fs/ext4/super.c

   1 /*
   2  *  linux/fs/ext4/super.c
   3  *
   4  * Copyright (C) 1992, 1993, 1994, 1995
   5  * Remy Card (card@masi.ibp.fr)
   6  * Laboratoire MASI - Institut Blaise Pascal
   7  * Universite Pierre et Marie Curie (Paris VI)
   8  *
   9  *  from
  10  *
  11  *  linux/fs/minix/inode.c
  12  *
  13  *  Copyright (C) 1991, 1992  Linus Torvalds
  14  *
  15  *  Big-endian to little-endian byte-swapping/bitmaps by
  16  *        David S. Miller (davem@caip.rutgers.edu), 1995
  17  */
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/fs.h>
  22 #include <linux/time.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/slab.h>
  25 #include <linux/init.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/parser.h>
  29 #include <linux/buffer_head.h>
  30 #include <linux/exportfs.h>
  31 #include <linux/vfs.h>
  32 #include <linux/random.h>
  33 #include <linux/mount.h>
  34 #include <linux/namei.h>
  35 #include <linux/quotaops.h>
  36 #include <linux/seq_file.h>
  37 #include <linux/ctype.h>
  38 #include <linux/log2.h>
  39 #include <linux/crc16.h>
  40 #include <linux/cleancache.h>
  41 #include <asm/uaccess.h>
  42
  43 #include <linux/kthread.h>
  44 #include <linux/freezer.h>
  45
  46 #include "ext4.h"
  47 #include "ext4_extents.h"       /* Needed for trace points definition */
  48 #include "ext4_jbd2.h"
  49 #include "xattr.h"
  50 #include "acl.h"
  51 #include "mballoc.h"
  52
  53 #define CREATE_TRACE_POINTS
  54 #include <trace/events/ext4.h>
  55
  56 static struct ext4_lazy_init *ext4_li_info;
  57 static struct mutex ext4_li_mtx;
  58 static int ext4_mballoc_ready;
  59 static struct ratelimit_state ext4_mount_msg_ratelimit;
  60
  61 static int ext4_load_journal(struct super_block *, struct ext4_super_block *,
  62                              unsigned long journal_devnum);
  63 static int ext4_show_options(struct seq_file *seq, struct dentry *root);
  64 static int ext4_commit_super(struct super_block *sb, int sync);
  65 static void ext4_mark_recovery_complete(struct super_block *sb,
  66                                         struct ext4_super_block *es);
  67 static void ext4_clear_journal_err(struct super_block *sb,
  68                                    struct ext4_super_block *es);
  69 static int ext4_sync_fs(struct super_block *sb, int wait);
  70 static int ext4_remount(struct super_block *sb, int *flags, char *data);
  71 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf);
  72 static int ext4_unfreeze(struct super_block *sb);
  73 static int ext4_freeze(struct super_block *sb);
  74 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
  75                        const char *dev_name, void *data);
  76 static inline int ext2_feature_set_ok(struct super_block *sb);
  77 static inline int ext3_feature_set_ok(struct super_block *sb);
  78 static int ext4_feature_set_ok(struct super_block *sb, int readonly);
  79 static void ext4_destroy_lazyinit_thread(void);
  80 static void ext4_unregister_li_request(struct super_block *sb);
  81 static void ext4_clear_request_list(void);
  82
  83 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
  84 static struct file_system_type ext2_fs_type = {
  85         .owner          = THIS_MODULE,
  86         .name           = "ext2",
  87         .mount          = ext4_mount,
  88         .kill_sb        = kill_block_super,
  89         .fs_flags       = FS_REQUIRES_DEV,
  90 };
  91 MODULE_ALIAS_FS("ext2");
  92 MODULE_ALIAS("ext2");
  93 #define IS_EXT2_SB(sb) ((sb)->s_bdev->bd_holder == &ext2_fs_type)
  94 #else
  95 #define IS_EXT2_SB(sb) (0)
  96 #endif
  97
  98
  99 static struct file_system_type ext3_fs_type = {
 100         .owner          = THIS_MODULE,
 101         .name           = "ext3",
 102         .mount          = ext4_mount,
 103         .kill_sb        = kill_block_super,
 104         .fs_flags       = FS_REQUIRES_DEV,
 105 };
 106 MODULE_ALIAS_FS("ext3");
 107 MODULE_ALIAS("ext3");
 108 #define IS_EXT3_SB(sb) ((sb)->s_bdev->bd_holder == &ext3_fs_type)
 109
 110 static int ext4_verify_csum_type(struct super_block *sb,
 111                                  struct ext4_super_block *es)
 112 {
 113         if (!ext4_has_feature_metadata_csum(sb))
 114                 return 1;
 115
 116         return es->s_checksum_type == EXT4_CRC32C_CHKSUM;
 117 }
 118
 119 static __le32 ext4_superblock_csum(struct super_block *sb,
 120                                    struct ext4_super_block *es)
 121 {
 122         struct ext4_sb_info *sbi = EXT4_SB(sb);
 123         int offset = offsetof(struct ext4_super_block, s_checksum);
 124         __u32 csum;
 125
 126         csum = ext4_chksum(sbi, ~0, (char *)es, offset);
 127
 128         return cpu_to_le32(csum);
 129 }
 130
 131 static int ext4_superblock_csum_verify(struct super_block *sb,
 132                                        struct ext4_super_block *es)
 133 {
 134         if (!ext4_has_metadata_csum(sb))
 135                 return 1;
 136
 137         return es->s_checksum == ext4_superblock_csum(sb, es);
 138 }
 139
 140 void ext4_superblock_csum_set(struct super_block *sb)
 141 {
 142         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 143
 144         if (!ext4_has_metadata_csum(sb))
 145                 return;
 146
 147         es->s_checksum = ext4_superblock_csum(sb, es);
 148 }
 149
 150 void *ext4_kvmalloc(size_t size, gfp_t flags)
 151 {
 152         void *ret;
 153
 154         ret = kmalloc(size, flags | __GFP_NOWARN);
 155         if (!ret)
 156                 ret = __vmalloc(size, flags, PAGE_KERNEL);
 157         return ret;
 158 }
 159
 160 void *ext4_kvzalloc(size_t size, gfp_t flags)
 161 {
 162         void *ret;
 163
 164         ret = kzalloc(size, flags | __GFP_NOWARN);
 165         if (!ret)
 166                 ret = __vmalloc(size, flags | __GFP_ZERO, PAGE_KERNEL);
 167         return ret;
 168 }
 169
 170 ext4_fsblk_t ext4_block_bitmap(struct super_block *sb,
 171                                struct ext4_group_desc *bg)
 172 {
 173         return le32_to_cpu(bg->bg_block_bitmap_lo) |
 174                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 175                  (ext4_fsblk_t)le32_to_cpu(bg->bg_block_bitmap_hi) << 32 : 0);
 176 }
 177
 178 ext4_fsblk_t ext4_inode_bitmap(struct super_block *sb,
 179                                struct ext4_group_desc *bg)
 180 {
 181         return le32_to_cpu(bg->bg_inode_bitmap_lo) |
 182                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 183                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_bitmap_hi) << 32 : 0);
 184 }
 185
 186 ext4_fsblk_t ext4_inode_table(struct super_block *sb,
 187                               struct ext4_group_desc *bg)
 188 {
 189         return le32_to_cpu(bg->bg_inode_table_lo) |
 190                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 191                  (ext4_fsblk_t)le32_to_cpu(bg->bg_inode_table_hi) << 32 : 0);
 192 }
 193
 194 __u32 ext4_free_group_clusters(struct super_block *sb,
 195                                struct ext4_group_desc *bg)
 196 {
 197         return le16_to_cpu(bg->bg_free_blocks_count_lo) |
 198                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 199                  (__u32)le16_to_cpu(bg->bg_free_blocks_count_hi) << 16 : 0);
 200 }
 201
 202 __u32 ext4_free_inodes_count(struct super_block *sb,
 203                               struct ext4_group_desc *bg)
 204 {
 205         return le16_to_cpu(bg->bg_free_inodes_count_lo) |
 206                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 207                  (__u32)le16_to_cpu(bg->bg_free_inodes_count_hi) << 16 : 0);
 208 }
 209
 210 __u32 ext4_used_dirs_count(struct super_block *sb,
 211                               struct ext4_group_desc *bg)
 212 {
 213         return le16_to_cpu(bg->bg_used_dirs_count_lo) |
 214                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 215                  (__u32)le16_to_cpu(bg->bg_used_dirs_count_hi) << 16 : 0);
 216 }
 217
 218 __u32 ext4_itable_unused_count(struct super_block *sb,
 219                               struct ext4_group_desc *bg)
 220 {
 221         return le16_to_cpu(bg->bg_itable_unused_lo) |
 222                 (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT ?
 223                  (__u32)le16_to_cpu(bg->bg_itable_unused_hi) << 16 : 0);
 224 }
 225
 226 void ext4_block_bitmap_set(struct super_block *sb,
 227                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 228 {
 229         bg->bg_block_bitmap_lo = cpu_to_le32((u32)blk);
 230         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 231                 bg->bg_block_bitmap_hi = cpu_to_le32(blk >> 32);
 232 }
 233
 234 void ext4_inode_bitmap_set(struct super_block *sb,
 235                            struct ext4_group_desc *bg, ext4_fsblk_t blk)
 236 {
 237         bg->bg_inode_bitmap_lo  = cpu_to_le32((u32)blk);
 238         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 239                 bg->bg_inode_bitmap_hi = cpu_to_le32(blk >> 32);
 240 }
 241
 242 void ext4_inode_table_set(struct super_block *sb,
 243                           struct ext4_group_desc *bg, ext4_fsblk_t blk)
 244 {
 245         bg->bg_inode_table_lo = cpu_to_le32((u32)blk);
 246         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 247                 bg->bg_inode_table_hi = cpu_to_le32(blk >> 32);
 248 }
 249
 250 void ext4_free_group_clusters_set(struct super_block *sb,
 251                                   struct ext4_group_desc *bg, __u32 count)
 252 {
 253         bg->bg_free_blocks_count_lo = cpu_to_le16((__u16)count);
 254         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 255                 bg->bg_free_blocks_count_hi = cpu_to_le16(count >> 16);
 256 }
 257
 258 void ext4_free_inodes_set(struct super_block *sb,
 259                           struct ext4_group_desc *bg, __u32 count)
 260 {
 261         bg->bg_free_inodes_count_lo = cpu_to_le16((__u16)count);
 262         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 263                 bg->bg_free_inodes_count_hi = cpu_to_le16(count >> 16);
 264 }
 265
 266 void ext4_used_dirs_set(struct super_block *sb,
 267                           struct ext4_group_desc *bg, __u32 count)
 268 {
 269         bg->bg_used_dirs_count_lo = cpu_to_le16((__u16)count);
 270         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 271                 bg->bg_used_dirs_count_hi = cpu_to_le16(count >> 16);
 272 }
 273
 274 void ext4_itable_unused_set(struct super_block *sb,
 275                           struct ext4_group_desc *bg, __u32 count)
 276 {
 277         bg->bg_itable_unused_lo = cpu_to_le16((__u16)count);
 278         if (EXT4_DESC_SIZE(sb) >= EXT4_MIN_DESC_SIZE_64BIT)
 279                 bg->bg_itable_unused_hi = cpu_to_le16(count >> 16);
 280 }
 281
 282
 283 static void __save_error_info(struct super_block *sb, const char *func,
 284                             unsigned int line)
 285 {
 286         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 287
 288         EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
 289         if (bdev_read_only(sb->s_bdev))
 290                 return;
 291         es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
 292         es->s_last_error_time = cpu_to_le32(get_seconds());
 293         strncpy(es->s_last_error_func, func, sizeof(es->s_last_error_func));
 294         es->s_last_error_line = cpu_to_le32(line);
 295         if (!es->s_first_error_time) {
 296                 es->s_first_error_time = es->s_last_error_time;
 297                 strncpy(es->s_first_error_func, func,
 298                         sizeof(es->s_first_error_func));
 299                 es->s_first_error_line = cpu_to_le32(line);
 300                 es->s_first_error_ino = es->s_last_error_ino;
 301                 es->s_first_error_block = es->s_last_error_block;
 302         }
 303         /*
 304          * Start the daily error reporting function if it hasn't been
 305          * started already
 306          */
 307         if (!es->s_error_count)
 308                 mod_timer(&EXT4_SB(sb)->s_err_report, jiffies + 24*60*60*HZ);
 309         le32_add_cpu(&es->s_error_count, 1);
 310 }
 311
 312 static void save_error_info(struct super_block *sb, const char *func,
 313                             unsigned int line)
 314 {
 315         __save_error_info(sb, func, line);
 316         ext4_commit_super(sb, 1);
 317 }
 318
 319 /*
 320  * The del_gendisk() function uninitializes the disk-specific data
 321  * structures, including the bdi structure, without telling anyone
 322  * else.  Once this happens, any attempt to call mark_buffer_dirty()
 323  * (for example, by ext4_commit_super), will cause a kernel OOPS.
 324  * This is a kludge to prevent these oops until we can put in a proper
 325  * hook in del_gendisk() to inform the VFS and file system layers.
 326  */
 327 static int block_device_ejected(struct super_block *sb)
 328 {
 329         struct inode *bd_inode = sb->s_bdev->bd_inode;
 330         struct backing_dev_info *bdi = inode_to_bdi(bd_inode);
 331
 332         return bdi->dev == NULL;
 333 }
 334
 335 static void ext4_journal_commit_callback(journal_t *journal, transaction_t *txn)
 336 {
 337         struct super_block              *sb = journal->j_private;
 338         struct ext4_sb_info             *sbi = EXT4_SB(sb);
 339         int                             error = is_journal_aborted(journal);
 340         struct ext4_journal_cb_entry    *jce;
 341
 342         BUG_ON(txn->t_state == T_FINISHED);
 343         spin_lock(&sbi->s_md_lock);
 344         while (!list_empty(&txn->t_private_list)) {
 345                 jce = list_entry(txn->t_private_list.next,
 346                                  struct ext4_journal_cb_entry, jce_list);
 347                 list_del_init(&jce->jce_list);
 348                 spin_unlock(&sbi->s_md_lock);
 349                 jce->jce_func(sb, jce, error);
 350                 spin_lock(&sbi->s_md_lock);
 351         }
 352         spin_unlock(&sbi->s_md_lock);
 353 }
 354
 355 /* Deal with the reporting of failure conditions on a filesystem such as
 356  * inconsistencies detected or read IO failures.
 357  *
 358  * On ext2, we can store the error state of the filesystem in the
 359  * superblock.  That is not possible on ext4, because we may have other
 360  * write ordering constraints on the superblock which prevent us from
 361  * writing it out straight away; and given that the journal is about to
 362  * be aborted, we can't rely on the current, or future, transactions to
 363  * write out the superblock safely.
 364  *
 365  * We'll just use the jbd2_journal_abort() error code to record an error in
 366  * the journal instead.  On recovery, the journal will complain about
 367  * that error until we've noted it down and cleared it.
 368  */
 369
 370 static void ext4_handle_error(struct super_block *sb)
 371 {
 372         if (sb->s_flags & MS_RDONLY)
 373                 return;
 374
 375         if (!test_opt(sb, ERRORS_CONT)) {
 376                 journal_t *journal = EXT4_SB(sb)->s_journal;
 377
 378                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 379                 if (journal)
 380                         jbd2_journal_abort(journal, -EIO);
 381         }
 382         if (test_opt(sb, ERRORS_RO)) {
 383                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 384                 /*
 385                  * Make sure updated value of ->s_mount_flags will be visible
 386                  * before ->s_flags update
 387                  */
 388                 smp_wmb();
 389                 sb->s_flags |= MS_RDONLY;
 390         }
 391         if (test_opt(sb, ERRORS_PANIC)) {
 392                 if (EXT4_SB(sb)->s_journal &&
 393                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 394                         return;
 395                 panic("EXT4-fs (device %s): panic forced after error\n",
 396                         sb->s_id);
 397         }
 398 }
 399
 400 #define ext4_error_ratelimit(sb)                                        \
 401                 ___ratelimit(&(EXT4_SB(sb)->s_err_ratelimit_state),     \
 402                              "EXT4-fs error")
 403
 404 void __ext4_error(struct super_block *sb, const char *function,
 405                   unsigned int line, const char *fmt, ...)
 406 {
 407         struct va_format vaf;
 408         va_list args;
 409
 410         if (ext4_error_ratelimit(sb)) {
 411                 va_start(args, fmt);
 412                 vaf.fmt = fmt;
 413                 vaf.va = &args;
 414                 printk(KERN_CRIT
 415                        "EXT4-fs error (device %s): %s:%d: comm %s: %pV\n",
 416                        sb->s_id, function, line, current->comm, &vaf);
 417                 va_end(args);
 418         }
 419         save_error_info(sb, function, line);
 420         ext4_handle_error(sb);
 421 }
 422
 423 void __ext4_error_inode(struct inode *inode, const char *function,
 424                         unsigned int line, ext4_fsblk_t block,
 425                         const char *fmt, ...)
 426 {
 427         va_list args;
 428         struct va_format vaf;
 429         struct ext4_super_block *es = EXT4_SB(inode->i_sb)->s_es;
 430
 431         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 432         es->s_last_error_block = cpu_to_le64(block);
 433         if (ext4_error_ratelimit(inode->i_sb)) {
 434                 va_start(args, fmt);
 435                 vaf.fmt = fmt;
 436                 vaf.va = &args;
 437                 if (block)
 438                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 439                                "inode #%lu: block %llu: comm %s: %pV\n",
 440                                inode->i_sb->s_id, function, line, inode->i_ino,
 441                                block, current->comm, &vaf);
 442                 else
 443                         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: "
 444                                "inode #%lu: comm %s: %pV\n",
 445                                inode->i_sb->s_id, function, line, inode->i_ino,
 446                                current->comm, &vaf);
 447                 va_end(args);
 448         }
 449         save_error_info(inode->i_sb, function, line);
 450         ext4_handle_error(inode->i_sb);
 451 }
 452
 453 void __ext4_error_file(struct file *file, const char *function,
 454                        unsigned int line, ext4_fsblk_t block,
 455                        const char *fmt, ...)
 456 {
 457         va_list args;
 458         struct va_format vaf;
 459         struct ext4_super_block *es;
 460         struct inode *inode = file_inode(file);
 461         char pathname[80], *path;
 462
 463         es = EXT4_SB(inode->i_sb)->s_es;
 464         es->s_last_error_ino = cpu_to_le32(inode->i_ino);
 465         if (ext4_error_ratelimit(inode->i_sb)) {
 466                 path = file_path(file, pathname, sizeof(pathname));
 467                 if (IS_ERR(path))
 468                         path = "(unknown)";
 469                 va_start(args, fmt);
 470                 vaf.fmt = fmt;
 471                 vaf.va = &args;
 472                 if (block)
 473                         printk(KERN_CRIT
 474                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 475                                "block %llu: comm %s: path %s: %pV\n",
 476                                inode->i_sb->s_id, function, line, inode->i_ino,
 477                                block, current->comm, path, &vaf);
 478                 else
 479                         printk(KERN_CRIT
 480                                "EXT4-fs error (device %s): %s:%d: inode #%lu: "
 481                                "comm %s: path %s: %pV\n",
 482                                inode->i_sb->s_id, function, line, inode->i_ino,
 483                                current->comm, path, &vaf);
 484                 va_end(args);
 485         }
 486         save_error_info(inode->i_sb, function, line);
 487         ext4_handle_error(inode->i_sb);
 488 }
 489
 490 const char *ext4_decode_error(struct super_block *sb, int errno,
 491                               char nbuf[16])
 492 {
 493         char *errstr = NULL;
 494
 495         switch (errno) {
 496         case -EFSCORRUPTED:
 497                 errstr = "Corrupt filesystem";
 498                 break;
 499         case -EFSBADCRC:
 500                 errstr = "Filesystem failed CRC";
 501                 break;
 502         case -EIO:
 503                 errstr = "IO failure";
 504                 break;
 505         case -ENOMEM:
 506                 errstr = "Out of memory";
 507                 break;
 508         case -EROFS:
 509                 if (!sb || (EXT4_SB(sb)->s_journal &&
 510                             EXT4_SB(sb)->s_journal->j_flags & JBD2_ABORT))
 511                         errstr = "Journal has aborted";
 512                 else
 513                         errstr = "Readonly filesystem";
 514                 break;
 515         default:
 516                 /* If the caller passed in an extra buffer for unknown
 517                  * errors, textualise them now.  Else we just return
 518                  * NULL. */
 519                 if (nbuf) {
 520                         /* Check for truncated error codes... */
 521                         if (snprintf(nbuf, 16, "error %d", -errno) >= 0)
 522                                 errstr = nbuf;
 523                 }
 524                 break;
 525         }
 526
 527         return errstr;
 528 }
 529
 530 /* __ext4_std_error decodes expected errors from journaling functions
 531  * automatically and invokes the appropriate error response.  */
 532
 533 void __ext4_std_error(struct super_block *sb, const char *function,
 534                       unsigned int line, int errno)
 535 {
 536         char nbuf[16];
 537         const char *errstr;
 538
 539         /* Special case: if the error is EROFS, and we're not already
 540          * inside a transaction, then there's really no point in logging
 541          * an error. */
 542         if (errno == -EROFS && journal_current_handle() == NULL &&
 543             (sb->s_flags & MS_RDONLY))
 544                 return;
 545
 546         if (ext4_error_ratelimit(sb)) {
 547                 errstr = ext4_decode_error(sb, errno, nbuf);
 548                 printk(KERN_CRIT "EXT4-fs error (device %s) in %s:%d: %s\n",
 549                        sb->s_id, function, line, errstr);
 550         }
 551
 552         save_error_info(sb, function, line);
 553         ext4_handle_error(sb);
 554 }
 555
 556 /*
 557  * ext4_abort is a much stronger failure handler than ext4_error.  The
 558  * abort function may be used to deal with unrecoverable failures such
 559  * as journal IO errors or ENOMEM at a critical moment in log management.
 560  *
 561  * We unconditionally force the filesystem into an ABORT|READONLY state,
 562  * unless the error response on the fs has been set to panic in which
 563  * case we take the easy way out and panic immediately.
 564  */
 565
 566 void __ext4_abort(struct super_block *sb, const char *function,
 567                 unsigned int line, const char *fmt, ...)
 568 {
 569         va_list args;
 570
 571         save_error_info(sb, function, line);
 572         va_start(args, fmt);
 573         printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: ", sb->s_id,
 574                function, line);
 575         vprintk(fmt, args);
 576         printk("\n");
 577         va_end(args);
 578
 579         if ((sb->s_flags & MS_RDONLY) == 0) {
 580                 ext4_msg(sb, KERN_CRIT, "Remounting filesystem read-only");
 581                 EXT4_SB(sb)->s_mount_flags |= EXT4_MF_FS_ABORTED;
 582                 /*
 583                  * Make sure updated value of ->s_mount_flags will be visible
 584                  * before ->s_flags update
 585                  */
 586                 smp_wmb();
 587                 sb->s_flags |= MS_RDONLY;
 588                 if (EXT4_SB(sb)->s_journal)
 589                         jbd2_journal_abort(EXT4_SB(sb)->s_journal, -EIO);
 590                 save_error_info(sb, function, line);
 591         }
 592         if (test_opt(sb, ERRORS_PANIC)) {
 593                 if (EXT4_SB(sb)->s_journal &&
 594                   !(EXT4_SB(sb)->s_journal->j_flags & JBD2_REC_ERR))
 595                         return;
 596                 panic("EXT4-fs panic from previous error\n");
 597         }
 598 }
 599
 600 void __ext4_msg(struct super_block *sb,
 601                 const char *prefix, const char *fmt, ...)
 602 {
 603         struct va_format vaf;
 604         va_list args;
 605
 606         if (!___ratelimit(&(EXT4_SB(sb)->s_msg_ratelimit_state), "EXT4-fs"))
 607                 return;
 608
 609         va_start(args, fmt);
 610         vaf.fmt = fmt;
 611         vaf.va = &args;
 612         printk("%sEXT4-fs (%s): %pV\n", prefix, sb->s_id, &vaf);
 613         va_end(args);
 614 }
 615
 616 #define ext4_warning_ratelimit(sb)                                      \
 617                 ___ratelimit(&(EXT4_SB(sb)->s_warning_ratelimit_state), \
 618                              "EXT4-fs warning")
 619
 620 void __ext4_warning(struct super_block *sb, const char *function,
 621                     unsigned int line, const char *fmt, ...)
 622 {
 623         struct va_format vaf;
 624         va_list args;
 625
 626         if (!ext4_warning_ratelimit(sb))
 627                 return;
 628
 629         va_start(args, fmt);
 630         vaf.fmt = fmt;
 631         vaf.va = &args;
 632         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: %pV\n",
 633                sb->s_id, function, line, &vaf);
 634         va_end(args);
 635 }
 636
 637 void __ext4_warning_inode(const struct inode *inode, const char *function,
 638                           unsigned int line, const char *fmt, ...)
 639 {
 640         struct va_format vaf;
 641         va_list args;
 642
 643         if (!ext4_warning_ratelimit(inode->i_sb))
 644                 return;
 645
 646         va_start(args, fmt);
 647         vaf.fmt = fmt;
 648         vaf.va = &args;
 649         printk(KERN_WARNING "EXT4-fs warning (device %s): %s:%d: "
 650                "inode #%lu: comm %s: %pV\n", inode->i_sb->s_id,
 651                function, line, inode->i_ino, current->comm, &vaf);
 652         va_end(args);
 653 }
 654
 655 void __ext4_grp_locked_error(const char *function, unsigned int line,
 656                              struct super_block *sb, ext4_group_t grp,
 657                              unsigned long ino, ext4_fsblk_t block,
 658                              const char *fmt, ...)
 659 __releases(bitlock)
 660 __acquires(bitlock)
 661 {
 662         struct va_format vaf;
 663         va_list args;
 664         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 665
 666         es->s_last_error_ino = cpu_to_le32(ino);
 667         es->s_last_error_block = cpu_to_le64(block);
 668         __save_error_info(sb, function, line);
 669
 670         if (ext4_error_ratelimit(sb)) {
 671                 va_start(args, fmt);
 672                 vaf.fmt = fmt;
 673                 vaf.va = &args;
 674                 printk(KERN_CRIT "EXT4-fs error (device %s): %s:%d: group %u, ",
 675                        sb->s_id, function, line, grp);
 676                 if (ino)
 677                         printk(KERN_CONT "inode %lu: ", ino);
 678                 if (block)
 679                         printk(KERN_CONT "block %llu:",
 680                                (unsigned long long) block);
 681                 printk(KERN_CONT "%pV\n", &vaf);
 682                 va_end(args);
 683         }
 684
 685         if (test_opt(sb, ERRORS_CONT)) {
 686                 ext4_commit_super(sb, 0);
 687                 return;
 688         }
 689
 690         ext4_unlock_group(sb, grp);
 691         ext4_handle_error(sb);
 692         /*
 693          * We only get here in the ERRORS_RO case; relocking the group
 694          * may be dangerous, but nothing bad will happen since the
 695          * filesystem will have already been marked read/only and the
 696          * journal has been aborted.  We return 1 as a hint to callers
 697          * who might what to use the return value from
 698          * ext4_grp_locked_error() to distinguish between the
 699          * ERRORS_CONT and ERRORS_RO case, and perhaps return more
 700          * aggressively from the ext4 function in question, with a
 701          * more appropriate error code.
 702          */
 703         ext4_lock_group(sb, grp);
 704         return;
 705 }
 706
 707 void ext4_update_dynamic_rev(struct super_block *sb)
 708 {
 709         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
 710
 711         if (le32_to_cpu(es->s_rev_level) > EXT4_GOOD_OLD_REV)
 712                 return;
 713
 714         ext4_warning(sb,
 715                      "updating to rev %d because of new feature flag, "
 716                      "running e2fsck is recommended",
 717                      EXT4_DYNAMIC_REV);
 718
 719         es->s_first_ino = cpu_to_le32(EXT4_GOOD_OLD_FIRST_INO);
 720         es->s_inode_size = cpu_to_le16(EXT4_GOOD_OLD_INODE_SIZE);
 721         es->s_rev_level = cpu_to_le32(EXT4_DYNAMIC_REV);
 722         /* leave es->s_feature_*compat flags alone */
 723         /* es->s_uuid will be set by e2fsck if empty */
 724
 725         /*
 726          * The rest of the superblock fields should be zero, and if not it
 727          * means they are likely already in use, so leave them alone.  We
 728          * can leave it up to e2fsck to clean up any inconsistencies there.
