err;
4902         handle_t *handle;
4903         struct inode *inode;
4904
4905         inode = dquot_to_inode(dquot);
4906         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4907                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4908         if (IS_ERR(handle))
4909                 return PTR_ERR(handle);
4910         ret = dquot_commit(dquot);
4911         err = ext4_journal_stop(handle);
4912         if (!ret)
4913                 ret = err;
4914         return ret;
4915 }
4916
4917 static int ext4_acquire_dquot(struct dquot *dquot)
4918 {
4919         int ret, err;
4920         handle_t *handle;
4921
4922         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4923                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4924         if (IS_ERR(handle))
4925                 return PTR_ERR(handle);
4926         ret = dquot_acquire(dquot);
4927         err = ext4_journal_stop(handle);
4928         if (!ret)
4929                 ret = err;
4930         return ret;
4931 }
4932
4933 static int ext4_release_dquot(struct dquot *dquot)
4934 {
4935         int ret, err;
4936         handle_t *handle;
4937
4938         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4939                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4940         if (IS_ERR(handle)) {
4941                 /* Release dquot anyway to avoid endless cycle in dqput() */
4942                 dquot_release(dquot);
4943                 return PTR_ERR(handle);
4944         }
4945         ret = dquot_release(dquot);
4946         err = ext4_journal_stop(handle);
4947         if (!ret)
4948                 ret = err;
4949         return ret;
4950 }
4951
4952 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4953 {
4954         struct super_block *sb = dquot->dq_sb;
4955         struct ext4_sb_info *sbi = EXT4_SB(sb);
4956
4957         /* Are we journaling quotas? */
4958         if (ext4_has_feature_quota(sb) ||
4959             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
4960                 dquot_mark_dquot_dirty(dquot);
4961                 return ext4_write_dquot(dquot);
4962         } else {
4963                 return dquot_mark_dquot_dirty(dquot);
4964         }
4965 }
4966
4967 static int ext4_write_info(struct super_block *sb, int type)
4968 {
4969         int ret, err;
4970         handle_t *handle;
4971
4972         /* Data block + inode block */
4973         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
4974         if (IS_ERR(handle))
4975                 return PTR_ERR(handle);
4976         ret = dquot_commit_info(sb, type);
4977         err = ext4_journal_stop(handle);
4978         if (!ret)
4979                 ret = err;
4980         return ret;
4981 }
4982
4983 /*
4984  * Turn on quotas during mount time - we need to find
4985  * the quota file and such...
4986  */
4987 static int ext4_quota_on_mount(struct super_block *sb, int type)
4988 {
4989         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
4990                                         EXT4_SB(sb)->s_jquota_fmt, type);
4991 }
4992
4993 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
4994 {
4995         struct ext4_inode_info *ei = EXT4_I(inode);
4996
4997         /* The first argument of lockdep_set_subclass has to be
4998          * *exactly* the same as the argument to init_rwsem() --- in
4999          * this case, in init_once() --- or lockdep gets unhappy
5000          * because the name of the lock is set using the
5001          * stringification of the argument to init_rwsem().
5002          */
5003         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5004         lockdep_set_subclass(&ei->i_data_sem, subclass);
5005 }
5006
5007 /*
5008  * Standard function to be called on quota_on
5009  */
5010 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5011                          struct path *path)
5012 {
5013         int err;
5014
5015         if (!test_opt(sb, QUOTA))
5016                 return -EINVAL;
5017
5018         /* Quotafile not on the same filesystem? */
5019         if (path->dentry->d_sb != sb)
5020                 return -EXDEV;
5021         /* Journaling quota? */
5022         if (EXT4_SB(sb)->s_qf_names[type]) {
5023                 /* Quotafile not in fs root? */
5024                 if (path->dentry->d_parent != sb->s_root)
5025                         ext4_msg(sb, KERN_WARNING,
5026                                 "Quota file not on filesystem root. "
5027                                 "Journaled quota will not work");
5028         }
5029
5030         /*
5031          * When we journal data on quota file, we have to flush journal to see
5032          * all updates to the file when we bypass pagecache...
5033          */
5034         if (EXT4_SB(sb)->s_journal &&
5035             ext4_should_journal_data(d_inode(path->dentry))) {
5036                 /*
5037                  * We don't need to lock updates but journal_flush() could
5038                  * otherwise be livelocked...
