fs/ext4/super.c

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