 729          */
 730 }
 731
 732 /*
 733  * Open the external journal device
 734  */
 735 static struct block_device *ext4_blkdev_get(dev_t dev, struct super_block *sb)
 736 {
 737         struct block_device *bdev;
 738         char b[BDEVNAME_SIZE];
 739
 740         bdev = blkdev_get_by_dev(dev, FMODE_READ|FMODE_WRITE|FMODE_EXCL, sb);
 741         if (IS_ERR(bdev))
 742                 goto fail;
 743         return bdev;
 744
 745 fail:
 746         ext4_msg(sb, KERN_ERR, "failed to open journal device %s: %ld",
 747                         __bdevname(dev, b), PTR_ERR(bdev));
 748         return NULL;
 749 }
 750
 751 /*
 752  * Release the journal device
 753  */
 754 static void ext4_blkdev_put(struct block_device *bdev)
 755 {
 756         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 757 }
 758
 759 static void ext4_blkdev_remove(struct ext4_sb_info *sbi)
 760 {
 761         struct block_device *bdev;
 762         bdev = sbi->journal_bdev;
 763         if (bdev) {
 764                 ext4_blkdev_put(bdev);
 765                 sbi->journal_bdev = NULL;
 766         }
 767 }
 768
 769 static inline struct inode *orphan_list_entry(struct list_head *l)
 770 {
 771         return &list_entry(l, struct ext4_inode_info, i_orphan)->vfs_inode;
 772 }
 773
 774 static void dump_orphan_list(struct super_block *sb, struct ext4_sb_info *sbi)
 775 {
 776         struct list_head *l;
 777
 778         ext4_msg(sb, KERN_ERR, "sb orphan head is %d",
 779                  le32_to_cpu(sbi->s_es->s_last_orphan));
 780
 781         printk(KERN_ERR "sb_info orphan list:\n");
 782         list_for_each(l, &sbi->s_orphan) {
 783                 struct inode *inode = orphan_list_entry(l);
 784                 printk(KERN_ERR "  "
 785                        "inode %s:%lu at %p: mode %o, nlink %d, next %d\n",
 786                        inode->i_sb->s_id, inode->i_ino, inode,
 787                        inode->i_mode, inode->i_nlink,
 788                        NEXT_ORPHAN(inode));
 789         }
 790 }
 791
 792 static void ext4_put_super(struct super_block *sb)
 793 {
 794         struct ext4_sb_info *sbi = EXT4_SB(sb);
 795         struct ext4_super_block *es = sbi->s_es;
 796         int aborted = 0;
 797         int i, err;
 798
 799         ext4_unregister_li_request(sb);
 800         dquot_disable(sb, -1, DQUOT_USAGE_ENABLED | DQUOT_LIMITS_ENABLED);
 801
 802         flush_workqueue(sbi->rsv_conversion_wq);
 803         destroy_workqueue(sbi->rsv_conversion_wq);
 804
 805         if (sbi->s_journal) {
 806                 aborted = is_journal_aborted(sbi->s_journal);
 807                 err = jbd2_journal_destroy(sbi->s_journal);
 808                 sbi->s_journal = NULL;
 809                 if ((err < 0) && !aborted)
 810                         ext4_abort(sb, "Couldn't clean up the journal");
 811         }
 812
 813         ext4_unregister_sysfs(sb);
 814         ext4_es_unregister_shrinker(sbi);
 815         del_timer_sync(&sbi->s_err_report);
 816         ext4_release_system_zone(sb);
 817         ext4_mb_release(sb);
 818         ext4_ext_release(sb);
 819
 820         if (!(sb->s_flags & MS_RDONLY) && !aborted) {
 821                 ext4_clear_feature_journal_needs_recovery(sb);
 822                 es->s_state = cpu_to_le16(sbi->s_mount_state);
 823         }
 824         if (!(sb->s_flags & MS_RDONLY))
 825                 ext4_commit_super(sb, 1);
 826
 827         for (i = 0; i < sbi->s_gdb_count; i++)
 828                 brelse(sbi->s_group_desc[i]);
 829         kvfree(sbi->s_group_desc);
 830         kvfree(sbi->s_flex_groups);
 831         percpu_counter_destroy(&sbi->s_freeclusters_counter);
 832         percpu_counter_destroy(&sbi->s_freeinodes_counter);
 833         percpu_counter_destroy(&sbi->s_dirs_counter);
 834         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
 835         brelse(sbi->s_sbh);
 836 #ifdef CONFIG_QUOTA
 837         for (i = 0; i < EXT4_MAXQUOTAS; i++)
 838                 kfree(sbi->s_qf_names[i]);
 839 #endif
 840
 841         /* Debugging code just in case the in-memory inode orphan list
 842          * isn't empty.  The on-disk one can be non-empty if we've
 843          * detected an error and taken the fs readonly, but the
 844          * in-memory list had better be clean by this point. */
 845         if (!list_empty(&sbi->s_orphan))
 846                 dump_orphan_list(sb, sbi);
 847         J_ASSERT(list_empty(&sbi->s_orphan));
 848
 849         sync_blockdev(sb->s_bdev);
 850         invalidate_bdev(sb->s_bdev);
 851         if (sbi->journal_bdev && sbi->journal_bdev != sb->s_bdev) {
 852                 /*
 853                  * Invalidate the journal device's buffers.  We don't want them
 854                  * floating about in memory - the physical journal device may
 855                  * hotswapped, and it breaks the `ro-after' testing code.
 856                  */
 857                 sync_blockdev(sbi->journal_bdev);
 858                 invalidate_bdev(sbi->journal_bdev);
 859                 ext4_blkdev_remove(sbi);
 860         }
 861         if (sbi->s_mb_cache) {
 862                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
 863                 sbi->s_mb_cache = NULL;
 864         }
 865         if (sbi->s_mmp_tsk)
 866                 kthread_stop(sbi->s_mmp_tsk);
 867         sb->s_fs_info = NULL;
 868         /*
 869          * Now that we are completely done shutting down the
 870          * superblock, we need to actually destroy the kobject.
 871          */
 872         kobject_put(&sbi->s_kobj);
 873         wait_for_completion(&sbi->s_kobj_unregister);
 874         if (sbi->s_chksum_driver)
 875                 crypto_free_shash(sbi->s_chksum_driver);
 876         kfree(sbi->s_blockgroup_lock);
 877         kfree(sbi);
 878 }
 879
 880 static struct kmem_cache *ext4_inode_cachep;
 881
 882 /*
 883  * Called inside transaction, so use GFP_NOFS
 884  */
 885 static struct inode *ext4_alloc_inode(struct super_block *sb)
 886 {
 887         struct ext4_inode_info *ei;
 888
 889         ei = kmem_cache_alloc(ext4_inode_cachep, GFP_NOFS);
 890         if (!ei)
 891                 return NULL;
 892
 893         ei->vfs_inode.i_version = 1;
 894         spin_lock_init(&ei->i_raw_lock);
 895         INIT_LIST_HEAD(&ei->i_prealloc_list);
 896         spin_lock_init(&ei->i_prealloc_lock);
 897         ext4_es_init_tree(&ei->i_es_tree);
 898         rwlock_init(&ei->i_es_lock);
 899         INIT_LIST_HEAD(&ei->i_es_list);
 900         ei->i_es_all_nr = 0;
 901         ei->i_es_shk_nr = 0;
 902         ei->i_es_shrink_lblk = 0;
 903         ei->i_reserved_data_blocks = 0;
 904         ei->i_reserved_meta_blocks = 0;
 905         ei->i_allocated_meta_blocks = 0;
 906         ei->i_da_metadata_calc_len = 0;
 907         ei->i_da_metadata_calc_last_lblock = 0;
 908         spin_lock_init(&(ei->i_block_reservation_lock));
 909 #ifdef CONFIG_QUOTA
 910         ei->i_reserved_quota = 0;
 911         memset(&ei->i_dquot, 0, sizeof(ei->i_dquot));
 912 #endif
 913         ei->jinode = NULL;
 914         INIT_LIST_HEAD(&ei->i_rsv_conversion_list);
 915         spin_lock_init(&ei->i_completed_io_lock);
 916         ei->i_sync_tid = 0;
 917         ei->i_datasync_tid = 0;
 918         atomic_set(&ei->i_ioend_count, 0);
 919         atomic_set(&ei->i_unwritten, 0);
 920         INIT_WORK(&ei->i_rsv_conversion_work, ext4_end_io_rsv_work);
 921 #ifdef CONFIG_EXT4_FS_ENCRYPTION
 922         ei->i_crypt_info = NULL;
 923 #endif
 924         return &ei->vfs_inode;
 925 }
 926
 927 static int ext4_drop_inode(struct inode *inode)
 928 {
 929         int drop = generic_drop_inode(inode);
 930
 931         trace_ext4_drop_inode(inode, drop);
 932         return drop;
 933 }
 934
 935 static void ext4_i_callback(struct rcu_head *head)
 936 {
 937         struct inode *inode = container_of(head, struct inode, i_rcu);
 938         kmem_cache_free(ext4_inode_cachep, EXT4_I(inode));
 939 }
 940
 941 static void ext4_destroy_inode(struct inode *inode)
 942 {
 943         if (!list_empty(&(EXT4_I(inode)->i_orphan))) {
 944                 ext4_msg(inode->i_sb, KERN_ERR,
 945                          "Inode %lu (%p): orphan list check failed!",
 946                          inode->i_ino, EXT4_I(inode));
 947                 print_hex_dump(KERN_INFO, "", DUMP_PREFIX_ADDRESS, 16, 4,
 948                                 EXT4_I(inode), sizeof(struct ext4_inode_info),
 949                                 true);
 950                 dump_stack();
 951         }
 952         call_rcu(&inode->i_rcu, ext4_i_callback);
 953 }
 954
 955 static void init_once(void *foo)
 956 {
 957         struct ext4_inode_info *ei = (struct ext4_inode_info *) foo;
 958
 959         INIT_LIST_HEAD(&ei->i_orphan);
 960         init_rwsem(&ei->xattr_sem);
 961         init_rwsem(&ei->i_data_sem);
 962         init_rwsem(&ei->i_mmap_sem);
 963         inode_init_once(&ei->vfs_inode);
 964 }
 965
 966 static int __init init_inodecache(void)
 967 {
 968         ext4_inode_cachep = kmem_cache_create("ext4_inode_cache",
 969                                              sizeof(struct ext4_inode_info),
 970                                              0, (SLAB_RECLAIM_ACCOUNT|
 971                                                 SLAB_MEM_SPREAD),
 972                                              init_once);
 973         if (ext4_inode_cachep == NULL)
 974                 return -ENOMEM;
 975         return 0;
 976 }
 977
 978 static void destroy_inodecache(void)
 979 {
 980         /*
 981          * Make sure all delayed rcu free inodes are flushed before we
 982          * destroy cache.
 983          */
 984         rcu_barrier();
 985         kmem_cache_destroy(ext4_inode_cachep);
 986 }
 987
 988 void ext4_clear_inode(struct inode *inode)
 989 {
 990         invalidate_inode_buffers(inode);
 991         clear_inode(inode);
 992         dquot_drop(inode);
 993         ext4_discard_preallocations(inode);
 994         ext4_es_remove_extent(inode, 0, EXT_MAX_BLOCKS);
 995         if (EXT4_I(inode)->jinode) {
 996                 jbd2_journal_release_jbd_inode(EXT4_JOURNAL(inode),
 997                                                EXT4_I(inode)->jinode);
 998                 jbd2_free_inode(EXT4_I(inode)->jinode);
 999                 EXT4_I(inode)->jinode = NULL;
1000         }
1001 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1002         if (EXT4_I(inode)->i_crypt_info)
1003                 ext4_free_encryption_info(inode, EXT4_I(inode)->i_crypt_info);
1004 #endif
1005 }
1006
1007 static struct inode *ext4_nfs_get_inode(struct super_block *sb,
1008                                         u64 ino, u32 generation)
1009 {
1010         struct inode *inode;
1011
1012         if (ino < EXT4_FIRST_INO(sb) && ino != EXT4_ROOT_INO)
1013                 return ERR_PTR(-ESTALE);
1014         if (ino > le32_to_cpu(EXT4_SB(sb)->s_es->s_inodes_count))
1015                 return ERR_PTR(-ESTALE);
1016
1017         /* iget isn't really right if the inode is currently unallocated!!
1018          *
1019          * ext4_read_inode will return a bad_inode if the inode had been
1020          * deleted, so we should be safe.
1021          *
1022          * Currently we don't know the generation for parent directory, so
1023          * a generation of 0 means "accept any"
1024          */
1025         inode = ext4_iget_normal(sb, ino);
1026         if (IS_ERR(inode))
1027                 return ERR_CAST(inode);
1028         if (generation && inode->i_generation != generation) {
1029                 iput(inode);
1030                 return ERR_PTR(-ESTALE);
1031         }
1032
1033         return inode;
1034 }
1035
1036 static struct dentry *ext4_fh_to_dentry(struct super_block *sb, struct fid *fid,
1037                                         int fh_len, int fh_type)
1038 {
1039         return generic_fh_to_dentry(sb, fid, fh_len, fh_type,
1040                                     ext4_nfs_get_inode);
1041 }
1042
1043 static struct dentry *ext4_fh_to_parent(struct super_block *sb, struct fid *fid,
1044                                         int fh_len, int fh_type)
1045 {
1046         return generic_fh_to_parent(sb, fid, fh_len, fh_type,
1047                                     ext4_nfs_get_inode);
1048 }
1049
1050 /*
1051  * Try to release metadata pages (indirect blocks, directories) which are
1052  * mapped via the block device.  Since these pages could have journal heads
1053  * which would prevent try_to_free_buffers() from freeing them, we must use
1054  * jbd2 layer's try_to_free_buffers() function to release them.
1055  */
1056 static int bdev_try_to_free_page(struct super_block *sb, struct page *page,
1057                                  gfp_t wait)
1058 {
1059         journal_t *journal = EXT4_SB(sb)->s_journal;
1060
1061         WARN_ON(PageChecked(page));
1062         if (!page_has_buffers(page))
1063                 return 0;
1064         if (journal)
1065                 return jbd2_journal_try_to_free_buffers(journal, page,
1066                                                 wait & ~__GFP_DIRECT_RECLAIM);
1067         return try_to_free_buffers(page);
1068 }
1069
1070 #ifdef CONFIG_QUOTA
1071 #define QTYPE2NAME(t) ((t) == USRQUOTA ? "user" : "group")
1072 #define QTYPE2MOPT(on, t) ((t) == USRQUOTA?((on)##USRJQUOTA):((on)##GRPJQUOTA))
1073
1074 static int ext4_write_dquot(struct dquot *dquot);
1075 static int ext4_acquire_dquot(struct dquot *dquot);
1076 static int ext4_release_dquot(struct dquot *dquot);
1077 static int ext4_mark_dquot_dirty(struct dquot *dquot);
1078 static int ext4_write_info(struct super_block *sb, int type);
1079 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
1080                          struct path *path);
1081 static int ext4_quota_off(struct super_block *sb, int type);
1082 static int ext4_quota_on_mount(struct super_block *sb, int type);
1083 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
1084                                size_t len, loff_t off);
1085 static ssize_t ext4_quota_write(struct super_block *sb, int type,
1086                                 const char *data, size_t len, loff_t off);
1087 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
1088                              unsigned int flags);
1089 static int ext4_enable_quotas(struct super_block *sb);
1090
1091 static struct dquot **ext4_get_dquots(struct inode *inode)
1092 {
1093         return EXT4_I(inode)->i_dquot;
1094 }
1095
1096 static const struct dquot_operations ext4_quota_operations = {
1097         .get_reserved_space = ext4_get_reserved_space,
1098         .write_dquot    = ext4_write_dquot,
1099         .acquire_dquot  = ext4_acquire_dquot,
1100         .release_dquot  = ext4_release_dquot,
1101         .mark_dirty     = ext4_mark_dquot_dirty,
1102         .write_info     = ext4_write_info,
1103         .alloc_dquot    = dquot_alloc,
1104         .destroy_dquot  = dquot_destroy,
1105 };
1106
1107 static const struct quotactl_ops ext4_qctl_operations = {
1108         .quota_on       = ext4_quota_on,
1109         .quota_off      = ext4_quota_off,
1110         .quota_sync     = dquot_quota_sync,
1111         .get_state      = dquot_get_state,
1112         .set_info       = dquot_set_dqinfo,
1113         .get_dqblk      = dquot_get_dqblk,
1114         .set_dqblk      = dquot_set_dqblk
1115 };
1116 #endif
1117
1118 static const struct super_operations ext4_sops = {
1119         .alloc_inode    = ext4_alloc_inode,
1120         .destroy_inode  = ext4_destroy_inode,
1121         .write_inode    = ext4_write_inode,
1122         .dirty_inode    = ext4_dirty_inode,
1123         .drop_inode     = ext4_drop_inode,
1124         .evict_inode    = ext4_evict_inode,
1125         .put_super      = ext4_put_super,
1126         .sync_fs        = ext4_sync_fs,
1127         .freeze_fs      = ext4_freeze,
1128         .unfreeze_fs    = ext4_unfreeze,
1129         .statfs         = ext4_statfs,
1130         .remount_fs     = ext4_remount,
1131         .show_options   = ext4_show_options,
1132 #ifdef CONFIG_QUOTA
1133         .quota_read     = ext4_quota_read,
1134         .quota_write    = ext4_quota_write,
1135         .get_dquots     = ext4_get_dquots,
1136 #endif
1137         .bdev_try_to_free_page = bdev_try_to_free_page,
1138 };
1139
1140 static const struct export_operations ext4_export_ops = {
1141         .fh_to_dentry = ext4_fh_to_dentry,
1142         .fh_to_parent = ext4_fh_to_parent,
1143         .get_parent = ext4_get_parent,
1144 };
1145
1146 enum {
1147         Opt_bsd_df, Opt_minix_df, Opt_grpid, Opt_nogrpid,
1148         Opt_resgid, Opt_resuid, Opt_sb, Opt_err_cont, Opt_err_panic, Opt_err_ro,
1149         Opt_nouid32, Opt_debug, Opt_removed,
1150         Opt_user_xattr, Opt_nouser_xattr, Opt_acl, Opt_noacl,
1151         Opt_auto_da_alloc, Opt_noauto_da_alloc, Opt_noload,
1152         Opt_commit, Opt_min_batch_time, Opt_max_batch_time, Opt_journal_dev,
1153         Opt_journal_path, Opt_journal_checksum, Opt_journal_async_commit,
1154         Opt_abort, Opt_data_journal, Opt_data_ordered, Opt_data_writeback,
1155         Opt_data_err_abort, Opt_data_err_ignore, Opt_test_dummy_encryption,
1156         Opt_usrjquota, Opt_grpjquota, Opt_offusrjquota, Opt_offgrpjquota,
1157         Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, Opt_quota,
1158         Opt_noquota, Opt_barrier, Opt_nobarrier, Opt_err,
1159         Opt_usrquota, Opt_grpquota, Opt_i_version, Opt_dax,
1160         Opt_stripe, Opt_delalloc, Opt_nodelalloc, Opt_mblk_io_submit,
1161         Opt_lazytime, Opt_nolazytime,
1162         Opt_nomblk_io_submit, Opt_block_validity, Opt_noblock_validity,
1163         Opt_inode_readahead_blks, Opt_journal_ioprio,
1164         Opt_dioread_nolock, Opt_dioread_lock,
1165         Opt_discard, Opt_nodiscard, Opt_init_itable, Opt_noinit_itable,
1166         Opt_max_dir_size_kb, Opt_nojournal_checksum,
1167 };
1168
1169 static const match_table_t tokens = {
1170         {Opt_bsd_df, "bsddf"},
1171         {Opt_minix_df, "minixdf"},
1172         {Opt_grpid, "grpid"},
1173         {Opt_grpid, "bsdgroups"},
1174         {Opt_nogrpid, "nogrpid"},
1175         {Opt_nogrpid, "sysvgroups"},
1176         {Opt_resgid, "resgid=%u"},
1177         {Opt_resuid, "resuid=%u"},
1178         {Opt_sb, "sb=%u"},
1179         {Opt_err_cont, "errors=continue"},
1180         {Opt_err_panic, "errors=panic"},
1181         {Opt_err_ro, "errors=remount-ro"},
1182         {Opt_nouid32, "nouid32"},
1183         {Opt_debug, "debug"},
1184         {Opt_removed, "oldalloc"},
1185         {Opt_removed, "orlov"},
1186         {Opt_user_xattr, "user_xattr"},
1187         {Opt_nouser_xattr, "nouser_xattr"},
1188         {Opt_acl, "acl"},
1189         {Opt_noacl, "noacl"},
1190         {Opt_noload, "norecovery"},
1191         {Opt_noload, "noload"},
1192         {Opt_removed, "nobh"},
1193         {Opt_removed, "bh"},
1194         {Opt_commit, "commit=%u"},
1195         {Opt_min_batch_time, "min_batch_time=%u"},
1196         {Opt_max_batch_time, "max_batch_time=%u"},
1197         {Opt_journal_dev, "journal_dev=%u"},
1198         {Opt_journal_path, "journal_path=%s"},
1199         {Opt_journal_checksum, "journal_checksum"},
1200         {Opt_nojournal_checksum, "nojournal_checksum"},
1201         {Opt_journal_async_commit, "journal_async_commit"},
1202         {Opt_abort, "abort"},
1203         {Opt_data_journal, "data=journal"},
1204         {Opt_data_ordered, "data=ordered"},
1205         {Opt_data_writeback, "data=writeback"},
1206         {Opt_data_err_abort, "data_err=abort"},
1207         {Opt_data_err_ignore, "data_err=ignore"},
1208         {Opt_offusrjquota, "usrjquota="},
1209         {Opt_usrjquota, "usrjquota=%s"},
1210         {Opt_offgrpjquota, "grpjquota="},
1211         {Opt_grpjquota, "grpjquota=%s"},
1212         {Opt_jqfmt_vfsold, "jqfmt=vfsold"},
1213         {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"},
1214         {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"},
1215         {Opt_grpquota, "grpquota"},
1216         {Opt_noquota, "noquota"},
1217         {Opt_quota, "quota"},
1218         {Opt_usrquota, "usrquota"},
1219         {Opt_barrier, "barrier=%u"},
1220         {Opt_barrier, "barrier"},
1221         {Opt_nobarrier, "nobarrier"},
1222         {Opt_i_version, "i_version"},
1223         {Opt_dax, "dax"},
1224         {Opt_stripe, "stripe=%u"},
1225         {Opt_delalloc, "delalloc"},
1226         {Opt_lazytime, "lazytime"},
1227         {Opt_nolazytime, "nolazytime"},
1228         {Opt_nodelalloc, "nodelalloc"},
1229         {Opt_removed, "mblk_io_submit"},
1230         {Opt_removed, "nomblk_io_submit"},
1231         {Opt_block_validity, "block_validity"},
1232         {Opt_noblock_validity, "noblock_validity"},
1233         {Opt_inode_readahead_blks, "inode_readahead_blks=%u"},
1234         {Opt_journal_ioprio, "journal_ioprio=%u"},
1235         {Opt_auto_da_alloc, "auto_da_alloc=%u"},
1236         {Opt_auto_da_alloc, "auto_da_alloc"},
1237         {Opt_noauto_da_alloc, "noauto_da_alloc"},
1238         {Opt_dioread_nolock, "dioread_nolock"},
1239         {Opt_dioread_lock, "dioread_lock"},
1240         {Opt_discard, "discard"},
1241         {Opt_nodiscard, "nodiscard"},
1242         {Opt_init_itable, "init_itable=%u"},
1243         {Opt_init_itable, "init_itable"},
1244         {Opt_noinit_itable, "noinit_itable"},
1245         {Opt_max_dir_size_kb, "max_dir_size_kb=%u"},
1246         {Opt_test_dummy_encryption, "test_dummy_encryption"},
1247         {Opt_removed, "check=none"},    /* mount option from ext2/3 */
1248         {Opt_removed, "nocheck"},       /* mount option from ext2/3 */
1249         {Opt_removed, "reservation"},   /* mount option from ext2/3 */
1250         {Opt_removed, "noreservation"}, /* mount option from ext2/3 */
1251         {Opt_removed, "journal=%u"},    /* mount option from ext2/3 */
1252         {Opt_err, NULL},
1253 };
1254
1255 static ext4_fsblk_t get_sb_block(void **data)
1256 {
1257         ext4_fsblk_t    sb_block;
1258         char            *options = (char *) *data;
1259
1260         if (!options || strncmp(options, "sb=", 3) != 0)
1261                 return 1;       /* Default location */
1262
1263         options += 3;
1264         /* TODO: use simple_strtoll with >32bit ext4 */
1265         sb_block = simple_strtoul(options, &options, 0);
1266         if (*options && *options != ',') {
1267                 printk(KERN_ERR "EXT4-fs: Invalid sb specification: %s\n",
1268                        (char *) *data);
1269                 return 1;
1270         }
1271         if (*options == ',')
1272                 options++;
1273         *data = (void *) options;
1274
1275         return sb_block;
1276 }
1277
1278 #define DEFAULT_JOURNAL_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, 3))
1279 static char deprecated_msg[] = "Mount option \"%s\" will be removed by %s\n"
1280         "Contact linux-ext4@vger.kernel.org if you think we should keep it.\n";
1281
1282 #ifdef CONFIG_QUOTA
1283 static int set_qf_name(struct super_block *sb, int qtype, substring_t *args)
1284 {
1285         struct ext4_sb_info *sbi = EXT4_SB(sb);
1286         char *qname;
1287         int ret = -1;
1288
1289         if (sb_any_quota_loaded(sb) &&
1290                 !sbi->s_qf_names[qtype]) {
1291                 ext4_msg(sb, KERN_ERR,
1292                         "Cannot change journaled "
1293                         "quota options when quota turned on");
1294                 return -1;
1295         }
1296         if (ext4_has_feature_quota(sb)) {
1297                 ext4_msg(sb, KERN_INFO, "Journaled quota options "
1298                          "ignored when QUOTA feature is enabled");
1299                 return 1;
1300         }
1301         qname = match_strdup(args);
1302         if (!qname) {
1303                 ext4_msg(sb, KERN_ERR,
1304                         "Not enough memory for storing quotafile name");
1305                 return -1;
1306         }
1307         if (sbi->s_qf_names[qtype]) {
1308                 if (strcmp(sbi->s_qf_names[qtype], qname) == 0)
1309                         ret = 1;
1310                 else
1311                         ext4_msg(sb, KERN_ERR,
1312                                  "%s quota file already specified",
1313                                  QTYPE2NAME(qtype));
1314                 goto errout;
1315         }
1316         if (strchr(qname, '/')) {
1317                 ext4_msg(sb, KERN_ERR,
1318                         "quotafile must be on filesystem root");
1319                 goto errout;
1320         }
1321         sbi->s_qf_names[qtype] = qname;
1322         set_opt(sb, QUOTA);
1323         return 1;
1324 errout:
1325         kfree(qname);
1326         return ret;
1327 }
1328
1329 static int clear_qf_name(struct super_block *sb, int qtype)
1330 {
1331
1332         struct ext4_sb_info *sbi = EXT4_SB(sb);
1333
1334         if (sb_any_quota_loaded(sb) &&
1335                 sbi->s_qf_names[qtype]) {
1336                 ext4_msg(sb, KERN_ERR, "Cannot change journaled quota options"
1337                         " when quota turned on");
1338                 return -1;
1339         }
1340         kfree(sbi->s_qf_names[qtype]);
1341         sbi->s_qf_names[qtype] = NULL;
1342         return 1;
1343 }
1344 #endif
1345
1346 #define MOPT_SET        0x0001
1347 #define MOPT_CLEAR      0x0002
1348 #define MOPT_NOSUPPORT  0x0004
1349 #define MOPT_EXPLICIT   0x0008
1350 #define MOPT_CLEAR_ERR  0x0010
1351 #define MOPT_GTE0       0x0020
1352 #ifdef CONFIG_QUOTA
1353 #define MOPT_Q          0
1354 #define MOPT_QFMT       0x0040
1355 #else
1356 #define MOPT_Q          MOPT_NOSUPPORT
1357 #define MOPT_QFMT       MOPT_NOSUPPORT
1358 #endif
1359 #define MOPT_DATAJ      0x0080
1360 #define MOPT_NO_EXT2    0x0100
1361 #define MOPT_NO_EXT3    0x0200
1362 #define MOPT_EXT4_ONLY  (MOPT_NO_EXT2 | MOPT_NO_EXT3)
1363 #define MOPT_STRING     0x0400
1364
1365 static const struct mount_opts {
1366         int     token;
1367         int     mount_opt;
1368         int     flags;
1369 } ext4_mount_opts[] = {
1370         {Opt_minix_df, EXT4_MOUNT_MINIX_DF, MOPT_SET},
1371         {Opt_bsd_df, EXT4_MOUNT_MINIX_DF, MOPT_CLEAR},
1372         {Opt_grpid, EXT4_MOUNT_GRPID, MOPT_SET},
1373         {Opt_nogrpid, EXT4_MOUNT_GRPID, MOPT_CLEAR},
1374         {Opt_block_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_SET},
1375         {Opt_noblock_validity, EXT4_MOUNT_BLOCK_VALIDITY, MOPT_CLEAR},
1376         {Opt_dioread_nolock, EXT4_MOUNT_DIOREAD_NOLOCK,
1377          MOPT_EXT4_ONLY | MOPT_SET},
1378         {Opt_dioread_lock, EXT4_MOUNT_DIOREAD_NOLOCK,
1379          MOPT_EXT4_ONLY | MOPT_CLEAR},
1380         {Opt_discard, EXT4_MOUNT_DISCARD, MOPT_SET},
1381         {Opt_nodiscard, EXT4_MOUNT_DISCARD, MOPT_CLEAR},
1382         {Opt_delalloc, EXT4_MOUNT_DELALLOC,
1383          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1384         {Opt_nodelalloc, EXT4_MOUNT_DELALLOC,
1385          MOPT_EXT4_ONLY | MOPT_CLEAR},
1386         {Opt_nojournal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1387          MOPT_EXT4_ONLY | MOPT_CLEAR},
1388         {Opt_journal_checksum, EXT4_MOUNT_JOURNAL_CHECKSUM,
1389          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1390         {Opt_journal_async_commit, (EXT4_MOUNT_JOURNAL_ASYNC_COMMIT |
1391                                     EXT4_MOUNT_JOURNAL_CHECKSUM),
1392          MOPT_EXT4_ONLY | MOPT_SET | MOPT_EXPLICIT},
1393         {Opt_noload, EXT4_MOUNT_NOLOAD, MOPT_NO_EXT2 | MOPT_SET},
1394         {Opt_err_panic, EXT4_MOUNT_ERRORS_PANIC, MOPT_SET | MOPT_CLEAR_ERR},
1395         {Opt_err_ro, EXT4_MOUNT_ERRORS_RO, MOPT_SET | MOPT_CLEAR_ERR},
1396         {Opt_err_cont, EXT4_MOUNT_ERRORS_CONT, MOPT_SET | MOPT_CLEAR_ERR},
1397         {Opt_data_err_abort, EXT4_MOUNT_DATA_ERR_ABORT,
1398          MOPT_NO_EXT2 | MOPT_SET},
1399         {Opt_data_err_ignore, EXT4_MOUNT_DATA_ERR_ABORT,
1400          MOPT_NO_EXT2 | MOPT_CLEAR},
1401         {Opt_barrier, EXT4_MOUNT_BARRIER, MOPT_SET},
1402         {Opt_nobarrier, EXT4_MOUNT_BARRIER, MOPT_CLEAR},
1403         {Opt_noauto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_SET},
1404         {Opt_auto_da_alloc, EXT4_MOUNT_NO_AUTO_DA_ALLOC, MOPT_CLEAR},
1405         {Opt_noinit_itable, EXT4_MOUNT_INIT_INODE_TABLE, MOPT_CLEAR},
1406         {Opt_commit, 0, MOPT_GTE0},
1407         {Opt_max_batch_time, 0, MOPT_GTE0},
1408         {Opt_min_batch_time, 0, MOPT_GTE0},
1409         {Opt_inode_readahead_blks, 0, MOPT_GTE0},
1410         {Opt_init_itable, 0, MOPT_GTE0},
1411         {Opt_dax, EXT4_MOUNT_DAX, MOPT_SET},
1412         {Opt_stripe, 0, MOPT_GTE0},
1413         {Opt_resuid, 0, MOPT_GTE0},
1414         {Opt_resgid, 0, MOPT_GTE0},
1415         {Opt_journal_dev, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1416         {Opt_journal_path, 0, MOPT_NO_EXT2 | MOPT_STRING},