5039                  */
5040                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5041                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5042                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5043                 if (err)
5044                         return err;
5045         }
5046         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5047         err = dquot_quota_on(sb, type, format_id, path);
5048         if (err)
5049                 lockdep_set_quota_inode(path->dentry->d_inode,
5050                                              I_DATA_SEM_NORMAL);
5051         return err;
5052 }
5053
5054 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5055                              unsigned int flags)
5056 {
5057         int err;
5058         struct inode *qf_inode;
5059         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5060                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5061                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5062         };
5063
5064         BUG_ON(!ext4_has_feature_quota(sb));
5065
5066         if (!qf_inums[type])
5067                 return -EPERM;
5068
5069         qf_inode = ext4_iget(sb, qf_inums[type]);
5070         if (IS_ERR(qf_inode)) {
5071                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5072                 return PTR_ERR(qf_inode);
5073         }
5074
5075         /* Don't account quota for quota files to avoid recursion */
5076         qf_inode->i_flags |= S_NOQUOTA;
5077         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5078         err = dquot_enable(qf_inode, type, format_id, flags);
5079         iput(qf_inode);
5080         if (err)
5081                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5082
5083         return err;
5084 }
5085
5086 /* Enable usage tracking for all quota types. */
5087 static int ext4_enable_quotas(struct super_block *sb)
5088 {
5089         int type, err = 0;
5090         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5091                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5092                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5093         };
5094
5095         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5096         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5097                 if (qf_inums[type]) {
5098                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5099                                                 DQUOT_USAGE_ENABLED);
5100                         if (err) {
5101                                 ext4_warning(sb,
5102                                         "Failed to enable quota tracking "
5103                                         "(type=%d, err=%d). Please run "
5104                                         "e2fsck to fix.", type, err);
5105                                 return err;
5106                         }
5107                 }
5108         }
5109         return 0;
5110 }
5111
5112 static int ext4_quota_off(struct super_block *sb, int type)
5113 {
5114         struct inode *inode = sb_dqopt(sb)->files[type];
5115         handle_t *handle;
5116
5117         /* Force all delayed allocation blocks to be allocated.
5118          * Caller already holds s_umount sem */
5119         if (test_opt(sb, DELALLOC))
5120                 sync_filesystem(sb);
5121
5122         if (!inode)
5123                 goto out;
5124
5125         /* Update modification times of quota files when userspace can
5126          * start looking at them */
5127         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5128         if (IS_ERR(handle))
5129                 goto out;
5130         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5131         ext4_mark_inode_dirty(handle, inode);
5132         ext4_journal_stop(handle);
5133
5134 out:
5135         return dquot_quota_off(sb, type);
5136 }
5137
5138 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5139  * acquiring the locks... As quota files are never truncated and quota code
5140  * itself serializes the operations (and no one else should touch the files)
5141  * we don't have to be afraid of races */
5142 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5143                                size_t len, loff_t off)
5144 {
5145         struct inode *inode = sb_dqopt(sb)->files[type];
5146         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5147         int offset = off & (sb->s_blocksize - 1);
5148         int tocopy;
5149         size_t toread;
5150         struct buffer_head *bh;
5151         loff_t i_size = i_size_read(inode);
5152
5153         if (off > i_size)
5154                 return 0;
5155         if (off+len > i_size)
5156                 len = i_size-off;
5157         toread = len;
5158         while (toread > 0) {
5159                 tocopy = sb->s_blocksize - offset < toread ?
5160                                 sb->s_blocksize - offset : toread;
5161                 bh = ext4_bread(NULL, inode, blk, 0);
5162                 if (IS_ERR(bh))
5163                         return PTR_ERR(bh);
5164                 if (!bh)        /* A hole? */
5165                         memset(data, 0, tocopy);
5166                 else
5167                         memcpy(data, bh->b_data+offset, tocopy);
5168                 brelse(bh);
5169                 offset = 0;
5170                 toread -= tocopy;
5171                 data += tocopy;
5172                 blk++;
5173         }
5174         return len;
5175 }
5176
5177 /* Write to quotafile (we know the transaction is already started and has
5178  * enough credits) */
5179 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5180                                 const char *data, size_t len, loff_t off)
5181 {
5182         struct inode *inode = sb_dqopt(sb)->files[type];
5183         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5184         int err, offset = off & (sb->s_blocksize - 1);
5185         int retries = 0;
5186         struct buffer_head *bh;
5187         handle_t *handle = journal_current_handle();
5188
5189         if (EXT4_SB(sb)->s_journal && !handle) {
5190                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5191                         " cancelled because transaction is not started",
5192                         (unsigned long long)off, (unsigned long long)len);
5193                 return -EIO;
5194         }
5195         /*
5196          * Since we account only one data block in transaction credits,
5197          * then it is impossible to cross a block boundary.