1417         {Opt_journal_ioprio, 0, MOPT_NO_EXT2 | MOPT_GTE0},
1418         {Opt_data_journal, EXT4_MOUNT_JOURNAL_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1419         {Opt_data_ordered, EXT4_MOUNT_ORDERED_DATA, MOPT_NO_EXT2 | MOPT_DATAJ},
1420         {Opt_data_writeback, EXT4_MOUNT_WRITEBACK_DATA,
1421          MOPT_NO_EXT2 | MOPT_DATAJ},
1422         {Opt_user_xattr, EXT4_MOUNT_XATTR_USER, MOPT_SET},
1423         {Opt_nouser_xattr, EXT4_MOUNT_XATTR_USER, MOPT_CLEAR},
1424 #ifdef CONFIG_EXT4_FS_POSIX_ACL
1425         {Opt_acl, EXT4_MOUNT_POSIX_ACL, MOPT_SET},
1426         {Opt_noacl, EXT4_MOUNT_POSIX_ACL, MOPT_CLEAR},
1427 #else
1428         {Opt_acl, 0, MOPT_NOSUPPORT},
1429         {Opt_noacl, 0, MOPT_NOSUPPORT},
1430 #endif
1431         {Opt_nouid32, EXT4_MOUNT_NO_UID32, MOPT_SET},
1432         {Opt_debug, EXT4_MOUNT_DEBUG, MOPT_SET},
1433         {Opt_quota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA, MOPT_SET | MOPT_Q},
1434         {Opt_usrquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA,
1435                                                         MOPT_SET | MOPT_Q},
1436         {Opt_grpquota, EXT4_MOUNT_QUOTA | EXT4_MOUNT_GRPQUOTA,
1437                                                         MOPT_SET | MOPT_Q},
1438         {Opt_noquota, (EXT4_MOUNT_QUOTA | EXT4_MOUNT_USRQUOTA |
1439                        EXT4_MOUNT_GRPQUOTA), MOPT_CLEAR | MOPT_Q},
1440         {Opt_usrjquota, 0, MOPT_Q},
1441         {Opt_grpjquota, 0, MOPT_Q},
1442         {Opt_offusrjquota, 0, MOPT_Q},
1443         {Opt_offgrpjquota, 0, MOPT_Q},
1444         {Opt_jqfmt_vfsold, QFMT_VFS_OLD, MOPT_QFMT},
1445         {Opt_jqfmt_vfsv0, QFMT_VFS_V0, MOPT_QFMT},
1446         {Opt_jqfmt_vfsv1, QFMT_VFS_V1, MOPT_QFMT},
1447         {Opt_max_dir_size_kb, 0, MOPT_GTE0},
1448         {Opt_test_dummy_encryption, 0, MOPT_GTE0},
1449         {Opt_err, 0, 0}
1450 };
1451
1452 static int handle_mount_opt(struct super_block *sb, char *opt, int token,
1453                             substring_t *args, unsigned long *journal_devnum,
1454                             unsigned int *journal_ioprio, int is_remount)
1455 {
1456         struct ext4_sb_info *sbi = EXT4_SB(sb);
1457         const struct mount_opts *m;
1458         kuid_t uid;
1459         kgid_t gid;
1460         int arg = 0;
1461
1462 #ifdef CONFIG_QUOTA
1463         if (token == Opt_usrjquota)
1464                 return set_qf_name(sb, USRQUOTA, &args[0]);
1465         else if (token == Opt_grpjquota)
1466                 return set_qf_name(sb, GRPQUOTA, &args[0]);
1467         else if (token == Opt_offusrjquota)
1468                 return clear_qf_name(sb, USRQUOTA);
1469         else if (token == Opt_offgrpjquota)
1470                 return clear_qf_name(sb, GRPQUOTA);
1471 #endif
1472         switch (token) {
1473         case Opt_noacl:
1474         case Opt_nouser_xattr:
1475                 ext4_msg(sb, KERN_WARNING, deprecated_msg, opt, "3.5");
1476                 break;
1477         case Opt_sb:
1478                 return 1;       /* handled by get_sb_block() */
1479         case Opt_removed:
1480                 ext4_msg(sb, KERN_WARNING, "Ignoring removed %s option", opt);
1481                 return 1;
1482         case Opt_abort:
1483                 sbi->s_mount_flags |= EXT4_MF_FS_ABORTED;
1484                 return 1;
1485         case Opt_i_version:
1486                 sb->s_flags |= MS_I_VERSION;
1487                 return 1;
1488         case Opt_lazytime:
1489                 sb->s_flags |= MS_LAZYTIME;
1490                 return 1;
1491         case Opt_nolazytime:
1492                 sb->s_flags &= ~MS_LAZYTIME;
1493                 return 1;
1494         }
1495
1496         for (m = ext4_mount_opts; m->token != Opt_err; m++)
1497                 if (token == m->token)
1498                         break;
1499
1500         if (m->token == Opt_err) {
1501                 ext4_msg(sb, KERN_ERR, "Unrecognized mount option \"%s\" "
1502                          "or missing value", opt);
1503                 return -1;
1504         }
1505
1506         if ((m->flags & MOPT_NO_EXT2) && IS_EXT2_SB(sb)) {
1507                 ext4_msg(sb, KERN_ERR,
1508                          "Mount option \"%s\" incompatible with ext2", opt);
1509                 return -1;
1510         }
1511         if ((m->flags & MOPT_NO_EXT3) && IS_EXT3_SB(sb)) {
1512                 ext4_msg(sb, KERN_ERR,
1513                          "Mount option \"%s\" incompatible with ext3", opt);
1514                 return -1;
1515         }
1516
1517         if (args->from && !(m->flags & MOPT_STRING) && match_int(args, &arg))
1518                 return -1;
1519         if (args->from && (m->flags & MOPT_GTE0) && (arg < 0))
1520                 return -1;
1521         if (m->flags & MOPT_EXPLICIT) {
1522                 if (m->mount_opt & EXT4_MOUNT_DELALLOC) {
1523                         set_opt2(sb, EXPLICIT_DELALLOC);
1524                 } else if (m->mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) {
1525                         set_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM);
1526                 } else
1527                         return -1;
1528         }
1529         if (m->flags & MOPT_CLEAR_ERR)
1530                 clear_opt(sb, ERRORS_MASK);
1531         if (token == Opt_noquota && sb_any_quota_loaded(sb)) {
1532                 ext4_msg(sb, KERN_ERR, "Cannot change quota "
1533                          "options when quota turned on");
1534                 return -1;
1535         }
1536
1537         if (m->flags & MOPT_NOSUPPORT) {
1538                 ext4_msg(sb, KERN_ERR, "%s option not supported", opt);
1539         } else if (token == Opt_commit) {
1540                 if (arg == 0)
1541                         arg = JBD2_DEFAULT_MAX_COMMIT_AGE;
1542                 sbi->s_commit_interval = HZ * arg;
1543         } else if (token == Opt_max_batch_time) {
1544                 sbi->s_max_batch_time = arg;
1545         } else if (token == Opt_min_batch_time) {
1546                 sbi->s_min_batch_time = arg;
1547         } else if (token == Opt_inode_readahead_blks) {
1548                 if (arg && (arg > (1 << 30) || !is_power_of_2(arg))) {
1549                         ext4_msg(sb, KERN_ERR,
1550                                  "EXT4-fs: inode_readahead_blks must be "
1551                                  "0 or a power of 2 smaller than 2^31");
1552                         return -1;
1553                 }
1554                 sbi->s_inode_readahead_blks = arg;
1555         } else if (token == Opt_init_itable) {
1556                 set_opt(sb, INIT_INODE_TABLE);
1557                 if (!args->from)
1558                         arg = EXT4_DEF_LI_WAIT_MULT;
1559                 sbi->s_li_wait_mult = arg;
1560         } else if (token == Opt_max_dir_size_kb) {
1561                 sbi->s_max_dir_size_kb = arg;
1562         } else if (token == Opt_stripe) {
1563                 sbi->s_stripe = arg;
1564         } else if (token == Opt_resuid) {
1565                 uid = make_kuid(current_user_ns(), arg);
1566                 if (!uid_valid(uid)) {
1567                         ext4_msg(sb, KERN_ERR, "Invalid uid value %d", arg);
1568                         return -1;
1569                 }
1570                 sbi->s_resuid = uid;
1571         } else if (token == Opt_resgid) {
1572                 gid = make_kgid(current_user_ns(), arg);
1573                 if (!gid_valid(gid)) {
1574                         ext4_msg(sb, KERN_ERR, "Invalid gid value %d", arg);
1575                         return -1;
1576                 }
1577                 sbi->s_resgid = gid;
1578         } else if (token == Opt_journal_dev) {
1579                 if (is_remount) {
1580                         ext4_msg(sb, KERN_ERR,
1581                                  "Cannot specify journal on remount");
1582                         return -1;
1583                 }
1584                 *journal_devnum = arg;
1585         } else if (token == Opt_journal_path) {
1586                 char *journal_path;
1587                 struct inode *journal_inode;
1588                 struct path path;
1589                 int error;
1590
1591                 if (is_remount) {
1592                         ext4_msg(sb, KERN_ERR,
1593                                  "Cannot specify journal on remount");
1594                         return -1;
1595                 }
1596                 journal_path = match_strdup(&args[0]);
1597                 if (!journal_path) {
1598                         ext4_msg(sb, KERN_ERR, "error: could not dup "
1599                                 "journal device string");
1600                         return -1;
1601                 }
1602
1603                 error = kern_path(journal_path, LOOKUP_FOLLOW, &path);
1604                 if (error) {
1605                         ext4_msg(sb, KERN_ERR, "error: could not find "
1606                                 "journal device path: error %d", error);
1607                         kfree(journal_path);
1608                         return -1;
1609                 }
1610
1611                 journal_inode = d_inode(path.dentry);
1612                 if (!S_ISBLK(journal_inode->i_mode)) {
1613                         ext4_msg(sb, KERN_ERR, "error: journal path %s "
1614                                 "is not a block device", journal_path);
1615                         path_put(&path);
1616                         kfree(journal_path);
1617                         return -1;
1618                 }
1619
1620                 *journal_devnum = new_encode_dev(journal_inode->i_rdev);
1621                 path_put(&path);
1622                 kfree(journal_path);
1623         } else if (token == Opt_journal_ioprio) {
1624                 if (arg > 7) {
1625                         ext4_msg(sb, KERN_ERR, "Invalid journal IO priority"
1626                                  " (must be 0-7)");
1627                         return -1;
1628                 }
1629                 *journal_ioprio =
1630                         IOPRIO_PRIO_VALUE(IOPRIO_CLASS_BE, arg);
1631         } else if (token == Opt_test_dummy_encryption) {
1632 #ifdef CONFIG_EXT4_FS_ENCRYPTION
1633                 sbi->s_mount_flags |= EXT4_MF_TEST_DUMMY_ENCRYPTION;
1634                 ext4_msg(sb, KERN_WARNING,
1635                          "Test dummy encryption mode enabled");
1636 #else
1637                 ext4_msg(sb, KERN_WARNING,
1638                          "Test dummy encryption mount option ignored");
1639 #endif
1640         } else if (m->flags & MOPT_DATAJ) {
1641                 if (is_remount) {
1642                         if (!sbi->s_journal)
1643                                 ext4_msg(sb, KERN_WARNING, "Remounting file system with no journal so ignoring journalled data option");
1644                         else if (test_opt(sb, DATA_FLAGS) != m->mount_opt) {
1645                                 ext4_msg(sb, KERN_ERR,
1646                                          "Cannot change data mode on remount");
1647                                 return -1;
1648                         }
1649                 } else {
1650                         clear_opt(sb, DATA_FLAGS);
1651                         sbi->s_mount_opt |= m->mount_opt;
1652                 }
1653 #ifdef CONFIG_QUOTA
1654         } else if (m->flags & MOPT_QFMT) {
1655                 if (sb_any_quota_loaded(sb) &&
1656                     sbi->s_jquota_fmt != m->mount_opt) {
1657                         ext4_msg(sb, KERN_ERR, "Cannot change journaled "
1658                                  "quota options when quota turned on");
1659                         return -1;
1660                 }
1661                 if (ext4_has_feature_quota(sb)) {
1662                         ext4_msg(sb, KERN_INFO,
1663                                  "Quota format mount options ignored "
1664                                  "when QUOTA feature is enabled");
1665                         return 1;
1666                 }
1667                 sbi->s_jquota_fmt = m->mount_opt;
1668 #endif
1669         } else if (token == Opt_dax) {
1670 #ifdef CONFIG_FS_DAX
1671                 ext4_msg(sb, KERN_WARNING,
1672                 "DAX enabled. Warning: EXPERIMENTAL, use at your own risk");
1673                         sbi->s_mount_opt |= m->mount_opt;
1674 #else
1675                 ext4_msg(sb, KERN_INFO, "dax option not supported");
1676                 return -1;
1677 #endif
1678         } else {
1679                 if (!args->from)
1680                         arg = 1;
1681                 if (m->flags & MOPT_CLEAR)
1682                         arg = !arg;
1683                 else if (unlikely(!(m->flags & MOPT_SET))) {
1684                         ext4_msg(sb, KERN_WARNING,
1685                                  "buggy handling of option %s", opt);
1686                         WARN_ON(1);
1687                         return -1;
1688                 }
1689                 if (arg != 0)
1690                         sbi->s_mount_opt |= m->mount_opt;
1691                 else
1692                         sbi->s_mount_opt &= ~m->mount_opt;
1693         }
1694         return 1;
1695 }
1696
1697 static int parse_options(char *options, struct super_block *sb,
1698                          unsigned long *journal_devnum,
1699                          unsigned int *journal_ioprio,
1700                          int is_remount)
1701 {
1702         struct ext4_sb_info *sbi = EXT4_SB(sb);
1703         char *p;
1704         substring_t args[MAX_OPT_ARGS];
1705         int token;
1706
1707         if (!options)
1708                 return 1;
1709
1710         while ((p = strsep(&options, ",")) != NULL) {
1711                 if (!*p)
1712                         continue;
1713                 /*
1714                  * Initialize args struct so we know whether arg was
1715                  * found; some options take optional arguments.
1716                  */
1717                 args[0].to = args[0].from = NULL;
1718                 token = match_token(p, tokens, args);
1719                 if (handle_mount_opt(sb, p, token, args, journal_devnum,
1720                                      journal_ioprio, is_remount) < 0)
1721                         return 0;
1722         }
1723 #ifdef CONFIG_QUOTA
1724         if (ext4_has_feature_quota(sb) &&
1725             (test_opt(sb, USRQUOTA) || test_opt(sb, GRPQUOTA))) {
1726                 ext4_msg(sb, KERN_INFO, "Quota feature enabled, usrquota and grpquota "
1727                          "mount options ignored.");
1728                 clear_opt(sb, USRQUOTA);
1729                 clear_opt(sb, GRPQUOTA);
1730         } else if (sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
1731                 if (test_opt(sb, USRQUOTA) && sbi->s_qf_names[USRQUOTA])
1732                         clear_opt(sb, USRQUOTA);
1733
1734                 if (test_opt(sb, GRPQUOTA) && sbi->s_qf_names[GRPQUOTA])
1735                         clear_opt(sb, GRPQUOTA);
1736
1737                 if (test_opt(sb, GRPQUOTA) || test_opt(sb, USRQUOTA)) {
1738                         ext4_msg(sb, KERN_ERR, "old and new quota "
1739                                         "format mixing");
1740                         return 0;
1741                 }
1742
1743                 if (!sbi->s_jquota_fmt) {
1744                         ext4_msg(sb, KERN_ERR, "journaled quota format "
1745                                         "not specified");
1746                         return 0;
1747                 }
1748         }
1749 #endif
1750         if (test_opt(sb, DIOREAD_NOLOCK)) {
1751                 int blocksize =
1752                         BLOCK_SIZE << le32_to_cpu(sbi->s_es->s_log_block_size);
1753
1754                 if (blocksize < PAGE_CACHE_SIZE) {
1755                         ext4_msg(sb, KERN_ERR, "can't mount with "
1756                                  "dioread_nolock if block size != PAGE_SIZE");
1757                         return 0;
1758                 }
1759         }
1760         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA &&
1761             test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
1762                 ext4_msg(sb, KERN_ERR, "can't mount with journal_async_commit "
1763                          "in data=ordered mode");
1764                 return 0;
1765         }
1766         return 1;
1767 }
1768
1769 static inline void ext4_show_quota_options(struct seq_file *seq,
1770                                            struct super_block *sb)
1771 {
1772 #if defined(CONFIG_QUOTA)
1773         struct ext4_sb_info *sbi = EXT4_SB(sb);
1774
1775         if (sbi->s_jquota_fmt) {
1776                 char *fmtname = "";
1777
1778                 switch (sbi->s_jquota_fmt) {
1779                 case QFMT_VFS_OLD:
1780                         fmtname = "vfsold";
1781                         break;
1782                 case QFMT_VFS_V0:
1783                         fmtname = "vfsv0";
1784                         break;
1785                 case QFMT_VFS_V1:
1786                         fmtname = "vfsv1";
1787                         break;
1788                 }
1789                 seq_printf(seq, ",jqfmt=%s", fmtname);
1790         }
1791
1792         if (sbi->s_qf_names[USRQUOTA])
1793                 seq_show_option(seq, "usrjquota", sbi->s_qf_names[USRQUOTA]);
1794
1795         if (sbi->s_qf_names[GRPQUOTA])
1796                 seq_show_option(seq, "grpjquota", sbi->s_qf_names[GRPQUOTA]);
1797 #endif
1798 }
1799
1800 static const char *token2str(int token)
1801 {
1802         const struct match_token *t;
1803
1804         for (t = tokens; t->token != Opt_err; t++)
1805                 if (t->token == token && !strchr(t->pattern, '='))
1806                         break;
1807         return t->pattern;
1808 }
1809
1810 /*
1811  * Show an option if
1812  *  - it's set to a non-default value OR
1813  *  - if the per-sb default is different from the global default
1814  */
1815 static int _ext4_show_options(struct seq_file *seq, struct super_block *sb,
1816                               int nodefs)
1817 {
1818         struct ext4_sb_info *sbi = EXT4_SB(sb);
1819         struct ext4_super_block *es = sbi->s_es;
1820         int def_errors, def_mount_opt = nodefs ? 0 : sbi->s_def_mount_opt;
1821         const struct mount_opts *m;
1822         char sep = nodefs ? '\n' : ',';
1823
1824 #define SEQ_OPTS_PUTS(str) seq_printf(seq, "%c" str, sep)
1825 #define SEQ_OPTS_PRINT(str, arg) seq_printf(seq, "%c" str, sep, arg)
1826
1827         if (sbi->s_sb_block != 1)
1828                 SEQ_OPTS_PRINT("sb=%llu", sbi->s_sb_block);
1829
1830         for (m = ext4_mount_opts; m->token != Opt_err; m++) {
1831                 int want_set = m->flags & MOPT_SET;
1832                 if (((m->flags & (MOPT_SET|MOPT_CLEAR)) == 0) ||
1833                     (m->flags & MOPT_CLEAR_ERR))
1834                         continue;
1835                 if (!(m->mount_opt & (sbi->s_mount_opt ^ def_mount_opt)))
1836                         continue; /* skip if same as the default */
1837                 if ((want_set &&
1838                      (sbi->s_mount_opt & m->mount_opt) != m->mount_opt) ||
1839                     (!want_set && (sbi->s_mount_opt & m->mount_opt)))
1840                         continue; /* select Opt_noFoo vs Opt_Foo */
1841                 SEQ_OPTS_PRINT("%s", token2str(m->token));
1842         }
1843
1844         if (nodefs || !uid_eq(sbi->s_resuid, make_kuid(&init_user_ns, EXT4_DEF_RESUID)) ||
1845             le16_to_cpu(es->s_def_resuid) != EXT4_DEF_RESUID)
1846                 SEQ_OPTS_PRINT("resuid=%u",
1847                                 from_kuid_munged(&init_user_ns, sbi->s_resuid));
1848         if (nodefs || !gid_eq(sbi->s_resgid, make_kgid(&init_user_ns, EXT4_DEF_RESGID)) ||
1849             le16_to_cpu(es->s_def_resgid) != EXT4_DEF_RESGID)
1850                 SEQ_OPTS_PRINT("resgid=%u",
1851                                 from_kgid_munged(&init_user_ns, sbi->s_resgid));
1852         def_errors = nodefs ? -1 : le16_to_cpu(es->s_errors);
1853         if (test_opt(sb, ERRORS_RO) && def_errors != EXT4_ERRORS_RO)
1854                 SEQ_OPTS_PUTS("errors=remount-ro");
1855         if (test_opt(sb, ERRORS_CONT) && def_errors != EXT4_ERRORS_CONTINUE)
1856                 SEQ_OPTS_PUTS("errors=continue");
1857         if (test_opt(sb, ERRORS_PANIC) && def_errors != EXT4_ERRORS_PANIC)
1858                 SEQ_OPTS_PUTS("errors=panic");
1859         if (nodefs || sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ)
1860                 SEQ_OPTS_PRINT("commit=%lu", sbi->s_commit_interval / HZ);
1861         if (nodefs || sbi->s_min_batch_time != EXT4_DEF_MIN_BATCH_TIME)
1862                 SEQ_OPTS_PRINT("min_batch_time=%u", sbi->s_min_batch_time);
1863         if (nodefs || sbi->s_max_batch_time != EXT4_DEF_MAX_BATCH_TIME)
1864                 SEQ_OPTS_PRINT("max_batch_time=%u", sbi->s_max_batch_time);
1865         if (sb->s_flags & MS_I_VERSION)
1866                 SEQ_OPTS_PUTS("i_version");
1867         if (nodefs || sbi->s_stripe)
1868                 SEQ_OPTS_PRINT("stripe=%lu", sbi->s_stripe);
1869         if (EXT4_MOUNT_DATA_FLAGS & (sbi->s_mount_opt ^ def_mount_opt)) {
1870                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
1871                         SEQ_OPTS_PUTS("data=journal");
1872                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
1873                         SEQ_OPTS_PUTS("data=ordered");
1874                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_WRITEBACK_DATA)
1875                         SEQ_OPTS_PUTS("data=writeback");
1876         }
1877         if (nodefs ||
1878             sbi->s_inode_readahead_blks != EXT4_DEF_INODE_READAHEAD_BLKS)
1879                 SEQ_OPTS_PRINT("inode_readahead_blks=%u",
1880                                sbi->s_inode_readahead_blks);
1881
1882         if (nodefs || (test_opt(sb, INIT_INODE_TABLE) &&
1883                        (sbi->s_li_wait_mult != EXT4_DEF_LI_WAIT_MULT)))
1884                 SEQ_OPTS_PRINT("init_itable=%u", sbi->s_li_wait_mult);
1885         if (nodefs || sbi->s_max_dir_size_kb)
1886                 SEQ_OPTS_PRINT("max_dir_size_kb=%u", sbi->s_max_dir_size_kb);
1887
1888         ext4_show_quota_options(seq, sb);
1889         return 0;
1890 }
1891
1892 static int ext4_show_options(struct seq_file *seq, struct dentry *root)
1893 {
1894         return _ext4_show_options(seq, root->d_sb, 0);
1895 }
1896
1897 int ext4_seq_options_show(struct seq_file *seq, void *offset)
1898 {
1899         struct super_block *sb = seq->private;
1900         int rc;
1901
1902         seq_puts(seq, (sb->s_flags & MS_RDONLY) ? "ro" : "rw");
1903         rc = _ext4_show_options(seq, sb, 1);
1904         seq_puts(seq, "\n");
1905         return rc;
1906 }
1907
1908 static int ext4_setup_super(struct super_block *sb, struct ext4_super_block *es,
1909                             int read_only)
1910 {
1911         struct ext4_sb_info *sbi = EXT4_SB(sb);
1912         int res = 0;
1913
1914         if (le32_to_cpu(es->s_rev_level) > EXT4_MAX_SUPP_REV) {
1915                 ext4_msg(sb, KERN_ERR, "revision level too high, "
1916                          "forcing read-only mode");
1917                 res = MS_RDONLY;
1918         }
1919         if (read_only)
1920                 goto done;
1921         if (!(sbi->s_mount_state & EXT4_VALID_FS))
1922                 ext4_msg(sb, KERN_WARNING, "warning: mounting unchecked fs, "
1923                          "running e2fsck is recommended");
1924         else if (sbi->s_mount_state & EXT4_ERROR_FS)
1925                 ext4_msg(sb, KERN_WARNING,
1926                          "warning: mounting fs with errors, "
1927                          "running e2fsck is recommended");
1928         else if ((__s16) le16_to_cpu(es->s_max_mnt_count) > 0 &&
1929                  le16_to_cpu(es->s_mnt_count) >=
1930                  (unsigned short) (__s16) le16_to_cpu(es->s_max_mnt_count))
1931                 ext4_msg(sb, KERN_WARNING,
1932                          "warning: maximal mount count reached, "
1933                          "running e2fsck is recommended");
1934         else if (le32_to_cpu(es->s_checkinterval) &&
1935                 (le32_to_cpu(es->s_lastcheck) +
1936                         le32_to_cpu(es->s_checkinterval) <= get_seconds()))
1937                 ext4_msg(sb, KERN_WARNING,
1938                          "warning: checktime reached, "
1939                          "running e2fsck is recommended");
1940         if (!sbi->s_journal)
1941                 es->s_state &= cpu_to_le16(~EXT4_VALID_FS);
1942         if (!(__s16) le16_to_cpu(es->s_max_mnt_count))
1943                 es->s_max_mnt_count = cpu_to_le16(EXT4_DFL_MAX_MNT_COUNT);
1944         le16_add_cpu(&es->s_mnt_count, 1);
1945         es->s_mtime = cpu_to_le32(get_seconds());
1946         ext4_update_dynamic_rev(sb);
1947         if (sbi->s_journal)
1948                 ext4_set_feature_journal_needs_recovery(sb);
1949
1950         ext4_commit_super(sb, 1);
1951 done:
1952         if (test_opt(sb, DEBUG))
1953                 printk(KERN_INFO "[EXT4 FS bs=%lu, gc=%u, "
1954                                 "bpg=%lu, ipg=%lu, mo=%04x, mo2=%04x]\n",
1955                         sb->s_blocksize,
1956                         sbi->s_groups_count,
1957                         EXT4_BLOCKS_PER_GROUP(sb),
1958                         EXT4_INODES_PER_GROUP(sb),
1959                         sbi->s_mount_opt, sbi->s_mount_opt2);
1960
1961         cleancache_init_fs(sb);
1962         return res;
1963 }
1964
1965 int ext4_alloc_flex_bg_array(struct super_block *sb, ext4_group_t ngroup)
1966 {
1967         struct ext4_sb_info *sbi = EXT4_SB(sb);
1968         struct flex_groups *new_groups;
1969         int size;
1970
1971         if (!sbi->s_log_groups_per_flex)
1972                 return 0;
1973
1974         size = ext4_flex_group(sbi, ngroup - 1) + 1;
1975         if (size <= sbi->s_flex_groups_allocated)
1976                 return 0;
1977
1978         size = roundup_pow_of_two(size * sizeof(struct flex_groups));
1979         new_groups = ext4_kvzalloc(size, GFP_KERNEL);
1980         if (!new_groups) {
1981                 ext4_msg(sb, KERN_ERR, "not enough memory for %d flex groups",
1982                          size / (int) sizeof(struct flex_groups));
1983                 return -ENOMEM;
1984         }
1985
1986         if (sbi->s_flex_groups) {
1987                 memcpy(new_groups, sbi->s_flex_groups,
1988                        (sbi->s_flex_groups_allocated *
1989                         sizeof(struct flex_groups)));
1990                 kvfree(sbi->s_flex_groups);
1991         }
1992         sbi->s_flex_groups = new_groups;
1993         sbi->s_flex_groups_allocated = size / sizeof(struct flex_groups);
1994         return 0;
1995 }
1996
1997 static int ext4_fill_flex_info(struct super_block *sb)
1998 {
1999         struct ext4_sb_info *sbi = EXT4_SB(sb);
2000         struct ext4_group_desc *gdp = NULL;
2001         ext4_group_t flex_group;
2002         int i, err;
2003
2004         sbi->s_log_groups_per_flex = sbi->s_es->s_log_groups_per_flex;
2005         if (sbi->s_log_groups_per_flex < 1 || sbi->s_log_groups_per_flex > 31) {
2006                 sbi->s_log_groups_per_flex = 0;
2007                 return 1;
2008         }
2009
2010         err = ext4_alloc_flex_bg_array(sb, sbi->s_groups_count);
2011         if (err)
2012                 goto failed;
2013
2014         for (i = 0; i < sbi->s_groups_count; i++) {
2015                 gdp = ext4_get_group_desc(sb, i, NULL);
2016
2017                 flex_group = ext4_flex_group(sbi, i);
2018                 atomic_add(ext4_free_inodes_count(sb, gdp),
2019                            &sbi->s_flex_groups[flex_group].free_inodes);
2020                 atomic64_add(ext4_free_group_clusters(sb, gdp),
2021                              &sbi->s_flex_groups[flex_group].free_clusters);
2022                 atomic_add(ext4_used_dirs_count(sb, gdp),
2023                            &sbi->s_flex_groups[flex_group].used_dirs);
2024         }
2025
2026         return 1;
2027 failed:
2028         return 0;
2029 }
2030
2031 static __le16 ext4_group_desc_csum(struct super_block *sb, __u32 block_group,
2032                                    struct ext4_group_desc *gdp)
2033 {
2034         int offset = offsetof(struct ext4_group_desc, bg_checksum);
2035         __u16 crc = 0;
2036         __le32 le_group = cpu_to_le32(block_group);
2037         struct ext4_sb_info *sbi = EXT4_SB(sb);
2038
2039         if (ext4_has_metadata_csum(sbi->s_sb)) {
2040                 /* Use new metadata_csum algorithm */
2041                 __u32 csum32;
2042                 __u16 dummy_csum = 0;
2043
2044                 csum32 = ext4_chksum(sbi, sbi->s_csum_seed, (__u8 *)&le_group,
2045                                      sizeof(le_group));
2046                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp, offset);
2047                 csum32 = ext4_chksum(sbi, csum32, (__u8 *)&dummy_csum,
2048                                      sizeof(dummy_csum));
2049                 offset += sizeof(dummy_csum);
2050                 if (offset < sbi->s_desc_size)
2051                         csum32 = ext4_chksum(sbi, csum32, (__u8 *)gdp + offset,
2052                                              sbi->s_desc_size - offset);
2053
2054                 crc = csum32 & 0xFFFF;
2055                 goto out;
2056         }
2057
2058         /* old crc16 code */
2059         if (!ext4_has_feature_gdt_csum(sb))
2060                 return 0;
2061
2062         crc = crc16(~0, sbi->s_es->s_uuid, sizeof(sbi->s_es->s_uuid));
2063         crc = crc16(crc, (__u8 *)&le_group, sizeof(le_group));
2064         crc = crc16(crc, (__u8 *)gdp, offset);
2065         offset += sizeof(gdp->bg_checksum); /* skip checksum */
2066         /* for checksum of struct ext4_group_desc do the rest...*/
2067         if (ext4_has_feature_64bit(sb) &&
2068             offset < le16_to_cpu(sbi->s_es->s_desc_size))
2069                 crc = crc16(crc, (__u8 *)gdp + offset,
2070                             le16_to_cpu(sbi->s_es->s_desc_size) -
2071                                 offset);
2072
2073 out:
2074         return cpu_to_le16(crc);
2075 }
2076
2077 int ext4_group_desc_csum_verify(struct super_block *sb, __u32 block_group,
2078                                 struct ext4_group_desc *gdp)
2079 {
2080         if (ext4_has_group_desc_csum(sb) &&
2081             (gdp->bg_checksum != ext4_group_desc_csum(sb, block_group, gdp)))
2082                 return 0;
2083
2084         return 1;
2085 }
2086
2087 void ext4_group_desc_csum_set(struct super_block *sb, __u32 block_group,
2088                               struct ext4_group_desc *gdp)
2089 {
2090         if (!ext4_has_group_desc_csum(sb))
2091                 return;
2092         gdp->bg_checksum = ext4_group_desc_csum(sb, block_group, gdp);
2093 }
2094
2095 /* Called at mount-time, super-block is locked */
2096 static int ext4_check_descriptors(struct super_block *sb,
2097                                   ext4_fsblk_t sb_block,
2098                                   ext4_group_t *first_not_zeroed)
2099 {
2100         struct ext4_sb_info *sbi = EXT4_SB(sb);
2101         ext4_fsblk_t first_block = le32_to_cpu(sbi->s_es->s_first_data_block);
2102         ext4_fsblk_t last_block;
2103         ext4_fsblk_t block_bitmap;
2104         ext4_fsblk_t inode_bitmap;
2105         ext4_fsblk_t inode_table;
2106         int flexbg_flag = 0;
2107         ext4_group_t i, grp = sbi->s_groups_count;
2108
2109         if (ext4_has_feature_flex_bg(sb))
2110                 flexbg_flag = 1;
2111
2112         ext4_debug("Checking group descriptors");
2113
2114         for (i = 0; i < sbi->s_groups_count; i++) {
2115                 struct ext4_group_desc *gdp = ext4_get_group_desc(sb, i, NULL);
2116
2117                 if (i == sbi->s_groups_count - 1 || flexbg_flag)
2118                         last_block = ext4_blocks_count(sbi->s_es) - 1;
2119                 else
2120                         last_block = first_block +
2121                                 (EXT4_BLOCKS_PER_GROUP(sb) - 1);
2122
2123                 if ((grp == sbi->s_groups_count) &&
2124                    !(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2125                         grp = i;
2126
2127                 block_bitmap = ext4_block_bitmap(sb, gdp);
2128                 if (block_bitmap == sb_block) {
2129                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2130                                  "Block bitmap for group %u overlaps "
2131                                  "superblock", i);
2132                 }
2133                 if (block_bitmap < first_block || block_bitmap > last_block) {
2134                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2135                                "Block bitmap for group %u not in group "
2136                                "(block %llu)!", i, block_bitmap);
2137                         return 0;
2138                 }
2139                 inode_bitmap = ext4_inode_bitmap(sb, gdp);
2140                 if (inode_bitmap == sb_block) {
2141                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2142                                  "Inode bitmap for group %u overlaps "
2143                                  "superblock", i);
2144                 }
2145                 if (inode_bitmap < first_block || inode_bitmap > last_block) {
2146                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2147                                "Inode bitmap for group %u not in group "
2148                                "(block %llu)!", i, inode_bitmap);
2149                         return 0;
2150                 }
2151                 inode_table = ext4_inode_table(sb, gdp);
2152                 if (inode_table == sb_block) {
2153                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2154                                  "Inode table for group %u overlaps "
2155                                  "superblock", i);
2156                 }
2157                 if (inode_table < first_block ||
2158                     inode_table + sbi->s_itb_per_group - 1 > last_block) {
2159                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2160                                "Inode table for group %u not in group "
2161                                "(block %llu)!", i, inode_table);
2162                         return 0;
2163                 }
2164                 ext4_lock_group(sb, i);
2165                 if (!ext4_group_desc_csum_verify(sb, i, gdp)) {
2166                         ext4_msg(sb, KERN_ERR, "ext4_check_descriptors: "
2167                                  "Checksum for group %u failed (%u!=%u)",
2168                                  i, le16_to_cpu(ext4_group_desc_csum(sb, i,
2169                                      gdp)), le16_to_cpu(gdp->bg_checksum));
2170                         if (!(sb->s_flags & MS_RDONLY)) {
2171                                 ext4_unlock_group(sb, i);
2172                                 return 0;
2173                         }
2174                 }
2175                 ext4_unlock_group(sb, i);
2176                 if (!flexbg_flag)
2177                         first_block += EXT4_BLOCKS_PER_GROUP(sb);
2178         }
2179         if (NULL != first_not_zeroed)
2180                 *first_not_zeroed = grp;
2181         return 1;
2182 }
2183
2184 /* ext4_orphan_cleanup() walks a singly-linked list of inodes (starting at
2185  * the superblock) which were deleted from all directories, but held open by
2186  * a process at the time of a crash.  We walk the list and try to delete these
2187  * inodes at recovery time (only with a read-write filesystem).
2188  *
2189  * In order to keep the orphan inode chain consistent during traversal (in
2190  * case of crash during recovery), we link each inode into the superblock
2191  * orphan list_head and handle it the same way as an inode deletion during
2192  * normal operation (which journals the operations for us).
2193  *
2194  * We only do an iget() and an iput() on each inode, which is very safe if we
2195  * accidentally point at an in-use or already deleted inode.  The worst that
2196  * can happen in this case is that we get a "bit already cleared" message from
2197  * ext4_free_inode().  The only reason we would point at a wrong inode is if
2198  * e2fsck was run on this filesystem, and it must have already done the orphan
2199  * inode cleanup for us, so we can safely abort without any further action.
2200  */
2201 static void ext4_orphan_cleanup(struct super_block *sb,
2202                                 struct ext4_super_block *es)
2203 {
2204         unsigned int s_flags = sb->s_flags;
2205         int nr_orphans = 0, nr_truncates = 0;
2206 #ifdef CONFIG_QUOTA
2207         int quota_update = 0;
2208         int i;
2209 #endif
2210         if (!es->s_last_orphan) {
2211                 jbd_debug(4, "no orphan inodes to clean up\n");
2212                 return;
2213         }
2214
2215         if (bdev_read_only(sb->s_bdev)) {
2216                 ext4_msg(sb, KERN_ERR, "write access "
2217                         "unavailable, skipping orphan cleanup");
2218                 return;
2219         }
2220
2221         /* Check if feature set would not allow a r/w mount */
2222         if (!ext4_feature_set_ok(sb, 0)) {
2223                 ext4_msg(sb, KERN_INFO, "Skipping orphan cleanup due to "
2224                          "unknown ROCOMPAT features");
2225                 return;
2226         }
2227
2228         if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2229                 /* don't clear list on RO mount w/ errors */
2230                 if (es->s_last_orphan && !(s_flags & MS_RDONLY)) {
2231                         ext4_msg(sb, KERN_INFO, "Errors on filesystem, "
2232                                   "clearing orphan list.\n");
2233                         es->s_last_orphan = 0;
2234                 }
2235                 jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2236                 return;
2237         }
2238
2239         if (s_flags & MS_RDONLY) {
2240                 ext4_msg(sb, KERN_INFO, "orphan cleanup on readonly fs");
2241                 sb->s_flags &= ~MS_RDONLY;
2242         }
2243 #ifdef CONFIG_QUOTA
2244         /* Needed for iput() to work correctly and not trash data */
2245         sb->s_flags |= MS_ACTIVE;
2246
2247         /*
2248          * Turn on quotas which were not enabled for read-only mounts if
2249          * filesystem has quota feature, so that they are updated correctly.
2250          */
2251         if (ext4_has_feature_quota(sb) && (s_flags & MS_RDONLY)) {
2252                 int ret = ext4_enable_quotas(sb);
2253
2254                 if (!ret)
2255                         quota_update = 1;
2256                 else
2257                         ext4_msg(sb, KERN_ERR,
2258                                 "Cannot turn on quotas: error %d", ret);
2259         }
2260
2261         /* Turn on journaled quotas used for old sytle */
2262         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2263                 if (EXT4_SB(sb)->s_qf_names[i]) {
2264                         int ret = ext4_quota_on_mount(sb, i);
2265
2266                         if (!ret)
2267                                 quota_update = 1;
2268                         else
2269                                 ext4_msg(sb, KERN_ERR,
2270                                         "Cannot turn on journaled "
2271                                         "quota: type %d: error %d", i, ret);
2272                 }
2273         }
2274 #endif
2275
2276         while (es->s_last_orphan) {
2277                 struct inode *inode;
2278
2279                 /*
2280                  * We may have encountered an error during cleanup; if
2281                  * so, skip the rest.
2282                  */
2283                 if (EXT4_SB(sb)->s_mount_state & EXT4_ERROR_FS) {
2284                         jbd_debug(1, "Skipping orphan recovery on fs with errors.\n");
2285                         es->s_last_orphan = 0;
2286                         break;
2287                 }
2288
2289                 inode = ext4_orphan_get(sb, le32_to_cpu(es->s_last_orphan));
2290                 if (IS_ERR(inode)) {
2291                         es->s_last_orphan = 0;
2292                         break;
2293                 }
2294
2295                 list_add(&EXT4_I(inode)->i_orphan, &EXT4_SB(sb)->s_orphan);
2296                 dquot_initialize(inode);
2297                 if (inode->i_nlink) {
2298                         if (test_opt(sb, DEBUG))
2299                                 ext4_msg(sb, KERN_DEBUG,
2300                                         "%s: truncating inode %lu to %lld bytes",
2301                                         __func__, inode->i_ino, inode->i_size);
2302                         jbd_debug(2, "truncating inode %lu to %lld bytes\n",
2303                                   inode->i_ino, inode->i_size);
2304                         mutex_lock(&inode->i_mutex);
2305                         truncate_inode_pages(inode->i_mapping, inode->i_size);
2306                         ext4_truncate(inode);
2307                         mutex_unlock(&inode->i_mutex);
2308                         nr_truncates++;
2309                 } else {
2310                         if (test_opt(sb, DEBUG))
2311                                 ext4_msg(sb, KERN_DEBUG,
2312                                         "%s: deleting unreferenced inode %lu",
2313                                         __func__, inode->i_ino);
2314                         jbd_debug(2, "deleting unreferenced inode %lu\n",
2315                                   inode->i_ino);
2316                         nr_orphans++;
2317                 }
2318                 iput(inode);  /* The delete magic happens here! */
2319         }
2320
2321 #define PLURAL(x) (x), ((x) == 1) ? "" : "s"
2322
2323         if (nr_orphans)
2324                 ext4_msg(sb, KERN_INFO, "%d orphan inode%s deleted",
2325                        PLURAL(nr_orphans));
2326         if (nr_truncates)
2327                 ext4_msg(sb, KERN_INFO, "%d truncate%s cleaned up",
2328                        PLURAL(nr_truncates));
2329 #ifdef CONFIG_QUOTA
2330         /* Turn off quotas if they were enabled for orphan cleanup */
2331         if (quota_update) {
2332                 for (i = 0; i < EXT4_MAXQUOTAS; i++) {
2333                         if (sb_dqopt(sb)->files[i])
2334                                 dquot_quota_off(sb, i);
2335                 }
2336         }
2337 #endif
2338         sb->s_flags = s_flags; /* Restore MS_RDONLY status */
2339 }
2340
2341 /*
2342  * Maximal extent format file size.
2343  * Resulting logical blkno at s_maxbytes must fit in our on-disk
2344  * extent format containers, within a sector_t, and within i_blocks
2345  * in the vfs.  ext4 inode has 48 bits of i_block in fsblock units,
2346  * so that won't be a limiting factor.
2347  *
2348  * However there is other limiting factor. We do store extents in the form
2349  * of starting block and length, hence the resulting length of the extent
2350  * covering maximum file size must fit into on-disk format containers as
2351  * well. Given that length is always by 1 unit bigger than max unit (because
2352  * we count 0 as well) we have to lower the s_maxbytes by one fs block.
2353  *
2354  * Note, this does *not* consider any metadata overhead for vfs i_blocks.
2355  */
2356 static loff_t ext4_max_size(int blkbits, int has_huge_files)
2357 {
2358         loff_t res;
2359         loff_t upper_limit = MAX_LFS_FILESIZE;
2360
2361         /* small i_blocks in vfs inode? */
2362         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2363                 /*
2364                  * CONFIG_LBDAF is not enabled implies the inode
2365                  * i_block represent total blocks in 512 bytes
2366                  * 32 == size of vfs inode i_blocks * 8
2367                  */
2368                 upper_limit = (1LL << 32) - 1;
2369
2370                 /* total blocks in file system block size */
2371                 upper_limit >>= (blkbits - 9);
2372                 upper_limit <<= blkbits;
2373         }
2374
2375         /*
2376          * 32-bit extent-start container, ee_block. We lower the maxbytes
2377          * by one fs block, so ee_len can cover the extent of maximum file
2378          * size
2379          */
2380         res = (1LL << 32) - 1;
2381         res <<= blkbits;
2382
2383         /* Sanity check against vm- & vfs- imposed limits */
2384         if (res > upper_limit)
2385                 res = upper_limit;
2386
2387         return res;
2388 }
2389
2390 /*
2391  * Maximal bitmap file size.  There is a direct, and {,double-,triple-}indirect
2392  * block limit, and also a limit of (2^48 - 1) 512-byte sectors in i_blocks.
2393  * We need to be 1 filesystem block less than the 2^48 sector limit.
2394  */
2395 static loff_t ext4_max_bitmap_size(int bits, int has_huge_files)
2396 {
2397         loff_t res = EXT4_NDIR_BLOCKS;
2398         int meta_blocks;
2399         loff_t upper_limit;
2400         /* This is calculated to be the largest file size for a dense, block
2401          * mapped file such that the file's total number of 512-byte sectors,
2402          * including data and all indirect blocks, does not exceed (2^48 - 1).
2403          *
2404          * __u32 i_blocks_lo and _u16 i_blocks_high represent the total
2405          * number of 512-byte sectors of the file.
2406          */
2407
2408         if (!has_huge_files || sizeof(blkcnt_t) < sizeof(u64)) {
2409                 /*
2410                  * !has_huge_files or CONFIG_LBDAF not enabled implies that
2411                  * the inode i_block field represents total file blocks in
2412                  * 2^32 512-byte sectors == size of vfs inode i_blocks * 8
2413                  */
2414                 upper_limit = (1LL << 32) - 1;
2415
2416                 /* total blocks in file system block size */
2417                 upper_limit >>= (bits - 9);
2418
2419         } else {
2420                 /*
2421                  * We use 48 bit ext4_inode i_blocks
2422                  * With EXT4_HUGE_FILE_FL set the i_blocks
2423                  * represent total number of blocks in
2424                  * file system block size
2425                  */
2426                 upper_limit = (1LL << 48) - 1;
2427
2428         }
2429
2430         /* indirect blocks */
2431         meta_blocks = 1;
2432         /* double indirect blocks */
2433         meta_blocks += 1 + (1LL << (bits-2));
2434         /* tripple indirect blocks */
2435         meta_blocks += 1 + (1LL << (bits-2)) + (1LL << (2*(bits-2)));
2436
2437         upper_limit -= meta_blocks;
2438         upper_limit <<= bits;
2439
2440         res += 1LL << (bits-2);
2441         res += 1LL << (2*(bits-2));
2442         res += 1LL << (3*(bits-2));
2443         res <<= bits;
2444         if (res > upper_limit)
2445                 res = upper_limit;
2446
2447         if (res > MAX_LFS_FILESIZE)
2448                 res = MAX_LFS_FILESIZE;
2449
2450         return res;
2451 }
2452
2453 static ext4_fsblk_t descriptor_loc(struct super_block *sb,
2454                                    ext4_fsblk_t logical_sb_block, int nr)
2455 {
2456         struct ext4_sb_info *sbi = EXT4_SB(sb);
2457         ext4_group_t bg, first_meta_bg;
2458         int has_super = 0;
2459
2460         first_meta_bg = le32_to_cpu(sbi->s_es->s_first_meta_bg);
2461
2462         if (!ext4_has_feature_meta_bg(sb) || nr < first_meta_bg)
2463                 return logical_sb_block + nr + 1;
2464         bg = sbi->s_desc_per_block * nr;
2465         if (ext4_bg_has_super(sb, bg))
2466                 has_super = 1;
2467
2468         /*
2469          * If we have a meta_bg fs with 1k blocks, group 0's GDT is at
2470          * block 2, not 1.  If s_first_data_block == 0 (bigalloc is enabled
2471          * on modern mke2fs or blksize > 1k on older mke2fs) then we must
2472          * compensate.
2473          */
2474         if (sb->s_blocksize == 1024 && nr == 0 &&
2475             le32_to_cpu(EXT4_SB(sb)->s_es->s_first_data_block) == 0)
2476                 has_super++;
2477
2478         return (has_super + ext4_group_first_block_no(sb, bg));
2479 }
2480
2481 /**
2482  * ext4_get_stripe_size: Get the stripe size.
2483  * @sbi: In memory super block info
2484  *
2485  * If we have specified it via mount option, then
2486  * use the mount option value. If the value specified at mount time is
2487  * greater than the blocks per group use the super block value.
2488  * If the super block value is greater than blocks per group return 0.
2489  * Allocator needs it be less than blocks per group.
2490  *
2491  */
2492 static unsigned long ext4_get_stripe_size(struct ext4_sb_info *sbi)
2493 {
2494         unsigned long stride = le16_to_cpu(sbi->s_es->s_raid_stride);
2495         unsigned long stripe_width =
2496                         le32_to_cpu(sbi->s_es->s_raid_stripe_width);
2497         int ret;
2498
2499         if (sbi->s_stripe && sbi->s_stripe <= sbi->s_blocks_per_group)
2500                 ret = sbi->s_stripe;
2501         else if (stripe_width && stripe_width <= sbi->s_blocks_per_group)
2502                 ret = stripe_width;
2503         else if (stride && stride <= sbi->s_blocks_per_group)
2504                 ret = stride;
2505         else
2506                 ret = 0;
2507
2508         /*
2509          * If the stripe width is 1, this makes no sense and
2510          * we set it to 0 to turn off stripe handling code.
2511          */
2512         if (ret <= 1)
2513                 ret = 0;
2514
2515         return ret;
2516 }
2517
2518 /*
2519  * Check whether this filesystem can be mounted based on
2520  * the features present and the RDONLY/RDWR mount requested.
2521  * Returns 1 if this filesystem can be mounted as requested,
2522  * 0 if it cannot be.
2523  */
2524 static int ext4_feature_set_ok(struct super_block *sb, int readonly)
2525 {
2526         if (ext4_has_unknown_ext4_incompat_features(sb)) {
2527                 ext4_msg(sb, KERN_ERR,
2528                         "Couldn't mount because of "
2529                         "unsupported optional features (%x)",
2530                         (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_incompat) &
2531                         ~EXT4_FEATURE_INCOMPAT_SUPP));
2532                 return 0;
2533         }
2534
2535         if (readonly)
2536                 return 1;
2537
2538         if (ext4_has_feature_readonly(sb)) {
2539                 ext4_msg(sb, KERN_INFO, "filesystem is read-only");
2540                 sb->s_flags |= MS_RDONLY;
2541                 return 1;
2542         }
2543
2544         /* Check that feature set is OK for a read-write mount */
2545         if (ext4_has_unknown_ext4_ro_compat_features(sb)) {
2546                 ext4_msg(sb, KERN_ERR, "couldn't mount RDWR because of "
2547                          "unsupported optional features (%x)",
2548                          (le32_to_cpu(EXT4_SB(sb)->s_es->s_feature_ro_compat) &
2549                                 ~EXT4_FEATURE_RO_COMPAT_SUPP));
2550                 return 0;
2551         }
2552         /*
2553          * Large file size enabled file system can only be mounted
2554          * read-write on 32-bit systems if kernel is built with CONFIG_LBDAF
2555          */
2556         if (ext4_has_feature_huge_file(sb)) {
2557                 if (sizeof(blkcnt_t) < sizeof(u64)) {
2558                         ext4_msg(sb, KERN_ERR, "Filesystem with huge files "
2559                                  "cannot be mounted RDWR without "
2560                                  "CONFIG_LBDAF");
2561                         return 0;
2562                 }
2563         }
2564         if (ext4_has_feature_bigalloc(sb) && !ext4_has_feature_extents(sb)) {
2565                 ext4_msg(sb, KERN_ERR,
2566                          "Can't support bigalloc feature without "
2567                          "extents feature\n");
2568                 return 0;
2569         }
2570
2571 #ifndef CONFIG_QUOTA
2572         if (ext4_has_feature_quota(sb) && !readonly) {
2573                 ext4_msg(sb, KERN_ERR,
2574                          "Filesystem with quota feature cannot be mounted RDWR "
2575                          "without CONFIG_QUOTA");
2576                 return 0;
2577         }
2578 #endif  /* CONFIG_QUOTA */
2579         return 1;
2580 }
2581
2582 /*
2583  * This function is called once a day if we have errors logged
2584  * on the file system
2585  */
2586 static void print_daily_error_info(unsigned long arg)
2587 {
2588         struct super_block *sb = (struct super_block *) arg;
2589         struct ext4_sb_info *sbi;
2590         struct ext4_super_block *es;
2591
2592         sbi = EXT4_SB(sb);
2593         es = sbi->s_es;
2594
2595         if (es->s_error_count)
2596                 /* fsck newer than v1.41.13 is needed to clean this condition. */
2597                 ext4_msg(sb, KERN_NOTICE, "error count since last fsck: %u",
2598                          le32_to_cpu(es->s_error_count));
2599         if (es->s_first_error_time) {
2600                 printk(KERN_NOTICE "EXT4-fs (%s): initial error at time %u: %.*s:%d",
2601                        sb->s_id, le32_to_cpu(es->s_first_error_time),
2602                        (int) sizeof(es->s_first_error_func),
2603                        es->s_first_error_func,
2604                        le32_to_cpu(es->s_first_error_line));
2605                 if (es->s_first_error_ino)
2606                         printk(": inode %u",
2607                                le32_to_cpu(es->s_first_error_ino));
2608                 if (es->s_first_error_block)
2609                         printk(": block %llu", (unsigned long long)
2610                                le64_to_cpu(es->s_first_error_block));
2611                 printk("\n");
2612         }
2613         if (es->s_last_error_time) {
2614                 printk(KERN_NOTICE "EXT4-fs (%s): last error at time %u: %.*s:%d",
2615                        sb->s_id, le32_to_cpu(es->s_last_error_time),
2616                        (int) sizeof(es->s_last_error_func),
2617                        es->s_last_error_func,
2618                        le32_to_cpu(es->s_last_error_line));
2619                 if (es->s_last_error_ino)
2620                         printk(": inode %u",
2621                                le32_to_cpu(es->s_last_error_ino));
2622                 if (es->s_last_error_block)
2623                         printk(": block %llu", (unsigned long long)
2624                                le64_to_cpu(es->s_last_error_block));
2625                 printk("\n");
2626         }
2627         mod_timer(&sbi->s_err_report, jiffies + 24*60*60*HZ);  /* Once a day */
2628 }
2629
2630 /* Find next suitable group and run ext4_init_inode_table */
2631 static int ext4_run_li_request(struct ext4_li_request *elr)
2632 {
2633         struct ext4_group_desc *gdp = NULL;
2634         ext4_group_t group, ngroups;
2635         struct super_block *sb;
2636         unsigned long timeout = 0;
2637         int ret = 0;
2638
2639         sb = elr->lr_super;
2640         ngroups = EXT4_SB(sb)->s_groups_count;
2641
2642         sb_start_write(sb);
2643         for (group = elr->lr_next_group; group < ngroups; group++) {
2644                 gdp = ext4_get_group_desc(sb, group, NULL);
2645                 if (!gdp) {
2646                         ret = 1;
2647                         break;
2648                 }
2649
2650                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2651                         break;
2652         }
2653
2654         if (group >= ngroups)
2655                 ret = 1;
2656
2657         if (!ret) {
2658                 timeout = jiffies;
2659                 ret = ext4_init_inode_table(sb, group,
2660                                             elr->lr_timeout ? 0 : 1);
2661                 if (elr->lr_timeout == 0) {
2662                         timeout = (jiffies - timeout) *
2663                                   elr->lr_sbi->s_li_wait_mult;
2664                         elr->lr_timeout = timeout;
2665                 }
2666                 elr->lr_next_sched = jiffies + elr->lr_timeout;
2667                 elr->lr_next_group = group + 1;
2668         }
2669         sb_end_write(sb);
2670
2671         return ret;
2672 }
2673
2674 /*
2675  * Remove lr_request from the list_request and free the
2676  * request structure. Should be called with li_list_mtx held
2677  */
2678 static void ext4_remove_li_request(struct ext4_li_request *elr)
2679 {
2680         struct ext4_sb_info *sbi;
2681
2682         if (!elr)
2683                 return;
2684
2685         sbi = elr->lr_sbi;
2686
2687         list_del(&elr->lr_request);
2688         sbi->s_li_request = NULL;
2689         kfree(elr);
2690 }
2691
2692 static void ext4_unregister_li_request(struct super_block *sb)
2693 {
2694         mutex_lock(&ext4_li_mtx);
2695         if (!ext4_li_info) {
2696                 mutex_unlock(&ext4_li_mtx);
2697                 return;
2698         }
2699
2700         mutex_lock(&ext4_li_info->li_list_mtx);
2701         ext4_remove_li_request(EXT4_SB(sb)->s_li_request);
2702         mutex_unlock(&ext4_li_info->li_list_mtx);
2703         mutex_unlock(&ext4_li_mtx);
2704 }
2705
2706 static struct task_struct *ext4_lazyinit_task;
2707
2708 /*
2709  * This is the function where ext4lazyinit thread lives. It walks
2710  * through the request list searching for next scheduled filesystem.