5198          */
5199         if (sb->s_blocksize - offset < len) {
5200                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5201                         " cancelled because not block aligned",
5202                         (unsigned long long)off, (unsigned long long)len);
5203                 return -EIO;
5204         }
5205
5206         do {
5207                 bh = ext4_bread(handle, inode, blk,
5208                                 EXT4_GET_BLOCKS_CREATE |
5209                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5210         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5211                  ext4_should_retry_alloc(inode->i_sb, &retries));
5212         if (IS_ERR(bh))
5213                 return PTR_ERR(bh);
5214         if (!bh)
5215                 goto out;
5216         BUFFER_TRACE(bh, "get write access");
5217         err = ext4_journal_get_write_access(handle, bh);
5218         if (err) {
5219                 brelse(bh);
5220                 return err;
5221         }
5222         lock_buffer(bh);
5223         memcpy(bh->b_data+offset, data, len);
5224         flush_dcache_page(bh->b_page);
5225         unlock_buffer(bh);
5226         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5227         brelse(bh);
5228 out:
5229         if (inode->i_size < off + len) {
5230                 i_size_write(inode, off + len);
5231                 EXT4_I(inode)->i_disksize = inode->i_size;
5232                 ext4_mark_inode_dirty(handle, inode);
5233         }
5234         return len;
5235 }
5236
5237 #endif
5238
5239 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5240                        const char *dev_name, void *data)
5241 {
5242         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5243 }
5244
5245 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5246 static inline void register_as_ext2(void)
5247 {
5248         int err = register_filesystem(&ext2_fs_type);
5249         if (err)
5250                 printk(KERN_WARNING
5251                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5252 }
5253
5254 static inline void unregister_as_ext2(void)
5255 {
5256         unregister_filesystem(&ext2_fs_type);
5257 }
5258
5259 static inline int ext2_feature_set_ok(struct super_block *sb)
5260 {
5261         if (ext4_has_unknown_ext2_incompat_features(sb))
5262                 return 0;
5263         if (sb->s_flags & MS_RDONLY)
5264                 return 1;
5265         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5266                 return 0;
5267         return 1;
5268 }
5269 #else
5270 static inline void register_as_ext2(void) { }
5271 static inline void unregister_as_ext2(void) { }
5272 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5273 #endif
5274
5275 static inline void register_as_ext3(void)
5276 {
5277         int err = register_filesystem(&ext3_fs_type);
5278         if (err)
5279                 printk(KERN_WARNING
5280                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5281 }
5282
5283 static inline void unregister_as_ext3(void)
5284 {
5285         unregister_filesystem(&ext3_fs_type);
5286 }
5287
5288 static inline int ext3_feature_set_ok(struct super_block *sb)
5289 {
5290         if (ext4_has_unknown_ext3_incompat_features(sb))
5291                 return 0;
5292         if (!ext4_has_feature_journal(sb))
5293                 return 0;
5294         if (sb->s_flags & MS_RDONLY)
5295                 return 1;
5296         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5297                 return 0;
5298         return 1;
5299 }
5300
5301 static struct file_system_type ext4_fs_type = {
5302         .owner          = THIS_MODULE,
5303         .name           = "ext4",
5304         .mount          = ext4_mount,
5305         .kill_sb        = kill_block_super,
5306         .fs_flags       = FS_REQUIRES_DEV,
5307 };
5308 MODULE_ALIAS_FS("ext4");
5309
5310 /* Shared across all ext4 file systems */
5311 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5312 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5313
5314 static int __init ext4_init_fs(void)
5315 {
5316         int i, err;
5317
5318         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5319         ext4_li_info = NULL;
5320         mutex_init(&ext4_li_mtx);
5321
5322         /* Build-time check for flags consistency */
5323         ext4_check_flag_values();
5324
5325         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5326                 mutex_init(&ext4__aio_mutex[i]);
5327                 init_waitqueue_head(&ext4__ioend_wq[i]);
5328         }
5329
5330         err = ext4_init_es();
5331         if (err)
5332                 return err;
5333
5334         err = ext4_init_pageio();
5335         if (err)
5336                 goto out5;
5337
5338         err = ext4_init_system_zone();
5339         if (err)
5340                 goto out4;
5341
5342         err = ext4_init_sysfs();
5343         if (err)
5344                 goto out3;
5345
5346         err = ext4_init_mballoc();
5347         if (err)
5348                 goto out2;
5349         else
5350                 ext4_mballoc_ready = 1;
5351         err = init_inodecache();
5352         if (err)
5353                 goto out1;
5354         register_as_ext3();
5355         register_as_ext2();
5356         err = register_filesystem(&ext4_fs_type);
5357         if (err)
5358                 goto out;
5359
5360         return 0;
5361 out:
5362         unregister_as_ext2();
5363         unregister_as_ext3();
5364         destroy_inodecache();
5365 out1:
5366         ext4_mballoc_ready = 0;
5367         ext4_exit_mballoc();
5368 out2:
5369         ext4_exit_sysfs();
5370 out3:
5371         ext4_exit_system_zone();
5372 out4:
5373         ext4_exit_pageio();
5374 out5:
5375         ext4_exit_es();
5376
5377         return err;
5378 }
5379
5380 static void __exit ext4_exit_fs(void)
5381 {
5382         ext4_exit_crypto();
5383         ext4_destroy_lazyinit_thread();
5384         unregister_as_ext2();
5385         unregister_as_ext3();
5386         unregister_filesystem(&ext4_fs_type);
5387         destroy_inodecache();
5388         ext4_exit_mballoc();
5389         ext4_exit_sysfs();
5390         ext4_exit_system_zone();
5391         ext4_exit_pageio();
5392         ext4_exit_es();
5393 }
5394
5395 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5396 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5397 MODULE_LICENSE("GPL");
5398 module_init(ext4_init_fs)
5399 module_exit(ext4_exit_fs)