2711  * When such a fs is found, run the lazy initialization request
2712  * (ext4_rn_li_request) and keep track of the time spend in this
2713  * function. Based on that time we compute next schedule time of
2714  * the request. When walking through the list is complete, compute
2715  * next waking time and put itself into sleep.
2716  */
2717 static int ext4_lazyinit_thread(void *arg)
2718 {
2719         struct ext4_lazy_init *eli = (struct ext4_lazy_init *)arg;
2720         struct list_head *pos, *n;
2721         struct ext4_li_request *elr;
2722         unsigned long next_wakeup, cur;
2723
2724         BUG_ON(NULL == eli);
2725
2726 cont_thread:
2727         while (true) {
2728                 next_wakeup = MAX_JIFFY_OFFSET;
2729
2730                 mutex_lock(&eli->li_list_mtx);
2731                 if (list_empty(&eli->li_request_list)) {
2732                         mutex_unlock(&eli->li_list_mtx);
2733                         goto exit_thread;
2734                 }
2735
2736                 list_for_each_safe(pos, n, &eli->li_request_list) {
2737                         elr = list_entry(pos, struct ext4_li_request,
2738                                          lr_request);
2739
2740                         if (time_after_eq(jiffies, elr->lr_next_sched)) {
2741                                 if (ext4_run_li_request(elr) != 0) {
2742                                         /* error, remove the lazy_init job */
2743                                         ext4_remove_li_request(elr);
2744                                         continue;
2745                                 }
2746                         }
2747
2748                         if (time_before(elr->lr_next_sched, next_wakeup))
2749                                 next_wakeup = elr->lr_next_sched;
2750                 }
2751                 mutex_unlock(&eli->li_list_mtx);
2752
2753                 try_to_freeze();
2754
2755                 cur = jiffies;
2756                 if ((time_after_eq(cur, next_wakeup)) ||
2757                     (MAX_JIFFY_OFFSET == next_wakeup)) {
2758                         cond_resched();
2759                         continue;
2760                 }
2761
2762                 schedule_timeout_interruptible(next_wakeup - cur);
2763
2764                 if (kthread_should_stop()) {
2765                         ext4_clear_request_list();
2766                         goto exit_thread;
2767                 }
2768         }
2769
2770 exit_thread:
2771         /*
2772          * It looks like the request list is empty, but we need
2773          * to check it under the li_list_mtx lock, to prevent any
2774          * additions into it, and of course we should lock ext4_li_mtx
2775          * to atomically free the list and ext4_li_info, because at
2776          * this point another ext4 filesystem could be registering
2777          * new one.
2778          */
2779         mutex_lock(&ext4_li_mtx);
2780         mutex_lock(&eli->li_list_mtx);
2781         if (!list_empty(&eli->li_request_list)) {
2782                 mutex_unlock(&eli->li_list_mtx);
2783                 mutex_unlock(&ext4_li_mtx);
2784                 goto cont_thread;
2785         }
2786         mutex_unlock(&eli->li_list_mtx);
2787         kfree(ext4_li_info);
2788         ext4_li_info = NULL;
2789         mutex_unlock(&ext4_li_mtx);
2790
2791         return 0;
2792 }
2793
2794 static void ext4_clear_request_list(void)
2795 {
2796         struct list_head *pos, *n;
2797         struct ext4_li_request *elr;
2798
2799         mutex_lock(&ext4_li_info->li_list_mtx);
2800         list_for_each_safe(pos, n, &ext4_li_info->li_request_list) {
2801                 elr = list_entry(pos, struct ext4_li_request,
2802                                  lr_request);
2803                 ext4_remove_li_request(elr);
2804         }
2805         mutex_unlock(&ext4_li_info->li_list_mtx);
2806 }
2807
2808 static int ext4_run_lazyinit_thread(void)
2809 {
2810         ext4_lazyinit_task = kthread_run(ext4_lazyinit_thread,
2811                                          ext4_li_info, "ext4lazyinit");
2812         if (IS_ERR(ext4_lazyinit_task)) {
2813                 int err = PTR_ERR(ext4_lazyinit_task);
2814                 ext4_clear_request_list();
2815                 kfree(ext4_li_info);
2816                 ext4_li_info = NULL;
2817                 printk(KERN_CRIT "EXT4-fs: error %d creating inode table "
2818                                  "initialization thread\n",
2819                                  err);
2820                 return err;
2821         }
2822         ext4_li_info->li_state |= EXT4_LAZYINIT_RUNNING;
2823         return 0;
2824 }
2825
2826 /*
2827  * Check whether it make sense to run itable init. thread or not.
2828  * If there is at least one uninitialized inode table, return
2829  * corresponding group number, else the loop goes through all
2830  * groups and return total number of groups.
2831  */
2832 static ext4_group_t ext4_has_uninit_itable(struct super_block *sb)
2833 {
2834         ext4_group_t group, ngroups = EXT4_SB(sb)->s_groups_count;
2835         struct ext4_group_desc *gdp = NULL;
2836
2837         for (group = 0; group < ngroups; group++) {
2838                 gdp = ext4_get_group_desc(sb, group, NULL);
2839                 if (!gdp)
2840                         continue;
2841
2842                 if (!(gdp->bg_flags & cpu_to_le16(EXT4_BG_INODE_ZEROED)))
2843                         break;
2844         }
2845
2846         return group;
2847 }
2848
2849 static int ext4_li_info_new(void)
2850 {
2851         struct ext4_lazy_init *eli = NULL;
2852
2853         eli = kzalloc(sizeof(*eli), GFP_KERNEL);
2854         if (!eli)
2855                 return -ENOMEM;
2856
2857         INIT_LIST_HEAD(&eli->li_request_list);
2858         mutex_init(&eli->li_list_mtx);
2859
2860         eli->li_state |= EXT4_LAZYINIT_QUIT;
2861
2862         ext4_li_info = eli;
2863
2864         return 0;
2865 }
2866
2867 static struct ext4_li_request *ext4_li_request_new(struct super_block *sb,
2868                                             ext4_group_t start)
2869 {
2870         struct ext4_sb_info *sbi = EXT4_SB(sb);
2871         struct ext4_li_request *elr;
2872
2873         elr = kzalloc(sizeof(*elr), GFP_KERNEL);
2874         if (!elr)
2875                 return NULL;
2876
2877         elr->lr_super = sb;
2878         elr->lr_sbi = sbi;
2879         elr->lr_next_group = start;
2880
2881         /*
2882          * Randomize first schedule time of the request to
2883          * spread the inode table initialization requests
2884          * better.
2885          */
2886         elr->lr_next_sched = jiffies + (prandom_u32() %
2887                                 (EXT4_DEF_LI_MAX_START_DELAY * HZ));
2888         return elr;
2889 }
2890
2891 int ext4_register_li_request(struct super_block *sb,
2892                              ext4_group_t first_not_zeroed)
2893 {
2894         struct ext4_sb_info *sbi = EXT4_SB(sb);
2895         struct ext4_li_request *elr = NULL;
2896         ext4_group_t ngroups = EXT4_SB(sb)->s_groups_count;
2897         int ret = 0;
2898
2899         mutex_lock(&ext4_li_mtx);
2900         if (sbi->s_li_request != NULL) {
2901                 /*
2902                  * Reset timeout so it can be computed again, because
2903                  * s_li_wait_mult might have changed.
2904                  */
2905                 sbi->s_li_request->lr_timeout = 0;
2906                 goto out;
2907         }
2908
2909         if (first_not_zeroed == ngroups ||
2910             (sb->s_flags & MS_RDONLY) ||
2911             !test_opt(sb, INIT_INODE_TABLE))
2912                 goto out;
2913
2914         elr = ext4_li_request_new(sb, first_not_zeroed);
2915         if (!elr) {
2916                 ret = -ENOMEM;
2917                 goto out;
2918         }
2919
2920         if (NULL == ext4_li_info) {
2921                 ret = ext4_li_info_new();
2922                 if (ret)
2923                         goto out;
2924         }
2925
2926         mutex_lock(&ext4_li_info->li_list_mtx);
2927         list_add(&elr->lr_request, &ext4_li_info->li_request_list);
2928         mutex_unlock(&ext4_li_info->li_list_mtx);
2929
2930         sbi->s_li_request = elr;
2931         /*
2932          * set elr to NULL here since it has been inserted to
2933          * the request_list and the removal and free of it is
2934          * handled by ext4_clear_request_list from now on.
2935          */
2936         elr = NULL;
2937
2938         if (!(ext4_li_info->li_state & EXT4_LAZYINIT_RUNNING)) {
2939                 ret = ext4_run_lazyinit_thread();
2940                 if (ret)
2941                         goto out;
2942         }
2943 out:
2944         mutex_unlock(&ext4_li_mtx);
2945         if (ret)
2946                 kfree(elr);
2947         return ret;
2948 }
2949
2950 /*
2951  * We do not need to lock anything since this is called on
2952  * module unload.
2953  */
2954 static void ext4_destroy_lazyinit_thread(void)
2955 {
2956         /*
2957          * If thread exited earlier
2958          * there's nothing to be done.
2959          */
2960         if (!ext4_li_info || !ext4_lazyinit_task)
2961                 return;
2962
2963         kthread_stop(ext4_lazyinit_task);
2964 }
2965
2966 static int set_journal_csum_feature_set(struct super_block *sb)
2967 {
2968         int ret = 1;
2969         int compat, incompat;
2970         struct ext4_sb_info *sbi = EXT4_SB(sb);
2971
2972         if (ext4_has_metadata_csum(sb)) {
2973                 /* journal checksum v3 */
2974                 compat = 0;
2975                 incompat = JBD2_FEATURE_INCOMPAT_CSUM_V3;
2976         } else {
2977                 /* journal checksum v1 */
2978                 compat = JBD2_FEATURE_COMPAT_CHECKSUM;
2979                 incompat = 0;
2980         }
2981
2982         jbd2_journal_clear_features(sbi->s_journal,
2983                         JBD2_FEATURE_COMPAT_CHECKSUM, 0,
2984                         JBD2_FEATURE_INCOMPAT_CSUM_V3 |
2985                         JBD2_FEATURE_INCOMPAT_CSUM_V2);
2986         if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
2987                 ret = jbd2_journal_set_features(sbi->s_journal,
2988                                 compat, 0,
2989                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT |
2990                                 incompat);
2991         } else if (test_opt(sb, JOURNAL_CHECKSUM)) {
2992                 ret = jbd2_journal_set_features(sbi->s_journal,
2993                                 compat, 0,
2994                                 incompat);
2995                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2996                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
2997         } else {
2998                 jbd2_journal_clear_features(sbi->s_journal, 0, 0,
2999                                 JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT);
3000         }
3001
3002         return ret;
3003 }
3004
3005 /*
3006  * Note: calculating the overhead so we can be compatible with
3007  * historical BSD practice is quite difficult in the face of
3008  * clusters/bigalloc.  This is because multiple metadata blocks from
3009  * different block group can end up in the same allocation cluster.
3010  * Calculating the exact overhead in the face of clustered allocation
3011  * requires either O(all block bitmaps) in memory or O(number of block
3012  * groups**2) in time.  We will still calculate the superblock for
3013  * older file systems --- and if we come across with a bigalloc file
3014  * system with zero in s_overhead_clusters the estimate will be close to
3015  * correct especially for very large cluster sizes --- but for newer
3016  * file systems, it's better to calculate this figure once at mkfs
3017  * time, and store it in the superblock.  If the superblock value is
3018  * present (even for non-bigalloc file systems), we will use it.
3019  */
3020 static int count_overhead(struct super_block *sb, ext4_group_t grp,
3021                           char *buf)
3022 {
3023         struct ext4_sb_info     *sbi = EXT4_SB(sb);
3024         struct ext4_group_desc  *gdp;
3025         ext4_fsblk_t            first_block, last_block, b;
3026         ext4_group_t            i, ngroups = ext4_get_groups_count(sb);
3027         int                     s, j, count = 0;
3028
3029         if (!ext4_has_feature_bigalloc(sb))
3030                 return (ext4_bg_has_super(sb, grp) + ext4_bg_num_gdb(sb, grp) +
3031                         sbi->s_itb_per_group + 2);
3032
3033         first_block = le32_to_cpu(sbi->s_es->s_first_data_block) +
3034                 (grp * EXT4_BLOCKS_PER_GROUP(sb));
3035         last_block = first_block + EXT4_BLOCKS_PER_GROUP(sb) - 1;
3036         for (i = 0; i < ngroups; i++) {
3037                 gdp = ext4_get_group_desc(sb, i, NULL);
3038                 b = ext4_block_bitmap(sb, gdp);
3039                 if (b >= first_block && b <= last_block) {
3040                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3041                         count++;
3042                 }
3043                 b = ext4_inode_bitmap(sb, gdp);
3044                 if (b >= first_block && b <= last_block) {
3045                         ext4_set_bit(EXT4_B2C(sbi, b - first_block), buf);
3046                         count++;
3047                 }
3048                 b = ext4_inode_table(sb, gdp);
3049                 if (b >= first_block && b + sbi->s_itb_per_group <= last_block)
3050                         for (j = 0; j < sbi->s_itb_per_group; j++, b++) {
3051                                 int c = EXT4_B2C(sbi, b - first_block);
3052                                 ext4_set_bit(c, buf);
3053                                 count++;
3054                         }
3055                 if (i != grp)
3056                         continue;
3057                 s = 0;
3058                 if (ext4_bg_has_super(sb, grp)) {
3059                         ext4_set_bit(s++, buf);
3060                         count++;
3061                 }
3062                 j = ext4_bg_num_gdb(sb, grp);
3063                 if (s + j > EXT4_BLOCKS_PER_GROUP(sb)) {
3064                         ext4_error(sb, "Invalid number of block group "
3065                                    "descriptor blocks: %d", j);
3066                         j = EXT4_BLOCKS_PER_GROUP(sb) - s;
3067                 }
3068                 count += j;
3069                 for (; j > 0; j--)
3070                         ext4_set_bit(EXT4_B2C(sbi, s++), buf);
3071         }
3072         if (!count)
3073                 return 0;
3074         return EXT4_CLUSTERS_PER_GROUP(sb) -
3075                 ext4_count_free(buf, EXT4_CLUSTERS_PER_GROUP(sb) / 8);
3076 }
3077
3078 /*
3079  * Compute the overhead and stash it in sbi->s_overhead
3080  */
3081 int ext4_calculate_overhead(struct super_block *sb)
3082 {
3083         struct ext4_sb_info *sbi = EXT4_SB(sb);
3084         struct ext4_super_block *es = sbi->s_es;
3085         ext4_group_t i, ngroups = ext4_get_groups_count(sb);
3086         ext4_fsblk_t overhead = 0;
3087         char *buf = (char *) get_zeroed_page(GFP_NOFS);
3088
3089         if (!buf)
3090                 return -ENOMEM;
3091
3092         /*
3093          * Compute the overhead (FS structures).  This is constant
3094          * for a given filesystem unless the number of block groups
3095          * changes so we cache the previous value until it does.
3096          */
3097
3098         /*
3099          * All of the blocks before first_data_block are overhead
3100          */
3101         overhead = EXT4_B2C(sbi, le32_to_cpu(es->s_first_data_block));
3102
3103         /*
3104          * Add the overhead found in each block group
3105          */
3106         for (i = 0; i < ngroups; i++) {
3107                 int blks;
3108
3109                 blks = count_overhead(sb, i, buf);
3110                 overhead += blks;
3111                 if (blks)
3112                         memset(buf, 0, PAGE_SIZE);
3113                 cond_resched();
3114         }
3115         /* Add the internal journal blocks as well */
3116         if (sbi->s_journal && !sbi->journal_bdev)
3117                 overhead += EXT4_NUM_B2C(sbi, sbi->s_journal->j_maxlen);
3118
3119         sbi->s_overhead = overhead;
3120         smp_wmb();
3121         free_page((unsigned long) buf);
3122         return 0;
3123 }
3124
3125 static void ext4_set_resv_clusters(struct super_block *sb)
3126 {
3127         ext4_fsblk_t resv_clusters;
3128         struct ext4_sb_info *sbi = EXT4_SB(sb);
3129
3130         /*
3131          * There's no need to reserve anything when we aren't using extents.
3132          * The space estimates are exact, there are no unwritten extents,
3133          * hole punching doesn't need new metadata... This is needed especially
3134          * to keep ext2/3 backward compatibility.
3135          */
3136         if (!ext4_has_feature_extents(sb))
3137                 return;
3138         /*
3139          * By default we reserve 2% or 4096 clusters, whichever is smaller.
3140          * This should cover the situations where we can not afford to run
3141          * out of space like for example punch hole, or converting
3142          * unwritten extents in delalloc path. In most cases such
3143          * allocation would require 1, or 2 blocks, higher numbers are
3144          * very rare.
3145          */
3146         resv_clusters = (ext4_blocks_count(sbi->s_es) >>
3147                          sbi->s_cluster_bits);
3148
3149         do_div(resv_clusters, 50);
3150         resv_clusters = min_t(ext4_fsblk_t, resv_clusters, 4096);
3151
3152         atomic64_set(&sbi->s_resv_clusters, resv_clusters);
3153 }
3154
3155 static int ext4_fill_super(struct super_block *sb, void *data, int silent)
3156 {
3157         char *orig_data = kstrdup(data, GFP_KERNEL);
3158         struct buffer_head *bh;
3159         struct ext4_super_block *es = NULL;
3160         struct ext4_sb_info *sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
3161         ext4_fsblk_t block;
3162         ext4_fsblk_t sb_block = get_sb_block(&data);
3163         ext4_fsblk_t logical_sb_block;
3164         unsigned long offset = 0;
3165         unsigned long journal_devnum = 0;
3166         unsigned long def_mount_opts;
3167         struct inode *root;
3168         const char *descr;
3169         int ret = -ENOMEM;
3170         int blocksize, clustersize;
3171         unsigned int db_count;
3172         unsigned int i;
3173         int needs_recovery, has_huge_files, has_bigalloc;
3174         __u64 blocks_count;
3175         int err = 0;
3176         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
3177         ext4_group_t first_not_zeroed;
3178
3179         if ((data && !orig_data) || !sbi)
3180                 goto out_free_base;
3181
3182         sbi->s_blockgroup_lock =
3183                 kzalloc(sizeof(struct blockgroup_lock), GFP_KERNEL);
3184         if (!sbi->s_blockgroup_lock)
3185                 goto out_free_base;
3186
3187         sb->s_fs_info = sbi;
3188         sbi->s_sb = sb;
3189         sbi->s_inode_readahead_blks = EXT4_DEF_INODE_READAHEAD_BLKS;
3190         sbi->s_sb_block = sb_block;
3191         if (sb->s_bdev->bd_part)
3192                 sbi->s_sectors_written_start =
3193                         part_stat_read(sb->s_bdev->bd_part, sectors[1]);
3194
3195         /* Cleanup superblock name */
3196         strreplace(sb->s_id, '/', '!');
3197
3198         /* -EINVAL is default */
3199         ret = -EINVAL;
3200         blocksize = sb_min_blocksize(sb, EXT4_MIN_BLOCK_SIZE);
3201         if (!blocksize) {
3202                 ext4_msg(sb, KERN_ERR, "unable to set blocksize");
3203                 goto out_fail;
3204         }
3205
3206         /*
3207          * The ext4 superblock will not be buffer aligned for other than 1kB
3208          * block sizes.  We need to calculate the offset from buffer start.
3209          */
3210         if (blocksize != EXT4_MIN_BLOCK_SIZE) {
3211                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3212                 offset = do_div(logical_sb_block, blocksize);
3213         } else {
3214                 logical_sb_block = sb_block;
3215         }
3216
3217         if (!(bh = sb_bread_unmovable(sb, logical_sb_block))) {
3218                 ext4_msg(sb, KERN_ERR, "unable to read superblock");
3219                 goto out_fail;
3220         }
3221         /*
3222          * Note: s_es must be initialized as soon as possible because
3223          *       some ext4 macro-instructions depend on its value
3224          */
3225         es = (struct ext4_super_block *) (bh->b_data + offset);
3226         sbi->s_es = es;
3227         sb->s_magic = le16_to_cpu(es->s_magic);
3228         if (sb->s_magic != EXT4_SUPER_MAGIC)
3229                 goto cantfind_ext4;
3230         sbi->s_kbytes_written = le64_to_cpu(es->s_kbytes_written);
3231
3232         /* Warn if metadata_csum and gdt_csum are both set. */
3233         if (ext4_has_feature_metadata_csum(sb) &&
3234             ext4_has_feature_gdt_csum(sb))
3235                 ext4_warning(sb, "metadata_csum and uninit_bg are "
3236                              "redundant flags; please run fsck.");
3237
3238         /* Check for a known checksum algorithm */
3239         if (!ext4_verify_csum_type(sb, es)) {
3240                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3241                          "unknown checksum algorithm.");
3242                 silent = 1;
3243                 goto cantfind_ext4;
3244         }
3245
3246         /* Load the checksum driver */
3247         if (ext4_has_feature_metadata_csum(sb)) {
3248                 sbi->s_chksum_driver = crypto_alloc_shash("crc32c", 0, 0);
3249                 if (IS_ERR(sbi->s_chksum_driver)) {
3250                         ext4_msg(sb, KERN_ERR, "Cannot load crc32c driver.");
3251                         ret = PTR_ERR(sbi->s_chksum_driver);
3252                         sbi->s_chksum_driver = NULL;
3253                         goto failed_mount;
3254                 }
3255         }
3256
3257         /* Check superblock checksum */
3258         if (!ext4_superblock_csum_verify(sb, es)) {
3259                 ext4_msg(sb, KERN_ERR, "VFS: Found ext4 filesystem with "
3260                          "invalid superblock checksum.  Run e2fsck?");
3261                 silent = 1;
3262                 ret = -EFSBADCRC;
3263                 goto cantfind_ext4;
3264         }
3265
3266         /* Precompute checksum seed for all metadata */
3267         if (ext4_has_feature_csum_seed(sb))
3268                 sbi->s_csum_seed = le32_to_cpu(es->s_checksum_seed);
3269         else if (ext4_has_metadata_csum(sb))
3270                 sbi->s_csum_seed = ext4_chksum(sbi, ~0, es->s_uuid,
3271                                                sizeof(es->s_uuid));
3272
3273         /* Set defaults before we parse the mount options */
3274         def_mount_opts = le32_to_cpu(es->s_default_mount_opts);
3275         set_opt(sb, INIT_INODE_TABLE);
3276         if (def_mount_opts & EXT4_DEFM_DEBUG)
3277                 set_opt(sb, DEBUG);
3278         if (def_mount_opts & EXT4_DEFM_BSDGROUPS)
3279                 set_opt(sb, GRPID);
3280         if (def_mount_opts & EXT4_DEFM_UID16)
3281                 set_opt(sb, NO_UID32);
3282         /* xattr user namespace & acls are now defaulted on */
3283         set_opt(sb, XATTR_USER);
3284 #ifdef CONFIG_EXT4_FS_POSIX_ACL
3285         set_opt(sb, POSIX_ACL);
3286 #endif
3287         /* don't forget to enable journal_csum when metadata_csum is enabled. */
3288         if (ext4_has_metadata_csum(sb))
3289                 set_opt(sb, JOURNAL_CHECKSUM);
3290
3291         if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_DATA)
3292                 set_opt(sb, JOURNAL_DATA);
3293         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_ORDERED)
3294                 set_opt(sb, ORDERED_DATA);
3295         else if ((def_mount_opts & EXT4_DEFM_JMODE) == EXT4_DEFM_JMODE_WBACK)
3296                 set_opt(sb, WRITEBACK_DATA);
3297
3298         if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_PANIC)
3299                 set_opt(sb, ERRORS_PANIC);
3300         else if (le16_to_cpu(sbi->s_es->s_errors) == EXT4_ERRORS_CONTINUE)
3301                 set_opt(sb, ERRORS_CONT);
3302         else
3303                 set_opt(sb, ERRORS_RO);
3304         /* block_validity enabled by default; disable with noblock_validity */
3305         set_opt(sb, BLOCK_VALIDITY);
3306         if (def_mount_opts & EXT4_DEFM_DISCARD)
3307                 set_opt(sb, DISCARD);
3308
3309         sbi->s_resuid = make_kuid(&init_user_ns, le16_to_cpu(es->s_def_resuid));
3310         sbi->s_resgid = make_kgid(&init_user_ns, le16_to_cpu(es->s_def_resgid));
3311         sbi->s_commit_interval = JBD2_DEFAULT_MAX_COMMIT_AGE * HZ;
3312         sbi->s_min_batch_time = EXT4_DEF_MIN_BATCH_TIME;
3313         sbi->s_max_batch_time = EXT4_DEF_MAX_BATCH_TIME;
3314
3315         if ((def_mount_opts & EXT4_DEFM_NOBARRIER) == 0)
3316                 set_opt(sb, BARRIER);
3317
3318         /*
3319          * enable delayed allocation by default
3320          * Use -o nodelalloc to turn it off
3321          */
3322         if (!IS_EXT3_SB(sb) && !IS_EXT2_SB(sb) &&
3323             ((def_mount_opts & EXT4_DEFM_NODELALLOC) == 0))
3324                 set_opt(sb, DELALLOC);
3325
3326         /*
3327          * set default s_li_wait_mult for lazyinit, for the case there is
3328          * no mount option specified.
3329          */
3330         sbi->s_li_wait_mult = EXT4_DEF_LI_WAIT_MULT;
3331
3332         if (sbi->s_es->s_mount_opts[0]) {
3333                 char *s_mount_opts = kstrndup(sbi->s_es->s_mount_opts,
3334                                               sizeof(sbi->s_es->s_mount_opts),
3335                                               GFP_KERNEL);
3336                 if (!s_mount_opts)
3337                         goto failed_mount;
3338                 if (!parse_options(s_mount_opts, sb, &journal_devnum,
3339                                    &journal_ioprio, 0)) {
3340                         ext4_msg(sb, KERN_WARNING,
3341                                  "failed to parse options in superblock: %s",
3342                                  s_mount_opts);
3343                 }
3344                 kfree(s_mount_opts);
3345         }
3346         sbi->s_def_mount_opt = sbi->s_mount_opt;
3347         if (!parse_options((char *) data, sb, &journal_devnum,
3348                            &journal_ioprio, 0))
3349                 goto failed_mount;
3350
3351         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
3352                 printk_once(KERN_WARNING "EXT4-fs: Warning: mounting "
3353                             "with data=journal disables delayed "
3354                             "allocation and O_DIRECT support!\n");
3355                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
3356                         ext4_msg(sb, KERN_ERR, "can't mount with "
3357                                  "both data=journal and delalloc");
3358                         goto failed_mount;
3359                 }
3360                 if (test_opt(sb, DIOREAD_NOLOCK)) {
3361                         ext4_msg(sb, KERN_ERR, "can't mount with "
3362                                  "both data=journal and dioread_nolock");
3363                         goto failed_mount;
3364                 }
3365                 if (test_opt(sb, DAX)) {
3366                         ext4_msg(sb, KERN_ERR, "can't mount with "
3367                                  "both data=journal and dax");
3368                         goto failed_mount;
3369                 }
3370                 if (ext4_has_feature_encrypt(sb)) {
3371                         ext4_msg(sb, KERN_WARNING,
3372                                  "encrypted files will use data=ordered "
3373                                  "instead of data journaling mode");
3374                 }
3375                 if (test_opt(sb, DELALLOC))
3376                         clear_opt(sb, DELALLOC);
3377         } else {
3378                 sb->s_iflags |= SB_I_CGROUPWB;
3379         }
3380
3381         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
3382                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
3383
3384         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV &&
3385             (ext4_has_compat_features(sb) ||
3386              ext4_has_ro_compat_features(sb) ||
3387              ext4_has_incompat_features(sb)))
3388                 ext4_msg(sb, KERN_WARNING,
3389                        "feature flags set on rev 0 fs, "
3390                        "running e2fsck is recommended");
3391
3392         if (es->s_creator_os == cpu_to_le32(EXT4_OS_HURD)) {
3393                 set_opt2(sb, HURD_COMPAT);
3394                 if (ext4_has_feature_64bit(sb)) {
3395                         ext4_msg(sb, KERN_ERR,
3396                                  "The Hurd can't support 64-bit file systems");
3397                         goto failed_mount;
3398                 }
3399         }
3400
3401         if (IS_EXT2_SB(sb)) {
3402                 if (ext2_feature_set_ok(sb))
3403                         ext4_msg(sb, KERN_INFO, "mounting ext2 file system "
3404                                  "using the ext4 subsystem");
3405                 else {
3406                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext2 due "
3407                                  "to feature incompatibilities");
3408                         goto failed_mount;
3409                 }
3410         }
3411
3412         if (IS_EXT3_SB(sb)) {
3413                 if (ext3_feature_set_ok(sb))
3414                         ext4_msg(sb, KERN_INFO, "mounting ext3 file system "
3415                                  "using the ext4 subsystem");
3416                 else {
3417                         ext4_msg(sb, KERN_ERR, "couldn't mount as ext3 due "
3418                                  "to feature incompatibilities");
3419                         goto failed_mount;
3420                 }
3421         }
3422
3423         /*
3424          * Check feature flags regardless of the revision level, since we
3425          * previously didn't change the revision level when setting the flags,
3426          * so there is a chance incompat flags are set on a rev 0 filesystem.
3427          */
3428         if (!ext4_feature_set_ok(sb, (sb->s_flags & MS_RDONLY)))
3429                 goto failed_mount;
3430
3431         blocksize = BLOCK_SIZE << le32_to_cpu(es->s_log_block_size);
3432         if (blocksize < EXT4_MIN_BLOCK_SIZE ||
3433             blocksize > EXT4_MAX_BLOCK_SIZE) {
3434                 ext4_msg(sb, KERN_ERR,
3435                        "Unsupported filesystem blocksize %d (%d log_block_size)",
3436                          blocksize, le32_to_cpu(es->s_log_block_size));
3437                 goto failed_mount;
3438         }
3439         if (le32_to_cpu(es->s_log_block_size) >
3440             (EXT4_MAX_BLOCK_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3441                 ext4_msg(sb, KERN_ERR,
3442                          "Invalid log block size: %u",
3443                          le32_to_cpu(es->s_log_block_size));
3444                 goto failed_mount;
3445         }
3446
3447         if (le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks) > (blocksize / 4)) {
3448                 ext4_msg(sb, KERN_ERR,
3449                          "Number of reserved GDT blocks insanely large: %d",
3450                          le16_to_cpu(sbi->s_es->s_reserved_gdt_blocks));
3451                 goto failed_mount;
3452         }
3453
3454         if (sbi->s_mount_opt & EXT4_MOUNT_DAX) {
3455                 if (blocksize != PAGE_SIZE) {
3456                         ext4_msg(sb, KERN_ERR,
3457                                         "error: unsupported blocksize for dax");
3458                         goto failed_mount;
3459                 }
3460                 if (!sb->s_bdev->bd_disk->fops->direct_access) {
3461                         ext4_msg(sb, KERN_ERR,
3462                                         "error: device does not support dax");
3463                         goto failed_mount;
3464                 }
3465         }
3466
3467         if (ext4_has_feature_encrypt(sb) && es->s_encryption_level) {
3468                 ext4_msg(sb, KERN_ERR, "Unsupported encryption level %d",
3469                          es->s_encryption_level);
3470                 goto failed_mount;
3471         }
3472
3473         if (sb->s_blocksize != blocksize) {
3474                 /* Validate the filesystem blocksize */
3475                 if (!sb_set_blocksize(sb, blocksize)) {
3476                         ext4_msg(sb, KERN_ERR, "bad block size %d",
3477                                         blocksize);
3478                         goto failed_mount;
3479                 }
3480
3481                 brelse(bh);
3482                 logical_sb_block = sb_block * EXT4_MIN_BLOCK_SIZE;
3483                 offset = do_div(logical_sb_block, blocksize);
3484                 bh = sb_bread_unmovable(sb, logical_sb_block);
3485                 if (!bh) {
3486                         ext4_msg(sb, KERN_ERR,
3487                                "Can't read superblock on 2nd try");
3488                         goto failed_mount;
3489                 }
3490                 es = (struct ext4_super_block *)(bh->b_data + offset);
3491                 sbi->s_es = es;
3492                 if (es->s_magic != cpu_to_le16(EXT4_SUPER_MAGIC)) {
3493                         ext4_msg(sb, KERN_ERR,
3494                                "Magic mismatch, very weird!");
3495                         goto failed_mount;
3496                 }
3497         }
3498
3499         has_huge_files = ext4_has_feature_huge_file(sb);
3500         sbi->s_bitmap_maxbytes = ext4_max_bitmap_size(sb->s_blocksize_bits,
3501                                                       has_huge_files);
3502         sb->s_maxbytes = ext4_max_size(sb->s_blocksize_bits, has_huge_files);
3503
3504         if (le32_to_cpu(es->s_rev_level) == EXT4_GOOD_OLD_REV) {
3505                 sbi->s_inode_size = EXT4_GOOD_OLD_INODE_SIZE;
3506                 sbi->s_first_ino = EXT4_GOOD_OLD_FIRST_INO;
3507         } else {
3508                 sbi->s_inode_size = le16_to_cpu(es->s_inode_size);
3509                 sbi->s_first_ino = le32_to_cpu(es->s_first_ino);
3510                 if ((sbi->s_inode_size < EXT4_GOOD_OLD_INODE_SIZE) ||
3511                     (!is_power_of_2(sbi->s_inode_size)) ||
3512                     (sbi->s_inode_size > blocksize)) {
3513                         ext4_msg(sb, KERN_ERR,
3514                                "unsupported inode size: %d",
3515                                sbi->s_inode_size);
3516                         goto failed_mount;
3517                 }
3518                 if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE)
3519                         sb->s_time_gran = 1 << (EXT4_EPOCH_BITS - 2);
3520         }
3521
3522         sbi->s_desc_size = le16_to_cpu(es->s_desc_size);
3523         if (ext4_has_feature_64bit(sb)) {
3524                 if (sbi->s_desc_size < EXT4_MIN_DESC_SIZE_64BIT ||
3525                     sbi->s_desc_size > EXT4_MAX_DESC_SIZE ||
3526                     !is_power_of_2(sbi->s_desc_size)) {
3527                         ext4_msg(sb, KERN_ERR,
3528                                "unsupported descriptor size %lu",
3529                                sbi->s_desc_size);
3530                         goto failed_mount;
3531                 }
3532         } else
3533                 sbi->s_desc_size = EXT4_MIN_DESC_SIZE;
3534
3535         sbi->s_blocks_per_group = le32_to_cpu(es->s_blocks_per_group);
3536         sbi->s_inodes_per_group = le32_to_cpu(es->s_inodes_per_group);
3537
3538         sbi->s_inodes_per_block = blocksize / EXT4_INODE_SIZE(sb);
3539         if (sbi->s_inodes_per_block == 0)
3540                 goto cantfind_ext4;
3541         if (sbi->s_inodes_per_group < sbi->s_inodes_per_block ||
3542             sbi->s_inodes_per_group > blocksize * 8) {
3543                 ext4_msg(sb, KERN_ERR, "invalid inodes per group: %lu\n",
3544                          sbi->s_blocks_per_group);
3545                 goto failed_mount;
3546         }
3547         sbi->s_itb_per_group = sbi->s_inodes_per_group /
3548                                         sbi->s_inodes_per_block;
3549         sbi->s_desc_per_block = blocksize / EXT4_DESC_SIZE(sb);
3550         sbi->s_sbh = bh;
3551         sbi->s_mount_state = le16_to_cpu(es->s_state);
3552         sbi->s_addr_per_block_bits = ilog2(EXT4_ADDR_PER_BLOCK(sb));
3553         sbi->s_desc_per_block_bits = ilog2(EXT4_DESC_PER_BLOCK(sb));
3554
3555         for (i = 0; i < 4; i++)
3556                 sbi->s_hash_seed[i] = le32_to_cpu(es->s_hash_seed[i]);
3557         sbi->s_def_hash_version = es->s_def_hash_version;
3558         if (ext4_has_feature_dir_index(sb)) {
3559                 i = le32_to_cpu(es->s_flags);
3560                 if (i & EXT2_FLAGS_UNSIGNED_HASH)
3561                         sbi->s_hash_unsigned = 3;
3562                 else if ((i & EXT2_FLAGS_SIGNED_HASH) == 0) {
3563 #ifdef __CHAR_UNSIGNED__
3564                         if (!(sb->s_flags & MS_RDONLY))
3565                                 es->s_flags |=
3566                                         cpu_to_le32(EXT2_FLAGS_UNSIGNED_HASH);
3567                         sbi->s_hash_unsigned = 3;
3568 #else
3569                         if (!(sb->s_flags & MS_RDONLY))
3570                                 es->s_flags |=
3571                                         cpu_to_le32(EXT2_FLAGS_SIGNED_HASH);
3572 #endif
3573                 }
3574         }
3575
3576         /* Handle clustersize */
3577         clustersize = BLOCK_SIZE << le32_to_cpu(es->s_log_cluster_size);
3578         has_bigalloc = ext4_has_feature_bigalloc(sb);
3579         if (has_bigalloc) {
3580                 if (clustersize < blocksize) {
3581                         ext4_msg(sb, KERN_ERR,
3582                                  "cluster size (%d) smaller than "
3583                                  "block size (%d)", clustersize, blocksize);
3584                         goto failed_mount;
3585                 }
3586                 if (le32_to_cpu(es->s_log_cluster_size) >
3587                     (EXT4_MAX_CLUSTER_LOG_SIZE - EXT4_MIN_BLOCK_LOG_SIZE)) {
3588                         ext4_msg(sb, KERN_ERR,
3589                                  "Invalid log cluster size: %u",
3590                                  le32_to_cpu(es->s_log_cluster_size));
3591                         goto failed_mount;
3592                 }
3593                 sbi->s_cluster_bits = le32_to_cpu(es->s_log_cluster_size) -
3594                         le32_to_cpu(es->s_log_block_size);
3595                 sbi->s_clusters_per_group =
3596                         le32_to_cpu(es->s_clusters_per_group);
3597                 if (sbi->s_clusters_per_group > blocksize * 8) {
3598                         ext4_msg(sb, KERN_ERR,
3599                                  "#clusters per group too big: %lu",
3600                                  sbi->s_clusters_per_group);
3601                         goto failed_mount;
3602                 }
3603                 if (sbi->s_blocks_per_group !=
3604                     (sbi->s_clusters_per_group * (clustersize / blocksize))) {
3605                         ext4_msg(sb, KERN_ERR, "blocks per group (%lu) and "
3606                                  "clusters per group (%lu) inconsistent",
3607                                  sbi->s_blocks_per_group,
3608                                  sbi->s_clusters_per_group);
3609                         goto failed_mount;
3610                 }
3611         } else {
3612                 if (clustersize != blocksize) {
3613                         ext4_warning(sb, "fragment/cluster size (%d) != "
3614                                      "block size (%d)", clustersize,
3615                                      blocksize);
3616                         clustersize = blocksize;
3617                 }
3618                 if (sbi->s_blocks_per_group > blocksize * 8) {
3619                         ext4_msg(sb, KERN_ERR,
3620                                  "#blocks per group too big: %lu",
3621                                  sbi->s_blocks_per_group);
3622                         goto failed_mount;
3623                 }
3624                 sbi->s_clusters_per_group = sbi->s_blocks_per_group;
3625                 sbi->s_cluster_bits = 0;
3626         }
3627         sbi->s_cluster_ratio = clustersize / blocksize;
3628
3629         /* Do we have standard group size of clustersize * 8 blocks ? */
3630         if (sbi->s_blocks_per_group == clustersize << 3)
3631                 set_opt2(sb, STD_GROUP_SIZE);
3632
3633         /*
3634          * Test whether we have more sectors than will fit in sector_t,
3635          * and whether the max offset is addressable by the page cache.
3636          */
3637         err = generic_check_addressable(sb->s_blocksize_bits,
3638                                         ext4_blocks_count(es));
3639         if (err) {
3640                 ext4_msg(sb, KERN_ERR, "filesystem"
3641                          " too large to mount safely on this system");
3642                 if (sizeof(sector_t) < 8)
3643                         ext4_msg(sb, KERN_WARNING, "CONFIG_LBDAF not enabled");
3644                 goto failed_mount;
3645         }
3646
3647         if (EXT4_BLOCKS_PER_GROUP(sb) == 0)
3648                 goto cantfind_ext4;
3649
3650         /* check blocks count against device size */
3651         blocks_count = sb->s_bdev->bd_inode->i_size >> sb->s_blocksize_bits;
3652         if (blocks_count && ext4_blocks_count(es) > blocks_count) {
3653                 ext4_msg(sb, KERN_WARNING, "bad geometry: block count %llu "
3654                        "exceeds size of device (%llu blocks)",
3655                        ext4_blocks_count(es), blocks_count);
3656                 goto failed_mount;
3657         }
3658
3659         /*
3660          * It makes no sense for the first data block to be beyond the end
3661          * of the filesystem.
3662          */
3663         if (le32_to_cpu(es->s_first_data_block) >= ext4_blocks_count(es)) {
3664                 ext4_msg(sb, KERN_WARNING, "bad geometry: first data "
3665                          "block %u is beyond end of filesystem (%llu)",
3666                          le32_to_cpu(es->s_first_data_block),
3667                          ext4_blocks_count(es));
3668                 goto failed_mount;
3669         }
3670         blocks_count = (ext4_blocks_count(es) -
3671                         le32_to_cpu(es->s_first_data_block) +
3672                         EXT4_BLOCKS_PER_GROUP(sb) - 1);
3673         do_div(blocks_count, EXT4_BLOCKS_PER_GROUP(sb));
3674         if (blocks_count > ((uint64_t)1<<32) - EXT4_DESC_PER_BLOCK(sb)) {
3675                 ext4_msg(sb, KERN_WARNING, "groups count too large: %u "
3676                        "(block count %llu, first data block %u, "
3677                        "blocks per group %lu)", sbi->s_groups_count,
3678                        ext4_blocks_count(es),
3679                        le32_to_cpu(es->s_first_data_block),
3680                        EXT4_BLOCKS_PER_GROUP(sb));
3681                 goto failed_mount;
3682         }
3683         sbi->s_groups_count = blocks_count;
3684         sbi->s_blockfile_groups = min_t(ext4_group_t, sbi->s_groups_count,
3685                         (EXT4_MAX_BLOCK_FILE_PHYS / EXT4_BLOCKS_PER_GROUP(sb)));
3686         db_count = (sbi->s_groups_count + EXT4_DESC_PER_BLOCK(sb) - 1) /
3687                    EXT4_DESC_PER_BLOCK(sb);
3688         if (ext4_has_feature_meta_bg(sb)) {
3689                 if (le32_to_cpu(es->s_first_meta_bg) > db_count) {
3690                         ext4_msg(sb, KERN_WARNING,
3691                                  "first meta block group too large: %u "
3692                                  "(group descriptor block count %u)",
3693                                  le32_to_cpu(es->s_first_meta_bg), db_count);
3694                         goto failed_mount;
3695                 }
3696         }
3697         sbi->s_group_desc = ext4_kvmalloc(db_count *
3698                                           sizeof(struct buffer_head *),
3699                                           GFP_KERNEL);
3700         if (sbi->s_group_desc == NULL) {
3701                 ext4_msg(sb, KERN_ERR, "not enough memory");
3702                 ret = -ENOMEM;
3703                 goto failed_mount;
3704         }
3705
3706         bgl_lock_init(sbi->s_blockgroup_lock);
3707
3708         for (i = 0; i < db_count; i++) {
3709                 block = descriptor_loc(sb, logical_sb_block, i);
3710                 sbi->s_group_desc[i] = sb_bread_unmovable(sb, block);
3711                 if (!sbi->s_group_desc[i]) {
3712                         ext4_msg(sb, KERN_ERR,
3713                                "can't read group descriptor %d", i);
3714                         db_count = i;
3715                         goto failed_mount2;
3716                 }
3717         }
3718         if (!ext4_check_descriptors(sb, logical_sb_block, &first_not_zeroed)) {
3719                 ext4_msg(sb, KERN_ERR, "group descriptors corrupted!");
3720                 ret = -EFSCORRUPTED;
3721                 goto failed_mount2;
3722         }
3723
3724         sbi->s_gdb_count = db_count;
3725         get_random_bytes(&sbi->s_next_generation, sizeof(u32));
3726         spin_lock_init(&sbi->s_next_gen_lock);
3727
3728         setup_timer(&sbi->s_err_report, print_daily_error_info,
3729                 (unsigned long) sb);
3730
3731         /* Register extent status tree shrinker */
3732         if (ext4_es_register_shrinker(sbi))
3733                 goto failed_mount3;
3734
3735         sbi->s_stripe = ext4_get_stripe_size(sbi);
3736         sbi->s_extent_max_zeroout_kb = 32;
3737
3738         /*
3739          * set up enough so that it can read an inode
3740          */
3741         sb->s_op = &ext4_sops;
3742         sb->s_export_op = &ext4_export_ops;
3743         sb->s_xattr = ext4_xattr_handlers;
3744 #ifdef CONFIG_QUOTA
3745         sb->dq_op = &ext4_quota_operations;
3746         if (ext4_has_feature_quota(sb))
3747                 sb->s_qcop = &dquot_quotactl_sysfile_ops;
3748         else
3749                 sb->s_qcop = &ext4_qctl_operations;
3750         sb->s_quota_types = QTYPE_MASK_USR | QTYPE_MASK_GRP;
3751 #endif
3752         memcpy(sb->s_uuid, es->s_uuid, sizeof(es->s_uuid));
3753
3754         INIT_LIST_HEAD(&sbi->s_orphan); /* unlinked but open files */
3755         mutex_init(&sbi->s_orphan_lock);
3756
3757         sb->s_root = NULL;
3758
3759         needs_recovery = (es->s_last_orphan != 0 ||
3760                           ext4_has_feature_journal_needs_recovery(sb));
3761
3762         if (ext4_has_feature_mmp(sb) && !(sb->s_flags & MS_RDONLY))
3763                 if (ext4_multi_mount_protect(sb, le64_to_cpu(es->s_mmp_block)))
3764                         goto failed_mount3a;
3765
3766         /*
3767          * The first inode we look at is the journal inode.  Don't try
3768          * root first: it may be modified in the journal!
3769          */
3770         if (!test_opt(sb, NOLOAD) && ext4_has_feature_journal(sb)) {
3771                 err = ext4_load_journal(sb, es, journal_devnum);
3772                 if (err)
3773                         goto failed_mount3a;
3774         } else if (test_opt(sb, NOLOAD) && !(sb->s_flags & MS_RDONLY) &&
3775                    ext4_has_feature_journal_needs_recovery(sb)) {
3776                 ext4_msg(sb, KERN_ERR, "required journal recovery "
3777                        "suppressed and not mounted read-only");
3778                 goto failed_mount_wq;
3779         } else {
3780                 /* Nojournal mode, all journal mount options are illegal */
3781                 if (test_opt2(sb, EXPLICIT_JOURNAL_CHECKSUM)) {
3782                         ext4_msg(sb, KERN_ERR, "can't mount with "
3783                                  "journal_checksum, fs mounted w/o journal");
3784                         goto failed_mount_wq;
3785                 }
3786                 if (test_opt(sb, JOURNAL_ASYNC_COMMIT)) {
3787                         ext4_msg(sb, KERN_ERR, "can't mount with "
3788                                  "journal_async_commit, fs mounted w/o journal");
3789                         goto failed_mount_wq;
3790                 }
3791                 if (sbi->s_commit_interval != JBD2_DEFAULT_MAX_COMMIT_AGE*HZ) {
3792                         ext4_msg(sb, KERN_ERR, "can't mount with "
3793                                  "commit=%lu, fs mounted w/o journal",
3794                                  sbi->s_commit_interval / HZ);
3795                         goto failed_mount_wq;
3796                 }
3797                 if (EXT4_MOUNT_DATA_FLAGS &
3798                     (sbi->s_mount_opt ^ sbi->s_def_mount_opt)) {
3799                         ext4_msg(sb, KERN_ERR, "can't mount with "
3800                                  "data=, fs mounted w/o journal");
3801                         goto failed_mount_wq;
3802                 }
3803                 sbi->s_def_mount_opt &= EXT4_MOUNT_JOURNAL_CHECKSUM;
3804                 clear_opt(sb, JOURNAL_CHECKSUM);
3805                 clear_opt(sb, DATA_FLAGS);
3806                 sbi->s_journal = NULL;
3807                 needs_recovery = 0;
3808                 goto no_journal;
3809         }
3810
3811         if (ext4_has_feature_64bit(sb) &&
3812             !jbd2_journal_set_features(EXT4_SB(sb)->s_journal, 0, 0,
3813                                        JBD2_FEATURE_INCOMPAT_64BIT)) {
3814                 ext4_msg(sb, KERN_ERR, "Failed to set 64-bit journal feature");
3815                 goto failed_mount_wq;
3816         }
3817
3818         if (!set_journal_csum_feature_set(sb)) {
3819                 ext4_msg(sb, KERN_ERR, "Failed to set journal checksum "
3820                          "feature set");
3821                 goto failed_mount_wq;
3822         }
3823
3824         /* We have now updated the journal if required, so we can
3825          * validate the data journaling mode. */
3826         switch (test_opt(sb, DATA_FLAGS)) {
3827         case 0:
3828                 /* No mode set, assume a default based on the journal
3829                  * capabilities: ORDERED_DATA if the journal can
3830                  * cope, else JOURNAL_DATA
3831                  */
3832                 if (jbd2_journal_check_available_features
3833                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE))
3834                         set_opt(sb, ORDERED_DATA);
3835                 else
3836                         set_opt(sb, JOURNAL_DATA);
3837                 break;
3838
3839         case EXT4_MOUNT_ORDERED_DATA:
3840         case EXT4_MOUNT_WRITEBACK_DATA:
3841                 if (!jbd2_journal_check_available_features
3842                     (sbi->s_journal, 0, 0, JBD2_FEATURE_INCOMPAT_REVOKE)) {
3843                         ext4_msg(sb, KERN_ERR, "Journal does not support "
3844                                "requested data journaling mode");
3845                         goto failed_mount_wq;
3846                 }
3847         default:
3848                 break;
3849         }
3850         set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
3851
3852         sbi->s_journal->j_commit_callback = ext4_journal_commit_callback;
3853
3854 no_journal:
3855         if (ext4_mballoc_ready) {
3856                 sbi->s_mb_cache = ext4_xattr_create_cache();
3857                 if (!sbi->s_mb_cache) {
3858                         ext4_msg(sb, KERN_ERR, "Failed to create an mb_cache");
3859                         goto failed_mount_wq;
3860                 }
3861         }
3862
3863         if ((DUMMY_ENCRYPTION_ENABLED(sbi) || ext4_has_feature_encrypt(sb)) &&
3864             (blocksize != PAGE_CACHE_SIZE)) {
3865                 ext4_msg(sb, KERN_ERR,
3866                          "Unsupported blocksize for fs encryption");
3867                 goto failed_mount_wq;
3868         }
3869
3870         if (DUMMY_ENCRYPTION_ENABLED(sbi) && !(sb->s_flags & MS_RDONLY) &&
3871             !ext4_has_feature_encrypt(sb)) {
3872                 ext4_set_feature_encrypt(sb);
3873                 ext4_commit_super(sb, 1);
3874         }
3875
3876         /*
3877          * Get the # of file system overhead blocks from the
3878          * superblock if present.
3879          */
3880         if (es->s_overhead_clusters)
3881                 sbi->s_overhead = le32_to_cpu(es->s_overhead_clusters);
3882         else {
3883                 err = ext4_calculate_overhead(sb);
3884                 if (err)
3885                         goto failed_mount_wq;
3886         }
3887
3888         /*
3889          * The maximum number of concurrent works can be high and
3890          * concurrency isn't really necessary.  Limit it to 1.
3891          */
3892         EXT4_SB(sb)->rsv_conversion_wq =
3893                 alloc_workqueue("ext4-rsv-conversion", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3894         if (!EXT4_SB(sb)->rsv_conversion_wq) {
3895                 printk(KERN_ERR "EXT4-fs: failed to create workqueue\n");
3896                 ret = -ENOMEM;
3897                 goto failed_mount4;
3898         }
3899
3900         /*
3901          * The jbd2_journal_load will have done any necessary log recovery,
3902          * so we can safely mount the rest of the filesystem now.
3903          */
3904
3905         root = ext4_iget(sb, EXT4_ROOT_INO);
3906         if (IS_ERR(root)) {
3907                 ext4_msg(sb, KERN_ERR, "get root inode failed");
3908                 ret = PTR_ERR(root);
3909                 root = NULL;
3910                 goto failed_mount4;
3911         }
3912         if (!S_ISDIR(root->i_mode) || !root->i_blocks || !root->i_size) {
3913                 ext4_msg(sb, KERN_ERR, "corrupt root inode, run e2fsck");
3914                 iput(root);
3915                 goto failed_mount4;
3916         }
3917         sb->s_root = d_make_root(root);
3918         if (!sb->s_root) {
3919                 ext4_msg(sb, KERN_ERR, "get root dentry failed");
3920                 ret = -ENOMEM;
3921                 goto failed_mount4;
3922         }
3923
3924         if (ext4_setup_super(sb, es, sb->s_flags & MS_RDONLY))
3925                 sb->s_flags |= MS_RDONLY;
3926
3927         /* determine the minimum size of new large inodes, if present */
3928         if (sbi->s_inode_size > EXT4_GOOD_OLD_INODE_SIZE) {
3929                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3930                                                      EXT4_GOOD_OLD_INODE_SIZE;
3931                 if (ext4_has_feature_extra_isize(sb)) {
3932                         if (sbi->s_want_extra_isize <
3933                             le16_to_cpu(es->s_want_extra_isize))
3934                                 sbi->s_want_extra_isize =
3935                                         le16_to_cpu(es->s_want_extra_isize);
3936                         if (sbi->s_want_extra_isize <
3937                             le16_to_cpu(es->s_min_extra_isize))
3938                                 sbi->s_want_extra_isize =
3939                                         le16_to_cpu(es->s_min_extra_isize);
3940                 }
3941         }
3942         /* Check if enough inode space is available */
3943         if (EXT4_GOOD_OLD_INODE_SIZE + sbi->s_want_extra_isize >
3944                                                         sbi->s_inode_size) {
3945                 sbi->s_want_extra_isize = sizeof(struct ext4_inode) -
3946                                                        EXT4_GOOD_OLD_INODE_SIZE;
3947                 ext4_msg(sb, KERN_INFO, "required extra inode space not"
3948                          "available");
3949         }
3950
3951         ext4_set_resv_clusters(sb);
3952
3953         err = ext4_setup_system_zone(sb);
3954         if (err) {
3955                 ext4_msg(sb, KERN_ERR, "failed to initialize system "
3956                          "zone (%d)", err);
3957                 goto failed_mount4a;
3958         }
3959
3960         ext4_ext_init(sb);
3961         err = ext4_mb_init(sb);
3962         if (err) {
3963                 ext4_msg(sb, KERN_ERR, "failed to initialize mballoc (%d)",
3964                          err);
3965                 goto failed_mount5;
3966         }
3967
3968         block = ext4_count_free_clusters(sb);
3969         ext4_free_blocks_count_set(sbi->s_es,
3970                                    EXT4_C2B(sbi, block));
3971         err = percpu_counter_init(&sbi->s_freeclusters_counter, block,
3972                                   GFP_KERNEL);
3973         if (!err) {
3974                 unsigned long freei = ext4_count_free_inodes(sb);
3975                 sbi->s_es->s_free_inodes_count = cpu_to_le32(freei);
3976                 err = percpu_counter_init(&sbi->s_freeinodes_counter, freei,
3977                                           GFP_KERNEL);
3978         }
3979         if (!err)
3980                 err = percpu_counter_init(&sbi->s_dirs_counter,
3981                                           ext4_count_dirs(sb), GFP_KERNEL);
3982         if (!err)
3983                 err = percpu_counter_init(&sbi->s_dirtyclusters_counter, 0,
3984                                           GFP_KERNEL);
3985         if (err) {
3986                 ext4_msg(sb, KERN_ERR, "insufficient memory");
3987                 goto failed_mount6;
3988         }
3989
3990         if (ext4_has_feature_flex_bg(sb))
3991                 if (!ext4_fill_flex_info(sb)) {
3992                         ext4_msg(sb, KERN_ERR,
3993                                "unable to initialize "
3994                                "flex_bg meta info!");
3995                         goto failed_mount6;
3996                 }
3997
3998         err = ext4_register_li_request(sb, first_not_zeroed);
3999         if (err)
4000                 goto failed_mount6;
4001
4002         err = ext4_register_sysfs(sb);
4003         if (err)
4004                 goto failed_mount7;
4005
4006 #ifdef CONFIG_QUOTA
4007         /* Enable quota usage during mount. */
4008         if (ext4_has_feature_quota(sb) && !(sb->s_flags & MS_RDONLY)) {
4009                 err = ext4_enable_quotas(sb);
4010                 if (err)
4011                         goto failed_mount8;
4012         }
4013 #endif  /* CONFIG_QUOTA */
4014
4015         EXT4_SB(sb)->s_mount_state |= EXT4_ORPHAN_FS;
4016         ext4_orphan_cleanup(sb, es);
4017         EXT4_SB(sb)->s_mount_state &= ~EXT4_ORPHAN_FS;
4018         if (needs_recovery) {
4019                 ext4_msg(sb, KERN_INFO, "recovery complete");
4020                 ext4_mark_recovery_complete(sb, es);
4021         }
4022         if (EXT4_SB(sb)->s_journal) {
4023                 if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA)
4024                         descr = " journalled data mode";
4025                 else if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_ORDERED_DATA)
4026                         descr = " ordered data mode";
4027                 else
4028                         descr = " writeback data mode";
4029         } else
4030                 descr = "out journal";
4031
4032         if (test_opt(sb, DISCARD)) {
4033                 struct request_queue *q = bdev_get_queue(sb->s_bdev);
4034                 if (!blk_queue_discard(q))
4035                         ext4_msg(sb, KERN_WARNING,
4036                                  "mounting with \"discard\" option, but "
4037                                  "the device does not support discard");
4038         }
4039
4040         if (___ratelimit(&ext4_mount_msg_ratelimit, "EXT4-fs mount"))
4041                 ext4_msg(sb, KERN_INFO, "mounted filesystem with%s. "
4042                          "Opts: %.*s%s%s", descr,
4043                          (int) sizeof(sbi->s_es->s_mount_opts),
4044                          sbi->s_es->s_mount_opts,
4045                          *sbi->s_es->s_mount_opts ? "; " : "", orig_data);
4046
4047         if (es->s_error_count)
4048                 mod_timer(&sbi->s_err_report, jiffies + 300*HZ); /* 5 minutes */
4049
4050         /* Enable message ratelimiting. Default is 10 messages per 5 secs. */
4051         ratelimit_state_init(&sbi->s_err_ratelimit_state, 5 * HZ, 10);
4052         ratelimit_state_init(&sbi->s_warning_ratelimit_state, 5 * HZ, 10);
4053         ratelimit_state_init(&sbi->s_msg_ratelimit_state, 5 * HZ, 10);
4054
4055         kfree(orig_data);
4056         return 0;
4057
4058 cantfind_ext4:
4059         if (!silent)
4060                 ext4_msg(sb, KERN_ERR, "VFS: Can't find ext4 filesystem");
4061         goto failed_mount;
4062
4063 #ifdef CONFIG_QUOTA
4064 failed_mount8:
4065         ext4_unregister_sysfs(sb);
4066 #endif
4067 failed_mount7:
4068         ext4_unregister_li_request(sb);
4069 failed_mount6:
4070         ext4_mb_release(sb);
4071         if (sbi->s_flex_groups)
4072                 kvfree(sbi->s_flex_groups);
4073         percpu_counter_destroy(&sbi->s_freeclusters_counter);
4074         percpu_counter_destroy(&sbi->s_freeinodes_counter);
4075         percpu_counter_destroy(&sbi->s_dirs_counter);
4076         percpu_counter_destroy(&sbi->s_dirtyclusters_counter);
4077 failed_mount5:
4078         ext4_ext_release(sb);
4079         ext4_release_system_zone(sb);
4080 failed_mount4a:
4081         dput(sb->s_root);
4082         sb->s_root = NULL;
4083 failed_mount4:
4084         ext4_msg(sb, KERN_ERR, "mount failed");
4085         if (EXT4_SB(sb)->rsv_conversion_wq)
4086                 destroy_workqueue(EXT4_SB(sb)->rsv_conversion_wq);
4087 failed_mount_wq:
4088         if (sbi->s_mb_cache) {
4089                 ext4_xattr_destroy_cache(sbi->s_mb_cache);
4090                 sbi->s_mb_cache = NULL;
4091         }
4092         if (sbi->s_journal) {
4093                 jbd2_journal_destroy(sbi->s_journal);
4094                 sbi->s_journal = NULL;
4095         }
4096 failed_mount3a:
4097         ext4_es_unregister_shrinker(sbi);
4098 failed_mount3:
4099         del_timer_sync(&sbi->s_err_report);
4100         if (sbi->s_mmp_tsk)
4101                 kthread_stop(sbi->s_mmp_tsk);
4102 failed_mount2:
4103         for (i = 0; i < db_count; i++)
4104                 brelse(sbi->s_group_desc[i]);
4105         kvfree(sbi->s_group_desc);
4106 failed_mount:
4107         if (sbi->s_chksum_driver)
4108                 crypto_free_shash(sbi->s_chksum_driver);
4109 #ifdef CONFIG_QUOTA
4110         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4111                 kfree(sbi->s_qf_names[i]);
4112 #endif
4113         ext4_blkdev_remove(sbi);
4114         brelse(bh);
4115 out_fail:
4116         sb->s_fs_info = NULL;
4117         kfree(sbi->s_blockgroup_lock);
4118 out_free_base:
4119         kfree(sbi);
4120         kfree(orig_data);
4121         return err ? err : ret;
4122 }
4123
4124 /*
4125  * Setup any per-fs journal parameters now.  We'll do this both on
4126  * initial mount, once the journal has been initialised but before we've
4127  * done any recovery; and again on any subsequent remount.
4128  */
4129 static void ext4_init_journal_params(struct super_block *sb, journal_t *journal)
4130 {
4131         struct ext4_sb_info *sbi = EXT4_SB(sb);
4132
4133         journal->j_commit_interval = sbi->s_commit_interval;
4134         journal->j_min_batch_time = sbi->s_min_batch_time;
4135         journal->j_max_batch_time = sbi->s_max_batch_time;
4136
4137         write_lock(&journal->j_state_lock);
4138         if (test_opt(sb, BARRIER))
4139                 journal->j_flags |= JBD2_BARRIER;
4140         else
4141                 journal->j_flags &= ~JBD2_BARRIER;
4142         if (test_opt(sb, DATA_ERR_ABORT))
4143                 journal->j_flags |= JBD2_ABORT_ON_SYNCDATA_ERR;
4144         else
4145                 journal->j_flags &= ~JBD2_ABORT_ON_SYNCDATA_ERR;
4146         write_unlock(&journal->j_state_lock);
4147 }
4148
4149 static journal_t *ext4_get_journal(struct super_block *sb,
4150                                    unsigned int journal_inum)
4151 {
4152         struct inode *journal_inode;
4153         journal_t *journal;
4154
4155         BUG_ON(!ext4_has_feature_journal(sb));
4156
4157         /* First, test for the existence of a valid inode on disk.  Bad
4158          * things happen if we iget() an unused inode, as the subsequent
4159          * iput() will try to delete it. */
4160
4161         journal_inode = ext4_iget(sb, journal_inum);
4162         if (IS_ERR(journal_inode)) {
4163                 ext4_msg(sb, KERN_ERR, "no journal found");
4164                 return NULL;
4165         }
4166         if (!journal_inode->i_nlink) {
4167                 make_bad_inode(journal_inode);
4168                 iput(journal_inode);
4169                 ext4_msg(sb, KERN_ERR, "journal inode is deleted");
4170                 return NULL;
4171         }
4172
4173         jbd_debug(2, "Journal inode found at %p: %lld bytes\n",
4174                   journal_inode, journal_inode->i_size);
4175         if (!S_ISREG(journal_inode->i_mode)) {
4176                 ext4_msg(sb, KERN_ERR, "invalid journal inode");
4177                 iput(journal_inode);
4178                 return NULL;
4179         }
4180
4181         journal = jbd2_journal_init_inode(journal_inode);
4182         if (!journal) {
4183                 ext4_msg(sb, KERN_ERR, "Could not load journal inode");
4184                 iput(journal_inode);
4185                 return NULL;
4186         }
4187         journal->j_private = sb;
4188         ext4_init_journal_params(sb, journal);
4189         return journal;
4190 }
4191
4192 static journal_t *ext4_get_dev_journal(struct super_block *sb,
4193                                        dev_t j_dev)
4194 {
4195         struct buffer_head *bh;
4196         journal_t *journal;
4197         ext4_fsblk_t start;
4198         ext4_fsblk_t len;
4199         int hblock, blocksize;
4200         ext4_fsblk_t sb_block;
4201         unsigned long offset;
4202         struct ext4_super_block *es;
4203         struct block_device *bdev;
4204
4205         BUG_ON(!ext4_has_feature_journal(sb));
4206
4207         bdev = ext4_blkdev_get(j_dev, sb);
4208         if (bdev == NULL)
4209                 return NULL;
4210
4211         blocksize = sb->s_blocksize;
4212         hblock = bdev_logical_block_size(bdev);
4213         if (blocksize < hblock) {
4214                 ext4_msg(sb, KERN_ERR,
4215                         "blocksize too small for journal device");
4216                 goto out_bdev;
4217         }
4218
4219         sb_block = EXT4_MIN_BLOCK_SIZE / blocksize;
4220         offset = EXT4_MIN_BLOCK_SIZE % blocksize;
4221         set_blocksize(bdev, blocksize);
4222         if (!(bh = __bread(bdev, sb_block, blocksize))) {
4223                 ext4_msg(sb, KERN_ERR, "couldn't read superblock of "
4224                        "external journal");
4225                 goto out_bdev;
4226         }
4227
4228         es = (struct ext4_super_block *) (bh->b_data + offset);
4229         if ((le16_to_cpu(es->s_magic) != EXT4_SUPER_MAGIC) ||
4230             !(le32_to_cpu(es->s_feature_incompat) &
4231               EXT4_FEATURE_INCOMPAT_JOURNAL_DEV)) {
4232                 ext4_msg(sb, KERN_ERR, "external journal has "
4233                                         "bad superblock");
4234                 brelse(bh);
4235                 goto out_bdev;
4236         }
4237
4238         if ((le32_to_cpu(es->s_feature_ro_compat) &
4239              EXT4_FEATURE_RO_COMPAT_METADATA_CSUM) &&
4240             es->s_checksum != ext4_superblock_csum(sb, es)) {
4241                 ext4_msg(sb, KERN_ERR, "external journal has "
4242                                        "corrupt superblock");
4243                 brelse(bh);
4244                 goto out_bdev;
4245         }
4246
4247         if (memcmp(EXT4_SB(sb)->s_es->s_journal_uuid, es->s_uuid, 16)) {
4248                 ext4_msg(sb, KERN_ERR, "journal UUID does not match");
4249                 brelse(bh);
4250                 goto out_bdev;
4251         }
4252
4253         len = ext4_blocks_count(es);
4254         start = sb_block + 1;
4255         brelse(bh);     /* we're done with the superblock */
4256
4257         journal = jbd2_journal_init_dev(bdev, sb->s_bdev,
4258                                         start, len, blocksize);
4259         if (!journal) {
4260                 ext4_msg(sb, KERN_ERR, "failed to create device journal");
4261                 goto out_bdev;
4262         }
4263         journal->j_private = sb;
4264         ll_rw_block(READ | REQ_META | REQ_PRIO, 1, &journal->j_sb_buffer);
4265         wait_on_buffer(journal->j_sb_buffer);
4266         if (!buffer_uptodate(journal->j_sb_buffer)) {
4267                 ext4_msg(sb, KERN_ERR, "I/O error on journal device");
4268                 goto out_journal;
4269         }
4270         if (be32_to_cpu(journal->j_superblock->s_nr_users) != 1) {
4271                 ext4_msg(sb, KERN_ERR, "External journal has more than one "
4272                                         "user (unsupported) - %d",
4273                         be32_to_cpu(journal->j_superblock->s_nr_users));
4274                 goto out_journal;
4275         }
4276         EXT4_SB(sb)->journal_bdev = bdev;
4277         ext4_init_journal_params(sb, journal);
4278         return journal;
4279
4280 out_journal:
4281         jbd2_journal_destroy(journal);
4282 out_bdev:
4283         ext4_blkdev_put(bdev);
4284         return NULL;
4285 }
4286
4287 static int ext4_load_journal(struct super_block *sb,
4288                              struct ext4_super_block *es,
4289                              unsigned long journal_devnum)
4290 {
4291         journal_t *journal;
4292         unsigned int journal_inum = le32_to_cpu(es->s_journal_inum);
4293         dev_t journal_dev;
4294         int err = 0;
4295         int really_read_only;
4296
4297         BUG_ON(!ext4_has_feature_journal(sb));
4298
4299         if (journal_devnum &&
4300             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4301                 ext4_msg(sb, KERN_INFO, "external journal device major/minor "
4302                         "numbers have changed");
4303                 journal_dev = new_decode_dev(journal_devnum);
4304         } else
4305                 journal_dev = new_decode_dev(le32_to_cpu(es->s_journal_dev));
4306
4307         really_read_only = bdev_read_only(sb->s_bdev);
4308
4309         /*
4310          * Are we loading a blank journal or performing recovery after a
4311          * crash?  For recovery, we need to check in advance whether we
4312          * can get read-write access to the device.
4313          */
4314         if (ext4_has_feature_journal_needs_recovery(sb)) {
4315                 if (sb->s_flags & MS_RDONLY) {
4316                         ext4_msg(sb, KERN_INFO, "INFO: recovery "
4317                                         "required on readonly filesystem");
4318                         if (really_read_only) {
4319                                 ext4_msg(sb, KERN_ERR, "write access "
4320                                         "unavailable, cannot proceed");
4321                                 return -EROFS;
4322                         }
4323                         ext4_msg(sb, KERN_INFO, "write access will "
4324                                "be enabled during recovery");
4325                 }
4326         }
4327
4328         if (journal_inum && journal_dev) {
4329                 ext4_msg(sb, KERN_ERR, "filesystem has both journal "
4330                        "and inode journals!");
4331                 return -EINVAL;
4332         }
4333
4334         if (journal_inum) {
4335                 if (!(journal = ext4_get_journal(sb, journal_inum)))
4336                         return -EINVAL;
4337         } else {
4338                 if (!(journal = ext4_get_dev_journal(sb, journal_dev)))
4339                         return -EINVAL;
4340         }
4341
4342         if (!(journal->j_flags & JBD2_BARRIER))
4343                 ext4_msg(sb, KERN_INFO, "barriers disabled");
4344
4345         if (!ext4_has_feature_journal_needs_recovery(sb))
4346                 err = jbd2_journal_wipe(journal, !really_read_only);
4347         if (!err) {
4348                 char *save = kmalloc(EXT4_S_ERR_LEN, GFP_KERNEL);
4349                 if (save)
4350                         memcpy(save, ((char *) es) +
4351                                EXT4_S_ERR_START, EXT4_S_ERR_LEN);
4352                 err = jbd2_journal_load(journal);
4353                 if (save)
4354                         memcpy(((char *) es) + EXT4_S_ERR_START,
4355                                save, EXT4_S_ERR_LEN);
4356                 kfree(save);
4357         }
4358
4359         if (err) {
4360                 ext4_msg(sb, KERN_ERR, "error loading journal");
4361                 jbd2_journal_destroy(journal);
4362                 return err;
4363         }
4364
4365         EXT4_SB(sb)->s_journal = journal;
4366         ext4_clear_journal_err(sb, es);
4367
4368         if (!really_read_only && journal_devnum &&
4369             journal_devnum != le32_to_cpu(es->s_journal_dev)) {
4370                 es->s_journal_dev = cpu_to_le32(journal_devnum);
4371
4372                 /* Make sure we flush the recovery flag to disk. */
4373                 ext4_commit_super(sb, 1);
4374         }
4375
4376         return 0;
4377 }
4378
4379 static int ext4_commit_super(struct super_block *sb, int sync)
4380 {
4381         struct ext4_super_block *es = EXT4_SB(sb)->s_es;
4382         struct buffer_head *sbh = EXT4_SB(sb)->s_sbh;
4383         int error = 0;
4384
4385         if (!sbh || block_device_ejected(sb))
4386                 return error;
4387         if (buffer_write_io_error(sbh)) {
4388                 /*
4389                  * Oh, dear.  A previous attempt to write the
4390                  * superblock failed.  This could happen because the
4391                  * USB device was yanked out.  Or it could happen to
4392                  * be a transient write error and maybe the block will
4393                  * be remapped.  Nothing we can do but to retry the
4394                  * write and hope for the best.
4395                  */
4396                 ext4_msg(sb, KERN_ERR, "previous I/O error to "
4397                        "superblock detected");
4398                 clear_buffer_write_io_error(sbh);
4399                 set_buffer_uptodate(sbh);
4400         }
4401         /*
4402          * If the file system is mounted read-only, don't update the
4403          * superblock write time.  This avoids updating the superblock
4404          * write time when we are mounting the root file system
4405          * read/only but we need to replay the journal; at that point,
4406          * for people who are east of GMT and who make their clock
4407          * tick in localtime for Windows bug-for-bug compatibility,
4408          * the clock is set in the future, and this will cause e2fsck
4409          * to complain and force a full file system check.
4410          */
4411         if (!(sb->s_flags & MS_RDONLY))
4412                 es->s_wtime = cpu_to_le32(get_seconds());
4413         if (sb->s_bdev->bd_part)
4414                 es->s_kbytes_written =
4415                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written +
4416                             ((part_stat_read(sb->s_bdev->bd_part, sectors[1]) -
4417                               EXT4_SB(sb)->s_sectors_written_start) >> 1));
4418         else
4419                 es->s_kbytes_written =
4420                         cpu_to_le64(EXT4_SB(sb)->s_kbytes_written);
4421         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeclusters_counter))
4422                 ext4_free_blocks_count_set(es,
4423                         EXT4_C2B(EXT4_SB(sb), percpu_counter_sum_positive(
4424                                 &EXT4_SB(sb)->s_freeclusters_counter)));
4425         if (percpu_counter_initialized(&EXT4_SB(sb)->s_freeinodes_counter))
4426                 es->s_free_inodes_count =
4427                         cpu_to_le32(percpu_counter_sum_positive(
4428                                 &EXT4_SB(sb)->s_freeinodes_counter));
4429         BUFFER_TRACE(sbh, "marking dirty");
4430         ext4_superblock_csum_set(sb);
4431         mark_buffer_dirty(sbh);
4432         if (sync) {
4433                 error = __sync_dirty_buffer(sbh,
4434                         test_opt(sb, BARRIER) ? WRITE_FUA : WRITE_SYNC);
4435                 if (error)
4436                         return error;
4437
4438                 error = buffer_write_io_error(sbh);
4439                 if (error) {
4440                         ext4_msg(sb, KERN_ERR, "I/O error while writing "
4441                                "superblock");
4442                         clear_buffer_write_io_error(sbh);
4443                         set_buffer_uptodate(sbh);
4444                 }
4445         }
4446         return error;
4447 }
4448
4449 /*
4450  * Have we just finished recovery?  If so, and if we are mounting (or
4451  * remounting) the filesystem readonly, then we will end up with a
4452  * consistent fs on disk.  Record that fact.
4453  */
4454 static void ext4_mark_recovery_complete(struct super_block *sb,
4455                                         struct ext4_super_block *es)
4456 {
4457         journal_t *journal = EXT4_SB(sb)->s_journal;
4458
4459         if (!ext4_has_feature_journal(sb)) {
4460                 BUG_ON(journal != NULL);
4461                 return;
4462         }
4463         jbd2_journal_lock_updates(journal);
4464         if (jbd2_journal_flush(journal) < 0)
4465                 goto out;
4466
4467         if (ext4_has_feature_journal_needs_recovery(sb) &&
4468             sb->s_flags & MS_RDONLY) {
4469                 ext4_clear_feature_journal_needs_recovery(sb);
4470                 ext4_commit_super(sb, 1);
4471         }
4472
4473 out:
4474         jbd2_journal_unlock_updates(journal);
4475 }
4476
4477 /*
4478  * If we are mounting (or read-write remounting) a filesystem whose journal
4479  * has recorded an error from a previous lifetime, move that error to the
4480  * main filesystem now.
4481  */
4482 static void ext4_clear_journal_err(struct super_block *sb,
4483                                    struct ext4_super_block *es)
4484 {
4485         journal_t *journal;
4486         int j_errno;
4487         const char *errstr;
4488
4489         BUG_ON(!ext4_has_feature_journal(sb));
4490
4491         journal = EXT4_SB(sb)->s_journal;
4492
4493         /*
4494          * Now check for any error status which may have been recorded in the
4495          * journal by a prior ext4_error() or ext4_abort()
4496          */
4497
4498         j_errno = jbd2_journal_errno(journal);
4499         if (j_errno) {
4500                 char nbuf[16];
4501
4502                 errstr = ext4_decode_error(sb, j_errno, nbuf);
4503                 ext4_warning(sb, "Filesystem error recorded "
4504                              "from previous mount: %s", errstr);
4505                 ext4_warning(sb, "Marking fs in need of filesystem check.");
4506
4507                 EXT4_SB(sb)->s_mount_state |= EXT4_ERROR_FS;
4508                 es->s_state |= cpu_to_le16(EXT4_ERROR_FS);
4509                 ext4_commit_super(sb, 1);
4510
4511                 jbd2_journal_clear_err(journal);
4512                 jbd2_journal_update_sb_errno(journal);
4513         }
4514 }
4515
4516 /*
4517  * Force the running and committing transactions to commit,
4518  * and wait on the commit.
4519  */
4520 int ext4_force_commit(struct super_block *sb)
4521 {
4522         journal_t *journal;
4523
4524         if (sb->s_flags & MS_RDONLY)
4525                 return 0;
4526
4527         journal = EXT4_SB(sb)->s_journal;
4528         return ext4_journal_force_commit(journal);
4529 }
4530
4531 static int ext4_sync_fs(struct super_block *sb, int wait)
4532 {
4533         int ret = 0;
4534         tid_t target;
4535         bool needs_barrier = false;
4536         struct ext4_sb_info *sbi = EXT4_SB(sb);
4537
4538         trace_ext4_sync_fs(sb, wait);
4539         flush_workqueue(sbi->rsv_conversion_wq);
4540         /*
4541          * Writeback quota in non-journalled quota case - journalled quota has
4542          * no dirty dquots
4543          */
4544         dquot_writeback_dquots(sb, -1);
4545         /*
4546          * Data writeback is possible w/o journal transaction, so barrier must
4547          * being sent at the end of the function. But we can skip it if
4548          * transaction_commit will do it for us.
4549          */
4550         if (sbi->s_journal) {
4551                 target = jbd2_get_latest_transaction(sbi->s_journal);
4552                 if (wait && sbi->s_journal->j_flags & JBD2_BARRIER &&
4553                     !jbd2_trans_will_send_data_barrier(sbi->s_journal, target))
4554                         needs_barrier = true;
4555
4556                 if (jbd2_journal_start_commit(sbi->s_journal, &target)) {
4557                         if (wait)
4558                                 ret = jbd2_log_wait_commit(sbi->s_journal,
4559                                                            target);
4560                 }
4561         } else if (wait && test_opt(sb, BARRIER))
4562                 needs_barrier = true;
4563         if (needs_barrier) {
4564                 int err;
4565                 err = blkdev_issue_flush(sb->s_bdev, GFP_KERNEL, NULL);
4566                 if (!ret)
4567                         ret = err;
4568         }
4569
4570         return ret;
4571 }
4572
4573 /*
4574  * LVM calls this function before a (read-only) snapshot is created.  This
4575  * gives us a chance to flush the journal completely and mark the fs clean.
4576  *
4577  * Note that only this function cannot bring a filesystem to be in a clean
4578  * state independently. It relies on upper layer to stop all data & metadata
4579  * modifications.
4580  */
4581 static int ext4_freeze(struct super_block *sb)
4582 {
4583         int error = 0;
4584         journal_t *journal;
4585
4586         if (sb->s_flags & MS_RDONLY)
4587                 return 0;
4588
4589         journal = EXT4_SB(sb)->s_journal;
4590
4591         if (journal) {
4592                 /* Now we set up the journal barrier. */
4593                 jbd2_journal_lock_updates(journal);
4594
4595                 /*
4596                  * Don't clear the needs_recovery flag if we failed to
4597                  * flush the journal.
4598                  */
4599                 error = jbd2_journal_flush(journal);
4600                 if (error < 0)
4601                         goto out;
4602
4603                 /* Journal blocked and flushed, clear needs_recovery flag. */
4604                 ext4_clear_feature_journal_needs_recovery(sb);
4605         }
4606
4607         error = ext4_commit_super(sb, 1);
4608 out:
4609         if (journal)
4610                 /* we rely on upper layer to stop further updates */
4611                 jbd2_journal_unlock_updates(journal);
4612         return error;
4613 }
4614
4615 /*
4616  * Called by LVM after the snapshot is done.  We need to reset the RECOVER
4617  * flag here, even though the filesystem is not technically dirty yet.
4618  */
4619 static int ext4_unfreeze(struct super_block *sb)
4620 {
4621         if (sb->s_flags & MS_RDONLY)
4622                 return 0;
4623
4624         if (EXT4_SB(sb)->s_journal) {
4625                 /* Reset the needs_recovery flag before the fs is unlocked. */
4626                 ext4_set_feature_journal_needs_recovery(sb);
4627         }
4628
4629         ext4_commit_super(sb, 1);
4630         return 0;
4631 }
4632
4633 /*
4634  * Structure to save mount options for ext4_remount's benefit
4635  */
4636 struct ext4_mount_options {
4637         unsigned long s_mount_opt;
4638         unsigned long s_mount_opt2;
4639         kuid_t s_resuid;
4640         kgid_t s_resgid;
4641         unsigned long s_commit_interval;
4642         u32 s_min_batch_time, s_max_batch_time;
4643 #ifdef CONFIG_QUOTA
4644         int s_jquota_fmt;
4645         char *s_qf_names[EXT4_MAXQUOTAS];
4646 #endif
4647 };
4648
4649 static int ext4_remount(struct super_block *sb, int *flags, char *data)
4650 {
4651         struct ext4_super_block *es;
4652         struct ext4_sb_info *sbi = EXT4_SB(sb);
4653         unsigned long old_sb_flags;
4654         struct ext4_mount_options old_opts;
4655         int enable_quota = 0;
4656         ext4_group_t g;
4657         unsigned int journal_ioprio = DEFAULT_JOURNAL_IOPRIO;
4658         int err = 0;
4659 #ifdef CONFIG_QUOTA
4660         int i, j;
4661 #endif
4662         char *orig_data = kstrdup(data, GFP_KERNEL);
4663
4664         /* Store the original options */
4665         old_sb_flags = sb->s_flags;
4666         old_opts.s_mount_opt = sbi->s_mount_opt;
4667         old_opts.s_mount_opt2 = sbi->s_mount_opt2;
4668         old_opts.s_resuid = sbi->s_resuid;
4669         old_opts.s_resgid = sbi->s_resgid;
4670         old_opts.s_commit_interval = sbi->s_commit_interval;
4671         old_opts.s_min_batch_time = sbi->s_min_batch_time;
4672         old_opts.s_max_batch_time = sbi->s_max_batch_time;
4673 #ifdef CONFIG_QUOTA
4674         old_opts.s_jquota_fmt = sbi->s_jquota_fmt;
4675         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4676                 if (sbi->s_qf_names[i]) {
4677                         old_opts.s_qf_names[i] = kstrdup(sbi->s_qf_names[i],
4678                                                          GFP_KERNEL);
4679                         if (!old_opts.s_qf_names[i]) {
4680                                 for (j = 0; j < i; j++)
4681                                         kfree(old_opts.s_qf_names[j]);
4682                                 kfree(orig_data);
4683                                 return -ENOMEM;
4684                         }
4685                 } else
4686                         old_opts.s_qf_names[i] = NULL;
4687 #endif
4688         if (sbi->s_journal && sbi->s_journal->j_task->io_context)
4689                 journal_ioprio = sbi->s_journal->j_task->io_context->ioprio;
4690
4691         if (!parse_options(data, sb, NULL, &journal_ioprio, 1)) {
4692                 err = -EINVAL;
4693                 goto restore_opts;
4694         }
4695
4696         if ((old_opts.s_mount_opt & EXT4_MOUNT_JOURNAL_CHECKSUM) ^
4697             test_opt(sb, JOURNAL_CHECKSUM)) {
4698                 ext4_msg(sb, KERN_ERR, "changing journal_checksum "
4699                          "during remount not supported; ignoring");
4700                 sbi->s_mount_opt ^= EXT4_MOUNT_JOURNAL_CHECKSUM;
4701         }
4702
4703         if (test_opt(sb, DATA_FLAGS) == EXT4_MOUNT_JOURNAL_DATA) {
4704                 if (test_opt2(sb, EXPLICIT_DELALLOC)) {
4705                         ext4_msg(sb, KERN_ERR, "can't mount with "
4706                                  "both data=journal and delalloc");
4707                         err = -EINVAL;
4708                         goto restore_opts;
4709                 }
4710                 if (test_opt(sb, DIOREAD_NOLOCK)) {
4711                         ext4_msg(sb, KERN_ERR, "can't mount with "
4712                                  "both data=journal and dioread_nolock");
4713                         err = -EINVAL;
4714                         goto restore_opts;
4715                 }
4716                 if (test_opt(sb, DAX)) {
4717                         ext4_msg(sb, KERN_ERR, "can't mount with "
4718                                  "both data=journal and dax");
4719                         err = -EINVAL;
4720                         goto restore_opts;
4721                 }
4722         }
4723
4724         if ((sbi->s_mount_opt ^ old_opts.s_mount_opt) & EXT4_MOUNT_DAX) {
4725                 ext4_msg(sb, KERN_WARNING, "warning: refusing change of "
4726                         "dax flag with busy inodes while remounting");
4727                 sbi->s_mount_opt ^= EXT4_MOUNT_DAX;
4728         }
4729
4730         if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED)
4731                 ext4_abort(sb, "Abort forced by user");
4732
4733         sb->s_flags = (sb->s_flags & ~MS_POSIXACL) |
4734                 (test_opt(sb, POSIX_ACL) ? MS_POSIXACL : 0);
4735
4736         es = sbi->s_es;
4737
4738         if (sbi->s_journal) {
4739                 ext4_init_journal_params(sb, sbi->s_journal);
4740                 set_task_ioprio(sbi->s_journal->j_task, journal_ioprio);
4741         }
4742
4743         if (*flags & MS_LAZYTIME)
4744                 sb->s_flags |= MS_LAZYTIME;
4745
4746         if ((*flags & MS_RDONLY) != (sb->s_flags & MS_RDONLY)) {
4747                 if (sbi->s_mount_flags & EXT4_MF_FS_ABORTED) {
4748                         err = -EROFS;
4749                         goto restore_opts;
4750                 }
4751
4752                 if (*flags & MS_RDONLY) {
4753                         err = sync_filesystem(sb);
4754                         if (err < 0)
4755                                 goto restore_opts;
4756                         err = dquot_suspend(sb, -1);
4757                         if (err < 0)
4758                                 goto restore_opts;
4759
4760                         /*
4761                          * First of all, the unconditional stuff we have to do
4762                          * to disable replay of the journal when we next remount
4763                          */
4764                         sb->s_flags |= MS_RDONLY;
4765
4766                         /*
4767                          * OK, test if we are remounting a valid rw partition
4768                          * readonly, and if so set the rdonly flag and then
4769                          * mark the partition as valid again.
4770                          */
4771                         if (!(es->s_state & cpu_to_le16(EXT4_VALID_FS)) &&
4772                             (sbi->s_mount_state & EXT4_VALID_FS))
4773                                 es->s_state = cpu_to_le16(sbi->s_mount_state);
4774
4775                         if (sbi->s_journal)
4776                                 ext4_mark_recovery_complete(sb, es);
4777                 } else {
4778                         /* Make sure we can mount this feature set readwrite */
4779                         if (ext4_has_feature_readonly(sb) ||
4780                             !ext4_feature_set_ok(sb, 0)) {
4781                                 err = -EROFS;
4782                                 goto restore_opts;
4783                         }
4784                         /*
4785                          * Make sure the group descriptor checksums
4786                          * are sane.  If they aren't, refuse to remount r/w.
4787                          */
4788                         for (g = 0; g < sbi->s_groups_count; g++) {
4789                                 struct ext4_group_desc *gdp =
4790                                         ext4_get_group_desc(sb, g, NULL);
4791
4792                                 if (!ext4_group_desc_csum_verify(sb, g, gdp)) {
4793                                         ext4_msg(sb, KERN_ERR,
4794                "ext4_remount: Checksum for group %u failed (%u!=%u)",
4795                 g, le16_to_cpu(ext4_group_desc_csum(sb, g, gdp)),
4796                                                le16_to_cpu(gdp->bg_checksum));
4797                                         err = -EFSBADCRC;
4798                                         goto restore_opts;
4799                                 }
4800                         }
4801
4802                         /*
4803                          * If we have an unprocessed orphan list hanging
4804                          * around from a previously readonly bdev mount,
4805                          * require a full umount/remount for now.
4806                          */
4807                         if (es->s_last_orphan) {
4808                                 ext4_msg(sb, KERN_WARNING, "Couldn't "
4809                                        "remount RDWR because of unprocessed "
4810                                        "orphan inode list.  Please "
4811                                        "umount/remount instead");
4812                                 err = -EINVAL;
4813                                 goto restore_opts;
4814                         }
4815
4816                         /*
4817                          * Mounting a RDONLY partition read-write, so reread
4818                          * and store the current valid flag.  (It may have
4819                          * been changed by e2fsck since we originally mounted
4820                          * the partition.)
4821                          */
4822                         if (sbi->s_journal)
4823                                 ext4_clear_journal_err(sb, es);
4824                         sbi->s_mount_state = le16_to_cpu(es->s_state);
4825                         if (!ext4_setup_super(sb, es, 0))
4826                                 sb->s_flags &= ~MS_RDONLY;
4827                         if (ext4_has_feature_mmp(sb))
4828                                 if (ext4_multi_mount_protect(sb,
4829                                                 le64_to_cpu(es->s_mmp_block))) {
4830                                         err = -EROFS;
4831                                         goto restore_opts;
4832                                 }
4833                         enable_quota = 1;
4834                 }
4835         }
4836
4837         /*
4838          * Reinitialize lazy itable initialization thread based on
4839          * current settings
4840          */
4841         if ((sb->s_flags & MS_RDONLY) || !test_opt(sb, INIT_INODE_TABLE))
4842                 ext4_unregister_li_request(sb);
4843         else {
4844                 ext4_group_t first_not_zeroed;
4845                 first_not_zeroed = ext4_has_uninit_itable(sb);
4846                 ext4_register_li_request(sb, first_not_zeroed);
4847         }
4848
4849         ext4_setup_system_zone(sb);
4850         if (sbi->s_journal == NULL && !(old_sb_flags & MS_RDONLY))
4851                 ext4_commit_super(sb, 1);
4852
4853 #ifdef CONFIG_QUOTA
4854         /* Release old quota file names */
4855         for (i = 0; i < EXT4_MAXQUOTAS; i++)
4856                 kfree(old_opts.s_qf_names[i]);
4857         if (enable_quota) {
4858                 if (sb_any_quota_suspended(sb))
4859                         dquot_resume(sb, -1);
4860                 else if (ext4_has_feature_quota(sb)) {
4861                         err = ext4_enable_quotas(sb);
4862                         if (err)
4863                                 goto restore_opts;
4864                 }
4865         }
4866 #endif
4867
4868         *flags = (*flags & ~MS_LAZYTIME) | (sb->s_flags & MS_LAZYTIME);
4869         ext4_msg(sb, KERN_INFO, "re-mounted. Opts: %s", orig_data);
4870         kfree(orig_data);
4871         return 0;
4872
4873 restore_opts:
4874         sb->s_flags = old_sb_flags;
4875         sbi->s_mount_opt = old_opts.s_mount_opt;
4876         sbi->s_mount_opt2 = old_opts.s_mount_opt2;
4877         sbi->s_resuid = old_opts.s_resuid;
4878         sbi->s_resgid = old_opts.s_resgid;
4879         sbi->s_commit_interval = old_opts.s_commit_interval;
4880         sbi->s_min_batch_time = old_opts.s_min_batch_time;
4881         sbi->s_max_batch_time = old_opts.s_max_batch_time;
4882 #ifdef CONFIG_QUOTA
4883         sbi->s_jquota_fmt = old_opts.s_jquota_fmt;
4884         for (i = 0; i < EXT4_MAXQUOTAS; i++) {
4885                 kfree(sbi->s_qf_names[i]);
4886                 sbi->s_qf_names[i] = old_opts.s_qf_names[i];
4887         }
4888 #endif
4889         kfree(orig_data);
4890         return err;
4891 }
4892
4893 static int ext4_statfs(struct dentry *dentry, struct kstatfs *buf)
4894 {
4895         struct super_block *sb = dentry->d_sb;
4896         struct ext4_sb_info *sbi = EXT4_SB(sb);
4897         struct ext4_super_block *es = sbi->s_es;
4898         ext4_fsblk_t overhead = 0, resv_blocks;
4899         u64 fsid;
4900         s64 bfree;
4901         resv_blocks = EXT4_C2B(sbi, atomic64_read(&sbi->s_resv_clusters));
4902
4903         if (!test_opt(sb, MINIX_DF))
4904                 overhead = sbi->s_overhead;
4905
4906         buf->f_type = EXT4_SUPER_MAGIC;
4907         buf->f_bsize = sb->s_blocksize;
4908         buf->f_blocks = ext4_blocks_count(es) - EXT4_C2B(sbi, overhead);
4909         bfree = percpu_counter_sum_positive(&sbi->s_freeclusters_counter) -
4910                 percpu_counter_sum_positive(&sbi->s_dirtyclusters_counter);
4911         /* prevent underflow in case that few free space is available */
4912         buf->f_bfree = EXT4_C2B(sbi, max_t(s64, bfree, 0));
4913         buf->f_bavail = buf->f_bfree -
4914                         (ext4_r_blocks_count(es) + resv_blocks);
4915         if (buf->f_bfree < (ext4_r_blocks_count(es) + resv_blocks))
4916                 buf->f_bavail = 0;
4917         buf->f_files = le32_to_cpu(es->s_inodes_count);
4918         buf->f_ffree = percpu_counter_sum_positive(&sbi->s_freeinodes_counter);
4919         buf->f_namelen = EXT4_NAME_LEN;
4920         fsid = le64_to_cpup((void *)es->s_uuid) ^
4921                le64_to_cpup((void *)es->s_uuid + sizeof(u64));
4922         buf->f_fsid.val[0] = fsid & 0xFFFFFFFFUL;
4923         buf->f_fsid.val[1] = (fsid >> 32) & 0xFFFFFFFFUL;
4924
4925         return 0;
4926 }
4927
4928 /* Helper function for writing quotas on sync - we need to start transaction
4929  * before quota file is locked for write. Otherwise the are possible deadlocks:
4930  * Process 1                         Process 2
4931  * ext4_create()                     quota_sync()
4932  *   jbd2_journal_start()                  write_dquot()
4933  *   dquot_initialize()                         down(dqio_mutex)
4934  *     down(dqio_mutex)                    jbd2_journal_start()
4935  *
4936  */
4937
4938 #ifdef CONFIG_QUOTA
4939
4940 static inline struct inode *dquot_to_inode(struct dquot *dquot)
4941 {
4942         return sb_dqopt(dquot->dq_sb)->files[dquot->dq_id.type];
4943 }
4944
4945 static int ext4_write_dquot(struct dquot *dquot)
4946 {
4947         int ret, err;
4948         handle_t *handle;
4949         struct inode *inode;
4950
4951         inode = dquot_to_inode(dquot);
4952         handle = ext4_journal_start(inode, EXT4_HT_QUOTA,
4953                                     EXT4_QUOTA_TRANS_BLOCKS(dquot->dq_sb));
4954         if (IS_ERR(handle))
4955                 return PTR_ERR(handle);
4956         ret = dquot_commit(dquot);
4957         err = ext4_journal_stop(handle);
4958         if (!ret)
4959                 ret = err;
4960         return ret;
4961 }
4962
4963 static int ext4_acquire_dquot(struct dquot *dquot)
4964 {
4965         int ret, err;
4966         handle_t *handle;
4967
4968         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4969                                     EXT4_QUOTA_INIT_BLOCKS(dquot->dq_sb));
4970         if (IS_ERR(handle))
4971                 return PTR_ERR(handle);
4972         ret = dquot_acquire(dquot);
4973         err = ext4_journal_stop(handle);
4974         if (!ret)
4975                 ret = err;
4976         return ret;
4977 }
4978
4979 static int ext4_release_dquot(struct dquot *dquot)
4980 {
4981         int ret, err;
4982         handle_t *handle;
4983
4984         handle = ext4_journal_start(dquot_to_inode(dquot), EXT4_HT_QUOTA,
4985                                     EXT4_QUOTA_DEL_BLOCKS(dquot->dq_sb));
4986         if (IS_ERR(handle)) {
4987                 /* Release dquot anyway to avoid endless cycle in dqput() */
4988                 dquot_release(dquot);
4989                 return PTR_ERR(handle);
4990         }
4991         ret = dquot_release(dquot);
4992         err = ext4_journal_stop(handle);
4993         if (!ret)
4994                 ret = err;
4995         return ret;
4996 }
4997
4998 static int ext4_mark_dquot_dirty(struct dquot *dquot)
4999 {
5000         struct super_block *sb = dquot->dq_sb;
5001         struct ext4_sb_info *sbi = EXT4_SB(sb);
5002
5003         /* Are we journaling quotas? */
5004         if (ext4_has_feature_quota(sb) ||
5005             sbi->s_qf_names[USRQUOTA] || sbi->s_qf_names[GRPQUOTA]) {
5006                 dquot_mark_dquot_dirty(dquot);
5007                 return ext4_write_dquot(dquot);
5008         } else {
5009                 return dquot_mark_dquot_dirty(dquot);
5010         }
5011 }
5012
5013 static int ext4_write_info(struct super_block *sb, int type)
5014 {
5015         int ret, err;
5016         handle_t *handle;
5017
5018         /* Data block + inode block */
5019         handle = ext4_journal_start(d_inode(sb->s_root), EXT4_HT_QUOTA, 2);
5020         if (IS_ERR(handle))
5021                 return PTR_ERR(handle);
5022         ret = dquot_commit_info(sb, type);
5023         err = ext4_journal_stop(handle);
5024         if (!ret)
5025                 ret = err;
5026         return ret;
5027 }
5028
5029 /*
5030  * Turn on quotas during mount time - we need to find
5031  * the quota file and such...
5032  */
5033 static int ext4_quota_on_mount(struct super_block *sb, int type)
5034 {
5035         return dquot_quota_on_mount(sb, EXT4_SB(sb)->s_qf_names[type],
5036                                         EXT4_SB(sb)->s_jquota_fmt, type);
5037 }
5038
5039 static void lockdep_set_quota_inode(struct inode *inode, int subclass)
5040 {
5041         struct ext4_inode_info *ei = EXT4_I(inode);
5042
5043         /* The first argument of lockdep_set_subclass has to be
5044          * *exactly* the same as the argument to init_rwsem() --- in
5045          * this case, in init_once() --- or lockdep gets unhappy
5046          * because the name of the lock is set using the
5047          * stringification of the argument to init_rwsem().
5048          */
5049         (void) ei;      /* shut up clang warning if !CONFIG_LOCKDEP */
5050         lockdep_set_subclass(&ei->i_data_sem, subclass);
5051 }
5052
5053 /*
5054  * Standard function to be called on quota_on
5055  */
5056 static int ext4_quota_on(struct super_block *sb, int type, int format_id,
5057                          struct path *path)
5058 {
5059         int err;
5060
5061         if (!test_opt(sb, QUOTA))
5062                 return -EINVAL;
5063
5064         /* Quotafile not on the same filesystem? */
5065         if (path->dentry->d_sb != sb)
5066                 return -EXDEV;
5067         /* Journaling quota? */
5068         if (EXT4_SB(sb)->s_qf_names[type]) {
5069                 /* Quotafile not in fs root? */
5070                 if (path->dentry->d_parent != sb->s_root)
5071                         ext4_msg(sb, KERN_WARNING,
5072                                 "Quota file not on filesystem root. "
5073                                 "Journaled quota will not work");
5074         }
5075
5076         /*
5077          * When we journal data on quota file, we have to flush journal to see
5078          * all updates to the file when we bypass pagecache...
5079          */
5080         if (EXT4_SB(sb)->s_journal &&
5081             ext4_should_journal_data(d_inode(path->dentry))) {
5082                 /*
5083                  * We don't need to lock updates but journal_flush() could
5084                  * otherwise be livelocked...
5085                  */
5086                 jbd2_journal_lock_updates(EXT4_SB(sb)->s_journal);
5087                 err = jbd2_journal_flush(EXT4_SB(sb)->s_journal);
5088                 jbd2_journal_unlock_updates(EXT4_SB(sb)->s_journal);
5089                 if (err)
5090                         return err;
5091         }
5092         lockdep_set_quota_inode(path->dentry->d_inode, I_DATA_SEM_QUOTA);
5093         err = dquot_quota_on(sb, type, format_id, path);
5094         if (err)
5095                 lockdep_set_quota_inode(path->dentry->d_inode,
5096                                              I_DATA_SEM_NORMAL);
5097         return err;
5098 }
5099
5100 static int ext4_quota_enable(struct super_block *sb, int type, int format_id,
5101                              unsigned int flags)
5102 {
5103         int err;
5104         struct inode *qf_inode;
5105         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5106                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5107                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5108         };
5109
5110         BUG_ON(!ext4_has_feature_quota(sb));
5111
5112         if (!qf_inums[type])
5113                 return -EPERM;
5114
5115         qf_inode = ext4_iget(sb, qf_inums[type]);
5116         if (IS_ERR(qf_inode)) {
5117                 ext4_error(sb, "Bad quota inode # %lu", qf_inums[type]);
5118                 return PTR_ERR(qf_inode);
5119         }
5120
5121         /* Don't account quota for quota files to avoid recursion */
5122         qf_inode->i_flags |= S_NOQUOTA;
5123         lockdep_set_quota_inode(qf_inode, I_DATA_SEM_QUOTA);
5124         err = dquot_enable(qf_inode, type, format_id, flags);
5125         iput(qf_inode);
5126         if (err)
5127                 lockdep_set_quota_inode(qf_inode, I_DATA_SEM_NORMAL);
5128
5129         return err;
5130 }
5131
5132 /* Enable usage tracking for all quota types. */
5133 static int ext4_enable_quotas(struct super_block *sb)
5134 {
5135         int type, err = 0;
5136         unsigned long qf_inums[EXT4_MAXQUOTAS] = {
5137                 le32_to_cpu(EXT4_SB(sb)->s_es->s_usr_quota_inum),
5138                 le32_to_cpu(EXT4_SB(sb)->s_es->s_grp_quota_inum)
5139         };
5140
5141         sb_dqopt(sb)->flags |= DQUOT_QUOTA_SYS_FILE;
5142         for (type = 0; type < EXT4_MAXQUOTAS; type++) {
5143                 if (qf_inums[type]) {
5144                         err = ext4_quota_enable(sb, type, QFMT_VFS_V1,
5145                                                 DQUOT_USAGE_ENABLED);
5146                         if (err) {
5147                                 for (type--; type >= 0; type--)
5148                                         dquot_quota_off(sb, type);
5149
5150                                 ext4_warning(sb,
5151                                         "Failed to enable quota tracking "
5152                                         "(type=%d, err=%d). Please run "
5153                                         "e2fsck to fix.", type, err);
5154                                 return err;
5155                         }
5156                 }
5157         }
5158         return 0;
5159 }
5160
5161 static int ext4_quota_off(struct super_block *sb, int type)
5162 {
5163         struct inode *inode = sb_dqopt(sb)->files[type];
5164         handle_t *handle;
5165
5166         /* Force all delayed allocation blocks to be allocated.
5167          * Caller already holds s_umount sem */
5168         if (test_opt(sb, DELALLOC))
5169                 sync_filesystem(sb);
5170
5171         if (!inode)
5172                 goto out;
5173
5174         /* Update modification times of quota files when userspace can
5175          * start looking at them */
5176         handle = ext4_journal_start(inode, EXT4_HT_QUOTA, 1);
5177         if (IS_ERR(handle))
5178                 goto out;
5179         inode->i_mtime = inode->i_ctime = CURRENT_TIME;
5180         ext4_mark_inode_dirty(handle, inode);
5181         ext4_journal_stop(handle);
5182
5183 out:
5184         return dquot_quota_off(sb, type);
5185 }
5186
5187 /* Read data from quotafile - avoid pagecache and such because we cannot afford
5188  * acquiring the locks... As quota files are never truncated and quota code
5189  * itself serializes the operations (and no one else should touch the files)
5190  * we don't have to be afraid of races */
5191 static ssize_t ext4_quota_read(struct super_block *sb, int type, char *data,
5192                                size_t len, loff_t off)
5193 {
5194         struct inode *inode = sb_dqopt(sb)->files[type];
5195         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5196         int offset = off & (sb->s_blocksize - 1);
5197         int tocopy;
5198         size_t toread;
5199         struct buffer_head *bh;
5200         loff_t i_size = i_size_read(inode);
5201
5202         if (off > i_size)
5203                 return 0;
5204         if (off+len > i_size)
5205                 len = i_size-off;
5206         toread = len;
5207         while (toread > 0) {
5208                 tocopy = sb->s_blocksize - offset < toread ?
5209                                 sb->s_blocksize - offset : toread;
5210                 bh = ext4_bread(NULL, inode, blk, 0);
5211                 if (IS_ERR(bh))
5212                         return PTR_ERR(bh);
5213                 if (!bh)        /* A hole? */
5214                         memset(data, 0, tocopy);
5215                 else
5216                         memcpy(data, bh->b_data+offset, tocopy);
5217                 brelse(bh);
5218                 offset = 0;
5219                 toread -= tocopy;
5220                 data += tocopy;
5221                 blk++;
5222         }
5223         return len;
5224 }
5225
5226 /* Write to quotafile (we know the transaction is already started and has
5227  * enough credits) */
5228 static ssize_t ext4_quota_write(struct super_block *sb, int type,
5229                                 const char *data, size_t len, loff_t off)
5230 {
5231         struct inode *inode = sb_dqopt(sb)->files[type];
5232         ext4_lblk_t blk = off >> EXT4_BLOCK_SIZE_BITS(sb);
5233         int err, offset = off & (sb->s_blocksize - 1);
5234         int retries = 0;
5235         struct buffer_head *bh;
5236         handle_t *handle = journal_current_handle();
5237
5238         if (EXT4_SB(sb)->s_journal && !handle) {
5239                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5240                         " cancelled because transaction is not started",
5241                         (unsigned long long)off, (unsigned long long)len);
5242                 return -EIO;
5243         }
5244         /*
5245          * Since we account only one data block in transaction credits,
5246          * then it is impossible to cross a block boundary.
5247          */
5248         if (sb->s_blocksize - offset < len) {
5249                 ext4_msg(sb, KERN_WARNING, "Quota write (off=%llu, len=%llu)"
5250                         " cancelled because not block aligned",
5251                         (unsigned long long)off, (unsigned long long)len);
5252                 return -EIO;
5253         }
5254
5255         do {
5256                 bh = ext4_bread(handle, inode, blk,
5257                                 EXT4_GET_BLOCKS_CREATE |
5258                                 EXT4_GET_BLOCKS_METADATA_NOFAIL);
5259         } while (IS_ERR(bh) && (PTR_ERR(bh) == -ENOSPC) &&
5260                  ext4_should_retry_alloc(inode->i_sb, &retries));
5261         if (IS_ERR(bh))
5262                 return PTR_ERR(bh);
5263         if (!bh)
5264                 goto out;
5265         BUFFER_TRACE(bh, "get write access");
5266         err = ext4_journal_get_write_access(handle, bh);
5267         if (err) {
5268                 brelse(bh);
5269                 return err;
5270         }
5271         lock_buffer(bh);
5272         memcpy(bh->b_data+offset, data, len);
5273         flush_dcache_page(bh->b_page);
5274         unlock_buffer(bh);
5275         err = ext4_handle_dirty_metadata(handle, NULL, bh);
5276         brelse(bh);
5277 out:
5278         if (inode->i_size < off + len) {
5279                 i_size_write(inode, off + len);
5280                 EXT4_I(inode)->i_disksize = inode->i_size;
5281                 ext4_mark_inode_dirty(handle, inode);
5282         }
5283         return len;
5284 }
5285
5286 #endif
5287
5288 static struct dentry *ext4_mount(struct file_system_type *fs_type, int flags,
5289                        const char *dev_name, void *data)
5290 {
5291         return mount_bdev(fs_type, flags, dev_name, data, ext4_fill_super);
5292 }
5293
5294 #if !defined(CONFIG_EXT2_FS) && !defined(CONFIG_EXT2_FS_MODULE) && defined(CONFIG_EXT4_USE_FOR_EXT2)
5295 static inline void register_as_ext2(void)
5296 {
5297         int err = register_filesystem(&ext2_fs_type);
5298         if (err)
5299                 printk(KERN_WARNING
5300                        "EXT4-fs: Unable to register as ext2 (%d)\n", err);
5301 }
5302
5303 static inline void unregister_as_ext2(void)
5304 {
5305         unregister_filesystem(&ext2_fs_type);
5306 }
5307
5308 static inline int ext2_feature_set_ok(struct super_block *sb)
5309 {
5310         if (ext4_has_unknown_ext2_incompat_features(sb))
5311                 return 0;
5312         if (sb->s_flags & MS_RDONLY)
5313                 return 1;
5314         if (ext4_has_unknown_ext2_ro_compat_features(sb))
5315                 return 0;
5316         return 1;
5317 }
5318 #else
5319 static inline void register_as_ext2(void) { }
5320 static inline void unregister_as_ext2(void) { }
5321 static inline int ext2_feature_set_ok(struct super_block *sb) { return 0; }
5322 #endif
5323
5324 static inline void register_as_ext3(void)
5325 {
5326         int err = register_filesystem(&ext3_fs_type);
5327         if (err)
5328                 printk(KERN_WARNING
5329                        "EXT4-fs: Unable to register as ext3 (%d)\n", err);
5330 }
5331
5332 static inline void unregister_as_ext3(void)
5333 {
5334         unregister_filesystem(&ext3_fs_type);
5335 }
5336
5337 static inline int ext3_feature_set_ok(struct super_block *sb)
5338 {
5339         if (ext4_has_unknown_ext3_incompat_features(sb))
5340                 return 0;
5341         if (!ext4_has_feature_journal(sb))
5342                 return 0;
5343         if (sb->s_flags & MS_RDONLY)
5344                 return 1;
5345         if (ext4_has_unknown_ext3_ro_compat_features(sb))
5346                 return 0;
5347         return 1;
5348 }
5349
5350 static struct file_system_type ext4_fs_type = {
5351         .owner          = THIS_MODULE,
5352         .name           = "ext4",
5353         .mount          = ext4_mount,
5354         .kill_sb        = kill_block_super,
5355         .fs_flags       = FS_REQUIRES_DEV,
5356 };
5357 MODULE_ALIAS_FS("ext4");
5358
5359 /* Shared across all ext4 file systems */
5360 wait_queue_head_t ext4__ioend_wq[EXT4_WQ_HASH_SZ];
5361 struct mutex ext4__aio_mutex[EXT4_WQ_HASH_SZ];
5362
5363 static int __init ext4_init_fs(void)
5364 {
5365         int i, err;
5366
5367         ratelimit_state_init(&ext4_mount_msg_ratelimit, 30 * HZ, 64);
5368         ext4_li_info = NULL;
5369         mutex_init(&ext4_li_mtx);
5370
5371         /* Build-time check for flags consistency */
5372         ext4_check_flag_values();
5373
5374         for (i = 0; i < EXT4_WQ_HASH_SZ; i++) {
5375                 mutex_init(&ext4__aio_mutex[i]);
5376                 init_waitqueue_head(&ext4__ioend_wq[i]);
5377         }
5378
5379         err = ext4_init_es();
5380         if (err)
5381                 return err;
5382
5383         err = ext4_init_pageio();
5384         if (err)
5385                 goto out5;
5386
5387         err = ext4_init_system_zone();
5388         if (err)
5389                 goto out4;
5390
5391         err = ext4_init_sysfs();
5392         if (err)
5393                 goto out3;
5394
5395         err = ext4_init_mballoc();
5396         if (err)
5397                 goto out2;
5398         else
5399                 ext4_mballoc_ready = 1;
5400         err = init_inodecache();
5401         if (err)
5402                 goto out1;
5403         register_as_ext3();
5404         register_as_ext2();
5405         err = register_filesystem(&ext4_fs_type);
5406         if (err)
5407                 goto out;
5408
5409         return 0;
5410 out:
5411         unregister_as_ext2();
5412         unregister_as_ext3();
5413         destroy_inodecache();
5414 out1:
5415         ext4_mballoc_ready = 0;
5416         ext4_exit_mballoc();
5417 out2:
5418         ext4_exit_sysfs();
5419 out3:
5420         ext4_exit_system_zone();
5421 out4:
5422         ext4_exit_pageio();
5423 out5:
5424         ext4_exit_es();
5425
5426         return err;
5427 }
5428
5429 static void __exit ext4_exit_fs(void)
5430 {
5431         ext4_exit_crypto();
5432         ext4_destroy_lazyinit_thread();
5433         unregister_as_ext2();
5434         unregister_as_ext3();
5435         unregister_filesystem(&ext4_fs_type);
5436         destroy_inodecache();
5437         ext4_exit_mballoc();
5438         ext4_exit_sysfs();
5439         ext4_exit_system_zone();
5440         ext4_exit_pageio();
5441         ext4_exit_es();
5442 }
5443
5444 MODULE_AUTHOR("Remy Card, Stephen Tweedie, Andrew Morton, Andreas Dilger, Theodore Ts'o and others");
5445 MODULE_DESCRIPTION("Fourth Extended Filesystem");
5446 MODULE_LICENSE("GPL");
5447 module_init(ext4_init_fs)
5448 module_exit(ext4_exit_fs)